]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * QEMU System Emulator block driver | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * | |
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 | */ | |
24 | #include "config-host.h" | |
25 | #include "qemu-common.h" | |
26 | #include "trace.h" | |
27 | #include "monitor.h" | |
28 | #include "block_int.h" | |
29 | #include "blockjob.h" | |
30 | #include "module.h" | |
31 | #include "qjson.h" | |
32 | #include "sysemu.h" | |
33 | #include "notify.h" | |
34 | #include "qemu-coroutine.h" | |
35 | #include "qmp-commands.h" | |
36 | #include "qemu-timer.h" | |
37 | ||
38 | #ifdef CONFIG_BSD | |
39 | #include <sys/types.h> | |
40 | #include <sys/stat.h> | |
41 | #include <sys/ioctl.h> | |
42 | #include <sys/queue.h> | |
43 | #ifndef __DragonFly__ | |
44 | #include <sys/disk.h> | |
45 | #endif | |
46 | #endif | |
47 | ||
48 | #ifdef _WIN32 | |
49 | #include <windows.h> | |
50 | #endif | |
51 | ||
52 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
53 | ||
54 | typedef enum { | |
55 | BDRV_REQ_COPY_ON_READ = 0x1, | |
56 | BDRV_REQ_ZERO_WRITE = 0x2, | |
57 | } BdrvRequestFlags; | |
58 | ||
59 | static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load); | |
60 | static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs, | |
61 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
62 | BlockDriverCompletionFunc *cb, void *opaque); | |
63 | static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs, | |
64 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
65 | BlockDriverCompletionFunc *cb, void *opaque); | |
66 | static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, | |
67 | int64_t sector_num, int nb_sectors, | |
68 | QEMUIOVector *iov); | |
69 | static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, | |
70 | int64_t sector_num, int nb_sectors, | |
71 | QEMUIOVector *iov); | |
72 | static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, | |
73 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
74 | BdrvRequestFlags flags); | |
75 | static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, | |
76 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
77 | BdrvRequestFlags flags); | |
78 | static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, | |
79 | int64_t sector_num, | |
80 | QEMUIOVector *qiov, | |
81 | int nb_sectors, | |
82 | BlockDriverCompletionFunc *cb, | |
83 | void *opaque, | |
84 | bool is_write); | |
85 | static void coroutine_fn bdrv_co_do_rw(void *opaque); | |
86 | static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, | |
87 | int64_t sector_num, int nb_sectors); | |
88 | ||
89 | static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors, | |
90 | bool is_write, double elapsed_time, uint64_t *wait); | |
91 | static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write, | |
92 | double elapsed_time, uint64_t *wait); | |
93 | static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors, | |
94 | bool is_write, int64_t *wait); | |
95 | ||
96 | static QTAILQ_HEAD(, BlockDriverState) bdrv_states = | |
97 | QTAILQ_HEAD_INITIALIZER(bdrv_states); | |
98 | ||
99 | static QLIST_HEAD(, BlockDriver) bdrv_drivers = | |
100 | QLIST_HEAD_INITIALIZER(bdrv_drivers); | |
101 | ||
102 | /* The device to use for VM snapshots */ | |
103 | static BlockDriverState *bs_snapshots; | |
104 | ||
105 | /* If non-zero, use only whitelisted block drivers */ | |
106 | static int use_bdrv_whitelist; | |
107 | ||
108 | #ifdef _WIN32 | |
109 | static int is_windows_drive_prefix(const char *filename) | |
110 | { | |
111 | return (((filename[0] >= 'a' && filename[0] <= 'z') || | |
112 | (filename[0] >= 'A' && filename[0] <= 'Z')) && | |
113 | filename[1] == ':'); | |
114 | } | |
115 | ||
116 | int is_windows_drive(const char *filename) | |
117 | { | |
118 | if (is_windows_drive_prefix(filename) && | |
119 | filename[2] == '\0') | |
120 | return 1; | |
121 | if (strstart(filename, "\\\\.\\", NULL) || | |
122 | strstart(filename, "//./", NULL)) | |
123 | return 1; | |
124 | return 0; | |
125 | } | |
126 | #endif | |
127 | ||
128 | /* throttling disk I/O limits */ | |
129 | void bdrv_io_limits_disable(BlockDriverState *bs) | |
130 | { | |
131 | bs->io_limits_enabled = false; | |
132 | ||
133 | while (qemu_co_queue_next(&bs->throttled_reqs)); | |
134 | ||
135 | if (bs->block_timer) { | |
136 | qemu_del_timer(bs->block_timer); | |
137 | qemu_free_timer(bs->block_timer); | |
138 | bs->block_timer = NULL; | |
139 | } | |
140 | ||
141 | bs->slice_start = 0; | |
142 | bs->slice_end = 0; | |
143 | bs->slice_time = 0; | |
144 | memset(&bs->io_base, 0, sizeof(bs->io_base)); | |
145 | } | |
146 | ||
147 | static void bdrv_block_timer(void *opaque) | |
148 | { | |
149 | BlockDriverState *bs = opaque; | |
150 | ||
151 | qemu_co_queue_next(&bs->throttled_reqs); | |
152 | } | |
153 | ||
154 | void bdrv_io_limits_enable(BlockDriverState *bs) | |
155 | { | |
156 | qemu_co_queue_init(&bs->throttled_reqs); | |
157 | bs->block_timer = qemu_new_timer_ns(vm_clock, bdrv_block_timer, bs); | |
158 | bs->slice_time = 5 * BLOCK_IO_SLICE_TIME; | |
159 | bs->slice_start = qemu_get_clock_ns(vm_clock); | |
160 | bs->slice_end = bs->slice_start + bs->slice_time; | |
161 | memset(&bs->io_base, 0, sizeof(bs->io_base)); | |
162 | bs->io_limits_enabled = true; | |
163 | } | |
164 | ||
165 | bool bdrv_io_limits_enabled(BlockDriverState *bs) | |
166 | { | |
167 | BlockIOLimit *io_limits = &bs->io_limits; | |
168 | return io_limits->bps[BLOCK_IO_LIMIT_READ] | |
169 | || io_limits->bps[BLOCK_IO_LIMIT_WRITE] | |
170 | || io_limits->bps[BLOCK_IO_LIMIT_TOTAL] | |
171 | || io_limits->iops[BLOCK_IO_LIMIT_READ] | |
172 | || io_limits->iops[BLOCK_IO_LIMIT_WRITE] | |
173 | || io_limits->iops[BLOCK_IO_LIMIT_TOTAL]; | |
174 | } | |
175 | ||
176 | static void bdrv_io_limits_intercept(BlockDriverState *bs, | |
177 | bool is_write, int nb_sectors) | |
178 | { | |
179 | int64_t wait_time = -1; | |
180 | ||
181 | if (!qemu_co_queue_empty(&bs->throttled_reqs)) { | |
182 | qemu_co_queue_wait(&bs->throttled_reqs); | |
183 | } | |
184 | ||
185 | /* In fact, we hope to keep each request's timing, in FIFO mode. The next | |
186 | * throttled requests will not be dequeued until the current request is | |
187 | * allowed to be serviced. So if the current request still exceeds the | |
188 | * limits, it will be inserted to the head. All requests followed it will | |
189 | * be still in throttled_reqs queue. | |
190 | */ | |
191 | ||
192 | while (bdrv_exceed_io_limits(bs, nb_sectors, is_write, &wait_time)) { | |
193 | qemu_mod_timer(bs->block_timer, | |
194 | wait_time + qemu_get_clock_ns(vm_clock)); | |
195 | qemu_co_queue_wait_insert_head(&bs->throttled_reqs); | |
196 | } | |
197 | ||
198 | qemu_co_queue_next(&bs->throttled_reqs); | |
199 | } | |
200 | ||
201 | /* check if the path starts with "<protocol>:" */ | |
202 | static int path_has_protocol(const char *path) | |
203 | { | |
204 | const char *p; | |
205 | ||
206 | #ifdef _WIN32 | |
207 | if (is_windows_drive(path) || | |
208 | is_windows_drive_prefix(path)) { | |
209 | return 0; | |
210 | } | |
211 | p = path + strcspn(path, ":/\\"); | |
212 | #else | |
213 | p = path + strcspn(path, ":/"); | |
214 | #endif | |
215 | ||
216 | return *p == ':'; | |
217 | } | |
218 | ||
219 | int path_is_absolute(const char *path) | |
220 | { | |
221 | #ifdef _WIN32 | |
222 | /* specific case for names like: "\\.\d:" */ | |
223 | if (is_windows_drive(path) || is_windows_drive_prefix(path)) { | |
224 | return 1; | |
225 | } | |
226 | return (*path == '/' || *path == '\\'); | |
227 | #else | |
228 | return (*path == '/'); | |
229 | #endif | |
230 | } | |
231 | ||
232 | /* if filename is absolute, just copy it to dest. Otherwise, build a | |
233 | path to it by considering it is relative to base_path. URL are | |
234 | supported. */ | |
235 | void path_combine(char *dest, int dest_size, | |
236 | const char *base_path, | |
237 | const char *filename) | |
238 | { | |
239 | const char *p, *p1; | |
240 | int len; | |
241 | ||
242 | if (dest_size <= 0) | |
243 | return; | |
244 | if (path_is_absolute(filename)) { | |
245 | pstrcpy(dest, dest_size, filename); | |
246 | } else { | |
247 | p = strchr(base_path, ':'); | |
248 | if (p) | |
249 | p++; | |
250 | else | |
251 | p = base_path; | |
252 | p1 = strrchr(base_path, '/'); | |
253 | #ifdef _WIN32 | |
254 | { | |
255 | const char *p2; | |
256 | p2 = strrchr(base_path, '\\'); | |
257 | if (!p1 || p2 > p1) | |
258 | p1 = p2; | |
259 | } | |
260 | #endif | |
261 | if (p1) | |
262 | p1++; | |
263 | else | |
264 | p1 = base_path; | |
265 | if (p1 > p) | |
266 | p = p1; | |
267 | len = p - base_path; | |
268 | if (len > dest_size - 1) | |
269 | len = dest_size - 1; | |
270 | memcpy(dest, base_path, len); | |
271 | dest[len] = '\0'; | |
272 | pstrcat(dest, dest_size, filename); | |
273 | } | |
274 | } | |
275 | ||
276 | void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz) | |
277 | { | |
278 | if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) { | |
279 | pstrcpy(dest, sz, bs->backing_file); | |
280 | } else { | |
281 | path_combine(dest, sz, bs->filename, bs->backing_file); | |
282 | } | |
283 | } | |
284 | ||
285 | void bdrv_register(BlockDriver *bdrv) | |
286 | { | |
287 | /* Block drivers without coroutine functions need emulation */ | |
288 | if (!bdrv->bdrv_co_readv) { | |
289 | bdrv->bdrv_co_readv = bdrv_co_readv_em; | |
290 | bdrv->bdrv_co_writev = bdrv_co_writev_em; | |
291 | ||
292 | /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if | |
293 | * the block driver lacks aio we need to emulate that too. | |
294 | */ | |
295 | if (!bdrv->bdrv_aio_readv) { | |
296 | /* add AIO emulation layer */ | |
297 | bdrv->bdrv_aio_readv = bdrv_aio_readv_em; | |
298 | bdrv->bdrv_aio_writev = bdrv_aio_writev_em; | |
299 | } | |
300 | } | |
301 | ||
302 | QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); | |
303 | } | |
304 | ||
305 | /* create a new block device (by default it is empty) */ | |
306 | BlockDriverState *bdrv_new(const char *device_name) | |
307 | { | |
308 | BlockDriverState *bs; | |
309 | ||
310 | bs = g_malloc0(sizeof(BlockDriverState)); | |
311 | pstrcpy(bs->device_name, sizeof(bs->device_name), device_name); | |
312 | if (device_name[0] != '\0') { | |
313 | QTAILQ_INSERT_TAIL(&bdrv_states, bs, list); | |
314 | } | |
315 | bdrv_iostatus_disable(bs); | |
316 | notifier_list_init(&bs->close_notifiers); | |
317 | ||
318 | return bs; | |
319 | } | |
320 | ||
321 | void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify) | |
322 | { | |
323 | notifier_list_add(&bs->close_notifiers, notify); | |
324 | } | |
325 | ||
326 | BlockDriver *bdrv_find_format(const char *format_name) | |
327 | { | |
328 | BlockDriver *drv1; | |
329 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
330 | if (!strcmp(drv1->format_name, format_name)) { | |
331 | return drv1; | |
332 | } | |
333 | } | |
334 | return NULL; | |
335 | } | |
336 | ||
337 | static int bdrv_is_whitelisted(BlockDriver *drv) | |
338 | { | |
339 | static const char *whitelist[] = { | |
340 | CONFIG_BDRV_WHITELIST | |
341 | }; | |
342 | const char **p; | |
343 | ||
344 | if (!whitelist[0]) | |
345 | return 1; /* no whitelist, anything goes */ | |
346 | ||
347 | for (p = whitelist; *p; p++) { | |
348 | if (!strcmp(drv->format_name, *p)) { | |
349 | return 1; | |
350 | } | |
351 | } | |
352 | return 0; | |
353 | } | |
354 | ||
355 | BlockDriver *bdrv_find_whitelisted_format(const char *format_name) | |
356 | { | |
357 | BlockDriver *drv = bdrv_find_format(format_name); | |
358 | return drv && bdrv_is_whitelisted(drv) ? drv : NULL; | |
359 | } | |
360 | ||
361 | typedef struct CreateCo { | |
362 | BlockDriver *drv; | |
363 | char *filename; | |
364 | QEMUOptionParameter *options; | |
365 | int ret; | |
366 | } CreateCo; | |
367 | ||
368 | static void coroutine_fn bdrv_create_co_entry(void *opaque) | |
369 | { | |
370 | CreateCo *cco = opaque; | |
371 | assert(cco->drv); | |
372 | ||
373 | cco->ret = cco->drv->bdrv_create(cco->filename, cco->options); | |
374 | } | |
375 | ||
376 | int bdrv_create(BlockDriver *drv, const char* filename, | |
377 | QEMUOptionParameter *options) | |
378 | { | |
379 | int ret; | |
380 | ||
381 | Coroutine *co; | |
382 | CreateCo cco = { | |
383 | .drv = drv, | |
384 | .filename = g_strdup(filename), | |
385 | .options = options, | |
386 | .ret = NOT_DONE, | |
387 | }; | |
388 | ||
389 | if (!drv->bdrv_create) { | |
390 | ret = -ENOTSUP; | |
391 | goto out; | |
392 | } | |
393 | ||
394 | if (qemu_in_coroutine()) { | |
395 | /* Fast-path if already in coroutine context */ | |
396 | bdrv_create_co_entry(&cco); | |
397 | } else { | |
398 | co = qemu_coroutine_create(bdrv_create_co_entry); | |
399 | qemu_coroutine_enter(co, &cco); | |
400 | while (cco.ret == NOT_DONE) { | |
401 | qemu_aio_wait(); | |
402 | } | |
403 | } | |
404 | ||
405 | ret = cco.ret; | |
406 | ||
407 | out: | |
408 | g_free(cco.filename); | |
409 | return ret; | |
410 | } | |
411 | ||
412 | int bdrv_create_file(const char* filename, QEMUOptionParameter *options) | |
413 | { | |
414 | BlockDriver *drv; | |
415 | ||
416 | drv = bdrv_find_protocol(filename); | |
417 | if (drv == NULL) { | |
418 | return -ENOENT; | |
419 | } | |
420 | ||
421 | return bdrv_create(drv, filename, options); | |
422 | } | |
423 | ||
424 | /* | |
425 | * Create a uniquely-named empty temporary file. | |
426 | * Return 0 upon success, otherwise a negative errno value. | |
427 | */ | |
428 | int get_tmp_filename(char *filename, int size) | |
429 | { | |
430 | #ifdef _WIN32 | |
431 | char temp_dir[MAX_PATH]; | |
432 | /* GetTempFileName requires that its output buffer (4th param) | |
433 | have length MAX_PATH or greater. */ | |
434 | assert(size >= MAX_PATH); | |
435 | return (GetTempPath(MAX_PATH, temp_dir) | |
436 | && GetTempFileName(temp_dir, "qem", 0, filename) | |
437 | ? 0 : -GetLastError()); | |
438 | #else | |
439 | int fd; | |
440 | const char *tmpdir; | |
441 | tmpdir = getenv("TMPDIR"); | |
442 | if (!tmpdir) | |
443 | tmpdir = "/tmp"; | |
444 | if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) { | |
445 | return -EOVERFLOW; | |
446 | } | |
447 | fd = mkstemp(filename); | |
448 | if (fd < 0) { | |
449 | return -errno; | |
450 | } | |
451 | if (close(fd) != 0) { | |
452 | unlink(filename); | |
453 | return -errno; | |
454 | } | |
455 | return 0; | |
456 | #endif | |
457 | } | |
458 | ||
459 | /* | |
460 | * Detect host devices. By convention, /dev/cdrom[N] is always | |
461 | * recognized as a host CDROM. | |
462 | */ | |
463 | static BlockDriver *find_hdev_driver(const char *filename) | |
464 | { | |
465 | int score_max = 0, score; | |
466 | BlockDriver *drv = NULL, *d; | |
467 | ||
468 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
469 | if (d->bdrv_probe_device) { | |
470 | score = d->bdrv_probe_device(filename); | |
471 | if (score > score_max) { | |
472 | score_max = score; | |
473 | drv = d; | |
474 | } | |
475 | } | |
476 | } | |
477 | ||
478 | return drv; | |
479 | } | |
480 | ||
481 | BlockDriver *bdrv_find_protocol(const char *filename) | |
482 | { | |
483 | BlockDriver *drv1; | |
484 | char protocol[128]; | |
485 | int len; | |
486 | const char *p; | |
487 | ||
488 | /* TODO Drivers without bdrv_file_open must be specified explicitly */ | |
489 | ||
490 | /* | |
491 | * XXX(hch): we really should not let host device detection | |
492 | * override an explicit protocol specification, but moving this | |
493 | * later breaks access to device names with colons in them. | |
494 | * Thanks to the brain-dead persistent naming schemes on udev- | |
495 | * based Linux systems those actually are quite common. | |
496 | */ | |
497 | drv1 = find_hdev_driver(filename); | |
498 | if (drv1) { | |
499 | return drv1; | |
500 | } | |
501 | ||
502 | if (!path_has_protocol(filename)) { | |
503 | return bdrv_find_format("file"); | |
504 | } | |
505 | p = strchr(filename, ':'); | |
506 | assert(p != NULL); | |
507 | len = p - filename; | |
508 | if (len > sizeof(protocol) - 1) | |
509 | len = sizeof(protocol) - 1; | |
510 | memcpy(protocol, filename, len); | |
511 | protocol[len] = '\0'; | |
512 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
513 | if (drv1->protocol_name && | |
514 | !strcmp(drv1->protocol_name, protocol)) { | |
515 | return drv1; | |
516 | } | |
517 | } | |
518 | return NULL; | |
519 | } | |
520 | ||
521 | static int find_image_format(BlockDriverState *bs, const char *filename, | |
522 | BlockDriver **pdrv) | |
523 | { | |
524 | int score, score_max; | |
525 | BlockDriver *drv1, *drv; | |
526 | uint8_t buf[2048]; | |
527 | int ret = 0; | |
528 | ||
529 | /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ | |
530 | if (bs->sg || !bdrv_is_inserted(bs)) { | |
531 | drv = bdrv_find_format("raw"); | |
532 | if (!drv) { | |
533 | ret = -ENOENT; | |
534 | } | |
535 | *pdrv = drv; | |
536 | return ret; | |
537 | } | |
538 | ||
539 | ret = bdrv_pread(bs, 0, buf, sizeof(buf)); | |
540 | if (ret < 0) { | |
541 | *pdrv = NULL; | |
542 | return ret; | |
543 | } | |
544 | ||
545 | score_max = 0; | |
546 | drv = NULL; | |
547 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
548 | if (drv1->bdrv_probe) { | |
549 | score = drv1->bdrv_probe(buf, ret, filename); | |
550 | if (score > score_max) { | |
551 | score_max = score; | |
552 | drv = drv1; | |
553 | } | |
554 | } | |
555 | } | |
556 | if (!drv) { | |
557 | ret = -ENOENT; | |
558 | } | |
559 | *pdrv = drv; | |
560 | return ret; | |
561 | } | |
562 | ||
563 | /** | |
564 | * Set the current 'total_sectors' value | |
565 | */ | |
566 | static int refresh_total_sectors(BlockDriverState *bs, int64_t hint) | |
567 | { | |
568 | BlockDriver *drv = bs->drv; | |
569 | ||
570 | /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */ | |
571 | if (bs->sg) | |
572 | return 0; | |
573 | ||
574 | /* query actual device if possible, otherwise just trust the hint */ | |
575 | if (drv->bdrv_getlength) { | |
576 | int64_t length = drv->bdrv_getlength(bs); | |
577 | if (length < 0) { | |
578 | return length; | |
579 | } | |
580 | hint = length >> BDRV_SECTOR_BITS; | |
581 | } | |
582 | ||
583 | bs->total_sectors = hint; | |
584 | return 0; | |
585 | } | |
586 | ||
587 | /** | |
588 | * Set open flags for a given cache mode | |
589 | * | |
590 | * Return 0 on success, -1 if the cache mode was invalid. | |
591 | */ | |
592 | int bdrv_parse_cache_flags(const char *mode, int *flags) | |
593 | { | |
594 | *flags &= ~BDRV_O_CACHE_MASK; | |
595 | ||
596 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
597 | *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB; | |
598 | } else if (!strcmp(mode, "directsync")) { | |
599 | *flags |= BDRV_O_NOCACHE; | |
600 | } else if (!strcmp(mode, "writeback")) { | |
601 | *flags |= BDRV_O_CACHE_WB; | |
602 | } else if (!strcmp(mode, "unsafe")) { | |
603 | *flags |= BDRV_O_CACHE_WB; | |
604 | *flags |= BDRV_O_NO_FLUSH; | |
605 | } else if (!strcmp(mode, "writethrough")) { | |
606 | /* this is the default */ | |
607 | } else { | |
608 | return -1; | |
609 | } | |
610 | ||
611 | return 0; | |
612 | } | |
613 | ||
614 | /** | |
615 | * The copy-on-read flag is actually a reference count so multiple users may | |
616 | * use the feature without worrying about clobbering its previous state. | |
617 | * Copy-on-read stays enabled until all users have called to disable it. | |
618 | */ | |
619 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
620 | { | |
621 | bs->copy_on_read++; | |
622 | } | |
623 | ||
624 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
625 | { | |
626 | assert(bs->copy_on_read > 0); | |
627 | bs->copy_on_read--; | |
628 | } | |
629 | ||
630 | static int bdrv_open_flags(BlockDriverState *bs, int flags) | |
631 | { | |
632 | int open_flags = flags | BDRV_O_CACHE_WB; | |
633 | ||
634 | /* | |
635 | * Clear flags that are internal to the block layer before opening the | |
636 | * image. | |
637 | */ | |
638 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
639 | ||
640 | /* | |
641 | * Snapshots should be writable. | |
642 | */ | |
643 | if (bs->is_temporary) { | |
644 | open_flags |= BDRV_O_RDWR; | |
645 | } | |
646 | ||
647 | return open_flags; | |
648 | } | |
649 | ||
650 | /* | |
651 | * Common part for opening disk images and files | |
652 | */ | |
653 | static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file, | |
654 | const char *filename, | |
655 | int flags, BlockDriver *drv) | |
656 | { | |
657 | int ret, open_flags; | |
658 | ||
659 | assert(drv != NULL); | |
660 | assert(bs->file == NULL); | |
661 | ||
662 | trace_bdrv_open_common(bs, filename, flags, drv->format_name); | |
663 | ||
664 | bs->open_flags = flags; | |
665 | bs->buffer_alignment = 512; | |
666 | ||
667 | assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */ | |
668 | if ((flags & BDRV_O_RDWR) && (flags & BDRV_O_COPY_ON_READ)) { | |
669 | bdrv_enable_copy_on_read(bs); | |
670 | } | |
671 | ||
672 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
673 | ||
674 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv)) { | |
675 | return -ENOTSUP; | |
676 | } | |
677 | ||
678 | bs->drv = drv; | |
679 | bs->opaque = g_malloc0(drv->instance_size); | |
680 | ||
681 | bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB); | |
682 | open_flags = bdrv_open_flags(bs, flags); | |
683 | ||
684 | bs->read_only = !(open_flags & BDRV_O_RDWR); | |
685 | ||
686 | /* Open the image, either directly or using a protocol */ | |
687 | if (drv->bdrv_file_open) { | |
688 | if (file != NULL) { | |
689 | bdrv_swap(file, bs); | |
690 | ret = 0; | |
691 | } else { | |
692 | ret = drv->bdrv_file_open(bs, filename, open_flags); | |
693 | } | |
694 | } else { | |
695 | assert(file != NULL); | |
696 | bs->file = file; | |
697 | ret = drv->bdrv_open(bs, open_flags); | |
698 | } | |
699 | ||
700 | if (ret < 0) { | |
701 | goto free_and_fail; | |
702 | } | |
703 | ||
704 | ret = refresh_total_sectors(bs, bs->total_sectors); | |
705 | if (ret < 0) { | |
706 | goto free_and_fail; | |
707 | } | |
708 | ||
709 | #ifndef _WIN32 | |
710 | if (bs->is_temporary) { | |
711 | unlink(filename); | |
712 | } | |
713 | #endif | |
714 | return 0; | |
715 | ||
716 | free_and_fail: | |
717 | bs->file = NULL; | |
718 | g_free(bs->opaque); | |
719 | bs->opaque = NULL; | |
720 | bs->drv = NULL; | |
721 | return ret; | |
722 | } | |
723 | ||
724 | /* | |
725 | * Opens a file using a protocol (file, host_device, nbd, ...) | |
726 | */ | |
727 | int bdrv_file_open(BlockDriverState **pbs, const char *filename, int flags) | |
728 | { | |
729 | BlockDriverState *bs; | |
730 | BlockDriver *drv; | |
731 | int ret; | |
732 | ||
733 | drv = bdrv_find_protocol(filename); | |
734 | if (!drv) { | |
735 | return -ENOENT; | |
736 | } | |
737 | ||
738 | bs = bdrv_new(""); | |
739 | ret = bdrv_open_common(bs, NULL, filename, flags, drv); | |
740 | if (ret < 0) { | |
741 | bdrv_delete(bs); | |
742 | return ret; | |
743 | } | |
744 | bs->growable = 1; | |
745 | *pbs = bs; | |
746 | return 0; | |
747 | } | |
748 | ||
749 | int bdrv_open_backing_file(BlockDriverState *bs) | |
750 | { | |
751 | char backing_filename[PATH_MAX]; | |
752 | int back_flags, ret; | |
753 | BlockDriver *back_drv = NULL; | |
754 | ||
755 | if (bs->backing_hd != NULL) { | |
756 | return 0; | |
757 | } | |
758 | ||
759 | bs->open_flags &= ~BDRV_O_NO_BACKING; | |
760 | if (bs->backing_file[0] == '\0') { | |
761 | return 0; | |
762 | } | |
763 | ||
764 | bs->backing_hd = bdrv_new(""); | |
765 | bdrv_get_full_backing_filename(bs, backing_filename, | |
766 | sizeof(backing_filename)); | |
767 | ||
768 | if (bs->backing_format[0] != '\0') { | |
769 | back_drv = bdrv_find_format(bs->backing_format); | |
770 | } | |
771 | ||
772 | /* backing files always opened read-only */ | |
773 | back_flags = bs->open_flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT); | |
774 | ||
775 | ret = bdrv_open(bs->backing_hd, backing_filename, back_flags, back_drv); | |
776 | if (ret < 0) { | |
777 | bdrv_delete(bs->backing_hd); | |
778 | bs->backing_hd = NULL; | |
779 | bs->open_flags |= BDRV_O_NO_BACKING; | |
780 | return ret; | |
781 | } | |
782 | return 0; | |
783 | } | |
784 | ||
785 | /* | |
786 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
787 | */ | |
788 | int bdrv_open(BlockDriverState *bs, const char *filename, int flags, | |
789 | BlockDriver *drv) | |
790 | { | |
791 | int ret; | |
792 | /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */ | |
793 | char tmp_filename[PATH_MAX + 1]; | |
794 | BlockDriverState *file = NULL; | |
795 | ||
796 | if (flags & BDRV_O_SNAPSHOT) { | |
797 | BlockDriverState *bs1; | |
798 | int64_t total_size; | |
799 | int is_protocol = 0; | |
800 | BlockDriver *bdrv_qcow2; | |
801 | QEMUOptionParameter *options; | |
802 | char backing_filename[PATH_MAX]; | |
803 | ||
804 | /* if snapshot, we create a temporary backing file and open it | |
805 | instead of opening 'filename' directly */ | |
806 | ||
807 | /* if there is a backing file, use it */ | |
808 | bs1 = bdrv_new(""); | |
809 | ret = bdrv_open(bs1, filename, 0, drv); | |
810 | if (ret < 0) { | |
811 | bdrv_delete(bs1); | |
812 | return ret; | |
813 | } | |
814 | total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK; | |
815 | ||
816 | if (bs1->drv && bs1->drv->protocol_name) | |
817 | is_protocol = 1; | |
818 | ||
819 | bdrv_delete(bs1); | |
820 | ||
821 | ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename)); | |
822 | if (ret < 0) { | |
823 | return ret; | |
824 | } | |
825 | ||
826 | /* Real path is meaningless for protocols */ | |
827 | if (is_protocol) | |
828 | snprintf(backing_filename, sizeof(backing_filename), | |
829 | "%s", filename); | |
830 | else if (!realpath(filename, backing_filename)) | |
831 | return -errno; | |
832 | ||
833 | bdrv_qcow2 = bdrv_find_format("qcow2"); | |
834 | options = parse_option_parameters("", bdrv_qcow2->create_options, NULL); | |
835 | ||
836 | set_option_parameter_int(options, BLOCK_OPT_SIZE, total_size); | |
837 | set_option_parameter(options, BLOCK_OPT_BACKING_FILE, backing_filename); | |
838 | if (drv) { | |
839 | set_option_parameter(options, BLOCK_OPT_BACKING_FMT, | |
840 | drv->format_name); | |
841 | } | |
842 | ||
843 | ret = bdrv_create(bdrv_qcow2, tmp_filename, options); | |
844 | free_option_parameters(options); | |
845 | if (ret < 0) { | |
846 | return ret; | |
847 | } | |
848 | ||
849 | filename = tmp_filename; | |
850 | drv = bdrv_qcow2; | |
851 | bs->is_temporary = 1; | |
852 | } | |
853 | ||
854 | /* Open image file without format layer */ | |
855 | if (flags & BDRV_O_RDWR) { | |
856 | flags |= BDRV_O_ALLOW_RDWR; | |
857 | } | |
858 | ||
859 | ret = bdrv_file_open(&file, filename, bdrv_open_flags(bs, flags)); | |
860 | if (ret < 0) { | |
861 | return ret; | |
862 | } | |
863 | ||
864 | /* Find the right image format driver */ | |
865 | if (!drv) { | |
866 | ret = find_image_format(file, filename, &drv); | |
867 | } | |
868 | ||
869 | if (!drv) { | |
870 | goto unlink_and_fail; | |
871 | } | |
872 | ||
873 | /* Open the image */ | |
874 | ret = bdrv_open_common(bs, file, filename, flags, drv); | |
875 | if (ret < 0) { | |
876 | goto unlink_and_fail; | |
877 | } | |
878 | ||
879 | if (bs->file != file) { | |
880 | bdrv_delete(file); | |
881 | file = NULL; | |
882 | } | |
883 | ||
884 | /* If there is a backing file, use it */ | |
885 | if ((flags & BDRV_O_NO_BACKING) == 0) { | |
886 | ret = bdrv_open_backing_file(bs); | |
887 | if (ret < 0) { | |
888 | bdrv_close(bs); | |
889 | return ret; | |
890 | } | |
891 | } | |
892 | ||
893 | if (!bdrv_key_required(bs)) { | |
894 | bdrv_dev_change_media_cb(bs, true); | |
895 | } | |
896 | ||
897 | /* throttling disk I/O limits */ | |
898 | if (bs->io_limits_enabled) { | |
899 | bdrv_io_limits_enable(bs); | |
900 | } | |
901 | ||
902 | return 0; | |
903 | ||
904 | unlink_and_fail: | |
905 | if (file != NULL) { | |
906 | bdrv_delete(file); | |
907 | } | |
908 | if (bs->is_temporary) { | |
909 | unlink(filename); | |
910 | } | |
911 | return ret; | |
912 | } | |
913 | ||
914 | typedef struct BlockReopenQueueEntry { | |
915 | bool prepared; | |
916 | BDRVReopenState state; | |
917 | QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry; | |
918 | } BlockReopenQueueEntry; | |
919 | ||
920 | /* | |
921 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
922 | * reopen of multiple devices. | |
923 | * | |
924 | * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT | |
925 | * already performed, or alternatively may be NULL a new BlockReopenQueue will | |
926 | * be created and initialized. This newly created BlockReopenQueue should be | |
927 | * passed back in for subsequent calls that are intended to be of the same | |
928 | * atomic 'set'. | |
929 | * | |
930 | * bs is the BlockDriverState to add to the reopen queue. | |
931 | * | |
932 | * flags contains the open flags for the associated bs | |
933 | * | |
934 | * returns a pointer to bs_queue, which is either the newly allocated | |
935 | * bs_queue, or the existing bs_queue being used. | |
936 | * | |
937 | */ | |
938 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, | |
939 | BlockDriverState *bs, int flags) | |
940 | { | |
941 | assert(bs != NULL); | |
942 | ||
943 | BlockReopenQueueEntry *bs_entry; | |
944 | if (bs_queue == NULL) { | |
945 | bs_queue = g_new0(BlockReopenQueue, 1); | |
946 | QSIMPLEQ_INIT(bs_queue); | |
947 | } | |
948 | ||
949 | if (bs->file) { | |
950 | bdrv_reopen_queue(bs_queue, bs->file, flags); | |
951 | } | |
952 | ||
953 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
954 | QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry); | |
955 | ||
956 | bs_entry->state.bs = bs; | |
957 | bs_entry->state.flags = flags; | |
958 | ||
959 | return bs_queue; | |
960 | } | |
961 | ||
962 | /* | |
963 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
964 | * | |
965 | * The queue passed in (bs_queue) must have been built up previous | |
966 | * via bdrv_reopen_queue(). | |
967 | * | |
968 | * Reopens all BDS specified in the queue, with the appropriate | |
969 | * flags. All devices are prepared for reopen, and failure of any | |
970 | * device will cause all device changes to be abandonded, and intermediate | |
971 | * data cleaned up. | |
972 | * | |
973 | * If all devices prepare successfully, then the changes are committed | |
974 | * to all devices. | |
975 | * | |
976 | */ | |
977 | int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) | |
978 | { | |
979 | int ret = -1; | |
980 | BlockReopenQueueEntry *bs_entry, *next; | |
981 | Error *local_err = NULL; | |
982 | ||
983 | assert(bs_queue != NULL); | |
984 | ||
985 | bdrv_drain_all(); | |
986 | ||
987 | QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { | |
988 | if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) { | |
989 | error_propagate(errp, local_err); | |
990 | goto cleanup; | |
991 | } | |
992 | bs_entry->prepared = true; | |
993 | } | |
994 | ||
995 | /* If we reach this point, we have success and just need to apply the | |
996 | * changes | |
997 | */ | |
998 | QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { | |
999 | bdrv_reopen_commit(&bs_entry->state); | |
1000 | } | |
1001 | ||
1002 | ret = 0; | |
1003 | ||
1004 | cleanup: | |
1005 | QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
1006 | if (ret && bs_entry->prepared) { | |
1007 | bdrv_reopen_abort(&bs_entry->state); | |
1008 | } | |
1009 | g_free(bs_entry); | |
1010 | } | |
1011 | g_free(bs_queue); | |
1012 | return ret; | |
1013 | } | |
1014 | ||
1015 | ||
1016 | /* Reopen a single BlockDriverState with the specified flags. */ | |
1017 | int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp) | |
1018 | { | |
1019 | int ret = -1; | |
1020 | Error *local_err = NULL; | |
1021 | BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags); | |
1022 | ||
1023 | ret = bdrv_reopen_multiple(queue, &local_err); | |
1024 | if (local_err != NULL) { | |
1025 | error_propagate(errp, local_err); | |
1026 | } | |
1027 | return ret; | |
1028 | } | |
1029 | ||
1030 | ||
1031 | /* | |
1032 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
1033 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
1034 | * the block driver layer .bdrv_reopen_prepare() | |
1035 | * | |
1036 | * bs is the BlockDriverState to reopen | |
1037 | * flags are the new open flags | |
1038 | * queue is the reopen queue | |
1039 | * | |
1040 | * Returns 0 on success, non-zero on error. On error errp will be set | |
1041 | * as well. | |
1042 | * | |
1043 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
1044 | * It is the responsibility of the caller to then call the abort() or | |
1045 | * commit() for any other BDS that have been left in a prepare() state | |
1046 | * | |
1047 | */ | |
1048 | int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue, | |
1049 | Error **errp) | |
1050 | { | |
1051 | int ret = -1; | |
1052 | Error *local_err = NULL; | |
1053 | BlockDriver *drv; | |
1054 | ||
1055 | assert(reopen_state != NULL); | |
1056 | assert(reopen_state->bs->drv != NULL); | |
1057 | drv = reopen_state->bs->drv; | |
1058 | ||
1059 | /* if we are to stay read-only, do not allow permission change | |
1060 | * to r/w */ | |
1061 | if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) && | |
1062 | reopen_state->flags & BDRV_O_RDWR) { | |
1063 | error_set(errp, QERR_DEVICE_IS_READ_ONLY, | |
1064 | reopen_state->bs->device_name); | |
1065 | goto error; | |
1066 | } | |
1067 | ||
1068 | ||
1069 | ret = bdrv_flush(reopen_state->bs); | |
1070 | if (ret) { | |
1071 | error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive", | |
1072 | strerror(-ret)); | |
1073 | goto error; | |
1074 | } | |
1075 | ||
1076 | if (drv->bdrv_reopen_prepare) { | |
1077 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); | |
1078 | if (ret) { | |
1079 | if (local_err != NULL) { | |
1080 | error_propagate(errp, local_err); | |
1081 | } else { | |
1082 | error_set(errp, QERR_OPEN_FILE_FAILED, | |
1083 | reopen_state->bs->filename); | |
1084 | } | |
1085 | goto error; | |
1086 | } | |
1087 | } else { | |
1088 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
1089 | * handler for each supported drv. */ | |
1090 | error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, | |
1091 | drv->format_name, reopen_state->bs->device_name, | |
1092 | "reopening of file"); | |
1093 | ret = -1; | |
1094 | goto error; | |
1095 | } | |
1096 | ||
1097 | ret = 0; | |
1098 | ||
1099 | error: | |
1100 | return ret; | |
1101 | } | |
1102 | ||
1103 | /* | |
1104 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
1105 | * makes them final by swapping the staging BlockDriverState contents into | |
1106 | * the active BlockDriverState contents. | |
1107 | */ | |
1108 | void bdrv_reopen_commit(BDRVReopenState *reopen_state) | |
1109 | { | |
1110 | BlockDriver *drv; | |
1111 | ||
1112 | assert(reopen_state != NULL); | |
1113 | drv = reopen_state->bs->drv; | |
1114 | assert(drv != NULL); | |
1115 | ||
1116 | /* If there are any driver level actions to take */ | |
1117 | if (drv->bdrv_reopen_commit) { | |
1118 | drv->bdrv_reopen_commit(reopen_state); | |
1119 | } | |
1120 | ||
1121 | /* set BDS specific flags now */ | |
1122 | reopen_state->bs->open_flags = reopen_state->flags; | |
1123 | reopen_state->bs->enable_write_cache = !!(reopen_state->flags & | |
1124 | BDRV_O_CACHE_WB); | |
1125 | reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
1126 | } | |
1127 | ||
1128 | /* | |
1129 | * Abort the reopen, and delete and free the staged changes in | |
1130 | * reopen_state | |
1131 | */ | |
1132 | void bdrv_reopen_abort(BDRVReopenState *reopen_state) | |
1133 | { | |
1134 | BlockDriver *drv; | |
1135 | ||
1136 | assert(reopen_state != NULL); | |
1137 | drv = reopen_state->bs->drv; | |
1138 | assert(drv != NULL); | |
1139 | ||
1140 | if (drv->bdrv_reopen_abort) { | |
1141 | drv->bdrv_reopen_abort(reopen_state); | |
1142 | } | |
1143 | } | |
1144 | ||
1145 | ||
1146 | void bdrv_close(BlockDriverState *bs) | |
1147 | { | |
1148 | bdrv_flush(bs); | |
1149 | if (bs->job) { | |
1150 | block_job_cancel_sync(bs->job); | |
1151 | } | |
1152 | bdrv_drain_all(); | |
1153 | notifier_list_notify(&bs->close_notifiers, bs); | |
1154 | ||
1155 | if (bs->drv) { | |
1156 | if (bs == bs_snapshots) { | |
1157 | bs_snapshots = NULL; | |
1158 | } | |
1159 | if (bs->backing_hd) { | |
1160 | bdrv_delete(bs->backing_hd); | |
1161 | bs->backing_hd = NULL; | |
1162 | } | |
1163 | bs->drv->bdrv_close(bs); | |
1164 | g_free(bs->opaque); | |
1165 | #ifdef _WIN32 | |
1166 | if (bs->is_temporary) { | |
1167 | unlink(bs->filename); | |
1168 | } | |
1169 | #endif | |
1170 | bs->opaque = NULL; | |
1171 | bs->drv = NULL; | |
1172 | bs->copy_on_read = 0; | |
1173 | bs->backing_file[0] = '\0'; | |
1174 | bs->backing_format[0] = '\0'; | |
1175 | bs->total_sectors = 0; | |
1176 | bs->encrypted = 0; | |
1177 | bs->valid_key = 0; | |
1178 | bs->sg = 0; | |
1179 | bs->growable = 0; | |
1180 | ||
1181 | if (bs->file != NULL) { | |
1182 | bdrv_delete(bs->file); | |
1183 | bs->file = NULL; | |
1184 | } | |
1185 | } | |
1186 | ||
1187 | bdrv_dev_change_media_cb(bs, false); | |
1188 | ||
1189 | /*throttling disk I/O limits*/ | |
1190 | if (bs->io_limits_enabled) { | |
1191 | bdrv_io_limits_disable(bs); | |
1192 | } | |
1193 | } | |
1194 | ||
1195 | void bdrv_close_all(void) | |
1196 | { | |
1197 | BlockDriverState *bs; | |
1198 | ||
1199 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
1200 | bdrv_close(bs); | |
1201 | } | |
1202 | } | |
1203 | ||
1204 | /* | |
1205 | * Wait for pending requests to complete across all BlockDriverStates | |
1206 | * | |
1207 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
1208 | * after calling this function. | |
1209 | * | |
1210 | * Note that completion of an asynchronous I/O operation can trigger any | |
1211 | * number of other I/O operations on other devices---for example a coroutine | |
1212 | * can be arbitrarily complex and a constant flow of I/O can come until the | |
1213 | * coroutine is complete. Because of this, it is not possible to have a | |
1214 | * function to drain a single device's I/O queue. | |
1215 | */ | |
1216 | void bdrv_drain_all(void) | |
1217 | { | |
1218 | BlockDriverState *bs; | |
1219 | bool busy; | |
1220 | ||
1221 | do { | |
1222 | busy = qemu_aio_wait(); | |
1223 | ||
1224 | /* FIXME: We do not have timer support here, so this is effectively | |
1225 | * a busy wait. | |
1226 | */ | |
1227 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
1228 | if (!qemu_co_queue_empty(&bs->throttled_reqs)) { | |
1229 | qemu_co_queue_restart_all(&bs->throttled_reqs); | |
1230 | busy = true; | |
1231 | } | |
1232 | } | |
1233 | } while (busy); | |
1234 | ||
1235 | /* If requests are still pending there is a bug somewhere */ | |
1236 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
1237 | assert(QLIST_EMPTY(&bs->tracked_requests)); | |
1238 | assert(qemu_co_queue_empty(&bs->throttled_reqs)); | |
1239 | } | |
1240 | } | |
1241 | ||
1242 | /* make a BlockDriverState anonymous by removing from bdrv_state list. | |
1243 | Also, NULL terminate the device_name to prevent double remove */ | |
1244 | void bdrv_make_anon(BlockDriverState *bs) | |
1245 | { | |
1246 | if (bs->device_name[0] != '\0') { | |
1247 | QTAILQ_REMOVE(&bdrv_states, bs, list); | |
1248 | } | |
1249 | bs->device_name[0] = '\0'; | |
1250 | } | |
1251 | ||
1252 | static void bdrv_rebind(BlockDriverState *bs) | |
1253 | { | |
1254 | if (bs->drv && bs->drv->bdrv_rebind) { | |
1255 | bs->drv->bdrv_rebind(bs); | |
1256 | } | |
1257 | } | |
1258 | ||
1259 | static void bdrv_move_feature_fields(BlockDriverState *bs_dest, | |
1260 | BlockDriverState *bs_src) | |
1261 | { | |
1262 | /* move some fields that need to stay attached to the device */ | |
1263 | bs_dest->open_flags = bs_src->open_flags; | |
1264 | ||
1265 | /* dev info */ | |
1266 | bs_dest->dev_ops = bs_src->dev_ops; | |
1267 | bs_dest->dev_opaque = bs_src->dev_opaque; | |
1268 | bs_dest->dev = bs_src->dev; | |
1269 | bs_dest->buffer_alignment = bs_src->buffer_alignment; | |
1270 | bs_dest->copy_on_read = bs_src->copy_on_read; | |
1271 | ||
1272 | bs_dest->enable_write_cache = bs_src->enable_write_cache; | |
1273 | ||
1274 | /* i/o timing parameters */ | |
1275 | bs_dest->slice_time = bs_src->slice_time; | |
1276 | bs_dest->slice_start = bs_src->slice_start; | |
1277 | bs_dest->slice_end = bs_src->slice_end; | |
1278 | bs_dest->io_limits = bs_src->io_limits; | |
1279 | bs_dest->io_base = bs_src->io_base; | |
1280 | bs_dest->throttled_reqs = bs_src->throttled_reqs; | |
1281 | bs_dest->block_timer = bs_src->block_timer; | |
1282 | bs_dest->io_limits_enabled = bs_src->io_limits_enabled; | |
1283 | ||
1284 | /* r/w error */ | |
1285 | bs_dest->on_read_error = bs_src->on_read_error; | |
1286 | bs_dest->on_write_error = bs_src->on_write_error; | |
1287 | ||
1288 | /* i/o status */ | |
1289 | bs_dest->iostatus_enabled = bs_src->iostatus_enabled; | |
1290 | bs_dest->iostatus = bs_src->iostatus; | |
1291 | ||
1292 | /* dirty bitmap */ | |
1293 | bs_dest->dirty_count = bs_src->dirty_count; | |
1294 | bs_dest->dirty_bitmap = bs_src->dirty_bitmap; | |
1295 | ||
1296 | /* job */ | |
1297 | bs_dest->in_use = bs_src->in_use; | |
1298 | bs_dest->job = bs_src->job; | |
1299 | ||
1300 | /* keep the same entry in bdrv_states */ | |
1301 | pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name), | |
1302 | bs_src->device_name); | |
1303 | bs_dest->list = bs_src->list; | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * Swap bs contents for two image chains while they are live, | |
1308 | * while keeping required fields on the BlockDriverState that is | |
1309 | * actually attached to a device. | |
1310 | * | |
1311 | * This will modify the BlockDriverState fields, and swap contents | |
1312 | * between bs_new and bs_old. Both bs_new and bs_old are modified. | |
1313 | * | |
1314 | * bs_new is required to be anonymous. | |
1315 | * | |
1316 | * This function does not create any image files. | |
1317 | */ | |
1318 | void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old) | |
1319 | { | |
1320 | BlockDriverState tmp; | |
1321 | ||
1322 | /* bs_new must be anonymous and shouldn't have anything fancy enabled */ | |
1323 | assert(bs_new->device_name[0] == '\0'); | |
1324 | assert(bs_new->dirty_bitmap == NULL); | |
1325 | assert(bs_new->job == NULL); | |
1326 | assert(bs_new->dev == NULL); | |
1327 | assert(bs_new->in_use == 0); | |
1328 | assert(bs_new->io_limits_enabled == false); | |
1329 | assert(bs_new->block_timer == NULL); | |
1330 | ||
1331 | tmp = *bs_new; | |
1332 | *bs_new = *bs_old; | |
1333 | *bs_old = tmp; | |
1334 | ||
1335 | /* there are some fields that should not be swapped, move them back */ | |
1336 | bdrv_move_feature_fields(&tmp, bs_old); | |
1337 | bdrv_move_feature_fields(bs_old, bs_new); | |
1338 | bdrv_move_feature_fields(bs_new, &tmp); | |
1339 | ||
1340 | /* bs_new shouldn't be in bdrv_states even after the swap! */ | |
1341 | assert(bs_new->device_name[0] == '\0'); | |
1342 | ||
1343 | /* Check a few fields that should remain attached to the device */ | |
1344 | assert(bs_new->dev == NULL); | |
1345 | assert(bs_new->job == NULL); | |
1346 | assert(bs_new->in_use == 0); | |
1347 | assert(bs_new->io_limits_enabled == false); | |
1348 | assert(bs_new->block_timer == NULL); | |
1349 | ||
1350 | bdrv_rebind(bs_new); | |
1351 | bdrv_rebind(bs_old); | |
1352 | } | |
1353 | ||
1354 | /* | |
1355 | * Add new bs contents at the top of an image chain while the chain is | |
1356 | * live, while keeping required fields on the top layer. | |
1357 | * | |
1358 | * This will modify the BlockDriverState fields, and swap contents | |
1359 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
1360 | * | |
1361 | * bs_new is required to be anonymous. | |
1362 | * | |
1363 | * This function does not create any image files. | |
1364 | */ | |
1365 | void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top) | |
1366 | { | |
1367 | bdrv_swap(bs_new, bs_top); | |
1368 | ||
1369 | /* The contents of 'tmp' will become bs_top, as we are | |
1370 | * swapping bs_new and bs_top contents. */ | |
1371 | bs_top->backing_hd = bs_new; | |
1372 | bs_top->open_flags &= ~BDRV_O_NO_BACKING; | |
1373 | pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file), | |
1374 | bs_new->filename); | |
1375 | pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format), | |
1376 | bs_new->drv ? bs_new->drv->format_name : ""); | |
1377 | } | |
1378 | ||
1379 | void bdrv_delete(BlockDriverState *bs) | |
1380 | { | |
1381 | assert(!bs->dev); | |
1382 | assert(!bs->job); | |
1383 | assert(!bs->in_use); | |
1384 | ||
1385 | /* remove from list, if necessary */ | |
1386 | bdrv_make_anon(bs); | |
1387 | ||
1388 | bdrv_close(bs); | |
1389 | ||
1390 | assert(bs != bs_snapshots); | |
1391 | g_free(bs); | |
1392 | } | |
1393 | ||
1394 | int bdrv_attach_dev(BlockDriverState *bs, void *dev) | |
1395 | /* TODO change to DeviceState *dev when all users are qdevified */ | |
1396 | { | |
1397 | if (bs->dev) { | |
1398 | return -EBUSY; | |
1399 | } | |
1400 | bs->dev = dev; | |
1401 | bdrv_iostatus_reset(bs); | |
1402 | return 0; | |
1403 | } | |
1404 | ||
1405 | /* TODO qdevified devices don't use this, remove when devices are qdevified */ | |
1406 | void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev) | |
1407 | { | |
1408 | if (bdrv_attach_dev(bs, dev) < 0) { | |
1409 | abort(); | |
1410 | } | |
1411 | } | |
1412 | ||
1413 | void bdrv_detach_dev(BlockDriverState *bs, void *dev) | |
1414 | /* TODO change to DeviceState *dev when all users are qdevified */ | |
1415 | { | |
1416 | assert(bs->dev == dev); | |
1417 | bs->dev = NULL; | |
1418 | bs->dev_ops = NULL; | |
1419 | bs->dev_opaque = NULL; | |
1420 | bs->buffer_alignment = 512; | |
1421 | } | |
1422 | ||
1423 | /* TODO change to return DeviceState * when all users are qdevified */ | |
1424 | void *bdrv_get_attached_dev(BlockDriverState *bs) | |
1425 | { | |
1426 | return bs->dev; | |
1427 | } | |
1428 | ||
1429 | void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops, | |
1430 | void *opaque) | |
1431 | { | |
1432 | bs->dev_ops = ops; | |
1433 | bs->dev_opaque = opaque; | |
1434 | if (bdrv_dev_has_removable_media(bs) && bs == bs_snapshots) { | |
1435 | bs_snapshots = NULL; | |
1436 | } | |
1437 | } | |
1438 | ||
1439 | void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv, | |
1440 | enum MonitorEvent ev, | |
1441 | BlockErrorAction action, bool is_read) | |
1442 | { | |
1443 | QObject *data; | |
1444 | const char *action_str; | |
1445 | ||
1446 | switch (action) { | |
1447 | case BDRV_ACTION_REPORT: | |
1448 | action_str = "report"; | |
1449 | break; | |
1450 | case BDRV_ACTION_IGNORE: | |
1451 | action_str = "ignore"; | |
1452 | break; | |
1453 | case BDRV_ACTION_STOP: | |
1454 | action_str = "stop"; | |
1455 | break; | |
1456 | default: | |
1457 | abort(); | |
1458 | } | |
1459 | ||
1460 | data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }", | |
1461 | bdrv->device_name, | |
1462 | action_str, | |
1463 | is_read ? "read" : "write"); | |
1464 | monitor_protocol_event(ev, data); | |
1465 | ||
1466 | qobject_decref(data); | |
1467 | } | |
1468 | ||
1469 | static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected) | |
1470 | { | |
1471 | QObject *data; | |
1472 | ||
1473 | data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }", | |
1474 | bdrv_get_device_name(bs), ejected); | |
1475 | monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data); | |
1476 | ||
1477 | qobject_decref(data); | |
1478 | } | |
1479 | ||
1480 | static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load) | |
1481 | { | |
1482 | if (bs->dev_ops && bs->dev_ops->change_media_cb) { | |
1483 | bool tray_was_closed = !bdrv_dev_is_tray_open(bs); | |
1484 | bs->dev_ops->change_media_cb(bs->dev_opaque, load); | |
1485 | if (tray_was_closed) { | |
1486 | /* tray open */ | |
1487 | bdrv_emit_qmp_eject_event(bs, true); | |
1488 | } | |
1489 | if (load) { | |
1490 | /* tray close */ | |
1491 | bdrv_emit_qmp_eject_event(bs, false); | |
1492 | } | |
1493 | } | |
1494 | } | |
1495 | ||
1496 | bool bdrv_dev_has_removable_media(BlockDriverState *bs) | |
1497 | { | |
1498 | return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb); | |
1499 | } | |
1500 | ||
1501 | void bdrv_dev_eject_request(BlockDriverState *bs, bool force) | |
1502 | { | |
1503 | if (bs->dev_ops && bs->dev_ops->eject_request_cb) { | |
1504 | bs->dev_ops->eject_request_cb(bs->dev_opaque, force); | |
1505 | } | |
1506 | } | |
1507 | ||
1508 | bool bdrv_dev_is_tray_open(BlockDriverState *bs) | |
1509 | { | |
1510 | if (bs->dev_ops && bs->dev_ops->is_tray_open) { | |
1511 | return bs->dev_ops->is_tray_open(bs->dev_opaque); | |
1512 | } | |
1513 | return false; | |
1514 | } | |
1515 | ||
1516 | static void bdrv_dev_resize_cb(BlockDriverState *bs) | |
1517 | { | |
1518 | if (bs->dev_ops && bs->dev_ops->resize_cb) { | |
1519 | bs->dev_ops->resize_cb(bs->dev_opaque); | |
1520 | } | |
1521 | } | |
1522 | ||
1523 | bool bdrv_dev_is_medium_locked(BlockDriverState *bs) | |
1524 | { | |
1525 | if (bs->dev_ops && bs->dev_ops->is_medium_locked) { | |
1526 | return bs->dev_ops->is_medium_locked(bs->dev_opaque); | |
1527 | } | |
1528 | return false; | |
1529 | } | |
1530 | ||
1531 | /* | |
1532 | * Run consistency checks on an image | |
1533 | * | |
1534 | * Returns 0 if the check could be completed (it doesn't mean that the image is | |
1535 | * free of errors) or -errno when an internal error occurred. The results of the | |
1536 | * check are stored in res. | |
1537 | */ | |
1538 | int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix) | |
1539 | { | |
1540 | if (bs->drv->bdrv_check == NULL) { | |
1541 | return -ENOTSUP; | |
1542 | } | |
1543 | ||
1544 | memset(res, 0, sizeof(*res)); | |
1545 | return bs->drv->bdrv_check(bs, res, fix); | |
1546 | } | |
1547 | ||
1548 | #define COMMIT_BUF_SECTORS 2048 | |
1549 | ||
1550 | /* commit COW file into the raw image */ | |
1551 | int bdrv_commit(BlockDriverState *bs) | |
1552 | { | |
1553 | BlockDriver *drv = bs->drv; | |
1554 | int64_t sector, total_sectors; | |
1555 | int n, ro, open_flags; | |
1556 | int ret = 0; | |
1557 | uint8_t *buf; | |
1558 | char filename[PATH_MAX]; | |
1559 | ||
1560 | if (!drv) | |
1561 | return -ENOMEDIUM; | |
1562 | ||
1563 | if (!bs->backing_hd) { | |
1564 | return -ENOTSUP; | |
1565 | } | |
1566 | ||
1567 | if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) { | |
1568 | return -EBUSY; | |
1569 | } | |
1570 | ||
1571 | ro = bs->backing_hd->read_only; | |
1572 | /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */ | |
1573 | pstrcpy(filename, sizeof(filename), bs->backing_hd->filename); | |
1574 | open_flags = bs->backing_hd->open_flags; | |
1575 | ||
1576 | if (ro) { | |
1577 | if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) { | |
1578 | return -EACCES; | |
1579 | } | |
1580 | } | |
1581 | ||
1582 | total_sectors = bdrv_getlength(bs) >> BDRV_SECTOR_BITS; | |
1583 | buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE); | |
1584 | ||
1585 | for (sector = 0; sector < total_sectors; sector += n) { | |
1586 | if (bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n)) { | |
1587 | ||
1588 | if (bdrv_read(bs, sector, buf, n) != 0) { | |
1589 | ret = -EIO; | |
1590 | goto ro_cleanup; | |
1591 | } | |
1592 | ||
1593 | if (bdrv_write(bs->backing_hd, sector, buf, n) != 0) { | |
1594 | ret = -EIO; | |
1595 | goto ro_cleanup; | |
1596 | } | |
1597 | } | |
1598 | } | |
1599 | ||
1600 | if (drv->bdrv_make_empty) { | |
1601 | ret = drv->bdrv_make_empty(bs); | |
1602 | bdrv_flush(bs); | |
1603 | } | |
1604 | ||
1605 | /* | |
1606 | * Make sure all data we wrote to the backing device is actually | |
1607 | * stable on disk. | |
1608 | */ | |
1609 | if (bs->backing_hd) | |
1610 | bdrv_flush(bs->backing_hd); | |
1611 | ||
1612 | ro_cleanup: | |
1613 | g_free(buf); | |
1614 | ||
1615 | if (ro) { | |
1616 | /* ignoring error return here */ | |
1617 | bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL); | |
1618 | } | |
1619 | ||
1620 | return ret; | |
1621 | } | |
1622 | ||
1623 | int bdrv_commit_all(void) | |
1624 | { | |
1625 | BlockDriverState *bs; | |
1626 | ||
1627 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
1628 | int ret = bdrv_commit(bs); | |
1629 | if (ret < 0) { | |
1630 | return ret; | |
1631 | } | |
1632 | } | |
1633 | return 0; | |
1634 | } | |
1635 | ||
1636 | struct BdrvTrackedRequest { | |
1637 | BlockDriverState *bs; | |
1638 | int64_t sector_num; | |
1639 | int nb_sectors; | |
1640 | bool is_write; | |
1641 | QLIST_ENTRY(BdrvTrackedRequest) list; | |
1642 | Coroutine *co; /* owner, used for deadlock detection */ | |
1643 | CoQueue wait_queue; /* coroutines blocked on this request */ | |
1644 | }; | |
1645 | ||
1646 | /** | |
1647 | * Remove an active request from the tracked requests list | |
1648 | * | |
1649 | * This function should be called when a tracked request is completing. | |
1650 | */ | |
1651 | static void tracked_request_end(BdrvTrackedRequest *req) | |
1652 | { | |
1653 | QLIST_REMOVE(req, list); | |
1654 | qemu_co_queue_restart_all(&req->wait_queue); | |
1655 | } | |
1656 | ||
1657 | /** | |
1658 | * Add an active request to the tracked requests list | |
1659 | */ | |
1660 | static void tracked_request_begin(BdrvTrackedRequest *req, | |
1661 | BlockDriverState *bs, | |
1662 | int64_t sector_num, | |
1663 | int nb_sectors, bool is_write) | |
1664 | { | |
1665 | *req = (BdrvTrackedRequest){ | |
1666 | .bs = bs, | |
1667 | .sector_num = sector_num, | |
1668 | .nb_sectors = nb_sectors, | |
1669 | .is_write = is_write, | |
1670 | .co = qemu_coroutine_self(), | |
1671 | }; | |
1672 | ||
1673 | qemu_co_queue_init(&req->wait_queue); | |
1674 | ||
1675 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); | |
1676 | } | |
1677 | ||
1678 | /** | |
1679 | * Round a region to cluster boundaries | |
1680 | */ | |
1681 | static void round_to_clusters(BlockDriverState *bs, | |
1682 | int64_t sector_num, int nb_sectors, | |
1683 | int64_t *cluster_sector_num, | |
1684 | int *cluster_nb_sectors) | |
1685 | { | |
1686 | BlockDriverInfo bdi; | |
1687 | ||
1688 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
1689 | *cluster_sector_num = sector_num; | |
1690 | *cluster_nb_sectors = nb_sectors; | |
1691 | } else { | |
1692 | int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE; | |
1693 | *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c); | |
1694 | *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num + | |
1695 | nb_sectors, c); | |
1696 | } | |
1697 | } | |
1698 | ||
1699 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, | |
1700 | int64_t sector_num, int nb_sectors) { | |
1701 | /* aaaa bbbb */ | |
1702 | if (sector_num >= req->sector_num + req->nb_sectors) { | |
1703 | return false; | |
1704 | } | |
1705 | /* bbbb aaaa */ | |
1706 | if (req->sector_num >= sector_num + nb_sectors) { | |
1707 | return false; | |
1708 | } | |
1709 | return true; | |
1710 | } | |
1711 | ||
1712 | static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs, | |
1713 | int64_t sector_num, int nb_sectors) | |
1714 | { | |
1715 | BdrvTrackedRequest *req; | |
1716 | int64_t cluster_sector_num; | |
1717 | int cluster_nb_sectors; | |
1718 | bool retry; | |
1719 | ||
1720 | /* If we touch the same cluster it counts as an overlap. This guarantees | |
1721 | * that allocating writes will be serialized and not race with each other | |
1722 | * for the same cluster. For example, in copy-on-read it ensures that the | |
1723 | * CoR read and write operations are atomic and guest writes cannot | |
1724 | * interleave between them. | |
1725 | */ | |
1726 | round_to_clusters(bs, sector_num, nb_sectors, | |
1727 | &cluster_sector_num, &cluster_nb_sectors); | |
1728 | ||
1729 | do { | |
1730 | retry = false; | |
1731 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
1732 | if (tracked_request_overlaps(req, cluster_sector_num, | |
1733 | cluster_nb_sectors)) { | |
1734 | /* Hitting this means there was a reentrant request, for | |
1735 | * example, a block driver issuing nested requests. This must | |
1736 | * never happen since it means deadlock. | |
1737 | */ | |
1738 | assert(qemu_coroutine_self() != req->co); | |
1739 | ||
1740 | qemu_co_queue_wait(&req->wait_queue); | |
1741 | retry = true; | |
1742 | break; | |
1743 | } | |
1744 | } | |
1745 | } while (retry); | |
1746 | } | |
1747 | ||
1748 | /* | |
1749 | * Return values: | |
1750 | * 0 - success | |
1751 | * -EINVAL - backing format specified, but no file | |
1752 | * -ENOSPC - can't update the backing file because no space is left in the | |
1753 | * image file header | |
1754 | * -ENOTSUP - format driver doesn't support changing the backing file | |
1755 | */ | |
1756 | int bdrv_change_backing_file(BlockDriverState *bs, | |
1757 | const char *backing_file, const char *backing_fmt) | |
1758 | { | |
1759 | BlockDriver *drv = bs->drv; | |
1760 | int ret; | |
1761 | ||
1762 | /* Backing file format doesn't make sense without a backing file */ | |
1763 | if (backing_fmt && !backing_file) { | |
1764 | return -EINVAL; | |
1765 | } | |
1766 | ||
1767 | if (drv->bdrv_change_backing_file != NULL) { | |
1768 | ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); | |
1769 | } else { | |
1770 | ret = -ENOTSUP; | |
1771 | } | |
1772 | ||
1773 | if (ret == 0) { | |
1774 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
1775 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
1776 | } | |
1777 | return ret; | |
1778 | } | |
1779 | ||
1780 | /* | |
1781 | * Finds the image layer in the chain that has 'bs' as its backing file. | |
1782 | * | |
1783 | * active is the current topmost image. | |
1784 | * | |
1785 | * Returns NULL if bs is not found in active's image chain, | |
1786 | * or if active == bs. | |
1787 | */ | |
1788 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
1789 | BlockDriverState *bs) | |
1790 | { | |
1791 | BlockDriverState *overlay = NULL; | |
1792 | BlockDriverState *intermediate; | |
1793 | ||
1794 | assert(active != NULL); | |
1795 | assert(bs != NULL); | |
1796 | ||
1797 | /* if bs is the same as active, then by definition it has no overlay | |
1798 | */ | |
1799 | if (active == bs) { | |
1800 | return NULL; | |
1801 | } | |
1802 | ||
1803 | intermediate = active; | |
1804 | while (intermediate->backing_hd) { | |
1805 | if (intermediate->backing_hd == bs) { | |
1806 | overlay = intermediate; | |
1807 | break; | |
1808 | } | |
1809 | intermediate = intermediate->backing_hd; | |
1810 | } | |
1811 | ||
1812 | return overlay; | |
1813 | } | |
1814 | ||
1815 | typedef struct BlkIntermediateStates { | |
1816 | BlockDriverState *bs; | |
1817 | QSIMPLEQ_ENTRY(BlkIntermediateStates) entry; | |
1818 | } BlkIntermediateStates; | |
1819 | ||
1820 | ||
1821 | /* | |
1822 | * Drops images above 'base' up to and including 'top', and sets the image | |
1823 | * above 'top' to have base as its backing file. | |
1824 | * | |
1825 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
1826 | * information in 'bs' can be properly updated. | |
1827 | * | |
1828 | * E.g., this will convert the following chain: | |
1829 | * bottom <- base <- intermediate <- top <- active | |
1830 | * | |
1831 | * to | |
1832 | * | |
1833 | * bottom <- base <- active | |
1834 | * | |
1835 | * It is allowed for bottom==base, in which case it converts: | |
1836 | * | |
1837 | * base <- intermediate <- top <- active | |
1838 | * | |
1839 | * to | |
1840 | * | |
1841 | * base <- active | |
1842 | * | |
1843 | * Error conditions: | |
1844 | * if active == top, that is considered an error | |
1845 | * | |
1846 | */ | |
1847 | int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top, | |
1848 | BlockDriverState *base) | |
1849 | { | |
1850 | BlockDriverState *intermediate; | |
1851 | BlockDriverState *base_bs = NULL; | |
1852 | BlockDriverState *new_top_bs = NULL; | |
1853 | BlkIntermediateStates *intermediate_state, *next; | |
1854 | int ret = -EIO; | |
1855 | ||
1856 | QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete; | |
1857 | QSIMPLEQ_INIT(&states_to_delete); | |
1858 | ||
1859 | if (!top->drv || !base->drv) { | |
1860 | goto exit; | |
1861 | } | |
1862 | ||
1863 | new_top_bs = bdrv_find_overlay(active, top); | |
1864 | ||
1865 | if (new_top_bs == NULL) { | |
1866 | /* we could not find the image above 'top', this is an error */ | |
1867 | goto exit; | |
1868 | } | |
1869 | ||
1870 | /* special case of new_top_bs->backing_hd already pointing to base - nothing | |
1871 | * to do, no intermediate images */ | |
1872 | if (new_top_bs->backing_hd == base) { | |
1873 | ret = 0; | |
1874 | goto exit; | |
1875 | } | |
1876 | ||
1877 | intermediate = top; | |
1878 | ||
1879 | /* now we will go down through the list, and add each BDS we find | |
1880 | * into our deletion queue, until we hit the 'base' | |
1881 | */ | |
1882 | while (intermediate) { | |
1883 | intermediate_state = g_malloc0(sizeof(BlkIntermediateStates)); | |
1884 | intermediate_state->bs = intermediate; | |
1885 | QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry); | |
1886 | ||
1887 | if (intermediate->backing_hd == base) { | |
1888 | base_bs = intermediate->backing_hd; | |
1889 | break; | |
1890 | } | |
1891 | intermediate = intermediate->backing_hd; | |
1892 | } | |
1893 | if (base_bs == NULL) { | |
1894 | /* something went wrong, we did not end at the base. safely | |
1895 | * unravel everything, and exit with error */ | |
1896 | goto exit; | |
1897 | } | |
1898 | ||
1899 | /* success - we can delete the intermediate states, and link top->base */ | |
1900 | ret = bdrv_change_backing_file(new_top_bs, base_bs->filename, | |
1901 | base_bs->drv ? base_bs->drv->format_name : ""); | |
1902 | if (ret) { | |
1903 | goto exit; | |
1904 | } | |
1905 | new_top_bs->backing_hd = base_bs; | |
1906 | ||
1907 | ||
1908 | QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) { | |
1909 | /* so that bdrv_close() does not recursively close the chain */ | |
1910 | intermediate_state->bs->backing_hd = NULL; | |
1911 | bdrv_delete(intermediate_state->bs); | |
1912 | } | |
1913 | ret = 0; | |
1914 | ||
1915 | exit: | |
1916 | QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) { | |
1917 | g_free(intermediate_state); | |
1918 | } | |
1919 | return ret; | |
1920 | } | |
1921 | ||
1922 | ||
1923 | static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, | |
1924 | size_t size) | |
1925 | { | |
1926 | int64_t len; | |
1927 | ||
1928 | if (!bdrv_is_inserted(bs)) | |
1929 | return -ENOMEDIUM; | |
1930 | ||
1931 | if (bs->growable) | |
1932 | return 0; | |
1933 | ||
1934 | len = bdrv_getlength(bs); | |
1935 | ||
1936 | if (offset < 0) | |
1937 | return -EIO; | |
1938 | ||
1939 | if ((offset > len) || (len - offset < size)) | |
1940 | return -EIO; | |
1941 | ||
1942 | return 0; | |
1943 | } | |
1944 | ||
1945 | static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num, | |
1946 | int nb_sectors) | |
1947 | { | |
1948 | return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE, | |
1949 | nb_sectors * BDRV_SECTOR_SIZE); | |
1950 | } | |
1951 | ||
1952 | typedef struct RwCo { | |
1953 | BlockDriverState *bs; | |
1954 | int64_t sector_num; | |
1955 | int nb_sectors; | |
1956 | QEMUIOVector *qiov; | |
1957 | bool is_write; | |
1958 | int ret; | |
1959 | } RwCo; | |
1960 | ||
1961 | static void coroutine_fn bdrv_rw_co_entry(void *opaque) | |
1962 | { | |
1963 | RwCo *rwco = opaque; | |
1964 | ||
1965 | if (!rwco->is_write) { | |
1966 | rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num, | |
1967 | rwco->nb_sectors, rwco->qiov, 0); | |
1968 | } else { | |
1969 | rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num, | |
1970 | rwco->nb_sectors, rwco->qiov, 0); | |
1971 | } | |
1972 | } | |
1973 | ||
1974 | /* | |
1975 | * Process a synchronous request using coroutines | |
1976 | */ | |
1977 | static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf, | |
1978 | int nb_sectors, bool is_write) | |
1979 | { | |
1980 | QEMUIOVector qiov; | |
1981 | struct iovec iov = { | |
1982 | .iov_base = (void *)buf, | |
1983 | .iov_len = nb_sectors * BDRV_SECTOR_SIZE, | |
1984 | }; | |
1985 | Coroutine *co; | |
1986 | RwCo rwco = { | |
1987 | .bs = bs, | |
1988 | .sector_num = sector_num, | |
1989 | .nb_sectors = nb_sectors, | |
1990 | .qiov = &qiov, | |
1991 | .is_write = is_write, | |
1992 | .ret = NOT_DONE, | |
1993 | }; | |
1994 | ||
1995 | qemu_iovec_init_external(&qiov, &iov, 1); | |
1996 | ||
1997 | /** | |
1998 | * In sync call context, when the vcpu is blocked, this throttling timer | |
1999 | * will not fire; so the I/O throttling function has to be disabled here | |
2000 | * if it has been enabled. | |
2001 | */ | |
2002 | if (bs->io_limits_enabled) { | |
2003 | fprintf(stderr, "Disabling I/O throttling on '%s' due " | |
2004 | "to synchronous I/O.\n", bdrv_get_device_name(bs)); | |
2005 | bdrv_io_limits_disable(bs); | |
2006 | } | |
2007 | ||
2008 | if (qemu_in_coroutine()) { | |
2009 | /* Fast-path if already in coroutine context */ | |
2010 | bdrv_rw_co_entry(&rwco); | |
2011 | } else { | |
2012 | co = qemu_coroutine_create(bdrv_rw_co_entry); | |
2013 | qemu_coroutine_enter(co, &rwco); | |
2014 | while (rwco.ret == NOT_DONE) { | |
2015 | qemu_aio_wait(); | |
2016 | } | |
2017 | } | |
2018 | return rwco.ret; | |
2019 | } | |
2020 | ||
2021 | /* return < 0 if error. See bdrv_write() for the return codes */ | |
2022 | int bdrv_read(BlockDriverState *bs, int64_t sector_num, | |
2023 | uint8_t *buf, int nb_sectors) | |
2024 | { | |
2025 | return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false); | |
2026 | } | |
2027 | ||
2028 | /* Just like bdrv_read(), but with I/O throttling temporarily disabled */ | |
2029 | int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num, | |
2030 | uint8_t *buf, int nb_sectors) | |
2031 | { | |
2032 | bool enabled; | |
2033 | int ret; | |
2034 | ||
2035 | enabled = bs->io_limits_enabled; | |
2036 | bs->io_limits_enabled = false; | |
2037 | ret = bdrv_read(bs, 0, buf, 1); | |
2038 | bs->io_limits_enabled = enabled; | |
2039 | return ret; | |
2040 | } | |
2041 | ||
2042 | #define BITS_PER_LONG (sizeof(unsigned long) * 8) | |
2043 | ||
2044 | static void set_dirty_bitmap(BlockDriverState *bs, int64_t sector_num, | |
2045 | int nb_sectors, int dirty) | |
2046 | { | |
2047 | int64_t start, end; | |
2048 | unsigned long val, idx, bit; | |
2049 | ||
2050 | start = sector_num / BDRV_SECTORS_PER_DIRTY_CHUNK; | |
2051 | end = (sector_num + nb_sectors - 1) / BDRV_SECTORS_PER_DIRTY_CHUNK; | |
2052 | ||
2053 | for (; start <= end; start++) { | |
2054 | idx = start / BITS_PER_LONG; | |
2055 | bit = start % BITS_PER_LONG; | |
2056 | val = bs->dirty_bitmap[idx]; | |
2057 | if (dirty) { | |
2058 | if (!(val & (1UL << bit))) { | |
2059 | bs->dirty_count++; | |
2060 | val |= 1UL << bit; | |
2061 | } | |
2062 | } else { | |
2063 | if (val & (1UL << bit)) { | |
2064 | bs->dirty_count--; | |
2065 | val &= ~(1UL << bit); | |
2066 | } | |
2067 | } | |
2068 | bs->dirty_bitmap[idx] = val; | |
2069 | } | |
2070 | } | |
2071 | ||
2072 | /* Return < 0 if error. Important errors are: | |
2073 | -EIO generic I/O error (may happen for all errors) | |
2074 | -ENOMEDIUM No media inserted. | |
2075 | -EINVAL Invalid sector number or nb_sectors | |
2076 | -EACCES Trying to write a read-only device | |
2077 | */ | |
2078 | int bdrv_write(BlockDriverState *bs, int64_t sector_num, | |
2079 | const uint8_t *buf, int nb_sectors) | |
2080 | { | |
2081 | return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true); | |
2082 | } | |
2083 | ||
2084 | int bdrv_pread(BlockDriverState *bs, int64_t offset, | |
2085 | void *buf, int count1) | |
2086 | { | |
2087 | uint8_t tmp_buf[BDRV_SECTOR_SIZE]; | |
2088 | int len, nb_sectors, count; | |
2089 | int64_t sector_num; | |
2090 | int ret; | |
2091 | ||
2092 | count = count1; | |
2093 | /* first read to align to sector start */ | |
2094 | len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1); | |
2095 | if (len > count) | |
2096 | len = count; | |
2097 | sector_num = offset >> BDRV_SECTOR_BITS; | |
2098 | if (len > 0) { | |
2099 | if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) | |
2100 | return ret; | |
2101 | memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len); | |
2102 | count -= len; | |
2103 | if (count == 0) | |
2104 | return count1; | |
2105 | sector_num++; | |
2106 | buf += len; | |
2107 | } | |
2108 | ||
2109 | /* read the sectors "in place" */ | |
2110 | nb_sectors = count >> BDRV_SECTOR_BITS; | |
2111 | if (nb_sectors > 0) { | |
2112 | if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0) | |
2113 | return ret; | |
2114 | sector_num += nb_sectors; | |
2115 | len = nb_sectors << BDRV_SECTOR_BITS; | |
2116 | buf += len; | |
2117 | count -= len; | |
2118 | } | |
2119 | ||
2120 | /* add data from the last sector */ | |
2121 | if (count > 0) { | |
2122 | if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) | |
2123 | return ret; | |
2124 | memcpy(buf, tmp_buf, count); | |
2125 | } | |
2126 | return count1; | |
2127 | } | |
2128 | ||
2129 | int bdrv_pwrite(BlockDriverState *bs, int64_t offset, | |
2130 | const void *buf, int count1) | |
2131 | { | |
2132 | uint8_t tmp_buf[BDRV_SECTOR_SIZE]; | |
2133 | int len, nb_sectors, count; | |
2134 | int64_t sector_num; | |
2135 | int ret; | |
2136 | ||
2137 | count = count1; | |
2138 | /* first write to align to sector start */ | |
2139 | len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1); | |
2140 | if (len > count) | |
2141 | len = count; | |
2142 | sector_num = offset >> BDRV_SECTOR_BITS; | |
2143 | if (len > 0) { | |
2144 | if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) | |
2145 | return ret; | |
2146 | memcpy(tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), buf, len); | |
2147 | if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) | |
2148 | return ret; | |
2149 | count -= len; | |
2150 | if (count == 0) | |
2151 | return count1; | |
2152 | sector_num++; | |
2153 | buf += len; | |
2154 | } | |
2155 | ||
2156 | /* write the sectors "in place" */ | |
2157 | nb_sectors = count >> BDRV_SECTOR_BITS; | |
2158 | if (nb_sectors > 0) { | |
2159 | if ((ret = bdrv_write(bs, sector_num, buf, nb_sectors)) < 0) | |
2160 | return ret; | |
2161 | sector_num += nb_sectors; | |
2162 | len = nb_sectors << BDRV_SECTOR_BITS; | |
2163 | buf += len; | |
2164 | count -= len; | |
2165 | } | |
2166 | ||
2167 | /* add data from the last sector */ | |
2168 | if (count > 0) { | |
2169 | if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) | |
2170 | return ret; | |
2171 | memcpy(tmp_buf, buf, count); | |
2172 | if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) | |
2173 | return ret; | |
2174 | } | |
2175 | return count1; | |
2176 | } | |
2177 | ||
2178 | /* | |
2179 | * Writes to the file and ensures that no writes are reordered across this | |
2180 | * request (acts as a barrier) | |
2181 | * | |
2182 | * Returns 0 on success, -errno in error cases. | |
2183 | */ | |
2184 | int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset, | |
2185 | const void *buf, int count) | |
2186 | { | |
2187 | int ret; | |
2188 | ||
2189 | ret = bdrv_pwrite(bs, offset, buf, count); | |
2190 | if (ret < 0) { | |
2191 | return ret; | |
2192 | } | |
2193 | ||
2194 | /* No flush needed for cache modes that already do it */ | |
2195 | if (bs->enable_write_cache) { | |
2196 | bdrv_flush(bs); | |
2197 | } | |
2198 | ||
2199 | return 0; | |
2200 | } | |
2201 | ||
2202 | static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs, | |
2203 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) | |
2204 | { | |
2205 | /* Perform I/O through a temporary buffer so that users who scribble over | |
2206 | * their read buffer while the operation is in progress do not end up | |
2207 | * modifying the image file. This is critical for zero-copy guest I/O | |
2208 | * where anything might happen inside guest memory. | |
2209 | */ | |
2210 | void *bounce_buffer; | |
2211 | ||
2212 | BlockDriver *drv = bs->drv; | |
2213 | struct iovec iov; | |
2214 | QEMUIOVector bounce_qiov; | |
2215 | int64_t cluster_sector_num; | |
2216 | int cluster_nb_sectors; | |
2217 | size_t skip_bytes; | |
2218 | int ret; | |
2219 | ||
2220 | /* Cover entire cluster so no additional backing file I/O is required when | |
2221 | * allocating cluster in the image file. | |
2222 | */ | |
2223 | round_to_clusters(bs, sector_num, nb_sectors, | |
2224 | &cluster_sector_num, &cluster_nb_sectors); | |
2225 | ||
2226 | trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, | |
2227 | cluster_sector_num, cluster_nb_sectors); | |
2228 | ||
2229 | iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE; | |
2230 | iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len); | |
2231 | qemu_iovec_init_external(&bounce_qiov, &iov, 1); | |
2232 | ||
2233 | ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors, | |
2234 | &bounce_qiov); | |
2235 | if (ret < 0) { | |
2236 | goto err; | |
2237 | } | |
2238 | ||
2239 | if (drv->bdrv_co_write_zeroes && | |
2240 | buffer_is_zero(bounce_buffer, iov.iov_len)) { | |
2241 | ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num, | |
2242 | cluster_nb_sectors); | |
2243 | } else { | |
2244 | /* This does not change the data on the disk, it is not necessary | |
2245 | * to flush even in cache=writethrough mode. | |
2246 | */ | |
2247 | ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors, | |
2248 | &bounce_qiov); | |
2249 | } | |
2250 | ||
2251 | if (ret < 0) { | |
2252 | /* It might be okay to ignore write errors for guest requests. If this | |
2253 | * is a deliberate copy-on-read then we don't want to ignore the error. | |
2254 | * Simply report it in all cases. | |
2255 | */ | |
2256 | goto err; | |
2257 | } | |
2258 | ||
2259 | skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE; | |
2260 | qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, | |
2261 | nb_sectors * BDRV_SECTOR_SIZE); | |
2262 | ||
2263 | err: | |
2264 | qemu_vfree(bounce_buffer); | |
2265 | return ret; | |
2266 | } | |
2267 | ||
2268 | /* | |
2269 | * Handle a read request in coroutine context | |
2270 | */ | |
2271 | static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, | |
2272 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
2273 | BdrvRequestFlags flags) | |
2274 | { | |
2275 | BlockDriver *drv = bs->drv; | |
2276 | BdrvTrackedRequest req; | |
2277 | int ret; | |
2278 | ||
2279 | if (!drv) { | |
2280 | return -ENOMEDIUM; | |
2281 | } | |
2282 | if (bdrv_check_request(bs, sector_num, nb_sectors)) { | |
2283 | return -EIO; | |
2284 | } | |
2285 | ||
2286 | /* throttling disk read I/O */ | |
2287 | if (bs->io_limits_enabled) { | |
2288 | bdrv_io_limits_intercept(bs, false, nb_sectors); | |
2289 | } | |
2290 | ||
2291 | if (bs->copy_on_read) { | |
2292 | flags |= BDRV_REQ_COPY_ON_READ; | |
2293 | } | |
2294 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
2295 | bs->copy_on_read_in_flight++; | |
2296 | } | |
2297 | ||
2298 | if (bs->copy_on_read_in_flight) { | |
2299 | wait_for_overlapping_requests(bs, sector_num, nb_sectors); | |
2300 | } | |
2301 | ||
2302 | tracked_request_begin(&req, bs, sector_num, nb_sectors, false); | |
2303 | ||
2304 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
2305 | int pnum; | |
2306 | ||
2307 | ret = bdrv_co_is_allocated(bs, sector_num, nb_sectors, &pnum); | |
2308 | if (ret < 0) { | |
2309 | goto out; | |
2310 | } | |
2311 | ||
2312 | if (!ret || pnum != nb_sectors) { | |
2313 | ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov); | |
2314 | goto out; | |
2315 | } | |
2316 | } | |
2317 | ||
2318 | ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); | |
2319 | ||
2320 | out: | |
2321 | tracked_request_end(&req); | |
2322 | ||
2323 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
2324 | bs->copy_on_read_in_flight--; | |
2325 | } | |
2326 | ||
2327 | return ret; | |
2328 | } | |
2329 | ||
2330 | int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num, | |
2331 | int nb_sectors, QEMUIOVector *qiov) | |
2332 | { | |
2333 | trace_bdrv_co_readv(bs, sector_num, nb_sectors); | |
2334 | ||
2335 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0); | |
2336 | } | |
2337 | ||
2338 | int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs, | |
2339 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) | |
2340 | { | |
2341 | trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors); | |
2342 | ||
2343 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, | |
2344 | BDRV_REQ_COPY_ON_READ); | |
2345 | } | |
2346 | ||
2347 | static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, | |
2348 | int64_t sector_num, int nb_sectors) | |
2349 | { | |
2350 | BlockDriver *drv = bs->drv; | |
2351 | QEMUIOVector qiov; | |
2352 | struct iovec iov; | |
2353 | int ret; | |
2354 | ||
2355 | /* TODO Emulate only part of misaligned requests instead of letting block | |
2356 | * drivers return -ENOTSUP and emulate everything */ | |
2357 | ||
2358 | /* First try the efficient write zeroes operation */ | |
2359 | if (drv->bdrv_co_write_zeroes) { | |
2360 | ret = drv->bdrv_co_write_zeroes(bs, sector_num, nb_sectors); | |
2361 | if (ret != -ENOTSUP) { | |
2362 | return ret; | |
2363 | } | |
2364 | } | |
2365 | ||
2366 | /* Fall back to bounce buffer if write zeroes is unsupported */ | |
2367 | iov.iov_len = nb_sectors * BDRV_SECTOR_SIZE; | |
2368 | iov.iov_base = qemu_blockalign(bs, iov.iov_len); | |
2369 | memset(iov.iov_base, 0, iov.iov_len); | |
2370 | qemu_iovec_init_external(&qiov, &iov, 1); | |
2371 | ||
2372 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, &qiov); | |
2373 | ||
2374 | qemu_vfree(iov.iov_base); | |
2375 | return ret; | |
2376 | } | |
2377 | ||
2378 | /* | |
2379 | * Handle a write request in coroutine context | |
2380 | */ | |
2381 | static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, | |
2382 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
2383 | BdrvRequestFlags flags) | |
2384 | { | |
2385 | BlockDriver *drv = bs->drv; | |
2386 | BdrvTrackedRequest req; | |
2387 | int ret; | |
2388 | ||
2389 | if (!bs->drv) { | |
2390 | return -ENOMEDIUM; | |
2391 | } | |
2392 | if (bs->read_only) { | |
2393 | return -EACCES; | |
2394 | } | |
2395 | if (bdrv_check_request(bs, sector_num, nb_sectors)) { | |
2396 | return -EIO; | |
2397 | } | |
2398 | ||
2399 | /* throttling disk write I/O */ | |
2400 | if (bs->io_limits_enabled) { | |
2401 | bdrv_io_limits_intercept(bs, true, nb_sectors); | |
2402 | } | |
2403 | ||
2404 | if (bs->copy_on_read_in_flight) { | |
2405 | wait_for_overlapping_requests(bs, sector_num, nb_sectors); | |
2406 | } | |
2407 | ||
2408 | tracked_request_begin(&req, bs, sector_num, nb_sectors, true); | |
2409 | ||
2410 | if (flags & BDRV_REQ_ZERO_WRITE) { | |
2411 | ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors); | |
2412 | } else { | |
2413 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov); | |
2414 | } | |
2415 | ||
2416 | if (ret == 0 && !bs->enable_write_cache) { | |
2417 | ret = bdrv_co_flush(bs); | |
2418 | } | |
2419 | ||
2420 | if (bs->dirty_bitmap) { | |
2421 | bdrv_set_dirty(bs, sector_num, nb_sectors); | |
2422 | } | |
2423 | ||
2424 | if (bs->wr_highest_sector < sector_num + nb_sectors - 1) { | |
2425 | bs->wr_highest_sector = sector_num + nb_sectors - 1; | |
2426 | } | |
2427 | ||
2428 | tracked_request_end(&req); | |
2429 | ||
2430 | return ret; | |
2431 | } | |
2432 | ||
2433 | int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num, | |
2434 | int nb_sectors, QEMUIOVector *qiov) | |
2435 | { | |
2436 | trace_bdrv_co_writev(bs, sector_num, nb_sectors); | |
2437 | ||
2438 | return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0); | |
2439 | } | |
2440 | ||
2441 | int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs, | |
2442 | int64_t sector_num, int nb_sectors) | |
2443 | { | |
2444 | trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors); | |
2445 | ||
2446 | return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL, | |
2447 | BDRV_REQ_ZERO_WRITE); | |
2448 | } | |
2449 | ||
2450 | /** | |
2451 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
2452 | */ | |
2453 | int bdrv_truncate(BlockDriverState *bs, int64_t offset) | |
2454 | { | |
2455 | BlockDriver *drv = bs->drv; | |
2456 | int ret; | |
2457 | if (!drv) | |
2458 | return -ENOMEDIUM; | |
2459 | if (!drv->bdrv_truncate) | |
2460 | return -ENOTSUP; | |
2461 | if (bs->read_only) | |
2462 | return -EACCES; | |
2463 | if (bdrv_in_use(bs)) | |
2464 | return -EBUSY; | |
2465 | ret = drv->bdrv_truncate(bs, offset); | |
2466 | if (ret == 0) { | |
2467 | ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); | |
2468 | bdrv_dev_resize_cb(bs); | |
2469 | } | |
2470 | return ret; | |
2471 | } | |
2472 | ||
2473 | /** | |
2474 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
2475 | * allocated space. Return < 0 if error or unknown. | |
2476 | */ | |
2477 | int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) | |
2478 | { | |
2479 | BlockDriver *drv = bs->drv; | |
2480 | if (!drv) { | |
2481 | return -ENOMEDIUM; | |
2482 | } | |
2483 | if (drv->bdrv_get_allocated_file_size) { | |
2484 | return drv->bdrv_get_allocated_file_size(bs); | |
2485 | } | |
2486 | if (bs->file) { | |
2487 | return bdrv_get_allocated_file_size(bs->file); | |
2488 | } | |
2489 | return -ENOTSUP; | |
2490 | } | |
2491 | ||
2492 | /** | |
2493 | * Length of a file in bytes. Return < 0 if error or unknown. | |
2494 | */ | |
2495 | int64_t bdrv_getlength(BlockDriverState *bs) | |
2496 | { | |
2497 | BlockDriver *drv = bs->drv; | |
2498 | if (!drv) | |
2499 | return -ENOMEDIUM; | |
2500 | ||
2501 | if (bs->growable || bdrv_dev_has_removable_media(bs)) { | |
2502 | if (drv->bdrv_getlength) { | |
2503 | return drv->bdrv_getlength(bs); | |
2504 | } | |
2505 | } | |
2506 | return bs->total_sectors * BDRV_SECTOR_SIZE; | |
2507 | } | |
2508 | ||
2509 | /* return 0 as number of sectors if no device present or error */ | |
2510 | void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr) | |
2511 | { | |
2512 | int64_t length; | |
2513 | length = bdrv_getlength(bs); | |
2514 | if (length < 0) | |
2515 | length = 0; | |
2516 | else | |
2517 | length = length >> BDRV_SECTOR_BITS; | |
2518 | *nb_sectors_ptr = length; | |
2519 | } | |
2520 | ||
2521 | /* throttling disk io limits */ | |
2522 | void bdrv_set_io_limits(BlockDriverState *bs, | |
2523 | BlockIOLimit *io_limits) | |
2524 | { | |
2525 | bs->io_limits = *io_limits; | |
2526 | bs->io_limits_enabled = bdrv_io_limits_enabled(bs); | |
2527 | } | |
2528 | ||
2529 | void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error, | |
2530 | BlockdevOnError on_write_error) | |
2531 | { | |
2532 | bs->on_read_error = on_read_error; | |
2533 | bs->on_write_error = on_write_error; | |
2534 | } | |
2535 | ||
2536 | BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read) | |
2537 | { | |
2538 | return is_read ? bs->on_read_error : bs->on_write_error; | |
2539 | } | |
2540 | ||
2541 | BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error) | |
2542 | { | |
2543 | BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error; | |
2544 | ||
2545 | switch (on_err) { | |
2546 | case BLOCKDEV_ON_ERROR_ENOSPC: | |
2547 | return (error == ENOSPC) ? BDRV_ACTION_STOP : BDRV_ACTION_REPORT; | |
2548 | case BLOCKDEV_ON_ERROR_STOP: | |
2549 | return BDRV_ACTION_STOP; | |
2550 | case BLOCKDEV_ON_ERROR_REPORT: | |
2551 | return BDRV_ACTION_REPORT; | |
2552 | case BLOCKDEV_ON_ERROR_IGNORE: | |
2553 | return BDRV_ACTION_IGNORE; | |
2554 | default: | |
2555 | abort(); | |
2556 | } | |
2557 | } | |
2558 | ||
2559 | /* This is done by device models because, while the block layer knows | |
2560 | * about the error, it does not know whether an operation comes from | |
2561 | * the device or the block layer (from a job, for example). | |
2562 | */ | |
2563 | void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action, | |
2564 | bool is_read, int error) | |
2565 | { | |
2566 | assert(error >= 0); | |
2567 | bdrv_emit_qmp_error_event(bs, QEVENT_BLOCK_IO_ERROR, action, is_read); | |
2568 | if (action == BDRV_ACTION_STOP) { | |
2569 | vm_stop(RUN_STATE_IO_ERROR); | |
2570 | bdrv_iostatus_set_err(bs, error); | |
2571 | } | |
2572 | } | |
2573 | ||
2574 | int bdrv_is_read_only(BlockDriverState *bs) | |
2575 | { | |
2576 | return bs->read_only; | |
2577 | } | |
2578 | ||
2579 | int bdrv_is_sg(BlockDriverState *bs) | |
2580 | { | |
2581 | return bs->sg; | |
2582 | } | |
2583 | ||
2584 | int bdrv_enable_write_cache(BlockDriverState *bs) | |
2585 | { | |
2586 | return bs->enable_write_cache; | |
2587 | } | |
2588 | ||
2589 | void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce) | |
2590 | { | |
2591 | bs->enable_write_cache = wce; | |
2592 | ||
2593 | /* so a reopen() will preserve wce */ | |
2594 | if (wce) { | |
2595 | bs->open_flags |= BDRV_O_CACHE_WB; | |
2596 | } else { | |
2597 | bs->open_flags &= ~BDRV_O_CACHE_WB; | |
2598 | } | |
2599 | } | |
2600 | ||
2601 | int bdrv_is_encrypted(BlockDriverState *bs) | |
2602 | { | |
2603 | if (bs->backing_hd && bs->backing_hd->encrypted) | |
2604 | return 1; | |
2605 | return bs->encrypted; | |
2606 | } | |
2607 | ||
2608 | int bdrv_key_required(BlockDriverState *bs) | |
2609 | { | |
2610 | BlockDriverState *backing_hd = bs->backing_hd; | |
2611 | ||
2612 | if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key) | |
2613 | return 1; | |
2614 | return (bs->encrypted && !bs->valid_key); | |
2615 | } | |
2616 | ||
2617 | int bdrv_set_key(BlockDriverState *bs, const char *key) | |
2618 | { | |
2619 | int ret; | |
2620 | if (bs->backing_hd && bs->backing_hd->encrypted) { | |
2621 | ret = bdrv_set_key(bs->backing_hd, key); | |
2622 | if (ret < 0) | |
2623 | return ret; | |
2624 | if (!bs->encrypted) | |
2625 | return 0; | |
2626 | } | |
2627 | if (!bs->encrypted) { | |
2628 | return -EINVAL; | |
2629 | } else if (!bs->drv || !bs->drv->bdrv_set_key) { | |
2630 | return -ENOMEDIUM; | |
2631 | } | |
2632 | ret = bs->drv->bdrv_set_key(bs, key); | |
2633 | if (ret < 0) { | |
2634 | bs->valid_key = 0; | |
2635 | } else if (!bs->valid_key) { | |
2636 | bs->valid_key = 1; | |
2637 | /* call the change callback now, we skipped it on open */ | |
2638 | bdrv_dev_change_media_cb(bs, true); | |
2639 | } | |
2640 | return ret; | |
2641 | } | |
2642 | ||
2643 | const char *bdrv_get_format_name(BlockDriverState *bs) | |
2644 | { | |
2645 | return bs->drv ? bs->drv->format_name : NULL; | |
2646 | } | |
2647 | ||
2648 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), | |
2649 | void *opaque) | |
2650 | { | |
2651 | BlockDriver *drv; | |
2652 | ||
2653 | QLIST_FOREACH(drv, &bdrv_drivers, list) { | |
2654 | it(opaque, drv->format_name); | |
2655 | } | |
2656 | } | |
2657 | ||
2658 | BlockDriverState *bdrv_find(const char *name) | |
2659 | { | |
2660 | BlockDriverState *bs; | |
2661 | ||
2662 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
2663 | if (!strcmp(name, bs->device_name)) { | |
2664 | return bs; | |
2665 | } | |
2666 | } | |
2667 | return NULL; | |
2668 | } | |
2669 | ||
2670 | BlockDriverState *bdrv_next(BlockDriverState *bs) | |
2671 | { | |
2672 | if (!bs) { | |
2673 | return QTAILQ_FIRST(&bdrv_states); | |
2674 | } | |
2675 | return QTAILQ_NEXT(bs, list); | |
2676 | } | |
2677 | ||
2678 | void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque) | |
2679 | { | |
2680 | BlockDriverState *bs; | |
2681 | ||
2682 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
2683 | it(opaque, bs); | |
2684 | } | |
2685 | } | |
2686 | ||
2687 | const char *bdrv_get_device_name(BlockDriverState *bs) | |
2688 | { | |
2689 | return bs->device_name; | |
2690 | } | |
2691 | ||
2692 | int bdrv_get_flags(BlockDriverState *bs) | |
2693 | { | |
2694 | return bs->open_flags; | |
2695 | } | |
2696 | ||
2697 | void bdrv_flush_all(void) | |
2698 | { | |
2699 | BlockDriverState *bs; | |
2700 | ||
2701 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
2702 | bdrv_flush(bs); | |
2703 | } | |
2704 | } | |
2705 | ||
2706 | int bdrv_has_zero_init(BlockDriverState *bs) | |
2707 | { | |
2708 | assert(bs->drv); | |
2709 | ||
2710 | if (bs->drv->bdrv_has_zero_init) { | |
2711 | return bs->drv->bdrv_has_zero_init(bs); | |
2712 | } | |
2713 | ||
2714 | return 1; | |
2715 | } | |
2716 | ||
2717 | typedef struct BdrvCoIsAllocatedData { | |
2718 | BlockDriverState *bs; | |
2719 | int64_t sector_num; | |
2720 | int nb_sectors; | |
2721 | int *pnum; | |
2722 | int ret; | |
2723 | bool done; | |
2724 | } BdrvCoIsAllocatedData; | |
2725 | ||
2726 | /* | |
2727 | * Returns true iff the specified sector is present in the disk image. Drivers | |
2728 | * not implementing the functionality are assumed to not support backing files, | |
2729 | * hence all their sectors are reported as allocated. | |
2730 | * | |
2731 | * If 'sector_num' is beyond the end of the disk image the return value is 0 | |
2732 | * and 'pnum' is set to 0. | |
2733 | * | |
2734 | * 'pnum' is set to the number of sectors (including and immediately following | |
2735 | * the specified sector) that are known to be in the same | |
2736 | * allocated/unallocated state. | |
2737 | * | |
2738 | * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes | |
2739 | * beyond the end of the disk image it will be clamped. | |
2740 | */ | |
2741 | int coroutine_fn bdrv_co_is_allocated(BlockDriverState *bs, int64_t sector_num, | |
2742 | int nb_sectors, int *pnum) | |
2743 | { | |
2744 | int64_t n; | |
2745 | ||
2746 | if (sector_num >= bs->total_sectors) { | |
2747 | *pnum = 0; | |
2748 | return 0; | |
2749 | } | |
2750 | ||
2751 | n = bs->total_sectors - sector_num; | |
2752 | if (n < nb_sectors) { | |
2753 | nb_sectors = n; | |
2754 | } | |
2755 | ||
2756 | if (!bs->drv->bdrv_co_is_allocated) { | |
2757 | *pnum = nb_sectors; | |
2758 | return 1; | |
2759 | } | |
2760 | ||
2761 | return bs->drv->bdrv_co_is_allocated(bs, sector_num, nb_sectors, pnum); | |
2762 | } | |
2763 | ||
2764 | /* Coroutine wrapper for bdrv_is_allocated() */ | |
2765 | static void coroutine_fn bdrv_is_allocated_co_entry(void *opaque) | |
2766 | { | |
2767 | BdrvCoIsAllocatedData *data = opaque; | |
2768 | BlockDriverState *bs = data->bs; | |
2769 | ||
2770 | data->ret = bdrv_co_is_allocated(bs, data->sector_num, data->nb_sectors, | |
2771 | data->pnum); | |
2772 | data->done = true; | |
2773 | } | |
2774 | ||
2775 | /* | |
2776 | * Synchronous wrapper around bdrv_co_is_allocated(). | |
2777 | * | |
2778 | * See bdrv_co_is_allocated() for details. | |
2779 | */ | |
2780 | int bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, int nb_sectors, | |
2781 | int *pnum) | |
2782 | { | |
2783 | Coroutine *co; | |
2784 | BdrvCoIsAllocatedData data = { | |
2785 | .bs = bs, | |
2786 | .sector_num = sector_num, | |
2787 | .nb_sectors = nb_sectors, | |
2788 | .pnum = pnum, | |
2789 | .done = false, | |
2790 | }; | |
2791 | ||
2792 | co = qemu_coroutine_create(bdrv_is_allocated_co_entry); | |
2793 | qemu_coroutine_enter(co, &data); | |
2794 | while (!data.done) { | |
2795 | qemu_aio_wait(); | |
2796 | } | |
2797 | return data.ret; | |
2798 | } | |
2799 | ||
2800 | /* | |
2801 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
2802 | * | |
2803 | * Return true if the given sector is allocated in any image between | |
2804 | * BASE and TOP (inclusive). BASE can be NULL to check if the given | |
2805 | * sector is allocated in any image of the chain. Return false otherwise. | |
2806 | * | |
2807 | * 'pnum' is set to the number of sectors (including and immediately following | |
2808 | * the specified sector) that are known to be in the same | |
2809 | * allocated/unallocated state. | |
2810 | * | |
2811 | */ | |
2812 | int coroutine_fn bdrv_co_is_allocated_above(BlockDriverState *top, | |
2813 | BlockDriverState *base, | |
2814 | int64_t sector_num, | |
2815 | int nb_sectors, int *pnum) | |
2816 | { | |
2817 | BlockDriverState *intermediate; | |
2818 | int ret, n = nb_sectors; | |
2819 | ||
2820 | intermediate = top; | |
2821 | while (intermediate && intermediate != base) { | |
2822 | int pnum_inter; | |
2823 | ret = bdrv_co_is_allocated(intermediate, sector_num, nb_sectors, | |
2824 | &pnum_inter); | |
2825 | if (ret < 0) { | |
2826 | return ret; | |
2827 | } else if (ret) { | |
2828 | *pnum = pnum_inter; | |
2829 | return 1; | |
2830 | } | |
2831 | ||
2832 | /* | |
2833 | * [sector_num, nb_sectors] is unallocated on top but intermediate | |
2834 | * might have | |
2835 | * | |
2836 | * [sector_num+x, nr_sectors] allocated. | |
2837 | */ | |
2838 | if (n > pnum_inter) { | |
2839 | n = pnum_inter; | |
2840 | } | |
2841 | ||
2842 | intermediate = intermediate->backing_hd; | |
2843 | } | |
2844 | ||
2845 | *pnum = n; | |
2846 | return 0; | |
2847 | } | |
2848 | ||
2849 | BlockInfo *bdrv_query_info(BlockDriverState *bs) | |
2850 | { | |
2851 | BlockInfo *info = g_malloc0(sizeof(*info)); | |
2852 | info->device = g_strdup(bs->device_name); | |
2853 | info->type = g_strdup("unknown"); | |
2854 | info->locked = bdrv_dev_is_medium_locked(bs); | |
2855 | info->removable = bdrv_dev_has_removable_media(bs); | |
2856 | ||
2857 | if (bdrv_dev_has_removable_media(bs)) { | |
2858 | info->has_tray_open = true; | |
2859 | info->tray_open = bdrv_dev_is_tray_open(bs); | |
2860 | } | |
2861 | ||
2862 | if (bdrv_iostatus_is_enabled(bs)) { | |
2863 | info->has_io_status = true; | |
2864 | info->io_status = bs->iostatus; | |
2865 | } | |
2866 | ||
2867 | if (bs->dirty_bitmap) { | |
2868 | info->has_dirty = true; | |
2869 | info->dirty = g_malloc0(sizeof(*info->dirty)); | |
2870 | info->dirty->count = bdrv_get_dirty_count(bs) * | |
2871 | BDRV_SECTORS_PER_DIRTY_CHUNK * BDRV_SECTOR_SIZE; | |
2872 | } | |
2873 | ||
2874 | if (bs->drv) { | |
2875 | info->has_inserted = true; | |
2876 | info->inserted = g_malloc0(sizeof(*info->inserted)); | |
2877 | info->inserted->file = g_strdup(bs->filename); | |
2878 | info->inserted->ro = bs->read_only; | |
2879 | info->inserted->drv = g_strdup(bs->drv->format_name); | |
2880 | info->inserted->encrypted = bs->encrypted; | |
2881 | info->inserted->encryption_key_missing = bdrv_key_required(bs); | |
2882 | ||
2883 | if (bs->backing_file[0]) { | |
2884 | info->inserted->has_backing_file = true; | |
2885 | info->inserted->backing_file = g_strdup(bs->backing_file); | |
2886 | } | |
2887 | ||
2888 | info->inserted->backing_file_depth = bdrv_get_backing_file_depth(bs); | |
2889 | ||
2890 | if (bs->io_limits_enabled) { | |
2891 | info->inserted->bps = | |
2892 | bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]; | |
2893 | info->inserted->bps_rd = | |
2894 | bs->io_limits.bps[BLOCK_IO_LIMIT_READ]; | |
2895 | info->inserted->bps_wr = | |
2896 | bs->io_limits.bps[BLOCK_IO_LIMIT_WRITE]; | |
2897 | info->inserted->iops = | |
2898 | bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]; | |
2899 | info->inserted->iops_rd = | |
2900 | bs->io_limits.iops[BLOCK_IO_LIMIT_READ]; | |
2901 | info->inserted->iops_wr = | |
2902 | bs->io_limits.iops[BLOCK_IO_LIMIT_WRITE]; | |
2903 | } | |
2904 | } | |
2905 | return info; | |
2906 | } | |
2907 | ||
2908 | BlockInfoList *qmp_query_block(Error **errp) | |
2909 | { | |
2910 | BlockInfoList *head = NULL, **p_next = &head; | |
2911 | BlockDriverState *bs; | |
2912 | ||
2913 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
2914 | BlockInfoList *info = g_malloc0(sizeof(*info)); | |
2915 | info->value = bdrv_query_info(bs); | |
2916 | ||
2917 | *p_next = info; | |
2918 | p_next = &info->next; | |
2919 | } | |
2920 | ||
2921 | return head; | |
2922 | } | |
2923 | ||
2924 | BlockStats *bdrv_query_stats(const BlockDriverState *bs) | |
2925 | { | |
2926 | BlockStats *s; | |
2927 | ||
2928 | s = g_malloc0(sizeof(*s)); | |
2929 | ||
2930 | if (bs->device_name[0]) { | |
2931 | s->has_device = true; | |
2932 | s->device = g_strdup(bs->device_name); | |
2933 | } | |
2934 | ||
2935 | s->stats = g_malloc0(sizeof(*s->stats)); | |
2936 | s->stats->rd_bytes = bs->nr_bytes[BDRV_ACCT_READ]; | |
2937 | s->stats->wr_bytes = bs->nr_bytes[BDRV_ACCT_WRITE]; | |
2938 | s->stats->rd_operations = bs->nr_ops[BDRV_ACCT_READ]; | |
2939 | s->stats->wr_operations = bs->nr_ops[BDRV_ACCT_WRITE]; | |
2940 | s->stats->wr_highest_offset = bs->wr_highest_sector * BDRV_SECTOR_SIZE; | |
2941 | s->stats->flush_operations = bs->nr_ops[BDRV_ACCT_FLUSH]; | |
2942 | s->stats->wr_total_time_ns = bs->total_time_ns[BDRV_ACCT_WRITE]; | |
2943 | s->stats->rd_total_time_ns = bs->total_time_ns[BDRV_ACCT_READ]; | |
2944 | s->stats->flush_total_time_ns = bs->total_time_ns[BDRV_ACCT_FLUSH]; | |
2945 | ||
2946 | if (bs->file) { | |
2947 | s->has_parent = true; | |
2948 | s->parent = bdrv_query_stats(bs->file); | |
2949 | } | |
2950 | ||
2951 | return s; | |
2952 | } | |
2953 | ||
2954 | BlockStatsList *qmp_query_blockstats(Error **errp) | |
2955 | { | |
2956 | BlockStatsList *head = NULL, **p_next = &head; | |
2957 | BlockDriverState *bs; | |
2958 | ||
2959 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
2960 | BlockStatsList *info = g_malloc0(sizeof(*info)); | |
2961 | info->value = bdrv_query_stats(bs); | |
2962 | ||
2963 | *p_next = info; | |
2964 | p_next = &info->next; | |
2965 | } | |
2966 | ||
2967 | return head; | |
2968 | } | |
2969 | ||
2970 | const char *bdrv_get_encrypted_filename(BlockDriverState *bs) | |
2971 | { | |
2972 | if (bs->backing_hd && bs->backing_hd->encrypted) | |
2973 | return bs->backing_file; | |
2974 | else if (bs->encrypted) | |
2975 | return bs->filename; | |
2976 | else | |
2977 | return NULL; | |
2978 | } | |
2979 | ||
2980 | void bdrv_get_backing_filename(BlockDriverState *bs, | |
2981 | char *filename, int filename_size) | |
2982 | { | |
2983 | pstrcpy(filename, filename_size, bs->backing_file); | |
2984 | } | |
2985 | ||
2986 | int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num, | |
2987 | const uint8_t *buf, int nb_sectors) | |
2988 | { | |
2989 | BlockDriver *drv = bs->drv; | |
2990 | if (!drv) | |
2991 | return -ENOMEDIUM; | |
2992 | if (!drv->bdrv_write_compressed) | |
2993 | return -ENOTSUP; | |
2994 | if (bdrv_check_request(bs, sector_num, nb_sectors)) | |
2995 | return -EIO; | |
2996 | ||
2997 | assert(!bs->dirty_bitmap); | |
2998 | ||
2999 | return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors); | |
3000 | } | |
3001 | ||
3002 | int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
3003 | { | |
3004 | BlockDriver *drv = bs->drv; | |
3005 | if (!drv) | |
3006 | return -ENOMEDIUM; | |
3007 | if (!drv->bdrv_get_info) | |
3008 | return -ENOTSUP; | |
3009 | memset(bdi, 0, sizeof(*bdi)); | |
3010 | return drv->bdrv_get_info(bs, bdi); | |
3011 | } | |
3012 | ||
3013 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, | |
3014 | int64_t pos, int size) | |
3015 | { | |
3016 | BlockDriver *drv = bs->drv; | |
3017 | if (!drv) | |
3018 | return -ENOMEDIUM; | |
3019 | if (drv->bdrv_save_vmstate) | |
3020 | return drv->bdrv_save_vmstate(bs, buf, pos, size); | |
3021 | if (bs->file) | |
3022 | return bdrv_save_vmstate(bs->file, buf, pos, size); | |
3023 | return -ENOTSUP; | |
3024 | } | |
3025 | ||
3026 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
3027 | int64_t pos, int size) | |
3028 | { | |
3029 | BlockDriver *drv = bs->drv; | |
3030 | if (!drv) | |
3031 | return -ENOMEDIUM; | |
3032 | if (drv->bdrv_load_vmstate) | |
3033 | return drv->bdrv_load_vmstate(bs, buf, pos, size); | |
3034 | if (bs->file) | |
3035 | return bdrv_load_vmstate(bs->file, buf, pos, size); | |
3036 | return -ENOTSUP; | |
3037 | } | |
3038 | ||
3039 | void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event) | |
3040 | { | |
3041 | BlockDriver *drv = bs->drv; | |
3042 | ||
3043 | if (!drv || !drv->bdrv_debug_event) { | |
3044 | return; | |
3045 | } | |
3046 | ||
3047 | drv->bdrv_debug_event(bs, event); | |
3048 | ||
3049 | } | |
3050 | ||
3051 | /**************************************************************/ | |
3052 | /* handling of snapshots */ | |
3053 | ||
3054 | int bdrv_can_snapshot(BlockDriverState *bs) | |
3055 | { | |
3056 | BlockDriver *drv = bs->drv; | |
3057 | if (!drv || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) { | |
3058 | return 0; | |
3059 | } | |
3060 | ||
3061 | if (!drv->bdrv_snapshot_create) { | |
3062 | if (bs->file != NULL) { | |
3063 | return bdrv_can_snapshot(bs->file); | |
3064 | } | |
3065 | return 0; | |
3066 | } | |
3067 | ||
3068 | return 1; | |
3069 | } | |
3070 | ||
3071 | int bdrv_is_snapshot(BlockDriverState *bs) | |
3072 | { | |
3073 | return !!(bs->open_flags & BDRV_O_SNAPSHOT); | |
3074 | } | |
3075 | ||
3076 | BlockDriverState *bdrv_snapshots(void) | |
3077 | { | |
3078 | BlockDriverState *bs; | |
3079 | ||
3080 | if (bs_snapshots) { | |
3081 | return bs_snapshots; | |
3082 | } | |
3083 | ||
3084 | bs = NULL; | |
3085 | while ((bs = bdrv_next(bs))) { | |
3086 | if (bdrv_can_snapshot(bs)) { | |
3087 | bs_snapshots = bs; | |
3088 | return bs; | |
3089 | } | |
3090 | } | |
3091 | return NULL; | |
3092 | } | |
3093 | ||
3094 | int bdrv_snapshot_create(BlockDriverState *bs, | |
3095 | QEMUSnapshotInfo *sn_info) | |
3096 | { | |
3097 | BlockDriver *drv = bs->drv; | |
3098 | if (!drv) | |
3099 | return -ENOMEDIUM; | |
3100 | if (drv->bdrv_snapshot_create) | |
3101 | return drv->bdrv_snapshot_create(bs, sn_info); | |
3102 | if (bs->file) | |
3103 | return bdrv_snapshot_create(bs->file, sn_info); | |
3104 | return -ENOTSUP; | |
3105 | } | |
3106 | ||
3107 | int bdrv_snapshot_goto(BlockDriverState *bs, | |
3108 | const char *snapshot_id) | |
3109 | { | |
3110 | BlockDriver *drv = bs->drv; | |
3111 | int ret, open_ret; | |
3112 | ||
3113 | if (!drv) | |
3114 | return -ENOMEDIUM; | |
3115 | if (drv->bdrv_snapshot_goto) | |
3116 | return drv->bdrv_snapshot_goto(bs, snapshot_id); | |
3117 | ||
3118 | if (bs->file) { | |
3119 | drv->bdrv_close(bs); | |
3120 | ret = bdrv_snapshot_goto(bs->file, snapshot_id); | |
3121 | open_ret = drv->bdrv_open(bs, bs->open_flags); | |
3122 | if (open_ret < 0) { | |
3123 | bdrv_delete(bs->file); | |
3124 | bs->drv = NULL; | |
3125 | return open_ret; | |
3126 | } | |
3127 | return ret; | |
3128 | } | |
3129 | ||
3130 | return -ENOTSUP; | |
3131 | } | |
3132 | ||
3133 | int bdrv_snapshot_delete(BlockDriverState *bs, const char *snapshot_id) | |
3134 | { | |
3135 | BlockDriver *drv = bs->drv; | |
3136 | if (!drv) | |
3137 | return -ENOMEDIUM; | |
3138 | if (drv->bdrv_snapshot_delete) | |
3139 | return drv->bdrv_snapshot_delete(bs, snapshot_id); | |
3140 | if (bs->file) | |
3141 | return bdrv_snapshot_delete(bs->file, snapshot_id); | |
3142 | return -ENOTSUP; | |
3143 | } | |
3144 | ||
3145 | int bdrv_snapshot_list(BlockDriverState *bs, | |
3146 | QEMUSnapshotInfo **psn_info) | |
3147 | { | |
3148 | BlockDriver *drv = bs->drv; | |
3149 | if (!drv) | |
3150 | return -ENOMEDIUM; | |
3151 | if (drv->bdrv_snapshot_list) | |
3152 | return drv->bdrv_snapshot_list(bs, psn_info); | |
3153 | if (bs->file) | |
3154 | return bdrv_snapshot_list(bs->file, psn_info); | |
3155 | return -ENOTSUP; | |
3156 | } | |
3157 | ||
3158 | int bdrv_snapshot_load_tmp(BlockDriverState *bs, | |
3159 | const char *snapshot_name) | |
3160 | { | |
3161 | BlockDriver *drv = bs->drv; | |
3162 | if (!drv) { | |
3163 | return -ENOMEDIUM; | |
3164 | } | |
3165 | if (!bs->read_only) { | |
3166 | return -EINVAL; | |
3167 | } | |
3168 | if (drv->bdrv_snapshot_load_tmp) { | |
3169 | return drv->bdrv_snapshot_load_tmp(bs, snapshot_name); | |
3170 | } | |
3171 | return -ENOTSUP; | |
3172 | } | |
3173 | ||
3174 | /* backing_file can either be relative, or absolute, or a protocol. If it is | |
3175 | * relative, it must be relative to the chain. So, passing in bs->filename | |
3176 | * from a BDS as backing_file should not be done, as that may be relative to | |
3177 | * the CWD rather than the chain. */ | |
3178 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
3179 | const char *backing_file) | |
3180 | { | |
3181 | char *filename_full = NULL; | |
3182 | char *backing_file_full = NULL; | |
3183 | char *filename_tmp = NULL; | |
3184 | int is_protocol = 0; | |
3185 | BlockDriverState *curr_bs = NULL; | |
3186 | BlockDriverState *retval = NULL; | |
3187 | ||
3188 | if (!bs || !bs->drv || !backing_file) { | |
3189 | return NULL; | |
3190 | } | |
3191 | ||
3192 | filename_full = g_malloc(PATH_MAX); | |
3193 | backing_file_full = g_malloc(PATH_MAX); | |
3194 | filename_tmp = g_malloc(PATH_MAX); | |
3195 | ||
3196 | is_protocol = path_has_protocol(backing_file); | |
3197 | ||
3198 | for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) { | |
3199 | ||
3200 | /* If either of the filename paths is actually a protocol, then | |
3201 | * compare unmodified paths; otherwise make paths relative */ | |
3202 | if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
3203 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { | |
3204 | retval = curr_bs->backing_hd; | |
3205 | break; | |
3206 | } | |
3207 | } else { | |
3208 | /* If not an absolute filename path, make it relative to the current | |
3209 | * image's filename path */ | |
3210 | path_combine(filename_tmp, PATH_MAX, curr_bs->filename, | |
3211 | backing_file); | |
3212 | ||
3213 | /* We are going to compare absolute pathnames */ | |
3214 | if (!realpath(filename_tmp, filename_full)) { | |
3215 | continue; | |
3216 | } | |
3217 | ||
3218 | /* We need to make sure the backing filename we are comparing against | |
3219 | * is relative to the current image filename (or absolute) */ | |
3220 | path_combine(filename_tmp, PATH_MAX, curr_bs->filename, | |
3221 | curr_bs->backing_file); | |
3222 | ||
3223 | if (!realpath(filename_tmp, backing_file_full)) { | |
3224 | continue; | |
3225 | } | |
3226 | ||
3227 | if (strcmp(backing_file_full, filename_full) == 0) { | |
3228 | retval = curr_bs->backing_hd; | |
3229 | break; | |
3230 | } | |
3231 | } | |
3232 | } | |
3233 | ||
3234 | g_free(filename_full); | |
3235 | g_free(backing_file_full); | |
3236 | g_free(filename_tmp); | |
3237 | return retval; | |
3238 | } | |
3239 | ||
3240 | int bdrv_get_backing_file_depth(BlockDriverState *bs) | |
3241 | { | |
3242 | if (!bs->drv) { | |
3243 | return 0; | |
3244 | } | |
3245 | ||
3246 | if (!bs->backing_hd) { | |
3247 | return 0; | |
3248 | } | |
3249 | ||
3250 | return 1 + bdrv_get_backing_file_depth(bs->backing_hd); | |
3251 | } | |
3252 | ||
3253 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
3254 | { | |
3255 | BlockDriverState *curr_bs = NULL; | |
3256 | ||
3257 | if (!bs) { | |
3258 | return NULL; | |
3259 | } | |
3260 | ||
3261 | curr_bs = bs; | |
3262 | ||
3263 | while (curr_bs->backing_hd) { | |
3264 | curr_bs = curr_bs->backing_hd; | |
3265 | } | |
3266 | return curr_bs; | |
3267 | } | |
3268 | ||
3269 | #define NB_SUFFIXES 4 | |
3270 | ||
3271 | char *get_human_readable_size(char *buf, int buf_size, int64_t size) | |
3272 | { | |
3273 | static const char suffixes[NB_SUFFIXES] = "KMGT"; | |
3274 | int64_t base; | |
3275 | int i; | |
3276 | ||
3277 | if (size <= 999) { | |
3278 | snprintf(buf, buf_size, "%" PRId64, size); | |
3279 | } else { | |
3280 | base = 1024; | |
3281 | for(i = 0; i < NB_SUFFIXES; i++) { | |
3282 | if (size < (10 * base)) { | |
3283 | snprintf(buf, buf_size, "%0.1f%c", | |
3284 | (double)size / base, | |
3285 | suffixes[i]); | |
3286 | break; | |
3287 | } else if (size < (1000 * base) || i == (NB_SUFFIXES - 1)) { | |
3288 | snprintf(buf, buf_size, "%" PRId64 "%c", | |
3289 | ((size + (base >> 1)) / base), | |
3290 | suffixes[i]); | |
3291 | break; | |
3292 | } | |
3293 | base = base * 1024; | |
3294 | } | |
3295 | } | |
3296 | return buf; | |
3297 | } | |
3298 | ||
3299 | char *bdrv_snapshot_dump(char *buf, int buf_size, QEMUSnapshotInfo *sn) | |
3300 | { | |
3301 | char buf1[128], date_buf[128], clock_buf[128]; | |
3302 | #ifdef _WIN32 | |
3303 | struct tm *ptm; | |
3304 | #else | |
3305 | struct tm tm; | |
3306 | #endif | |
3307 | time_t ti; | |
3308 | int64_t secs; | |
3309 | ||
3310 | if (!sn) { | |
3311 | snprintf(buf, buf_size, | |
3312 | "%-10s%-20s%7s%20s%15s", | |
3313 | "ID", "TAG", "VM SIZE", "DATE", "VM CLOCK"); | |
3314 | } else { | |
3315 | ti = sn->date_sec; | |
3316 | #ifdef _WIN32 | |
3317 | ptm = localtime(&ti); | |
3318 | strftime(date_buf, sizeof(date_buf), | |
3319 | "%Y-%m-%d %H:%M:%S", ptm); | |
3320 | #else | |
3321 | localtime_r(&ti, &tm); | |
3322 | strftime(date_buf, sizeof(date_buf), | |
3323 | "%Y-%m-%d %H:%M:%S", &tm); | |
3324 | #endif | |
3325 | secs = sn->vm_clock_nsec / 1000000000; | |
3326 | snprintf(clock_buf, sizeof(clock_buf), | |
3327 | "%02d:%02d:%02d.%03d", | |
3328 | (int)(secs / 3600), | |
3329 | (int)((secs / 60) % 60), | |
3330 | (int)(secs % 60), | |
3331 | (int)((sn->vm_clock_nsec / 1000000) % 1000)); | |
3332 | snprintf(buf, buf_size, | |
3333 | "%-10s%-20s%7s%20s%15s", | |
3334 | sn->id_str, sn->name, | |
3335 | get_human_readable_size(buf1, sizeof(buf1), sn->vm_state_size), | |
3336 | date_buf, | |
3337 | clock_buf); | |
3338 | } | |
3339 | return buf; | |
3340 | } | |
3341 | ||
3342 | /**************************************************************/ | |
3343 | /* async I/Os */ | |
3344 | ||
3345 | BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num, | |
3346 | QEMUIOVector *qiov, int nb_sectors, | |
3347 | BlockDriverCompletionFunc *cb, void *opaque) | |
3348 | { | |
3349 | trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque); | |
3350 | ||
3351 | return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, | |
3352 | cb, opaque, false); | |
3353 | } | |
3354 | ||
3355 | BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num, | |
3356 | QEMUIOVector *qiov, int nb_sectors, | |
3357 | BlockDriverCompletionFunc *cb, void *opaque) | |
3358 | { | |
3359 | trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque); | |
3360 | ||
3361 | return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, | |
3362 | cb, opaque, true); | |
3363 | } | |
3364 | ||
3365 | ||
3366 | typedef struct MultiwriteCB { | |
3367 | int error; | |
3368 | int num_requests; | |
3369 | int num_callbacks; | |
3370 | struct { | |
3371 | BlockDriverCompletionFunc *cb; | |
3372 | void *opaque; | |
3373 | QEMUIOVector *free_qiov; | |
3374 | } callbacks[]; | |
3375 | } MultiwriteCB; | |
3376 | ||
3377 | static void multiwrite_user_cb(MultiwriteCB *mcb) | |
3378 | { | |
3379 | int i; | |
3380 | ||
3381 | for (i = 0; i < mcb->num_callbacks; i++) { | |
3382 | mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error); | |
3383 | if (mcb->callbacks[i].free_qiov) { | |
3384 | qemu_iovec_destroy(mcb->callbacks[i].free_qiov); | |
3385 | } | |
3386 | g_free(mcb->callbacks[i].free_qiov); | |
3387 | } | |
3388 | } | |
3389 | ||
3390 | static void multiwrite_cb(void *opaque, int ret) | |
3391 | { | |
3392 | MultiwriteCB *mcb = opaque; | |
3393 | ||
3394 | trace_multiwrite_cb(mcb, ret); | |
3395 | ||
3396 | if (ret < 0 && !mcb->error) { | |
3397 | mcb->error = ret; | |
3398 | } | |
3399 | ||
3400 | mcb->num_requests--; | |
3401 | if (mcb->num_requests == 0) { | |
3402 | multiwrite_user_cb(mcb); | |
3403 | g_free(mcb); | |
3404 | } | |
3405 | } | |
3406 | ||
3407 | static int multiwrite_req_compare(const void *a, const void *b) | |
3408 | { | |
3409 | const BlockRequest *req1 = a, *req2 = b; | |
3410 | ||
3411 | /* | |
3412 | * Note that we can't simply subtract req2->sector from req1->sector | |
3413 | * here as that could overflow the return value. | |
3414 | */ | |
3415 | if (req1->sector > req2->sector) { | |
3416 | return 1; | |
3417 | } else if (req1->sector < req2->sector) { | |
3418 | return -1; | |
3419 | } else { | |
3420 | return 0; | |
3421 | } | |
3422 | } | |
3423 | ||
3424 | /* | |
3425 | * Takes a bunch of requests and tries to merge them. Returns the number of | |
3426 | * requests that remain after merging. | |
3427 | */ | |
3428 | static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs, | |
3429 | int num_reqs, MultiwriteCB *mcb) | |
3430 | { | |
3431 | int i, outidx; | |
3432 | ||
3433 | // Sort requests by start sector | |
3434 | qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare); | |
3435 | ||
3436 | // Check if adjacent requests touch the same clusters. If so, combine them, | |
3437 | // filling up gaps with zero sectors. | |
3438 | outidx = 0; | |
3439 | for (i = 1; i < num_reqs; i++) { | |
3440 | int merge = 0; | |
3441 | int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors; | |
3442 | ||
3443 | // Handle exactly sequential writes and overlapping writes. | |
3444 | if (reqs[i].sector <= oldreq_last) { | |
3445 | merge = 1; | |
3446 | } | |
3447 | ||
3448 | if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) { | |
3449 | merge = 0; | |
3450 | } | |
3451 | ||
3452 | if (merge) { | |
3453 | size_t size; | |
3454 | QEMUIOVector *qiov = g_malloc0(sizeof(*qiov)); | |
3455 | qemu_iovec_init(qiov, | |
3456 | reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1); | |
3457 | ||
3458 | // Add the first request to the merged one. If the requests are | |
3459 | // overlapping, drop the last sectors of the first request. | |
3460 | size = (reqs[i].sector - reqs[outidx].sector) << 9; | |
3461 | qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size); | |
3462 | ||
3463 | // We should need to add any zeros between the two requests | |
3464 | assert (reqs[i].sector <= oldreq_last); | |
3465 | ||
3466 | // Add the second request | |
3467 | qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size); | |
3468 | ||
3469 | reqs[outidx].nb_sectors = qiov->size >> 9; | |
3470 | reqs[outidx].qiov = qiov; | |
3471 | ||
3472 | mcb->callbacks[i].free_qiov = reqs[outidx].qiov; | |
3473 | } else { | |
3474 | outidx++; | |
3475 | reqs[outidx].sector = reqs[i].sector; | |
3476 | reqs[outidx].nb_sectors = reqs[i].nb_sectors; | |
3477 | reqs[outidx].qiov = reqs[i].qiov; | |
3478 | } | |
3479 | } | |
3480 | ||
3481 | return outidx + 1; | |
3482 | } | |
3483 | ||
3484 | /* | |
3485 | * Submit multiple AIO write requests at once. | |
3486 | * | |
3487 | * On success, the function returns 0 and all requests in the reqs array have | |
3488 | * been submitted. In error case this function returns -1, and any of the | |
3489 | * requests may or may not be submitted yet. In particular, this means that the | |
3490 | * callback will be called for some of the requests, for others it won't. The | |
3491 | * caller must check the error field of the BlockRequest to wait for the right | |
3492 | * callbacks (if error != 0, no callback will be called). | |
3493 | * | |
3494 | * The implementation may modify the contents of the reqs array, e.g. to merge | |
3495 | * requests. However, the fields opaque and error are left unmodified as they | |
3496 | * are used to signal failure for a single request to the caller. | |
3497 | */ | |
3498 | int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs) | |
3499 | { | |
3500 | MultiwriteCB *mcb; | |
3501 | int i; | |
3502 | ||
3503 | /* don't submit writes if we don't have a medium */ | |
3504 | if (bs->drv == NULL) { | |
3505 | for (i = 0; i < num_reqs; i++) { | |
3506 | reqs[i].error = -ENOMEDIUM; | |
3507 | } | |
3508 | return -1; | |
3509 | } | |
3510 | ||
3511 | if (num_reqs == 0) { | |
3512 | return 0; | |
3513 | } | |
3514 | ||
3515 | // Create MultiwriteCB structure | |
3516 | mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks)); | |
3517 | mcb->num_requests = 0; | |
3518 | mcb->num_callbacks = num_reqs; | |
3519 | ||
3520 | for (i = 0; i < num_reqs; i++) { | |
3521 | mcb->callbacks[i].cb = reqs[i].cb; | |
3522 | mcb->callbacks[i].opaque = reqs[i].opaque; | |
3523 | } | |
3524 | ||
3525 | // Check for mergable requests | |
3526 | num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb); | |
3527 | ||
3528 | trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs); | |
3529 | ||
3530 | /* Run the aio requests. */ | |
3531 | mcb->num_requests = num_reqs; | |
3532 | for (i = 0; i < num_reqs; i++) { | |
3533 | bdrv_aio_writev(bs, reqs[i].sector, reqs[i].qiov, | |
3534 | reqs[i].nb_sectors, multiwrite_cb, mcb); | |
3535 | } | |
3536 | ||
3537 | return 0; | |
3538 | } | |
3539 | ||
3540 | void bdrv_aio_cancel(BlockDriverAIOCB *acb) | |
3541 | { | |
3542 | acb->aiocb_info->cancel(acb); | |
3543 | } | |
3544 | ||
3545 | /* block I/O throttling */ | |
3546 | static bool bdrv_exceed_bps_limits(BlockDriverState *bs, int nb_sectors, | |
3547 | bool is_write, double elapsed_time, uint64_t *wait) | |
3548 | { | |
3549 | uint64_t bps_limit = 0; | |
3550 | double bytes_limit, bytes_base, bytes_res; | |
3551 | double slice_time, wait_time; | |
3552 | ||
3553 | if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) { | |
3554 | bps_limit = bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]; | |
3555 | } else if (bs->io_limits.bps[is_write]) { | |
3556 | bps_limit = bs->io_limits.bps[is_write]; | |
3557 | } else { | |
3558 | if (wait) { | |
3559 | *wait = 0; | |
3560 | } | |
3561 | ||
3562 | return false; | |
3563 | } | |
3564 | ||
3565 | slice_time = bs->slice_end - bs->slice_start; | |
3566 | slice_time /= (NANOSECONDS_PER_SECOND); | |
3567 | bytes_limit = bps_limit * slice_time; | |
3568 | bytes_base = bs->nr_bytes[is_write] - bs->io_base.bytes[is_write]; | |
3569 | if (bs->io_limits.bps[BLOCK_IO_LIMIT_TOTAL]) { | |
3570 | bytes_base += bs->nr_bytes[!is_write] - bs->io_base.bytes[!is_write]; | |
3571 | } | |
3572 | ||
3573 | /* bytes_base: the bytes of data which have been read/written; and | |
3574 | * it is obtained from the history statistic info. | |
3575 | * bytes_res: the remaining bytes of data which need to be read/written. | |
3576 | * (bytes_base + bytes_res) / bps_limit: used to calcuate | |
3577 | * the total time for completing reading/writting all data. | |
3578 | */ | |
3579 | bytes_res = (unsigned) nb_sectors * BDRV_SECTOR_SIZE; | |
3580 | ||
3581 | if (bytes_base + bytes_res <= bytes_limit) { | |
3582 | if (wait) { | |
3583 | *wait = 0; | |
3584 | } | |
3585 | ||
3586 | return false; | |
3587 | } | |
3588 | ||
3589 | /* Calc approx time to dispatch */ | |
3590 | wait_time = (bytes_base + bytes_res) / bps_limit - elapsed_time; | |
3591 | ||
3592 | /* When the I/O rate at runtime exceeds the limits, | |
3593 | * bs->slice_end need to be extended in order that the current statistic | |
3594 | * info can be kept until the timer fire, so it is increased and tuned | |
3595 | * based on the result of experiment. | |
3596 | */ | |
3597 | bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10; | |
3598 | bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME; | |
3599 | if (wait) { | |
3600 | *wait = wait_time * BLOCK_IO_SLICE_TIME * 10; | |
3601 | } | |
3602 | ||
3603 | return true; | |
3604 | } | |
3605 | ||
3606 | static bool bdrv_exceed_iops_limits(BlockDriverState *bs, bool is_write, | |
3607 | double elapsed_time, uint64_t *wait) | |
3608 | { | |
3609 | uint64_t iops_limit = 0; | |
3610 | double ios_limit, ios_base; | |
3611 | double slice_time, wait_time; | |
3612 | ||
3613 | if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) { | |
3614 | iops_limit = bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]; | |
3615 | } else if (bs->io_limits.iops[is_write]) { | |
3616 | iops_limit = bs->io_limits.iops[is_write]; | |
3617 | } else { | |
3618 | if (wait) { | |
3619 | *wait = 0; | |
3620 | } | |
3621 | ||
3622 | return false; | |
3623 | } | |
3624 | ||
3625 | slice_time = bs->slice_end - bs->slice_start; | |
3626 | slice_time /= (NANOSECONDS_PER_SECOND); | |
3627 | ios_limit = iops_limit * slice_time; | |
3628 | ios_base = bs->nr_ops[is_write] - bs->io_base.ios[is_write]; | |
3629 | if (bs->io_limits.iops[BLOCK_IO_LIMIT_TOTAL]) { | |
3630 | ios_base += bs->nr_ops[!is_write] - bs->io_base.ios[!is_write]; | |
3631 | } | |
3632 | ||
3633 | if (ios_base + 1 <= ios_limit) { | |
3634 | if (wait) { | |
3635 | *wait = 0; | |
3636 | } | |
3637 | ||
3638 | return false; | |
3639 | } | |
3640 | ||
3641 | /* Calc approx time to dispatch */ | |
3642 | wait_time = (ios_base + 1) / iops_limit; | |
3643 | if (wait_time > elapsed_time) { | |
3644 | wait_time = wait_time - elapsed_time; | |
3645 | } else { | |
3646 | wait_time = 0; | |
3647 | } | |
3648 | ||
3649 | bs->slice_time = wait_time * BLOCK_IO_SLICE_TIME * 10; | |
3650 | bs->slice_end += bs->slice_time - 3 * BLOCK_IO_SLICE_TIME; | |
3651 | if (wait) { | |
3652 | *wait = wait_time * BLOCK_IO_SLICE_TIME * 10; | |
3653 | } | |
3654 | ||
3655 | return true; | |
3656 | } | |
3657 | ||
3658 | static bool bdrv_exceed_io_limits(BlockDriverState *bs, int nb_sectors, | |
3659 | bool is_write, int64_t *wait) | |
3660 | { | |
3661 | int64_t now, max_wait; | |
3662 | uint64_t bps_wait = 0, iops_wait = 0; | |
3663 | double elapsed_time; | |
3664 | int bps_ret, iops_ret; | |
3665 | ||
3666 | now = qemu_get_clock_ns(vm_clock); | |
3667 | if ((bs->slice_start < now) | |
3668 | && (bs->slice_end > now)) { | |
3669 | bs->slice_end = now + bs->slice_time; | |
3670 | } else { | |
3671 | bs->slice_time = 5 * BLOCK_IO_SLICE_TIME; | |
3672 | bs->slice_start = now; | |
3673 | bs->slice_end = now + bs->slice_time; | |
3674 | ||
3675 | bs->io_base.bytes[is_write] = bs->nr_bytes[is_write]; | |
3676 | bs->io_base.bytes[!is_write] = bs->nr_bytes[!is_write]; | |
3677 | ||
3678 | bs->io_base.ios[is_write] = bs->nr_ops[is_write]; | |
3679 | bs->io_base.ios[!is_write] = bs->nr_ops[!is_write]; | |
3680 | } | |
3681 | ||
3682 | elapsed_time = now - bs->slice_start; | |
3683 | elapsed_time /= (NANOSECONDS_PER_SECOND); | |
3684 | ||
3685 | bps_ret = bdrv_exceed_bps_limits(bs, nb_sectors, | |
3686 | is_write, elapsed_time, &bps_wait); | |
3687 | iops_ret = bdrv_exceed_iops_limits(bs, is_write, | |
3688 | elapsed_time, &iops_wait); | |
3689 | if (bps_ret || iops_ret) { | |
3690 | max_wait = bps_wait > iops_wait ? bps_wait : iops_wait; | |
3691 | if (wait) { | |
3692 | *wait = max_wait; | |
3693 | } | |
3694 | ||
3695 | now = qemu_get_clock_ns(vm_clock); | |
3696 | if (bs->slice_end < now + max_wait) { | |
3697 | bs->slice_end = now + max_wait; | |
3698 | } | |
3699 | ||
3700 | return true; | |
3701 | } | |
3702 | ||
3703 | if (wait) { | |
3704 | *wait = 0; | |
3705 | } | |
3706 | ||
3707 | return false; | |
3708 | } | |
3709 | ||
3710 | /**************************************************************/ | |
3711 | /* async block device emulation */ | |
3712 | ||
3713 | typedef struct BlockDriverAIOCBSync { | |
3714 | BlockDriverAIOCB common; | |
3715 | QEMUBH *bh; | |
3716 | int ret; | |
3717 | /* vector translation state */ | |
3718 | QEMUIOVector *qiov; | |
3719 | uint8_t *bounce; | |
3720 | int is_write; | |
3721 | } BlockDriverAIOCBSync; | |
3722 | ||
3723 | static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb) | |
3724 | { | |
3725 | BlockDriverAIOCBSync *acb = | |
3726 | container_of(blockacb, BlockDriverAIOCBSync, common); | |
3727 | qemu_bh_delete(acb->bh); | |
3728 | acb->bh = NULL; | |
3729 | qemu_aio_release(acb); | |
3730 | } | |
3731 | ||
3732 | static const AIOCBInfo bdrv_em_aiocb_info = { | |
3733 | .aiocb_size = sizeof(BlockDriverAIOCBSync), | |
3734 | .cancel = bdrv_aio_cancel_em, | |
3735 | }; | |
3736 | ||
3737 | static void bdrv_aio_bh_cb(void *opaque) | |
3738 | { | |
3739 | BlockDriverAIOCBSync *acb = opaque; | |
3740 | ||
3741 | if (!acb->is_write) | |
3742 | qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size); | |
3743 | qemu_vfree(acb->bounce); | |
3744 | acb->common.cb(acb->common.opaque, acb->ret); | |
3745 | qemu_bh_delete(acb->bh); | |
3746 | acb->bh = NULL; | |
3747 | qemu_aio_release(acb); | |
3748 | } | |
3749 | ||
3750 | static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs, | |
3751 | int64_t sector_num, | |
3752 | QEMUIOVector *qiov, | |
3753 | int nb_sectors, | |
3754 | BlockDriverCompletionFunc *cb, | |
3755 | void *opaque, | |
3756 | int is_write) | |
3757 | ||
3758 | { | |
3759 | BlockDriverAIOCBSync *acb; | |
3760 | ||
3761 | acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque); | |
3762 | acb->is_write = is_write; | |
3763 | acb->qiov = qiov; | |
3764 | acb->bounce = qemu_blockalign(bs, qiov->size); | |
3765 | acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb); | |
3766 | ||
3767 | if (is_write) { | |
3768 | qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size); | |
3769 | acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors); | |
3770 | } else { | |
3771 | acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors); | |
3772 | } | |
3773 | ||
3774 | qemu_bh_schedule(acb->bh); | |
3775 | ||
3776 | return &acb->common; | |
3777 | } | |
3778 | ||
3779 | static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs, | |
3780 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
3781 | BlockDriverCompletionFunc *cb, void *opaque) | |
3782 | { | |
3783 | return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); | |
3784 | } | |
3785 | ||
3786 | static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs, | |
3787 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
3788 | BlockDriverCompletionFunc *cb, void *opaque) | |
3789 | { | |
3790 | return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); | |
3791 | } | |
3792 | ||
3793 | ||
3794 | typedef struct BlockDriverAIOCBCoroutine { | |
3795 | BlockDriverAIOCB common; | |
3796 | BlockRequest req; | |
3797 | bool is_write; | |
3798 | bool *done; | |
3799 | QEMUBH* bh; | |
3800 | } BlockDriverAIOCBCoroutine; | |
3801 | ||
3802 | static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb) | |
3803 | { | |
3804 | BlockDriverAIOCBCoroutine *acb = | |
3805 | container_of(blockacb, BlockDriverAIOCBCoroutine, common); | |
3806 | bool done = false; | |
3807 | ||
3808 | acb->done = &done; | |
3809 | while (!done) { | |
3810 | qemu_aio_wait(); | |
3811 | } | |
3812 | } | |
3813 | ||
3814 | static const AIOCBInfo bdrv_em_co_aiocb_info = { | |
3815 | .aiocb_size = sizeof(BlockDriverAIOCBCoroutine), | |
3816 | .cancel = bdrv_aio_co_cancel_em, | |
3817 | }; | |
3818 | ||
3819 | static void bdrv_co_em_bh(void *opaque) | |
3820 | { | |
3821 | BlockDriverAIOCBCoroutine *acb = opaque; | |
3822 | ||
3823 | acb->common.cb(acb->common.opaque, acb->req.error); | |
3824 | ||
3825 | if (acb->done) { | |
3826 | *acb->done = true; | |
3827 | } | |
3828 | ||
3829 | qemu_bh_delete(acb->bh); | |
3830 | qemu_aio_release(acb); | |
3831 | } | |
3832 | ||
3833 | /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */ | |
3834 | static void coroutine_fn bdrv_co_do_rw(void *opaque) | |
3835 | { | |
3836 | BlockDriverAIOCBCoroutine *acb = opaque; | |
3837 | BlockDriverState *bs = acb->common.bs; | |
3838 | ||
3839 | if (!acb->is_write) { | |
3840 | acb->req.error = bdrv_co_do_readv(bs, acb->req.sector, | |
3841 | acb->req.nb_sectors, acb->req.qiov, 0); | |
3842 | } else { | |
3843 | acb->req.error = bdrv_co_do_writev(bs, acb->req.sector, | |
3844 | acb->req.nb_sectors, acb->req.qiov, 0); | |
3845 | } | |
3846 | ||
3847 | acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); | |
3848 | qemu_bh_schedule(acb->bh); | |
3849 | } | |
3850 | ||
3851 | static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, | |
3852 | int64_t sector_num, | |
3853 | QEMUIOVector *qiov, | |
3854 | int nb_sectors, | |
3855 | BlockDriverCompletionFunc *cb, | |
3856 | void *opaque, | |
3857 | bool is_write) | |
3858 | { | |
3859 | Coroutine *co; | |
3860 | BlockDriverAIOCBCoroutine *acb; | |
3861 | ||
3862 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
3863 | acb->req.sector = sector_num; | |
3864 | acb->req.nb_sectors = nb_sectors; | |
3865 | acb->req.qiov = qiov; | |
3866 | acb->is_write = is_write; | |
3867 | acb->done = NULL; | |
3868 | ||
3869 | co = qemu_coroutine_create(bdrv_co_do_rw); | |
3870 | qemu_coroutine_enter(co, acb); | |
3871 | ||
3872 | return &acb->common; | |
3873 | } | |
3874 | ||
3875 | static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque) | |
3876 | { | |
3877 | BlockDriverAIOCBCoroutine *acb = opaque; | |
3878 | BlockDriverState *bs = acb->common.bs; | |
3879 | ||
3880 | acb->req.error = bdrv_co_flush(bs); | |
3881 | acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); | |
3882 | qemu_bh_schedule(acb->bh); | |
3883 | } | |
3884 | ||
3885 | BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs, | |
3886 | BlockDriverCompletionFunc *cb, void *opaque) | |
3887 | { | |
3888 | trace_bdrv_aio_flush(bs, opaque); | |
3889 | ||
3890 | Coroutine *co; | |
3891 | BlockDriverAIOCBCoroutine *acb; | |
3892 | ||
3893 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
3894 | acb->done = NULL; | |
3895 | ||
3896 | co = qemu_coroutine_create(bdrv_aio_flush_co_entry); | |
3897 | qemu_coroutine_enter(co, acb); | |
3898 | ||
3899 | return &acb->common; | |
3900 | } | |
3901 | ||
3902 | static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque) | |
3903 | { | |
3904 | BlockDriverAIOCBCoroutine *acb = opaque; | |
3905 | BlockDriverState *bs = acb->common.bs; | |
3906 | ||
3907 | acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors); | |
3908 | acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); | |
3909 | qemu_bh_schedule(acb->bh); | |
3910 | } | |
3911 | ||
3912 | BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs, | |
3913 | int64_t sector_num, int nb_sectors, | |
3914 | BlockDriverCompletionFunc *cb, void *opaque) | |
3915 | { | |
3916 | Coroutine *co; | |
3917 | BlockDriverAIOCBCoroutine *acb; | |
3918 | ||
3919 | trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque); | |
3920 | ||
3921 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
3922 | acb->req.sector = sector_num; | |
3923 | acb->req.nb_sectors = nb_sectors; | |
3924 | acb->done = NULL; | |
3925 | co = qemu_coroutine_create(bdrv_aio_discard_co_entry); | |
3926 | qemu_coroutine_enter(co, acb); | |
3927 | ||
3928 | return &acb->common; | |
3929 | } | |
3930 | ||
3931 | void bdrv_init(void) | |
3932 | { | |
3933 | module_call_init(MODULE_INIT_BLOCK); | |
3934 | } | |
3935 | ||
3936 | void bdrv_init_with_whitelist(void) | |
3937 | { | |
3938 | use_bdrv_whitelist = 1; | |
3939 | bdrv_init(); | |
3940 | } | |
3941 | ||
3942 | void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, | |
3943 | BlockDriverCompletionFunc *cb, void *opaque) | |
3944 | { | |
3945 | BlockDriverAIOCB *acb; | |
3946 | ||
3947 | acb = g_slice_alloc(aiocb_info->aiocb_size); | |
3948 | acb->aiocb_info = aiocb_info; | |
3949 | acb->bs = bs; | |
3950 | acb->cb = cb; | |
3951 | acb->opaque = opaque; | |
3952 | return acb; | |
3953 | } | |
3954 | ||
3955 | void qemu_aio_release(void *p) | |
3956 | { | |
3957 | BlockDriverAIOCB *acb = p; | |
3958 | g_slice_free1(acb->aiocb_info->aiocb_size, acb); | |
3959 | } | |
3960 | ||
3961 | /**************************************************************/ | |
3962 | /* Coroutine block device emulation */ | |
3963 | ||
3964 | typedef struct CoroutineIOCompletion { | |
3965 | Coroutine *coroutine; | |
3966 | int ret; | |
3967 | } CoroutineIOCompletion; | |
3968 | ||
3969 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
3970 | { | |
3971 | CoroutineIOCompletion *co = opaque; | |
3972 | ||
3973 | co->ret = ret; | |
3974 | qemu_coroutine_enter(co->coroutine, NULL); | |
3975 | } | |
3976 | ||
3977 | static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num, | |
3978 | int nb_sectors, QEMUIOVector *iov, | |
3979 | bool is_write) | |
3980 | { | |
3981 | CoroutineIOCompletion co = { | |
3982 | .coroutine = qemu_coroutine_self(), | |
3983 | }; | |
3984 | BlockDriverAIOCB *acb; | |
3985 | ||
3986 | if (is_write) { | |
3987 | acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors, | |
3988 | bdrv_co_io_em_complete, &co); | |
3989 | } else { | |
3990 | acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors, | |
3991 | bdrv_co_io_em_complete, &co); | |
3992 | } | |
3993 | ||
3994 | trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb); | |
3995 | if (!acb) { | |
3996 | return -EIO; | |
3997 | } | |
3998 | qemu_coroutine_yield(); | |
3999 | ||
4000 | return co.ret; | |
4001 | } | |
4002 | ||
4003 | static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, | |
4004 | int64_t sector_num, int nb_sectors, | |
4005 | QEMUIOVector *iov) | |
4006 | { | |
4007 | return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false); | |
4008 | } | |
4009 | ||
4010 | static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, | |
4011 | int64_t sector_num, int nb_sectors, | |
4012 | QEMUIOVector *iov) | |
4013 | { | |
4014 | return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true); | |
4015 | } | |
4016 | ||
4017 | static void coroutine_fn bdrv_flush_co_entry(void *opaque) | |
4018 | { | |
4019 | RwCo *rwco = opaque; | |
4020 | ||
4021 | rwco->ret = bdrv_co_flush(rwco->bs); | |
4022 | } | |
4023 | ||
4024 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) | |
4025 | { | |
4026 | int ret; | |
4027 | ||
4028 | if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) { | |
4029 | return 0; | |
4030 | } | |
4031 | ||
4032 | /* Write back cached data to the OS even with cache=unsafe */ | |
4033 | if (bs->drv->bdrv_co_flush_to_os) { | |
4034 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
4035 | if (ret < 0) { | |
4036 | return ret; | |
4037 | } | |
4038 | } | |
4039 | ||
4040 | /* But don't actually force it to the disk with cache=unsafe */ | |
4041 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
4042 | goto flush_parent; | |
4043 | } | |
4044 | ||
4045 | if (bs->drv->bdrv_co_flush_to_disk) { | |
4046 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
4047 | } else if (bs->drv->bdrv_aio_flush) { | |
4048 | BlockDriverAIOCB *acb; | |
4049 | CoroutineIOCompletion co = { | |
4050 | .coroutine = qemu_coroutine_self(), | |
4051 | }; | |
4052 | ||
4053 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
4054 | if (acb == NULL) { | |
4055 | ret = -EIO; | |
4056 | } else { | |
4057 | qemu_coroutine_yield(); | |
4058 | ret = co.ret; | |
4059 | } | |
4060 | } else { | |
4061 | /* | |
4062 | * Some block drivers always operate in either writethrough or unsafe | |
4063 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
4064 | * know how the server works (because the behaviour is hardcoded or | |
4065 | * depends on server-side configuration), so we can't ensure that | |
4066 | * everything is safe on disk. Returning an error doesn't work because | |
4067 | * that would break guests even if the server operates in writethrough | |
4068 | * mode. | |
4069 | * | |
4070 | * Let's hope the user knows what he's doing. | |
4071 | */ | |
4072 | ret = 0; | |
4073 | } | |
4074 | if (ret < 0) { | |
4075 | return ret; | |
4076 | } | |
4077 | ||
4078 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
4079 | * in the case of cache=unsafe, so there are no useless flushes. | |
4080 | */ | |
4081 | flush_parent: | |
4082 | return bdrv_co_flush(bs->file); | |
4083 | } | |
4084 | ||
4085 | void bdrv_invalidate_cache(BlockDriverState *bs) | |
4086 | { | |
4087 | if (bs->drv && bs->drv->bdrv_invalidate_cache) { | |
4088 | bs->drv->bdrv_invalidate_cache(bs); | |
4089 | } | |
4090 | } | |
4091 | ||
4092 | void bdrv_invalidate_cache_all(void) | |
4093 | { | |
4094 | BlockDriverState *bs; | |
4095 | ||
4096 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
4097 | bdrv_invalidate_cache(bs); | |
4098 | } | |
4099 | } | |
4100 | ||
4101 | void bdrv_clear_incoming_migration_all(void) | |
4102 | { | |
4103 | BlockDriverState *bs; | |
4104 | ||
4105 | QTAILQ_FOREACH(bs, &bdrv_states, list) { | |
4106 | bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING); | |
4107 | } | |
4108 | } | |
4109 | ||
4110 | int bdrv_flush(BlockDriverState *bs) | |
4111 | { | |
4112 | Coroutine *co; | |
4113 | RwCo rwco = { | |
4114 | .bs = bs, | |
4115 | .ret = NOT_DONE, | |
4116 | }; | |
4117 | ||
4118 | if (qemu_in_coroutine()) { | |
4119 | /* Fast-path if already in coroutine context */ | |
4120 | bdrv_flush_co_entry(&rwco); | |
4121 | } else { | |
4122 | co = qemu_coroutine_create(bdrv_flush_co_entry); | |
4123 | qemu_coroutine_enter(co, &rwco); | |
4124 | while (rwco.ret == NOT_DONE) { | |
4125 | qemu_aio_wait(); | |
4126 | } | |
4127 | } | |
4128 | ||
4129 | return rwco.ret; | |
4130 | } | |
4131 | ||
4132 | static void coroutine_fn bdrv_discard_co_entry(void *opaque) | |
4133 | { | |
4134 | RwCo *rwco = opaque; | |
4135 | ||
4136 | rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors); | |
4137 | } | |
4138 | ||
4139 | int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num, | |
4140 | int nb_sectors) | |
4141 | { | |
4142 | if (!bs->drv) { | |
4143 | return -ENOMEDIUM; | |
4144 | } else if (bdrv_check_request(bs, sector_num, nb_sectors)) { | |
4145 | return -EIO; | |
4146 | } else if (bs->read_only) { | |
4147 | return -EROFS; | |
4148 | } else if (bs->drv->bdrv_co_discard) { | |
4149 | return bs->drv->bdrv_co_discard(bs, sector_num, nb_sectors); | |
4150 | } else if (bs->drv->bdrv_aio_discard) { | |
4151 | BlockDriverAIOCB *acb; | |
4152 | CoroutineIOCompletion co = { | |
4153 | .coroutine = qemu_coroutine_self(), | |
4154 | }; | |
4155 | ||
4156 | acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors, | |
4157 | bdrv_co_io_em_complete, &co); | |
4158 | if (acb == NULL) { | |
4159 | return -EIO; | |
4160 | } else { | |
4161 | qemu_coroutine_yield(); | |
4162 | return co.ret; | |
4163 | } | |
4164 | } else { | |
4165 | return 0; | |
4166 | } | |
4167 | } | |
4168 | ||
4169 | int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors) | |
4170 | { | |
4171 | Coroutine *co; | |
4172 | RwCo rwco = { | |
4173 | .bs = bs, | |
4174 | .sector_num = sector_num, | |
4175 | .nb_sectors = nb_sectors, | |
4176 | .ret = NOT_DONE, | |
4177 | }; | |
4178 | ||
4179 | if (qemu_in_coroutine()) { | |
4180 | /* Fast-path if already in coroutine context */ | |
4181 | bdrv_discard_co_entry(&rwco); | |
4182 | } else { | |
4183 | co = qemu_coroutine_create(bdrv_discard_co_entry); | |
4184 | qemu_coroutine_enter(co, &rwco); | |
4185 | while (rwco.ret == NOT_DONE) { | |
4186 | qemu_aio_wait(); | |
4187 | } | |
4188 | } | |
4189 | ||
4190 | return rwco.ret; | |
4191 | } | |
4192 | ||
4193 | /**************************************************************/ | |
4194 | /* removable device support */ | |
4195 | ||
4196 | /** | |
4197 | * Return TRUE if the media is present | |
4198 | */ | |
4199 | int bdrv_is_inserted(BlockDriverState *bs) | |
4200 | { | |
4201 | BlockDriver *drv = bs->drv; | |
4202 | ||
4203 | if (!drv) | |
4204 | return 0; | |
4205 | if (!drv->bdrv_is_inserted) | |
4206 | return 1; | |
4207 | return drv->bdrv_is_inserted(bs); | |
4208 | } | |
4209 | ||
4210 | /** | |
4211 | * Return whether the media changed since the last call to this | |
4212 | * function, or -ENOTSUP if we don't know. Most drivers don't know. | |
4213 | */ | |
4214 | int bdrv_media_changed(BlockDriverState *bs) | |
4215 | { | |
4216 | BlockDriver *drv = bs->drv; | |
4217 | ||
4218 | if (drv && drv->bdrv_media_changed) { | |
4219 | return drv->bdrv_media_changed(bs); | |
4220 | } | |
4221 | return -ENOTSUP; | |
4222 | } | |
4223 | ||
4224 | /** | |
4225 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
4226 | */ | |
4227 | void bdrv_eject(BlockDriverState *bs, bool eject_flag) | |
4228 | { | |
4229 | BlockDriver *drv = bs->drv; | |
4230 | ||
4231 | if (drv && drv->bdrv_eject) { | |
4232 | drv->bdrv_eject(bs, eject_flag); | |
4233 | } | |
4234 | ||
4235 | if (bs->device_name[0] != '\0') { | |
4236 | bdrv_emit_qmp_eject_event(bs, eject_flag); | |
4237 | } | |
4238 | } | |
4239 | ||
4240 | /** | |
4241 | * Lock or unlock the media (if it is locked, the user won't be able | |
4242 | * to eject it manually). | |
4243 | */ | |
4244 | void bdrv_lock_medium(BlockDriverState *bs, bool locked) | |
4245 | { | |
4246 | BlockDriver *drv = bs->drv; | |
4247 | ||
4248 | trace_bdrv_lock_medium(bs, locked); | |
4249 | ||
4250 | if (drv && drv->bdrv_lock_medium) { | |
4251 | drv->bdrv_lock_medium(bs, locked); | |
4252 | } | |
4253 | } | |
4254 | ||
4255 | /* needed for generic scsi interface */ | |
4256 | ||
4257 | int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
4258 | { | |
4259 | BlockDriver *drv = bs->drv; | |
4260 | ||
4261 | if (drv && drv->bdrv_ioctl) | |
4262 | return drv->bdrv_ioctl(bs, req, buf); | |
4263 | return -ENOTSUP; | |
4264 | } | |
4265 | ||
4266 | BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs, | |
4267 | unsigned long int req, void *buf, | |
4268 | BlockDriverCompletionFunc *cb, void *opaque) | |
4269 | { | |
4270 | BlockDriver *drv = bs->drv; | |
4271 | ||
4272 | if (drv && drv->bdrv_aio_ioctl) | |
4273 | return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque); | |
4274 | return NULL; | |
4275 | } | |
4276 | ||
4277 | void bdrv_set_buffer_alignment(BlockDriverState *bs, int align) | |
4278 | { | |
4279 | bs->buffer_alignment = align; | |
4280 | } | |
4281 | ||
4282 | void *qemu_blockalign(BlockDriverState *bs, size_t size) | |
4283 | { | |
4284 | return qemu_memalign((bs && bs->buffer_alignment) ? bs->buffer_alignment : 512, size); | |
4285 | } | |
4286 | ||
4287 | void bdrv_set_dirty_tracking(BlockDriverState *bs, int enable) | |
4288 | { | |
4289 | int64_t bitmap_size; | |
4290 | ||
4291 | bs->dirty_count = 0; | |
4292 | if (enable) { | |
4293 | if (!bs->dirty_bitmap) { | |
4294 | bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS) + | |
4295 | BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG - 1; | |
4296 | bitmap_size /= BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG; | |
4297 | ||
4298 | bs->dirty_bitmap = g_new0(unsigned long, bitmap_size); | |
4299 | } | |
4300 | } else { | |
4301 | if (bs->dirty_bitmap) { | |
4302 | g_free(bs->dirty_bitmap); | |
4303 | bs->dirty_bitmap = NULL; | |
4304 | } | |
4305 | } | |
4306 | } | |
4307 | ||
4308 | int bdrv_get_dirty(BlockDriverState *bs, int64_t sector) | |
4309 | { | |
4310 | int64_t chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK; | |
4311 | ||
4312 | if (bs->dirty_bitmap && | |
4313 | (sector << BDRV_SECTOR_BITS) < bdrv_getlength(bs)) { | |
4314 | return !!(bs->dirty_bitmap[chunk / BITS_PER_LONG] & | |
4315 | (1UL << (chunk % BITS_PER_LONG))); | |
4316 | } else { | |
4317 | return 0; | |
4318 | } | |
4319 | } | |
4320 | ||
4321 | int64_t bdrv_get_next_dirty(BlockDriverState *bs, int64_t sector) | |
4322 | { | |
4323 | int64_t chunk; | |
4324 | int bit, elem; | |
4325 | ||
4326 | /* Avoid an infinite loop. */ | |
4327 | assert(bs->dirty_count > 0); | |
4328 | ||
4329 | sector = (sector | (BDRV_SECTORS_PER_DIRTY_CHUNK - 1)) + 1; | |
4330 | chunk = sector / (int64_t)BDRV_SECTORS_PER_DIRTY_CHUNK; | |
4331 | ||
4332 | QEMU_BUILD_BUG_ON(sizeof(bs->dirty_bitmap[0]) * 8 != BITS_PER_LONG); | |
4333 | elem = chunk / BITS_PER_LONG; | |
4334 | bit = chunk % BITS_PER_LONG; | |
4335 | for (;;) { | |
4336 | if (sector >= bs->total_sectors) { | |
4337 | sector = 0; | |
4338 | bit = elem = 0; | |
4339 | } | |
4340 | if (bit == 0 && bs->dirty_bitmap[elem] == 0) { | |
4341 | sector += BDRV_SECTORS_PER_DIRTY_CHUNK * BITS_PER_LONG; | |
4342 | elem++; | |
4343 | } else { | |
4344 | if (bs->dirty_bitmap[elem] & (1UL << bit)) { | |
4345 | return sector; | |
4346 | } | |
4347 | sector += BDRV_SECTORS_PER_DIRTY_CHUNK; | |
4348 | if (++bit == BITS_PER_LONG) { | |
4349 | bit = 0; | |
4350 | elem++; | |
4351 | } | |
4352 | } | |
4353 | } | |
4354 | } | |
4355 | ||
4356 | void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, | |
4357 | int nr_sectors) | |
4358 | { | |
4359 | set_dirty_bitmap(bs, cur_sector, nr_sectors, 1); | |
4360 | } | |
4361 | ||
4362 | void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, | |
4363 | int nr_sectors) | |
4364 | { | |
4365 | set_dirty_bitmap(bs, cur_sector, nr_sectors, 0); | |
4366 | } | |
4367 | ||
4368 | int64_t bdrv_get_dirty_count(BlockDriverState *bs) | |
4369 | { | |
4370 | return bs->dirty_count; | |
4371 | } | |
4372 | ||
4373 | void bdrv_set_in_use(BlockDriverState *bs, int in_use) | |
4374 | { | |
4375 | assert(bs->in_use != in_use); | |
4376 | bs->in_use = in_use; | |
4377 | } | |
4378 | ||
4379 | int bdrv_in_use(BlockDriverState *bs) | |
4380 | { | |
4381 | return bs->in_use; | |
4382 | } | |
4383 | ||
4384 | void bdrv_iostatus_enable(BlockDriverState *bs) | |
4385 | { | |
4386 | bs->iostatus_enabled = true; | |
4387 | bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; | |
4388 | } | |
4389 | ||
4390 | /* The I/O status is only enabled if the drive explicitly | |
4391 | * enables it _and_ the VM is configured to stop on errors */ | |
4392 | bool bdrv_iostatus_is_enabled(const BlockDriverState *bs) | |
4393 | { | |
4394 | return (bs->iostatus_enabled && | |
4395 | (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC || | |
4396 | bs->on_write_error == BLOCKDEV_ON_ERROR_STOP || | |
4397 | bs->on_read_error == BLOCKDEV_ON_ERROR_STOP)); | |
4398 | } | |
4399 | ||
4400 | void bdrv_iostatus_disable(BlockDriverState *bs) | |
4401 | { | |
4402 | bs->iostatus_enabled = false; | |
4403 | } | |
4404 | ||
4405 | void bdrv_iostatus_reset(BlockDriverState *bs) | |
4406 | { | |
4407 | if (bdrv_iostatus_is_enabled(bs)) { | |
4408 | bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; | |
4409 | if (bs->job) { | |
4410 | block_job_iostatus_reset(bs->job); | |
4411 | } | |
4412 | } | |
4413 | } | |
4414 | ||
4415 | void bdrv_iostatus_set_err(BlockDriverState *bs, int error) | |
4416 | { | |
4417 | assert(bdrv_iostatus_is_enabled(bs)); | |
4418 | if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { | |
4419 | bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE : | |
4420 | BLOCK_DEVICE_IO_STATUS_FAILED; | |
4421 | } | |
4422 | } | |
4423 | ||
4424 | void | |
4425 | bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes, | |
4426 | enum BlockAcctType type) | |
4427 | { | |
4428 | assert(type < BDRV_MAX_IOTYPE); | |
4429 | ||
4430 | cookie->bytes = bytes; | |
4431 | cookie->start_time_ns = get_clock(); | |
4432 | cookie->type = type; | |
4433 | } | |
4434 | ||
4435 | void | |
4436 | bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie) | |
4437 | { | |
4438 | assert(cookie->type < BDRV_MAX_IOTYPE); | |
4439 | ||
4440 | bs->nr_bytes[cookie->type] += cookie->bytes; | |
4441 | bs->nr_ops[cookie->type]++; | |
4442 | bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns; | |
4443 | } | |
4444 | ||
4445 | int bdrv_img_create(const char *filename, const char *fmt, | |
4446 | const char *base_filename, const char *base_fmt, | |
4447 | char *options, uint64_t img_size, int flags) | |
4448 | { | |
4449 | QEMUOptionParameter *param = NULL, *create_options = NULL; | |
4450 | QEMUOptionParameter *backing_fmt, *backing_file, *size; | |
4451 | BlockDriverState *bs = NULL; | |
4452 | BlockDriver *drv, *proto_drv; | |
4453 | BlockDriver *backing_drv = NULL; | |
4454 | int ret = 0; | |
4455 | ||
4456 | /* Find driver and parse its options */ | |
4457 | drv = bdrv_find_format(fmt); | |
4458 | if (!drv) { | |
4459 | error_report("Unknown file format '%s'", fmt); | |
4460 | ret = -EINVAL; | |
4461 | goto out; | |
4462 | } | |
4463 | ||
4464 | proto_drv = bdrv_find_protocol(filename); | |
4465 | if (!proto_drv) { | |
4466 | error_report("Unknown protocol '%s'", filename); | |
4467 | ret = -EINVAL; | |
4468 | goto out; | |
4469 | } | |
4470 | ||
4471 | create_options = append_option_parameters(create_options, | |
4472 | drv->create_options); | |
4473 | create_options = append_option_parameters(create_options, | |
4474 | proto_drv->create_options); | |
4475 | ||
4476 | /* Create parameter list with default values */ | |
4477 | param = parse_option_parameters("", create_options, param); | |
4478 | ||
4479 | set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size); | |
4480 | ||
4481 | /* Parse -o options */ | |
4482 | if (options) { | |
4483 | param = parse_option_parameters(options, create_options, param); | |
4484 | if (param == NULL) { | |
4485 | error_report("Invalid options for file format '%s'.", fmt); | |
4486 | ret = -EINVAL; | |
4487 | goto out; | |
4488 | } | |
4489 | } | |
4490 | ||
4491 | if (base_filename) { | |
4492 | if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE, | |
4493 | base_filename)) { | |
4494 | error_report("Backing file not supported for file format '%s'", | |
4495 | fmt); | |
4496 | ret = -EINVAL; | |
4497 | goto out; | |
4498 | } | |
4499 | } | |
4500 | ||
4501 | if (base_fmt) { | |
4502 | if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) { | |
4503 | error_report("Backing file format not supported for file " | |
4504 | "format '%s'", fmt); | |
4505 | ret = -EINVAL; | |
4506 | goto out; | |
4507 | } | |
4508 | } | |
4509 | ||
4510 | backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE); | |
4511 | if (backing_file && backing_file->value.s) { | |
4512 | if (!strcmp(filename, backing_file->value.s)) { | |
4513 | error_report("Error: Trying to create an image with the " | |
4514 | "same filename as the backing file"); | |
4515 | ret = -EINVAL; | |
4516 | goto out; | |
4517 | } | |
4518 | } | |
4519 | ||
4520 | backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT); | |
4521 | if (backing_fmt && backing_fmt->value.s) { | |
4522 | backing_drv = bdrv_find_format(backing_fmt->value.s); | |
4523 | if (!backing_drv) { | |
4524 | error_report("Unknown backing file format '%s'", | |
4525 | backing_fmt->value.s); | |
4526 | ret = -EINVAL; | |
4527 | goto out; | |
4528 | } | |
4529 | } | |
4530 | ||
4531 | // The size for the image must always be specified, with one exception: | |
4532 | // If we are using a backing file, we can obtain the size from there | |
4533 | size = get_option_parameter(param, BLOCK_OPT_SIZE); | |
4534 | if (size && size->value.n == -1) { | |
4535 | if (backing_file && backing_file->value.s) { | |
4536 | uint64_t size; | |
4537 | char buf[32]; | |
4538 | int back_flags; | |
4539 | ||
4540 | /* backing files always opened read-only */ | |
4541 | back_flags = | |
4542 | flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
4543 | ||
4544 | bs = bdrv_new(""); | |
4545 | ||
4546 | ret = bdrv_open(bs, backing_file->value.s, back_flags, backing_drv); | |
4547 | if (ret < 0) { | |
4548 | error_report("Could not open '%s'", backing_file->value.s); | |
4549 | goto out; | |
4550 | } | |
4551 | bdrv_get_geometry(bs, &size); | |
4552 | size *= 512; | |
4553 | ||
4554 | snprintf(buf, sizeof(buf), "%" PRId64, size); | |
4555 | set_option_parameter(param, BLOCK_OPT_SIZE, buf); | |
4556 | } else { | |
4557 | error_report("Image creation needs a size parameter"); | |
4558 | ret = -EINVAL; | |
4559 | goto out; | |
4560 | } | |
4561 | } | |
4562 | ||
4563 | printf("Formatting '%s', fmt=%s ", filename, fmt); | |
4564 | print_option_parameters(param); | |
4565 | puts(""); | |
4566 | ||
4567 | ret = bdrv_create(drv, filename, param); | |
4568 | ||
4569 | if (ret < 0) { | |
4570 | if (ret == -ENOTSUP) { | |
4571 | error_report("Formatting or formatting option not supported for " | |
4572 | "file format '%s'", fmt); | |
4573 | } else if (ret == -EFBIG) { | |
4574 | error_report("The image size is too large for file format '%s'", | |
4575 | fmt); | |
4576 | } else { | |
4577 | error_report("%s: error while creating %s: %s", filename, fmt, | |
4578 | strerror(-ret)); | |
4579 | } | |
4580 | } | |
4581 | ||
4582 | out: | |
4583 | free_option_parameters(create_options); | |
4584 | free_option_parameters(param); | |
4585 | ||
4586 | if (bs) { | |
4587 | bdrv_delete(bs); | |
4588 | } | |
4589 | ||
4590 | return ret; | |
4591 | } |