]>
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/monitor.h" | |
28 | #include "block/block_int.h" | |
29 | #include "block/blockjob.h" | |
30 | #include "qemu/module.h" | |
31 | #include "qapi/qmp/qjson.h" | |
32 | #include "sysemu/sysemu.h" | |
33 | #include "qemu/notify.h" | |
34 | #include "block/coroutine.h" | |
35 | #include "block/qapi.h" | |
36 | #include "qmp-commands.h" | |
37 | #include "qemu/timer.h" | |
38 | ||
39 | #ifdef CONFIG_BSD | |
40 | #include <sys/types.h> | |
41 | #include <sys/stat.h> | |
42 | #include <sys/ioctl.h> | |
43 | #include <sys/queue.h> | |
44 | #ifndef __DragonFly__ | |
45 | #include <sys/disk.h> | |
46 | #endif | |
47 | #endif | |
48 | ||
49 | #ifdef _WIN32 | |
50 | #include <windows.h> | |
51 | #endif | |
52 | ||
53 | struct BdrvDirtyBitmap { | |
54 | HBitmap *bitmap; | |
55 | QLIST_ENTRY(BdrvDirtyBitmap) list; | |
56 | }; | |
57 | ||
58 | #define NOT_DONE 0x7fffffff /* used while emulated sync operation in progress */ | |
59 | ||
60 | static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load); | |
61 | static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs, | |
62 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
63 | BlockDriverCompletionFunc *cb, void *opaque); | |
64 | static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs, | |
65 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
66 | BlockDriverCompletionFunc *cb, void *opaque); | |
67 | static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, | |
68 | int64_t sector_num, int nb_sectors, | |
69 | QEMUIOVector *iov); | |
70 | static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, | |
71 | int64_t sector_num, int nb_sectors, | |
72 | QEMUIOVector *iov); | |
73 | static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, | |
74 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
75 | BdrvRequestFlags flags); | |
76 | static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, | |
77 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
78 | BdrvRequestFlags flags); | |
79 | static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, | |
80 | int64_t sector_num, | |
81 | QEMUIOVector *qiov, | |
82 | int nb_sectors, | |
83 | BdrvRequestFlags flags, | |
84 | BlockDriverCompletionFunc *cb, | |
85 | void *opaque, | |
86 | bool is_write); | |
87 | static void coroutine_fn bdrv_co_do_rw(void *opaque); | |
88 | static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, | |
89 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags); | |
90 | ||
91 | static QTAILQ_HEAD(, BlockDriverState) bdrv_states = | |
92 | QTAILQ_HEAD_INITIALIZER(bdrv_states); | |
93 | ||
94 | static QTAILQ_HEAD(, BlockDriverState) graph_bdrv_states = | |
95 | QTAILQ_HEAD_INITIALIZER(graph_bdrv_states); | |
96 | ||
97 | static QLIST_HEAD(, BlockDriver) bdrv_drivers = | |
98 | QLIST_HEAD_INITIALIZER(bdrv_drivers); | |
99 | ||
100 | /* If non-zero, use only whitelisted block drivers */ | |
101 | static int use_bdrv_whitelist; | |
102 | ||
103 | #ifdef _WIN32 | |
104 | static int is_windows_drive_prefix(const char *filename) | |
105 | { | |
106 | return (((filename[0] >= 'a' && filename[0] <= 'z') || | |
107 | (filename[0] >= 'A' && filename[0] <= 'Z')) && | |
108 | filename[1] == ':'); | |
109 | } | |
110 | ||
111 | int is_windows_drive(const char *filename) | |
112 | { | |
113 | if (is_windows_drive_prefix(filename) && | |
114 | filename[2] == '\0') | |
115 | return 1; | |
116 | if (strstart(filename, "\\\\.\\", NULL) || | |
117 | strstart(filename, "//./", NULL)) | |
118 | return 1; | |
119 | return 0; | |
120 | } | |
121 | #endif | |
122 | ||
123 | /* throttling disk I/O limits */ | |
124 | void bdrv_set_io_limits(BlockDriverState *bs, | |
125 | ThrottleConfig *cfg) | |
126 | { | |
127 | int i; | |
128 | ||
129 | throttle_config(&bs->throttle_state, cfg); | |
130 | ||
131 | for (i = 0; i < 2; i++) { | |
132 | qemu_co_enter_next(&bs->throttled_reqs[i]); | |
133 | } | |
134 | } | |
135 | ||
136 | /* this function drain all the throttled IOs */ | |
137 | static bool bdrv_start_throttled_reqs(BlockDriverState *bs) | |
138 | { | |
139 | bool drained = false; | |
140 | bool enabled = bs->io_limits_enabled; | |
141 | int i; | |
142 | ||
143 | bs->io_limits_enabled = false; | |
144 | ||
145 | for (i = 0; i < 2; i++) { | |
146 | while (qemu_co_enter_next(&bs->throttled_reqs[i])) { | |
147 | drained = true; | |
148 | } | |
149 | } | |
150 | ||
151 | bs->io_limits_enabled = enabled; | |
152 | ||
153 | return drained; | |
154 | } | |
155 | ||
156 | void bdrv_io_limits_disable(BlockDriverState *bs) | |
157 | { | |
158 | bs->io_limits_enabled = false; | |
159 | ||
160 | bdrv_start_throttled_reqs(bs); | |
161 | ||
162 | throttle_destroy(&bs->throttle_state); | |
163 | } | |
164 | ||
165 | static void bdrv_throttle_read_timer_cb(void *opaque) | |
166 | { | |
167 | BlockDriverState *bs = opaque; | |
168 | qemu_co_enter_next(&bs->throttled_reqs[0]); | |
169 | } | |
170 | ||
171 | static void bdrv_throttle_write_timer_cb(void *opaque) | |
172 | { | |
173 | BlockDriverState *bs = opaque; | |
174 | qemu_co_enter_next(&bs->throttled_reqs[1]); | |
175 | } | |
176 | ||
177 | /* should be called before bdrv_set_io_limits if a limit is set */ | |
178 | void bdrv_io_limits_enable(BlockDriverState *bs) | |
179 | { | |
180 | assert(!bs->io_limits_enabled); | |
181 | throttle_init(&bs->throttle_state, | |
182 | QEMU_CLOCK_VIRTUAL, | |
183 | bdrv_throttle_read_timer_cb, | |
184 | bdrv_throttle_write_timer_cb, | |
185 | bs); | |
186 | bs->io_limits_enabled = true; | |
187 | } | |
188 | ||
189 | /* This function makes an IO wait if needed | |
190 | * | |
191 | * @nb_sectors: the number of sectors of the IO | |
192 | * @is_write: is the IO a write | |
193 | */ | |
194 | static void bdrv_io_limits_intercept(BlockDriverState *bs, | |
195 | int nb_sectors, | |
196 | bool is_write) | |
197 | { | |
198 | /* does this io must wait */ | |
199 | bool must_wait = throttle_schedule_timer(&bs->throttle_state, is_write); | |
200 | ||
201 | /* if must wait or any request of this type throttled queue the IO */ | |
202 | if (must_wait || | |
203 | !qemu_co_queue_empty(&bs->throttled_reqs[is_write])) { | |
204 | qemu_co_queue_wait(&bs->throttled_reqs[is_write]); | |
205 | } | |
206 | ||
207 | /* the IO will be executed, do the accounting */ | |
208 | throttle_account(&bs->throttle_state, | |
209 | is_write, | |
210 | nb_sectors * BDRV_SECTOR_SIZE); | |
211 | ||
212 | /* if the next request must wait -> do nothing */ | |
213 | if (throttle_schedule_timer(&bs->throttle_state, is_write)) { | |
214 | return; | |
215 | } | |
216 | ||
217 | /* else queue next request for execution */ | |
218 | qemu_co_queue_next(&bs->throttled_reqs[is_write]); | |
219 | } | |
220 | ||
221 | size_t bdrv_opt_mem_align(BlockDriverState *bs) | |
222 | { | |
223 | if (!bs || !bs->drv) { | |
224 | /* 4k should be on the safe side */ | |
225 | return 4096; | |
226 | } | |
227 | ||
228 | return bs->bl.opt_mem_alignment; | |
229 | } | |
230 | ||
231 | /* check if the path starts with "<protocol>:" */ | |
232 | static int path_has_protocol(const char *path) | |
233 | { | |
234 | const char *p; | |
235 | ||
236 | #ifdef _WIN32 | |
237 | if (is_windows_drive(path) || | |
238 | is_windows_drive_prefix(path)) { | |
239 | return 0; | |
240 | } | |
241 | p = path + strcspn(path, ":/\\"); | |
242 | #else | |
243 | p = path + strcspn(path, ":/"); | |
244 | #endif | |
245 | ||
246 | return *p == ':'; | |
247 | } | |
248 | ||
249 | int path_is_absolute(const char *path) | |
250 | { | |
251 | #ifdef _WIN32 | |
252 | /* specific case for names like: "\\.\d:" */ | |
253 | if (is_windows_drive(path) || is_windows_drive_prefix(path)) { | |
254 | return 1; | |
255 | } | |
256 | return (*path == '/' || *path == '\\'); | |
257 | #else | |
258 | return (*path == '/'); | |
259 | #endif | |
260 | } | |
261 | ||
262 | /* if filename is absolute, just copy it to dest. Otherwise, build a | |
263 | path to it by considering it is relative to base_path. URL are | |
264 | supported. */ | |
265 | void path_combine(char *dest, int dest_size, | |
266 | const char *base_path, | |
267 | const char *filename) | |
268 | { | |
269 | const char *p, *p1; | |
270 | int len; | |
271 | ||
272 | if (dest_size <= 0) | |
273 | return; | |
274 | if (path_is_absolute(filename)) { | |
275 | pstrcpy(dest, dest_size, filename); | |
276 | } else { | |
277 | p = strchr(base_path, ':'); | |
278 | if (p) | |
279 | p++; | |
280 | else | |
281 | p = base_path; | |
282 | p1 = strrchr(base_path, '/'); | |
283 | #ifdef _WIN32 | |
284 | { | |
285 | const char *p2; | |
286 | p2 = strrchr(base_path, '\\'); | |
287 | if (!p1 || p2 > p1) | |
288 | p1 = p2; | |
289 | } | |
290 | #endif | |
291 | if (p1) | |
292 | p1++; | |
293 | else | |
294 | p1 = base_path; | |
295 | if (p1 > p) | |
296 | p = p1; | |
297 | len = p - base_path; | |
298 | if (len > dest_size - 1) | |
299 | len = dest_size - 1; | |
300 | memcpy(dest, base_path, len); | |
301 | dest[len] = '\0'; | |
302 | pstrcat(dest, dest_size, filename); | |
303 | } | |
304 | } | |
305 | ||
306 | void bdrv_get_full_backing_filename(BlockDriverState *bs, char *dest, size_t sz) | |
307 | { | |
308 | if (bs->backing_file[0] == '\0' || path_has_protocol(bs->backing_file)) { | |
309 | pstrcpy(dest, sz, bs->backing_file); | |
310 | } else { | |
311 | path_combine(dest, sz, bs->filename, bs->backing_file); | |
312 | } | |
313 | } | |
314 | ||
315 | void bdrv_register(BlockDriver *bdrv) | |
316 | { | |
317 | /* Block drivers without coroutine functions need emulation */ | |
318 | if (!bdrv->bdrv_co_readv) { | |
319 | bdrv->bdrv_co_readv = bdrv_co_readv_em; | |
320 | bdrv->bdrv_co_writev = bdrv_co_writev_em; | |
321 | ||
322 | /* bdrv_co_readv_em()/brdv_co_writev_em() work in terms of aio, so if | |
323 | * the block driver lacks aio we need to emulate that too. | |
324 | */ | |
325 | if (!bdrv->bdrv_aio_readv) { | |
326 | /* add AIO emulation layer */ | |
327 | bdrv->bdrv_aio_readv = bdrv_aio_readv_em; | |
328 | bdrv->bdrv_aio_writev = bdrv_aio_writev_em; | |
329 | } | |
330 | } | |
331 | ||
332 | QLIST_INSERT_HEAD(&bdrv_drivers, bdrv, list); | |
333 | } | |
334 | ||
335 | /* create a new block device (by default it is empty) */ | |
336 | BlockDriverState *bdrv_new(const char *device_name) | |
337 | { | |
338 | BlockDriverState *bs; | |
339 | ||
340 | bs = g_malloc0(sizeof(BlockDriverState)); | |
341 | QLIST_INIT(&bs->dirty_bitmaps); | |
342 | pstrcpy(bs->device_name, sizeof(bs->device_name), device_name); | |
343 | if (device_name[0] != '\0') { | |
344 | QTAILQ_INSERT_TAIL(&bdrv_states, bs, device_list); | |
345 | } | |
346 | bdrv_iostatus_disable(bs); | |
347 | notifier_list_init(&bs->close_notifiers); | |
348 | notifier_with_return_list_init(&bs->before_write_notifiers); | |
349 | qemu_co_queue_init(&bs->throttled_reqs[0]); | |
350 | qemu_co_queue_init(&bs->throttled_reqs[1]); | |
351 | bs->refcnt = 1; | |
352 | ||
353 | return bs; | |
354 | } | |
355 | ||
356 | void bdrv_add_close_notifier(BlockDriverState *bs, Notifier *notify) | |
357 | { | |
358 | notifier_list_add(&bs->close_notifiers, notify); | |
359 | } | |
360 | ||
361 | BlockDriver *bdrv_find_format(const char *format_name) | |
362 | { | |
363 | BlockDriver *drv1; | |
364 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
365 | if (!strcmp(drv1->format_name, format_name)) { | |
366 | return drv1; | |
367 | } | |
368 | } | |
369 | return NULL; | |
370 | } | |
371 | ||
372 | static int bdrv_is_whitelisted(BlockDriver *drv, bool read_only) | |
373 | { | |
374 | static const char *whitelist_rw[] = { | |
375 | CONFIG_BDRV_RW_WHITELIST | |
376 | }; | |
377 | static const char *whitelist_ro[] = { | |
378 | CONFIG_BDRV_RO_WHITELIST | |
379 | }; | |
380 | const char **p; | |
381 | ||
382 | if (!whitelist_rw[0] && !whitelist_ro[0]) { | |
383 | return 1; /* no whitelist, anything goes */ | |
384 | } | |
385 | ||
386 | for (p = whitelist_rw; *p; p++) { | |
387 | if (!strcmp(drv->format_name, *p)) { | |
388 | return 1; | |
389 | } | |
390 | } | |
391 | if (read_only) { | |
392 | for (p = whitelist_ro; *p; p++) { | |
393 | if (!strcmp(drv->format_name, *p)) { | |
394 | return 1; | |
395 | } | |
396 | } | |
397 | } | |
398 | return 0; | |
399 | } | |
400 | ||
401 | BlockDriver *bdrv_find_whitelisted_format(const char *format_name, | |
402 | bool read_only) | |
403 | { | |
404 | BlockDriver *drv = bdrv_find_format(format_name); | |
405 | return drv && bdrv_is_whitelisted(drv, read_only) ? drv : NULL; | |
406 | } | |
407 | ||
408 | typedef struct CreateCo { | |
409 | BlockDriver *drv; | |
410 | char *filename; | |
411 | QEMUOptionParameter *options; | |
412 | int ret; | |
413 | Error *err; | |
414 | } CreateCo; | |
415 | ||
416 | static void coroutine_fn bdrv_create_co_entry(void *opaque) | |
417 | { | |
418 | Error *local_err = NULL; | |
419 | int ret; | |
420 | ||
421 | CreateCo *cco = opaque; | |
422 | assert(cco->drv); | |
423 | ||
424 | ret = cco->drv->bdrv_create(cco->filename, cco->options, &local_err); | |
425 | if (error_is_set(&local_err)) { | |
426 | error_propagate(&cco->err, local_err); | |
427 | } | |
428 | cco->ret = ret; | |
429 | } | |
430 | ||
431 | int bdrv_create(BlockDriver *drv, const char* filename, | |
432 | QEMUOptionParameter *options, Error **errp) | |
433 | { | |
434 | int ret; | |
435 | ||
436 | Coroutine *co; | |
437 | CreateCo cco = { | |
438 | .drv = drv, | |
439 | .filename = g_strdup(filename), | |
440 | .options = options, | |
441 | .ret = NOT_DONE, | |
442 | .err = NULL, | |
443 | }; | |
444 | ||
445 | if (!drv->bdrv_create) { | |
446 | error_setg(errp, "Driver '%s' does not support image creation", drv->format_name); | |
447 | ret = -ENOTSUP; | |
448 | goto out; | |
449 | } | |
450 | ||
451 | if (qemu_in_coroutine()) { | |
452 | /* Fast-path if already in coroutine context */ | |
453 | bdrv_create_co_entry(&cco); | |
454 | } else { | |
455 | co = qemu_coroutine_create(bdrv_create_co_entry); | |
456 | qemu_coroutine_enter(co, &cco); | |
457 | while (cco.ret == NOT_DONE) { | |
458 | qemu_aio_wait(); | |
459 | } | |
460 | } | |
461 | ||
462 | ret = cco.ret; | |
463 | if (ret < 0) { | |
464 | if (error_is_set(&cco.err)) { | |
465 | error_propagate(errp, cco.err); | |
466 | } else { | |
467 | error_setg_errno(errp, -ret, "Could not create image"); | |
468 | } | |
469 | } | |
470 | ||
471 | out: | |
472 | g_free(cco.filename); | |
473 | return ret; | |
474 | } | |
475 | ||
476 | int bdrv_create_file(const char* filename, QEMUOptionParameter *options, | |
477 | Error **errp) | |
478 | { | |
479 | BlockDriver *drv; | |
480 | Error *local_err = NULL; | |
481 | int ret; | |
482 | ||
483 | drv = bdrv_find_protocol(filename, true); | |
484 | if (drv == NULL) { | |
485 | error_setg(errp, "Could not find protocol for file '%s'", filename); | |
486 | return -ENOENT; | |
487 | } | |
488 | ||
489 | ret = bdrv_create(drv, filename, options, &local_err); | |
490 | if (error_is_set(&local_err)) { | |
491 | error_propagate(errp, local_err); | |
492 | } | |
493 | return ret; | |
494 | } | |
495 | ||
496 | int bdrv_refresh_limits(BlockDriverState *bs) | |
497 | { | |
498 | BlockDriver *drv = bs->drv; | |
499 | ||
500 | memset(&bs->bl, 0, sizeof(bs->bl)); | |
501 | ||
502 | if (!drv) { | |
503 | return 0; | |
504 | } | |
505 | ||
506 | /* Take some limits from the children as a default */ | |
507 | if (bs->file) { | |
508 | bdrv_refresh_limits(bs->file); | |
509 | bs->bl.opt_transfer_length = bs->file->bl.opt_transfer_length; | |
510 | bs->bl.opt_mem_alignment = bs->file->bl.opt_mem_alignment; | |
511 | } else { | |
512 | bs->bl.opt_mem_alignment = 512; | |
513 | } | |
514 | ||
515 | if (bs->backing_hd) { | |
516 | bdrv_refresh_limits(bs->backing_hd); | |
517 | bs->bl.opt_transfer_length = | |
518 | MAX(bs->bl.opt_transfer_length, | |
519 | bs->backing_hd->bl.opt_transfer_length); | |
520 | bs->bl.opt_mem_alignment = | |
521 | MAX(bs->bl.opt_mem_alignment, | |
522 | bs->backing_hd->bl.opt_mem_alignment); | |
523 | } | |
524 | ||
525 | /* Then let the driver override it */ | |
526 | if (drv->bdrv_refresh_limits) { | |
527 | return drv->bdrv_refresh_limits(bs); | |
528 | } | |
529 | ||
530 | return 0; | |
531 | } | |
532 | ||
533 | /* | |
534 | * Create a uniquely-named empty temporary file. | |
535 | * Return 0 upon success, otherwise a negative errno value. | |
536 | */ | |
537 | int get_tmp_filename(char *filename, int size) | |
538 | { | |
539 | #ifdef _WIN32 | |
540 | char temp_dir[MAX_PATH]; | |
541 | /* GetTempFileName requires that its output buffer (4th param) | |
542 | have length MAX_PATH or greater. */ | |
543 | assert(size >= MAX_PATH); | |
544 | return (GetTempPath(MAX_PATH, temp_dir) | |
545 | && GetTempFileName(temp_dir, "qem", 0, filename) | |
546 | ? 0 : -GetLastError()); | |
547 | #else | |
548 | int fd; | |
549 | const char *tmpdir; | |
550 | tmpdir = getenv("TMPDIR"); | |
551 | if (!tmpdir) | |
552 | tmpdir = "/tmp"; | |
553 | if (snprintf(filename, size, "%s/vl.XXXXXX", tmpdir) >= size) { | |
554 | return -EOVERFLOW; | |
555 | } | |
556 | fd = mkstemp(filename); | |
557 | if (fd < 0) { | |
558 | return -errno; | |
559 | } | |
560 | if (close(fd) != 0) { | |
561 | unlink(filename); | |
562 | return -errno; | |
563 | } | |
564 | return 0; | |
565 | #endif | |
566 | } | |
567 | ||
568 | /* | |
569 | * Detect host devices. By convention, /dev/cdrom[N] is always | |
570 | * recognized as a host CDROM. | |
571 | */ | |
572 | static BlockDriver *find_hdev_driver(const char *filename) | |
573 | { | |
574 | int score_max = 0, score; | |
575 | BlockDriver *drv = NULL, *d; | |
576 | ||
577 | QLIST_FOREACH(d, &bdrv_drivers, list) { | |
578 | if (d->bdrv_probe_device) { | |
579 | score = d->bdrv_probe_device(filename); | |
580 | if (score > score_max) { | |
581 | score_max = score; | |
582 | drv = d; | |
583 | } | |
584 | } | |
585 | } | |
586 | ||
587 | return drv; | |
588 | } | |
589 | ||
590 | BlockDriver *bdrv_find_protocol(const char *filename, | |
591 | bool allow_protocol_prefix) | |
592 | { | |
593 | BlockDriver *drv1; | |
594 | char protocol[128]; | |
595 | int len; | |
596 | const char *p; | |
597 | ||
598 | /* TODO Drivers without bdrv_file_open must be specified explicitly */ | |
599 | ||
600 | /* | |
601 | * XXX(hch): we really should not let host device detection | |
602 | * override an explicit protocol specification, but moving this | |
603 | * later breaks access to device names with colons in them. | |
604 | * Thanks to the brain-dead persistent naming schemes on udev- | |
605 | * based Linux systems those actually are quite common. | |
606 | */ | |
607 | drv1 = find_hdev_driver(filename); | |
608 | if (drv1) { | |
609 | return drv1; | |
610 | } | |
611 | ||
612 | if (!path_has_protocol(filename) || !allow_protocol_prefix) { | |
613 | return bdrv_find_format("file"); | |
614 | } | |
615 | ||
616 | p = strchr(filename, ':'); | |
617 | assert(p != NULL); | |
618 | len = p - filename; | |
619 | if (len > sizeof(protocol) - 1) | |
620 | len = sizeof(protocol) - 1; | |
621 | memcpy(protocol, filename, len); | |
622 | protocol[len] = '\0'; | |
623 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
624 | if (drv1->protocol_name && | |
625 | !strcmp(drv1->protocol_name, protocol)) { | |
626 | return drv1; | |
627 | } | |
628 | } | |
629 | return NULL; | |
630 | } | |
631 | ||
632 | static int find_image_format(BlockDriverState *bs, const char *filename, | |
633 | BlockDriver **pdrv, Error **errp) | |
634 | { | |
635 | int score, score_max; | |
636 | BlockDriver *drv1, *drv; | |
637 | uint8_t buf[2048]; | |
638 | int ret = 0; | |
639 | ||
640 | /* Return the raw BlockDriver * to scsi-generic devices or empty drives */ | |
641 | if (bs->sg || !bdrv_is_inserted(bs) || bdrv_getlength(bs) == 0) { | |
642 | drv = bdrv_find_format("raw"); | |
643 | if (!drv) { | |
644 | error_setg(errp, "Could not find raw image format"); | |
645 | ret = -ENOENT; | |
646 | } | |
647 | *pdrv = drv; | |
648 | return ret; | |
649 | } | |
650 | ||
651 | ret = bdrv_pread(bs, 0, buf, sizeof(buf)); | |
652 | if (ret < 0) { | |
653 | error_setg_errno(errp, -ret, "Could not read image for determining its " | |
654 | "format"); | |
655 | *pdrv = NULL; | |
656 | return ret; | |
657 | } | |
658 | ||
659 | score_max = 0; | |
660 | drv = NULL; | |
661 | QLIST_FOREACH(drv1, &bdrv_drivers, list) { | |
662 | if (drv1->bdrv_probe) { | |
663 | score = drv1->bdrv_probe(buf, ret, filename); | |
664 | if (score > score_max) { | |
665 | score_max = score; | |
666 | drv = drv1; | |
667 | } | |
668 | } | |
669 | } | |
670 | if (!drv) { | |
671 | error_setg(errp, "Could not determine image format: No compatible " | |
672 | "driver found"); | |
673 | ret = -ENOENT; | |
674 | } | |
675 | *pdrv = drv; | |
676 | return ret; | |
677 | } | |
678 | ||
679 | /** | |
680 | * Set the current 'total_sectors' value | |
681 | */ | |
682 | static int refresh_total_sectors(BlockDriverState *bs, int64_t hint) | |
683 | { | |
684 | BlockDriver *drv = bs->drv; | |
685 | ||
686 | /* Do not attempt drv->bdrv_getlength() on scsi-generic devices */ | |
687 | if (bs->sg) | |
688 | return 0; | |
689 | ||
690 | /* query actual device if possible, otherwise just trust the hint */ | |
691 | if (drv->bdrv_getlength) { | |
692 | int64_t length = drv->bdrv_getlength(bs); | |
693 | if (length < 0) { | |
694 | return length; | |
695 | } | |
696 | hint = DIV_ROUND_UP(length, BDRV_SECTOR_SIZE); | |
697 | } | |
698 | ||
699 | bs->total_sectors = hint; | |
700 | return 0; | |
701 | } | |
702 | ||
703 | /** | |
704 | * Set open flags for a given discard mode | |
705 | * | |
706 | * Return 0 on success, -1 if the discard mode was invalid. | |
707 | */ | |
708 | int bdrv_parse_discard_flags(const char *mode, int *flags) | |
709 | { | |
710 | *flags &= ~BDRV_O_UNMAP; | |
711 | ||
712 | if (!strcmp(mode, "off") || !strcmp(mode, "ignore")) { | |
713 | /* do nothing */ | |
714 | } else if (!strcmp(mode, "on") || !strcmp(mode, "unmap")) { | |
715 | *flags |= BDRV_O_UNMAP; | |
716 | } else { | |
717 | return -1; | |
718 | } | |
719 | ||
720 | return 0; | |
721 | } | |
722 | ||
723 | /** | |
724 | * Set open flags for a given cache mode | |
725 | * | |
726 | * Return 0 on success, -1 if the cache mode was invalid. | |
727 | */ | |
728 | int bdrv_parse_cache_flags(const char *mode, int *flags) | |
729 | { | |
730 | *flags &= ~BDRV_O_CACHE_MASK; | |
731 | ||
732 | if (!strcmp(mode, "off") || !strcmp(mode, "none")) { | |
733 | *flags |= BDRV_O_NOCACHE | BDRV_O_CACHE_WB; | |
734 | } else if (!strcmp(mode, "directsync")) { | |
735 | *flags |= BDRV_O_NOCACHE; | |
736 | } else if (!strcmp(mode, "writeback")) { | |
737 | *flags |= BDRV_O_CACHE_WB; | |
738 | } else if (!strcmp(mode, "unsafe")) { | |
739 | *flags |= BDRV_O_CACHE_WB; | |
740 | *flags |= BDRV_O_NO_FLUSH; | |
741 | } else if (!strcmp(mode, "writethrough")) { | |
742 | /* this is the default */ | |
743 | } else { | |
744 | return -1; | |
745 | } | |
746 | ||
747 | return 0; | |
748 | } | |
749 | ||
750 | /** | |
751 | * The copy-on-read flag is actually a reference count so multiple users may | |
752 | * use the feature without worrying about clobbering its previous state. | |
753 | * Copy-on-read stays enabled until all users have called to disable it. | |
754 | */ | |
755 | void bdrv_enable_copy_on_read(BlockDriverState *bs) | |
756 | { | |
757 | bs->copy_on_read++; | |
758 | } | |
759 | ||
760 | void bdrv_disable_copy_on_read(BlockDriverState *bs) | |
761 | { | |
762 | assert(bs->copy_on_read > 0); | |
763 | bs->copy_on_read--; | |
764 | } | |
765 | ||
766 | static int bdrv_open_flags(BlockDriverState *bs, int flags) | |
767 | { | |
768 | int open_flags = flags | BDRV_O_CACHE_WB; | |
769 | ||
770 | /* | |
771 | * Clear flags that are internal to the block layer before opening the | |
772 | * image. | |
773 | */ | |
774 | open_flags &= ~(BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
775 | ||
776 | /* | |
777 | * Snapshots should be writable. | |
778 | */ | |
779 | if (bs->is_temporary) { | |
780 | open_flags |= BDRV_O_RDWR; | |
781 | } | |
782 | ||
783 | return open_flags; | |
784 | } | |
785 | ||
786 | static int bdrv_assign_node_name(BlockDriverState *bs, | |
787 | const char *node_name, | |
788 | Error **errp) | |
789 | { | |
790 | if (!node_name) { | |
791 | return 0; | |
792 | } | |
793 | ||
794 | /* empty string node name is invalid */ | |
795 | if (node_name[0] == '\0') { | |
796 | error_setg(errp, "Empty node name"); | |
797 | return -EINVAL; | |
798 | } | |
799 | ||
800 | /* takes care of avoiding duplicates node names */ | |
801 | if (bdrv_find_node(node_name)) { | |
802 | error_setg(errp, "Duplicate node name"); | |
803 | return -EINVAL; | |
804 | } | |
805 | ||
806 | /* copy node name into the bs and insert it into the graph list */ | |
807 | pstrcpy(bs->node_name, sizeof(bs->node_name), node_name); | |
808 | QTAILQ_INSERT_TAIL(&graph_bdrv_states, bs, node_list); | |
809 | ||
810 | return 0; | |
811 | } | |
812 | ||
813 | /* | |
814 | * Common part for opening disk images and files | |
815 | * | |
816 | * Removes all processed options from *options. | |
817 | */ | |
818 | static int bdrv_open_common(BlockDriverState *bs, BlockDriverState *file, | |
819 | QDict *options, int flags, BlockDriver *drv, Error **errp) | |
820 | { | |
821 | int ret, open_flags; | |
822 | const char *filename; | |
823 | const char *node_name = NULL; | |
824 | Error *local_err = NULL; | |
825 | ||
826 | assert(drv != NULL); | |
827 | assert(bs->file == NULL); | |
828 | assert(options != NULL && bs->options != options); | |
829 | ||
830 | if (file != NULL) { | |
831 | filename = file->filename; | |
832 | } else { | |
833 | filename = qdict_get_try_str(options, "filename"); | |
834 | } | |
835 | ||
836 | trace_bdrv_open_common(bs, filename ?: "", flags, drv->format_name); | |
837 | ||
838 | node_name = qdict_get_try_str(options, "node-name"); | |
839 | ret = bdrv_assign_node_name(bs, node_name, errp); | |
840 | if (ret < 0) { | |
841 | return ret; | |
842 | } | |
843 | qdict_del(options, "node-name"); | |
844 | ||
845 | /* bdrv_open() with directly using a protocol as drv. This layer is already | |
846 | * opened, so assign it to bs (while file becomes a closed BlockDriverState) | |
847 | * and return immediately. */ | |
848 | if (file != NULL && drv->bdrv_file_open) { | |
849 | bdrv_swap(file, bs); | |
850 | return 0; | |
851 | } | |
852 | ||
853 | bs->open_flags = flags; | |
854 | bs->guest_block_size = 512; | |
855 | bs->request_alignment = 512; | |
856 | bs->zero_beyond_eof = true; | |
857 | open_flags = bdrv_open_flags(bs, flags); | |
858 | bs->read_only = !(open_flags & BDRV_O_RDWR); | |
859 | ||
860 | if (use_bdrv_whitelist && !bdrv_is_whitelisted(drv, bs->read_only)) { | |
861 | error_setg(errp, | |
862 | !bs->read_only && bdrv_is_whitelisted(drv, true) | |
863 | ? "Driver '%s' can only be used for read-only devices" | |
864 | : "Driver '%s' is not whitelisted", | |
865 | drv->format_name); | |
866 | return -ENOTSUP; | |
867 | } | |
868 | ||
869 | assert(bs->copy_on_read == 0); /* bdrv_new() and bdrv_close() make it so */ | |
870 | if (flags & BDRV_O_COPY_ON_READ) { | |
871 | if (!bs->read_only) { | |
872 | bdrv_enable_copy_on_read(bs); | |
873 | } else { | |
874 | error_setg(errp, "Can't use copy-on-read on read-only device"); | |
875 | return -EINVAL; | |
876 | } | |
877 | } | |
878 | ||
879 | if (filename != NULL) { | |
880 | pstrcpy(bs->filename, sizeof(bs->filename), filename); | |
881 | } else { | |
882 | bs->filename[0] = '\0'; | |
883 | } | |
884 | ||
885 | bs->drv = drv; | |
886 | bs->opaque = g_malloc0(drv->instance_size); | |
887 | ||
888 | bs->enable_write_cache = !!(flags & BDRV_O_CACHE_WB); | |
889 | ||
890 | /* Open the image, either directly or using a protocol */ | |
891 | if (drv->bdrv_file_open) { | |
892 | assert(file == NULL); | |
893 | assert(!drv->bdrv_needs_filename || filename != NULL); | |
894 | ret = drv->bdrv_file_open(bs, options, open_flags, &local_err); | |
895 | } else { | |
896 | if (file == NULL) { | |
897 | error_setg(errp, "Can't use '%s' as a block driver for the " | |
898 | "protocol level", drv->format_name); | |
899 | ret = -EINVAL; | |
900 | goto free_and_fail; | |
901 | } | |
902 | bs->file = file; | |
903 | ret = drv->bdrv_open(bs, options, open_flags, &local_err); | |
904 | } | |
905 | ||
906 | if (ret < 0) { | |
907 | if (error_is_set(&local_err)) { | |
908 | error_propagate(errp, local_err); | |
909 | } else if (bs->filename[0]) { | |
910 | error_setg_errno(errp, -ret, "Could not open '%s'", bs->filename); | |
911 | } else { | |
912 | error_setg_errno(errp, -ret, "Could not open image"); | |
913 | } | |
914 | goto free_and_fail; | |
915 | } | |
916 | ||
917 | ret = refresh_total_sectors(bs, bs->total_sectors); | |
918 | if (ret < 0) { | |
919 | error_setg_errno(errp, -ret, "Could not refresh total sector count"); | |
920 | goto free_and_fail; | |
921 | } | |
922 | ||
923 | bdrv_refresh_limits(bs); | |
924 | assert(bdrv_opt_mem_align(bs) != 0); | |
925 | assert(bs->request_alignment != 0); | |
926 | ||
927 | #ifndef _WIN32 | |
928 | if (bs->is_temporary) { | |
929 | assert(bs->filename[0] != '\0'); | |
930 | unlink(bs->filename); | |
931 | } | |
932 | #endif | |
933 | return 0; | |
934 | ||
935 | free_and_fail: | |
936 | bs->file = NULL; | |
937 | g_free(bs->opaque); | |
938 | bs->opaque = NULL; | |
939 | bs->drv = NULL; | |
940 | return ret; | |
941 | } | |
942 | ||
943 | /* | |
944 | * Opens a file using a protocol (file, host_device, nbd, ...) | |
945 | * | |
946 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
947 | * empty set of options. The reference to the QDict belongs to the block layer | |
948 | * after the call (even on failure), so if the caller intends to reuse the | |
949 | * dictionary, it needs to use QINCREF() before calling bdrv_file_open. | |
950 | */ | |
951 | int bdrv_file_open(BlockDriverState **pbs, const char *filename, | |
952 | const char *reference, QDict *options, int flags, | |
953 | Error **errp) | |
954 | { | |
955 | BlockDriverState *bs = NULL; | |
956 | BlockDriver *drv; | |
957 | const char *drvname; | |
958 | bool allow_protocol_prefix = false; | |
959 | Error *local_err = NULL; | |
960 | int ret; | |
961 | ||
962 | /* NULL means an empty set of options */ | |
963 | if (options == NULL) { | |
964 | options = qdict_new(); | |
965 | } | |
966 | ||
967 | if (reference) { | |
968 | if (filename || qdict_size(options)) { | |
969 | error_setg(errp, "Cannot reference an existing block device with " | |
970 | "additional options or a new filename"); | |
971 | return -EINVAL; | |
972 | } | |
973 | QDECREF(options); | |
974 | ||
975 | bs = bdrv_find(reference); | |
976 | if (!bs) { | |
977 | error_setg(errp, "Cannot find block device '%s'", reference); | |
978 | return -ENODEV; | |
979 | } | |
980 | bdrv_ref(bs); | |
981 | *pbs = bs; | |
982 | return 0; | |
983 | } | |
984 | ||
985 | bs = bdrv_new(""); | |
986 | bs->options = options; | |
987 | options = qdict_clone_shallow(options); | |
988 | ||
989 | /* Fetch the file name from the options QDict if necessary */ | |
990 | if (!filename) { | |
991 | filename = qdict_get_try_str(options, "filename"); | |
992 | } else if (filename && !qdict_haskey(options, "filename")) { | |
993 | qdict_put(options, "filename", qstring_from_str(filename)); | |
994 | allow_protocol_prefix = true; | |
995 | } else { | |
996 | error_setg(errp, "Can't specify 'file' and 'filename' options at the " | |
997 | "same time"); | |
998 | ret = -EINVAL; | |
999 | goto fail; | |
1000 | } | |
1001 | ||
1002 | /* Find the right block driver */ | |
1003 | drvname = qdict_get_try_str(options, "driver"); | |
1004 | if (drvname) { | |
1005 | drv = bdrv_find_format(drvname); | |
1006 | if (!drv) { | |
1007 | error_setg(errp, "Unknown driver '%s'", drvname); | |
1008 | } | |
1009 | qdict_del(options, "driver"); | |
1010 | } else if (filename) { | |
1011 | drv = bdrv_find_protocol(filename, allow_protocol_prefix); | |
1012 | if (!drv) { | |
1013 | error_setg(errp, "Unknown protocol"); | |
1014 | } | |
1015 | } else { | |
1016 | error_setg(errp, "Must specify either driver or file"); | |
1017 | drv = NULL; | |
1018 | } | |
1019 | ||
1020 | if (!drv) { | |
1021 | /* errp has been set already */ | |
1022 | ret = -ENOENT; | |
1023 | goto fail; | |
1024 | } | |
1025 | ||
1026 | /* Parse the filename and open it */ | |
1027 | if (drv->bdrv_parse_filename && filename) { | |
1028 | drv->bdrv_parse_filename(filename, options, &local_err); | |
1029 | if (error_is_set(&local_err)) { | |
1030 | error_propagate(errp, local_err); | |
1031 | ret = -EINVAL; | |
1032 | goto fail; | |
1033 | } | |
1034 | qdict_del(options, "filename"); | |
1035 | } else if (drv->bdrv_needs_filename && !filename) { | |
1036 | error_setg(errp, "The '%s' block driver requires a file name", | |
1037 | drv->format_name); | |
1038 | ret = -EINVAL; | |
1039 | goto fail; | |
1040 | } | |
1041 | ||
1042 | if (!drv->bdrv_file_open) { | |
1043 | ret = bdrv_open(bs, filename, options, flags, drv, &local_err); | |
1044 | options = NULL; | |
1045 | } else { | |
1046 | ret = bdrv_open_common(bs, NULL, options, flags, drv, &local_err); | |
1047 | } | |
1048 | if (ret < 0) { | |
1049 | error_propagate(errp, local_err); | |
1050 | goto fail; | |
1051 | } | |
1052 | ||
1053 | /* Check if any unknown options were used */ | |
1054 | if (options && (qdict_size(options) != 0)) { | |
1055 | const QDictEntry *entry = qdict_first(options); | |
1056 | error_setg(errp, "Block protocol '%s' doesn't support the option '%s'", | |
1057 | drv->format_name, entry->key); | |
1058 | ret = -EINVAL; | |
1059 | goto fail; | |
1060 | } | |
1061 | QDECREF(options); | |
1062 | ||
1063 | bs->growable = 1; | |
1064 | *pbs = bs; | |
1065 | return 0; | |
1066 | ||
1067 | fail: | |
1068 | QDECREF(options); | |
1069 | if (!bs->drv) { | |
1070 | QDECREF(bs->options); | |
1071 | } | |
1072 | bdrv_unref(bs); | |
1073 | return ret; | |
1074 | } | |
1075 | ||
1076 | /* | |
1077 | * Opens the backing file for a BlockDriverState if not yet open | |
1078 | * | |
1079 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
1080 | * empty set of options. The reference to the QDict is transferred to this | |
1081 | * function (even on failure), so if the caller intends to reuse the dictionary, | |
1082 | * it needs to use QINCREF() before calling bdrv_file_open. | |
1083 | */ | |
1084 | int bdrv_open_backing_file(BlockDriverState *bs, QDict *options, Error **errp) | |
1085 | { | |
1086 | char backing_filename[PATH_MAX]; | |
1087 | int back_flags, ret; | |
1088 | BlockDriver *back_drv = NULL; | |
1089 | Error *local_err = NULL; | |
1090 | ||
1091 | if (bs->backing_hd != NULL) { | |
1092 | QDECREF(options); | |
1093 | return 0; | |
1094 | } | |
1095 | ||
1096 | /* NULL means an empty set of options */ | |
1097 | if (options == NULL) { | |
1098 | options = qdict_new(); | |
1099 | } | |
1100 | ||
1101 | bs->open_flags &= ~BDRV_O_NO_BACKING; | |
1102 | if (qdict_haskey(options, "file.filename")) { | |
1103 | backing_filename[0] = '\0'; | |
1104 | } else if (bs->backing_file[0] == '\0' && qdict_size(options) == 0) { | |
1105 | QDECREF(options); | |
1106 | return 0; | |
1107 | } else { | |
1108 | bdrv_get_full_backing_filename(bs, backing_filename, | |
1109 | sizeof(backing_filename)); | |
1110 | } | |
1111 | ||
1112 | bs->backing_hd = bdrv_new(""); | |
1113 | ||
1114 | if (bs->backing_format[0] != '\0') { | |
1115 | back_drv = bdrv_find_format(bs->backing_format); | |
1116 | } | |
1117 | ||
1118 | /* backing files always opened read-only */ | |
1119 | back_flags = bs->open_flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | | |
1120 | BDRV_O_COPY_ON_READ); | |
1121 | ||
1122 | ret = bdrv_open(bs->backing_hd, | |
1123 | *backing_filename ? backing_filename : NULL, options, | |
1124 | back_flags, back_drv, &local_err); | |
1125 | if (ret < 0) { | |
1126 | bdrv_unref(bs->backing_hd); | |
1127 | bs->backing_hd = NULL; | |
1128 | bs->open_flags |= BDRV_O_NO_BACKING; | |
1129 | error_setg(errp, "Could not open backing file: %s", | |
1130 | error_get_pretty(local_err)); | |
1131 | error_free(local_err); | |
1132 | return ret; | |
1133 | } | |
1134 | ||
1135 | if (bs->backing_hd->file) { | |
1136 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), | |
1137 | bs->backing_hd->file->filename); | |
1138 | } | |
1139 | ||
1140 | /* Recalculate the BlockLimits with the backing file */ | |
1141 | bdrv_refresh_limits(bs); | |
1142 | ||
1143 | return 0; | |
1144 | } | |
1145 | ||
1146 | /* | |
1147 | * Opens a disk image whose options are given as BlockdevRef in another block | |
1148 | * device's options. | |
1149 | * | |
1150 | * If force_raw is true, bdrv_file_open() will be used, thereby preventing any | |
1151 | * image format auto-detection. If it is false and a filename is given, | |
1152 | * bdrv_open() will be used for auto-detection. | |
1153 | * | |
1154 | * If allow_none is true, no image will be opened if filename is false and no | |
1155 | * BlockdevRef is given. *pbs will remain unchanged and 0 will be returned. | |
1156 | * | |
1157 | * bdrev_key specifies the key for the image's BlockdevRef in the options QDict. | |
1158 | * That QDict has to be flattened; therefore, if the BlockdevRef is a QDict | |
1159 | * itself, all options starting with "${bdref_key}." are considered part of the | |
1160 | * BlockdevRef. | |
1161 | * | |
1162 | * The BlockdevRef will be removed from the options QDict. | |
1163 | */ | |
1164 | int bdrv_open_image(BlockDriverState **pbs, const char *filename, | |
1165 | QDict *options, const char *bdref_key, int flags, | |
1166 | bool force_raw, bool allow_none, Error **errp) | |
1167 | { | |
1168 | QDict *image_options; | |
1169 | int ret; | |
1170 | char *bdref_key_dot; | |
1171 | const char *reference; | |
1172 | ||
1173 | bdref_key_dot = g_strdup_printf("%s.", bdref_key); | |
1174 | qdict_extract_subqdict(options, &image_options, bdref_key_dot); | |
1175 | g_free(bdref_key_dot); | |
1176 | ||
1177 | reference = qdict_get_try_str(options, bdref_key); | |
1178 | if (!filename && !reference && !qdict_size(image_options)) { | |
1179 | if (allow_none) { | |
1180 | ret = 0; | |
1181 | } else { | |
1182 | error_setg(errp, "A block device must be specified for \"%s\"", | |
1183 | bdref_key); | |
1184 | ret = -EINVAL; | |
1185 | } | |
1186 | goto done; | |
1187 | } | |
1188 | ||
1189 | if (filename && !force_raw) { | |
1190 | /* If a filename is given and the block driver should be detected | |
1191 | automatically (instead of using none), use bdrv_open() in order to do | |
1192 | that auto-detection. */ | |
1193 | BlockDriverState *bs; | |
1194 | ||
1195 | if (reference) { | |
1196 | error_setg(errp, "Cannot reference an existing block device while " | |
1197 | "giving a filename"); | |
1198 | ret = -EINVAL; | |
1199 | goto done; | |
1200 | } | |
1201 | ||
1202 | bs = bdrv_new(""); | |
1203 | ret = bdrv_open(bs, filename, image_options, flags, NULL, errp); | |
1204 | if (ret < 0) { | |
1205 | bdrv_unref(bs); | |
1206 | } else { | |
1207 | *pbs = bs; | |
1208 | } | |
1209 | } else { | |
1210 | ret = bdrv_file_open(pbs, filename, reference, image_options, flags, | |
1211 | errp); | |
1212 | } | |
1213 | ||
1214 | done: | |
1215 | qdict_del(options, bdref_key); | |
1216 | return ret; | |
1217 | } | |
1218 | ||
1219 | /* | |
1220 | * Opens a disk image (raw, qcow2, vmdk, ...) | |
1221 | * | |
1222 | * options is a QDict of options to pass to the block drivers, or NULL for an | |
1223 | * empty set of options. The reference to the QDict belongs to the block layer | |
1224 | * after the call (even on failure), so if the caller intends to reuse the | |
1225 | * dictionary, it needs to use QINCREF() before calling bdrv_open. | |
1226 | */ | |
1227 | int bdrv_open(BlockDriverState *bs, const char *filename, QDict *options, | |
1228 | int flags, BlockDriver *drv, Error **errp) | |
1229 | { | |
1230 | int ret; | |
1231 | /* TODO: extra byte is a hack to ensure MAX_PATH space on Windows. */ | |
1232 | char tmp_filename[PATH_MAX + 1]; | |
1233 | BlockDriverState *file = NULL; | |
1234 | const char *drvname; | |
1235 | Error *local_err = NULL; | |
1236 | ||
1237 | /* NULL means an empty set of options */ | |
1238 | if (options == NULL) { | |
1239 | options = qdict_new(); | |
1240 | } | |
1241 | ||
1242 | bs->options = options; | |
1243 | options = qdict_clone_shallow(options); | |
1244 | ||
1245 | /* For snapshot=on, create a temporary qcow2 overlay */ | |
1246 | if (flags & BDRV_O_SNAPSHOT) { | |
1247 | BlockDriverState *bs1; | |
1248 | int64_t total_size; | |
1249 | BlockDriver *bdrv_qcow2; | |
1250 | QEMUOptionParameter *create_options; | |
1251 | QDict *snapshot_options; | |
1252 | ||
1253 | /* if snapshot, we create a temporary backing file and open it | |
1254 | instead of opening 'filename' directly */ | |
1255 | ||
1256 | /* Get the required size from the image */ | |
1257 | bs1 = bdrv_new(""); | |
1258 | QINCREF(options); | |
1259 | ret = bdrv_open(bs1, filename, options, BDRV_O_NO_BACKING, | |
1260 | drv, &local_err); | |
1261 | if (ret < 0) { | |
1262 | bdrv_unref(bs1); | |
1263 | goto fail; | |
1264 | } | |
1265 | total_size = bdrv_getlength(bs1) & BDRV_SECTOR_MASK; | |
1266 | ||
1267 | bdrv_unref(bs1); | |
1268 | ||
1269 | /* Create the temporary image */ | |
1270 | ret = get_tmp_filename(tmp_filename, sizeof(tmp_filename)); | |
1271 | if (ret < 0) { | |
1272 | error_setg_errno(errp, -ret, "Could not get temporary filename"); | |
1273 | goto fail; | |
1274 | } | |
1275 | ||
1276 | bdrv_qcow2 = bdrv_find_format("qcow2"); | |
1277 | create_options = parse_option_parameters("", bdrv_qcow2->create_options, | |
1278 | NULL); | |
1279 | ||
1280 | set_option_parameter_int(create_options, BLOCK_OPT_SIZE, total_size); | |
1281 | ||
1282 | ret = bdrv_create(bdrv_qcow2, tmp_filename, create_options, &local_err); | |
1283 | free_option_parameters(create_options); | |
1284 | if (ret < 0) { | |
1285 | error_setg_errno(errp, -ret, "Could not create temporary overlay " | |
1286 | "'%s': %s", tmp_filename, | |
1287 | error_get_pretty(local_err)); | |
1288 | error_free(local_err); | |
1289 | local_err = NULL; | |
1290 | goto fail; | |
1291 | } | |
1292 | ||
1293 | /* Prepare a new options QDict for the temporary file, where user | |
1294 | * options refer to the backing file */ | |
1295 | if (filename) { | |
1296 | qdict_put(options, "file.filename", qstring_from_str(filename)); | |
1297 | } | |
1298 | if (drv) { | |
1299 | qdict_put(options, "driver", qstring_from_str(drv->format_name)); | |
1300 | } | |
1301 | ||
1302 | snapshot_options = qdict_new(); | |
1303 | qdict_put(snapshot_options, "backing", options); | |
1304 | qdict_flatten(snapshot_options); | |
1305 | ||
1306 | bs->options = snapshot_options; | |
1307 | options = qdict_clone_shallow(bs->options); | |
1308 | ||
1309 | filename = tmp_filename; | |
1310 | drv = bdrv_qcow2; | |
1311 | bs->is_temporary = 1; | |
1312 | } | |
1313 | ||
1314 | /* Open image file without format layer */ | |
1315 | if (flags & BDRV_O_RDWR) { | |
1316 | flags |= BDRV_O_ALLOW_RDWR; | |
1317 | } | |
1318 | ||
1319 | ret = bdrv_open_image(&file, filename, options, "file", | |
1320 | bdrv_open_flags(bs, flags | BDRV_O_UNMAP), true, true, | |
1321 | &local_err); | |
1322 | if (ret < 0) { | |
1323 | goto fail; | |
1324 | } | |
1325 | ||
1326 | /* Find the right image format driver */ | |
1327 | drvname = qdict_get_try_str(options, "driver"); | |
1328 | if (drvname) { | |
1329 | drv = bdrv_find_format(drvname); | |
1330 | qdict_del(options, "driver"); | |
1331 | if (!drv) { | |
1332 | error_setg(errp, "Invalid driver: '%s'", drvname); | |
1333 | ret = -EINVAL; | |
1334 | goto unlink_and_fail; | |
1335 | } | |
1336 | } | |
1337 | ||
1338 | if (!drv) { | |
1339 | if (file) { | |
1340 | ret = find_image_format(file, filename, &drv, &local_err); | |
1341 | } else { | |
1342 | error_setg(errp, "Must specify either driver or file"); | |
1343 | ret = -EINVAL; | |
1344 | goto unlink_and_fail; | |
1345 | } | |
1346 | } | |
1347 | ||
1348 | if (!drv) { | |
1349 | goto unlink_and_fail; | |
1350 | } | |
1351 | ||
1352 | /* Open the image */ | |
1353 | ret = bdrv_open_common(bs, file, options, flags, drv, &local_err); | |
1354 | if (ret < 0) { | |
1355 | goto unlink_and_fail; | |
1356 | } | |
1357 | ||
1358 | if (file && (bs->file != file)) { | |
1359 | bdrv_unref(file); | |
1360 | file = NULL; | |
1361 | } | |
1362 | ||
1363 | /* If there is a backing file, use it */ | |
1364 | if ((flags & BDRV_O_NO_BACKING) == 0) { | |
1365 | QDict *backing_options; | |
1366 | ||
1367 | qdict_extract_subqdict(options, &backing_options, "backing."); | |
1368 | ret = bdrv_open_backing_file(bs, backing_options, &local_err); | |
1369 | if (ret < 0) { | |
1370 | goto close_and_fail; | |
1371 | } | |
1372 | } | |
1373 | ||
1374 | /* Check if any unknown options were used */ | |
1375 | if (qdict_size(options) != 0) { | |
1376 | const QDictEntry *entry = qdict_first(options); | |
1377 | error_setg(errp, "Block format '%s' used by device '%s' doesn't " | |
1378 | "support the option '%s'", drv->format_name, bs->device_name, | |
1379 | entry->key); | |
1380 | ||
1381 | ret = -EINVAL; | |
1382 | goto close_and_fail; | |
1383 | } | |
1384 | QDECREF(options); | |
1385 | ||
1386 | if (!bdrv_key_required(bs)) { | |
1387 | bdrv_dev_change_media_cb(bs, true); | |
1388 | } | |
1389 | ||
1390 | return 0; | |
1391 | ||
1392 | unlink_and_fail: | |
1393 | if (file != NULL) { | |
1394 | bdrv_unref(file); | |
1395 | } | |
1396 | if (bs->is_temporary) { | |
1397 | unlink(filename); | |
1398 | } | |
1399 | fail: | |
1400 | QDECREF(bs->options); | |
1401 | QDECREF(options); | |
1402 | bs->options = NULL; | |
1403 | if (error_is_set(&local_err)) { | |
1404 | error_propagate(errp, local_err); | |
1405 | } | |
1406 | return ret; | |
1407 | ||
1408 | close_and_fail: | |
1409 | bdrv_close(bs); | |
1410 | QDECREF(options); | |
1411 | if (error_is_set(&local_err)) { | |
1412 | error_propagate(errp, local_err); | |
1413 | } | |
1414 | return ret; | |
1415 | } | |
1416 | ||
1417 | typedef struct BlockReopenQueueEntry { | |
1418 | bool prepared; | |
1419 | BDRVReopenState state; | |
1420 | QSIMPLEQ_ENTRY(BlockReopenQueueEntry) entry; | |
1421 | } BlockReopenQueueEntry; | |
1422 | ||
1423 | /* | |
1424 | * Adds a BlockDriverState to a simple queue for an atomic, transactional | |
1425 | * reopen of multiple devices. | |
1426 | * | |
1427 | * bs_queue can either be an existing BlockReopenQueue that has had QSIMPLE_INIT | |
1428 | * already performed, or alternatively may be NULL a new BlockReopenQueue will | |
1429 | * be created and initialized. This newly created BlockReopenQueue should be | |
1430 | * passed back in for subsequent calls that are intended to be of the same | |
1431 | * atomic 'set'. | |
1432 | * | |
1433 | * bs is the BlockDriverState to add to the reopen queue. | |
1434 | * | |
1435 | * flags contains the open flags for the associated bs | |
1436 | * | |
1437 | * returns a pointer to bs_queue, which is either the newly allocated | |
1438 | * bs_queue, or the existing bs_queue being used. | |
1439 | * | |
1440 | */ | |
1441 | BlockReopenQueue *bdrv_reopen_queue(BlockReopenQueue *bs_queue, | |
1442 | BlockDriverState *bs, int flags) | |
1443 | { | |
1444 | assert(bs != NULL); | |
1445 | ||
1446 | BlockReopenQueueEntry *bs_entry; | |
1447 | if (bs_queue == NULL) { | |
1448 | bs_queue = g_new0(BlockReopenQueue, 1); | |
1449 | QSIMPLEQ_INIT(bs_queue); | |
1450 | } | |
1451 | ||
1452 | if (bs->file) { | |
1453 | bdrv_reopen_queue(bs_queue, bs->file, flags); | |
1454 | } | |
1455 | ||
1456 | bs_entry = g_new0(BlockReopenQueueEntry, 1); | |
1457 | QSIMPLEQ_INSERT_TAIL(bs_queue, bs_entry, entry); | |
1458 | ||
1459 | bs_entry->state.bs = bs; | |
1460 | bs_entry->state.flags = flags; | |
1461 | ||
1462 | return bs_queue; | |
1463 | } | |
1464 | ||
1465 | /* | |
1466 | * Reopen multiple BlockDriverStates atomically & transactionally. | |
1467 | * | |
1468 | * The queue passed in (bs_queue) must have been built up previous | |
1469 | * via bdrv_reopen_queue(). | |
1470 | * | |
1471 | * Reopens all BDS specified in the queue, with the appropriate | |
1472 | * flags. All devices are prepared for reopen, and failure of any | |
1473 | * device will cause all device changes to be abandonded, and intermediate | |
1474 | * data cleaned up. | |
1475 | * | |
1476 | * If all devices prepare successfully, then the changes are committed | |
1477 | * to all devices. | |
1478 | * | |
1479 | */ | |
1480 | int bdrv_reopen_multiple(BlockReopenQueue *bs_queue, Error **errp) | |
1481 | { | |
1482 | int ret = -1; | |
1483 | BlockReopenQueueEntry *bs_entry, *next; | |
1484 | Error *local_err = NULL; | |
1485 | ||
1486 | assert(bs_queue != NULL); | |
1487 | ||
1488 | bdrv_drain_all(); | |
1489 | ||
1490 | QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { | |
1491 | if (bdrv_reopen_prepare(&bs_entry->state, bs_queue, &local_err)) { | |
1492 | error_propagate(errp, local_err); | |
1493 | goto cleanup; | |
1494 | } | |
1495 | bs_entry->prepared = true; | |
1496 | } | |
1497 | ||
1498 | /* If we reach this point, we have success and just need to apply the | |
1499 | * changes | |
1500 | */ | |
1501 | QSIMPLEQ_FOREACH(bs_entry, bs_queue, entry) { | |
1502 | bdrv_reopen_commit(&bs_entry->state); | |
1503 | } | |
1504 | ||
1505 | ret = 0; | |
1506 | ||
1507 | cleanup: | |
1508 | QSIMPLEQ_FOREACH_SAFE(bs_entry, bs_queue, entry, next) { | |
1509 | if (ret && bs_entry->prepared) { | |
1510 | bdrv_reopen_abort(&bs_entry->state); | |
1511 | } | |
1512 | g_free(bs_entry); | |
1513 | } | |
1514 | g_free(bs_queue); | |
1515 | return ret; | |
1516 | } | |
1517 | ||
1518 | ||
1519 | /* Reopen a single BlockDriverState with the specified flags. */ | |
1520 | int bdrv_reopen(BlockDriverState *bs, int bdrv_flags, Error **errp) | |
1521 | { | |
1522 | int ret = -1; | |
1523 | Error *local_err = NULL; | |
1524 | BlockReopenQueue *queue = bdrv_reopen_queue(NULL, bs, bdrv_flags); | |
1525 | ||
1526 | ret = bdrv_reopen_multiple(queue, &local_err); | |
1527 | if (local_err != NULL) { | |
1528 | error_propagate(errp, local_err); | |
1529 | } | |
1530 | return ret; | |
1531 | } | |
1532 | ||
1533 | ||
1534 | /* | |
1535 | * Prepares a BlockDriverState for reopen. All changes are staged in the | |
1536 | * 'opaque' field of the BDRVReopenState, which is used and allocated by | |
1537 | * the block driver layer .bdrv_reopen_prepare() | |
1538 | * | |
1539 | * bs is the BlockDriverState to reopen | |
1540 | * flags are the new open flags | |
1541 | * queue is the reopen queue | |
1542 | * | |
1543 | * Returns 0 on success, non-zero on error. On error errp will be set | |
1544 | * as well. | |
1545 | * | |
1546 | * On failure, bdrv_reopen_abort() will be called to clean up any data. | |
1547 | * It is the responsibility of the caller to then call the abort() or | |
1548 | * commit() for any other BDS that have been left in a prepare() state | |
1549 | * | |
1550 | */ | |
1551 | int bdrv_reopen_prepare(BDRVReopenState *reopen_state, BlockReopenQueue *queue, | |
1552 | Error **errp) | |
1553 | { | |
1554 | int ret = -1; | |
1555 | Error *local_err = NULL; | |
1556 | BlockDriver *drv; | |
1557 | ||
1558 | assert(reopen_state != NULL); | |
1559 | assert(reopen_state->bs->drv != NULL); | |
1560 | drv = reopen_state->bs->drv; | |
1561 | ||
1562 | /* if we are to stay read-only, do not allow permission change | |
1563 | * to r/w */ | |
1564 | if (!(reopen_state->bs->open_flags & BDRV_O_ALLOW_RDWR) && | |
1565 | reopen_state->flags & BDRV_O_RDWR) { | |
1566 | error_set(errp, QERR_DEVICE_IS_READ_ONLY, | |
1567 | reopen_state->bs->device_name); | |
1568 | goto error; | |
1569 | } | |
1570 | ||
1571 | ||
1572 | ret = bdrv_flush(reopen_state->bs); | |
1573 | if (ret) { | |
1574 | error_set(errp, ERROR_CLASS_GENERIC_ERROR, "Error (%s) flushing drive", | |
1575 | strerror(-ret)); | |
1576 | goto error; | |
1577 | } | |
1578 | ||
1579 | if (drv->bdrv_reopen_prepare) { | |
1580 | ret = drv->bdrv_reopen_prepare(reopen_state, queue, &local_err); | |
1581 | if (ret) { | |
1582 | if (local_err != NULL) { | |
1583 | error_propagate(errp, local_err); | |
1584 | } else { | |
1585 | error_setg(errp, "failed while preparing to reopen image '%s'", | |
1586 | reopen_state->bs->filename); | |
1587 | } | |
1588 | goto error; | |
1589 | } | |
1590 | } else { | |
1591 | /* It is currently mandatory to have a bdrv_reopen_prepare() | |
1592 | * handler for each supported drv. */ | |
1593 | error_set(errp, QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, | |
1594 | drv->format_name, reopen_state->bs->device_name, | |
1595 | "reopening of file"); | |
1596 | ret = -1; | |
1597 | goto error; | |
1598 | } | |
1599 | ||
1600 | ret = 0; | |
1601 | ||
1602 | error: | |
1603 | return ret; | |
1604 | } | |
1605 | ||
1606 | /* | |
1607 | * Takes the staged changes for the reopen from bdrv_reopen_prepare(), and | |
1608 | * makes them final by swapping the staging BlockDriverState contents into | |
1609 | * the active BlockDriverState contents. | |
1610 | */ | |
1611 | void bdrv_reopen_commit(BDRVReopenState *reopen_state) | |
1612 | { | |
1613 | BlockDriver *drv; | |
1614 | ||
1615 | assert(reopen_state != NULL); | |
1616 | drv = reopen_state->bs->drv; | |
1617 | assert(drv != NULL); | |
1618 | ||
1619 | /* If there are any driver level actions to take */ | |
1620 | if (drv->bdrv_reopen_commit) { | |
1621 | drv->bdrv_reopen_commit(reopen_state); | |
1622 | } | |
1623 | ||
1624 | /* set BDS specific flags now */ | |
1625 | reopen_state->bs->open_flags = reopen_state->flags; | |
1626 | reopen_state->bs->enable_write_cache = !!(reopen_state->flags & | |
1627 | BDRV_O_CACHE_WB); | |
1628 | reopen_state->bs->read_only = !(reopen_state->flags & BDRV_O_RDWR); | |
1629 | ||
1630 | bdrv_refresh_limits(reopen_state->bs); | |
1631 | } | |
1632 | ||
1633 | /* | |
1634 | * Abort the reopen, and delete and free the staged changes in | |
1635 | * reopen_state | |
1636 | */ | |
1637 | void bdrv_reopen_abort(BDRVReopenState *reopen_state) | |
1638 | { | |
1639 | BlockDriver *drv; | |
1640 | ||
1641 | assert(reopen_state != NULL); | |
1642 | drv = reopen_state->bs->drv; | |
1643 | assert(drv != NULL); | |
1644 | ||
1645 | if (drv->bdrv_reopen_abort) { | |
1646 | drv->bdrv_reopen_abort(reopen_state); | |
1647 | } | |
1648 | } | |
1649 | ||
1650 | ||
1651 | void bdrv_close(BlockDriverState *bs) | |
1652 | { | |
1653 | if (bs->job) { | |
1654 | block_job_cancel_sync(bs->job); | |
1655 | } | |
1656 | bdrv_drain_all(); /* complete I/O */ | |
1657 | bdrv_flush(bs); | |
1658 | bdrv_drain_all(); /* in case flush left pending I/O */ | |
1659 | notifier_list_notify(&bs->close_notifiers, bs); | |
1660 | ||
1661 | if (bs->drv) { | |
1662 | if (bs->backing_hd) { | |
1663 | bdrv_unref(bs->backing_hd); | |
1664 | bs->backing_hd = NULL; | |
1665 | } | |
1666 | bs->drv->bdrv_close(bs); | |
1667 | g_free(bs->opaque); | |
1668 | #ifdef _WIN32 | |
1669 | if (bs->is_temporary) { | |
1670 | unlink(bs->filename); | |
1671 | } | |
1672 | #endif | |
1673 | bs->opaque = NULL; | |
1674 | bs->drv = NULL; | |
1675 | bs->copy_on_read = 0; | |
1676 | bs->backing_file[0] = '\0'; | |
1677 | bs->backing_format[0] = '\0'; | |
1678 | bs->total_sectors = 0; | |
1679 | bs->encrypted = 0; | |
1680 | bs->valid_key = 0; | |
1681 | bs->sg = 0; | |
1682 | bs->growable = 0; | |
1683 | bs->zero_beyond_eof = false; | |
1684 | QDECREF(bs->options); | |
1685 | bs->options = NULL; | |
1686 | ||
1687 | if (bs->file != NULL) { | |
1688 | bdrv_unref(bs->file); | |
1689 | bs->file = NULL; | |
1690 | } | |
1691 | } | |
1692 | ||
1693 | bdrv_dev_change_media_cb(bs, false); | |
1694 | ||
1695 | /*throttling disk I/O limits*/ | |
1696 | if (bs->io_limits_enabled) { | |
1697 | bdrv_io_limits_disable(bs); | |
1698 | } | |
1699 | } | |
1700 | ||
1701 | void bdrv_close_all(void) | |
1702 | { | |
1703 | BlockDriverState *bs; | |
1704 | ||
1705 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
1706 | bdrv_close(bs); | |
1707 | } | |
1708 | } | |
1709 | ||
1710 | /* Check if any requests are in-flight (including throttled requests) */ | |
1711 | static bool bdrv_requests_pending(BlockDriverState *bs) | |
1712 | { | |
1713 | if (!QLIST_EMPTY(&bs->tracked_requests)) { | |
1714 | return true; | |
1715 | } | |
1716 | if (!qemu_co_queue_empty(&bs->throttled_reqs[0])) { | |
1717 | return true; | |
1718 | } | |
1719 | if (!qemu_co_queue_empty(&bs->throttled_reqs[1])) { | |
1720 | return true; | |
1721 | } | |
1722 | if (bs->file && bdrv_requests_pending(bs->file)) { | |
1723 | return true; | |
1724 | } | |
1725 | if (bs->backing_hd && bdrv_requests_pending(bs->backing_hd)) { | |
1726 | return true; | |
1727 | } | |
1728 | return false; | |
1729 | } | |
1730 | ||
1731 | static bool bdrv_requests_pending_all(void) | |
1732 | { | |
1733 | BlockDriverState *bs; | |
1734 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
1735 | if (bdrv_requests_pending(bs)) { | |
1736 | return true; | |
1737 | } | |
1738 | } | |
1739 | return false; | |
1740 | } | |
1741 | ||
1742 | /* | |
1743 | * Wait for pending requests to complete across all BlockDriverStates | |
1744 | * | |
1745 | * This function does not flush data to disk, use bdrv_flush_all() for that | |
1746 | * after calling this function. | |
1747 | * | |
1748 | * Note that completion of an asynchronous I/O operation can trigger any | |
1749 | * number of other I/O operations on other devices---for example a coroutine | |
1750 | * can be arbitrarily complex and a constant flow of I/O can come until the | |
1751 | * coroutine is complete. Because of this, it is not possible to have a | |
1752 | * function to drain a single device's I/O queue. | |
1753 | */ | |
1754 | void bdrv_drain_all(void) | |
1755 | { | |
1756 | /* Always run first iteration so any pending completion BHs run */ | |
1757 | bool busy = true; | |
1758 | BlockDriverState *bs; | |
1759 | ||
1760 | while (busy) { | |
1761 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
1762 | bdrv_start_throttled_reqs(bs); | |
1763 | } | |
1764 | ||
1765 | busy = bdrv_requests_pending_all(); | |
1766 | busy |= aio_poll(qemu_get_aio_context(), busy); | |
1767 | } | |
1768 | } | |
1769 | ||
1770 | /* make a BlockDriverState anonymous by removing from bdrv_state and | |
1771 | * graph_bdrv_state list. | |
1772 | Also, NULL terminate the device_name to prevent double remove */ | |
1773 | void bdrv_make_anon(BlockDriverState *bs) | |
1774 | { | |
1775 | if (bs->device_name[0] != '\0') { | |
1776 | QTAILQ_REMOVE(&bdrv_states, bs, device_list); | |
1777 | } | |
1778 | bs->device_name[0] = '\0'; | |
1779 | if (bs->node_name[0] != '\0') { | |
1780 | QTAILQ_REMOVE(&graph_bdrv_states, bs, node_list); | |
1781 | } | |
1782 | bs->node_name[0] = '\0'; | |
1783 | } | |
1784 | ||
1785 | static void bdrv_rebind(BlockDriverState *bs) | |
1786 | { | |
1787 | if (bs->drv && bs->drv->bdrv_rebind) { | |
1788 | bs->drv->bdrv_rebind(bs); | |
1789 | } | |
1790 | } | |
1791 | ||
1792 | static void bdrv_move_feature_fields(BlockDriverState *bs_dest, | |
1793 | BlockDriverState *bs_src) | |
1794 | { | |
1795 | /* move some fields that need to stay attached to the device */ | |
1796 | bs_dest->open_flags = bs_src->open_flags; | |
1797 | ||
1798 | /* dev info */ | |
1799 | bs_dest->dev_ops = bs_src->dev_ops; | |
1800 | bs_dest->dev_opaque = bs_src->dev_opaque; | |
1801 | bs_dest->dev = bs_src->dev; | |
1802 | bs_dest->guest_block_size = bs_src->guest_block_size; | |
1803 | bs_dest->copy_on_read = bs_src->copy_on_read; | |
1804 | ||
1805 | bs_dest->enable_write_cache = bs_src->enable_write_cache; | |
1806 | ||
1807 | /* i/o throttled req */ | |
1808 | memcpy(&bs_dest->throttle_state, | |
1809 | &bs_src->throttle_state, | |
1810 | sizeof(ThrottleState)); | |
1811 | bs_dest->throttled_reqs[0] = bs_src->throttled_reqs[0]; | |
1812 | bs_dest->throttled_reqs[1] = bs_src->throttled_reqs[1]; | |
1813 | bs_dest->io_limits_enabled = bs_src->io_limits_enabled; | |
1814 | ||
1815 | /* r/w error */ | |
1816 | bs_dest->on_read_error = bs_src->on_read_error; | |
1817 | bs_dest->on_write_error = bs_src->on_write_error; | |
1818 | ||
1819 | /* i/o status */ | |
1820 | bs_dest->iostatus_enabled = bs_src->iostatus_enabled; | |
1821 | bs_dest->iostatus = bs_src->iostatus; | |
1822 | ||
1823 | /* dirty bitmap */ | |
1824 | bs_dest->dirty_bitmaps = bs_src->dirty_bitmaps; | |
1825 | ||
1826 | /* reference count */ | |
1827 | bs_dest->refcnt = bs_src->refcnt; | |
1828 | ||
1829 | /* job */ | |
1830 | bs_dest->in_use = bs_src->in_use; | |
1831 | bs_dest->job = bs_src->job; | |
1832 | ||
1833 | /* keep the same entry in bdrv_states */ | |
1834 | pstrcpy(bs_dest->device_name, sizeof(bs_dest->device_name), | |
1835 | bs_src->device_name); | |
1836 | bs_dest->device_list = bs_src->device_list; | |
1837 | ||
1838 | /* keep the same entry in graph_bdrv_states | |
1839 | * We do want to swap name but don't want to swap linked list entries | |
1840 | */ | |
1841 | bs_dest->node_list = bs_src->node_list; | |
1842 | } | |
1843 | ||
1844 | /* | |
1845 | * Swap bs contents for two image chains while they are live, | |
1846 | * while keeping required fields on the BlockDriverState that is | |
1847 | * actually attached to a device. | |
1848 | * | |
1849 | * This will modify the BlockDriverState fields, and swap contents | |
1850 | * between bs_new and bs_old. Both bs_new and bs_old are modified. | |
1851 | * | |
1852 | * bs_new is required to be anonymous. | |
1853 | * | |
1854 | * This function does not create any image files. | |
1855 | */ | |
1856 | void bdrv_swap(BlockDriverState *bs_new, BlockDriverState *bs_old) | |
1857 | { | |
1858 | BlockDriverState tmp; | |
1859 | ||
1860 | /* bs_new must be anonymous and shouldn't have anything fancy enabled */ | |
1861 | assert(bs_new->device_name[0] == '\0'); | |
1862 | assert(QLIST_EMPTY(&bs_new->dirty_bitmaps)); | |
1863 | assert(bs_new->job == NULL); | |
1864 | assert(bs_new->dev == NULL); | |
1865 | assert(bs_new->in_use == 0); | |
1866 | assert(bs_new->io_limits_enabled == false); | |
1867 | assert(!throttle_have_timer(&bs_new->throttle_state)); | |
1868 | ||
1869 | tmp = *bs_new; | |
1870 | *bs_new = *bs_old; | |
1871 | *bs_old = tmp; | |
1872 | ||
1873 | /* there are some fields that should not be swapped, move them back */ | |
1874 | bdrv_move_feature_fields(&tmp, bs_old); | |
1875 | bdrv_move_feature_fields(bs_old, bs_new); | |
1876 | bdrv_move_feature_fields(bs_new, &tmp); | |
1877 | ||
1878 | /* bs_new shouldn't be in bdrv_states even after the swap! */ | |
1879 | assert(bs_new->device_name[0] == '\0'); | |
1880 | ||
1881 | /* Check a few fields that should remain attached to the device */ | |
1882 | assert(bs_new->dev == NULL); | |
1883 | assert(bs_new->job == NULL); | |
1884 | assert(bs_new->in_use == 0); | |
1885 | assert(bs_new->io_limits_enabled == false); | |
1886 | assert(!throttle_have_timer(&bs_new->throttle_state)); | |
1887 | ||
1888 | bdrv_rebind(bs_new); | |
1889 | bdrv_rebind(bs_old); | |
1890 | } | |
1891 | ||
1892 | /* | |
1893 | * Add new bs contents at the top of an image chain while the chain is | |
1894 | * live, while keeping required fields on the top layer. | |
1895 | * | |
1896 | * This will modify the BlockDriverState fields, and swap contents | |
1897 | * between bs_new and bs_top. Both bs_new and bs_top are modified. | |
1898 | * | |
1899 | * bs_new is required to be anonymous. | |
1900 | * | |
1901 | * This function does not create any image files. | |
1902 | */ | |
1903 | void bdrv_append(BlockDriverState *bs_new, BlockDriverState *bs_top) | |
1904 | { | |
1905 | bdrv_swap(bs_new, bs_top); | |
1906 | ||
1907 | /* The contents of 'tmp' will become bs_top, as we are | |
1908 | * swapping bs_new and bs_top contents. */ | |
1909 | bs_top->backing_hd = bs_new; | |
1910 | bs_top->open_flags &= ~BDRV_O_NO_BACKING; | |
1911 | pstrcpy(bs_top->backing_file, sizeof(bs_top->backing_file), | |
1912 | bs_new->filename); | |
1913 | pstrcpy(bs_top->backing_format, sizeof(bs_top->backing_format), | |
1914 | bs_new->drv ? bs_new->drv->format_name : ""); | |
1915 | } | |
1916 | ||
1917 | static void bdrv_delete(BlockDriverState *bs) | |
1918 | { | |
1919 | assert(!bs->dev); | |
1920 | assert(!bs->job); | |
1921 | assert(!bs->in_use); | |
1922 | assert(!bs->refcnt); | |
1923 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
1924 | ||
1925 | bdrv_close(bs); | |
1926 | ||
1927 | /* remove from list, if necessary */ | |
1928 | bdrv_make_anon(bs); | |
1929 | ||
1930 | g_free(bs); | |
1931 | } | |
1932 | ||
1933 | int bdrv_attach_dev(BlockDriverState *bs, void *dev) | |
1934 | /* TODO change to DeviceState *dev when all users are qdevified */ | |
1935 | { | |
1936 | if (bs->dev) { | |
1937 | return -EBUSY; | |
1938 | } | |
1939 | bs->dev = dev; | |
1940 | bdrv_iostatus_reset(bs); | |
1941 | return 0; | |
1942 | } | |
1943 | ||
1944 | /* TODO qdevified devices don't use this, remove when devices are qdevified */ | |
1945 | void bdrv_attach_dev_nofail(BlockDriverState *bs, void *dev) | |
1946 | { | |
1947 | if (bdrv_attach_dev(bs, dev) < 0) { | |
1948 | abort(); | |
1949 | } | |
1950 | } | |
1951 | ||
1952 | void bdrv_detach_dev(BlockDriverState *bs, void *dev) | |
1953 | /* TODO change to DeviceState *dev when all users are qdevified */ | |
1954 | { | |
1955 | assert(bs->dev == dev); | |
1956 | bs->dev = NULL; | |
1957 | bs->dev_ops = NULL; | |
1958 | bs->dev_opaque = NULL; | |
1959 | bs->guest_block_size = 512; | |
1960 | } | |
1961 | ||
1962 | /* TODO change to return DeviceState * when all users are qdevified */ | |
1963 | void *bdrv_get_attached_dev(BlockDriverState *bs) | |
1964 | { | |
1965 | return bs->dev; | |
1966 | } | |
1967 | ||
1968 | void bdrv_set_dev_ops(BlockDriverState *bs, const BlockDevOps *ops, | |
1969 | void *opaque) | |
1970 | { | |
1971 | bs->dev_ops = ops; | |
1972 | bs->dev_opaque = opaque; | |
1973 | } | |
1974 | ||
1975 | void bdrv_emit_qmp_error_event(const BlockDriverState *bdrv, | |
1976 | enum MonitorEvent ev, | |
1977 | BlockErrorAction action, bool is_read) | |
1978 | { | |
1979 | QObject *data; | |
1980 | const char *action_str; | |
1981 | ||
1982 | switch (action) { | |
1983 | case BDRV_ACTION_REPORT: | |
1984 | action_str = "report"; | |
1985 | break; | |
1986 | case BDRV_ACTION_IGNORE: | |
1987 | action_str = "ignore"; | |
1988 | break; | |
1989 | case BDRV_ACTION_STOP: | |
1990 | action_str = "stop"; | |
1991 | break; | |
1992 | default: | |
1993 | abort(); | |
1994 | } | |
1995 | ||
1996 | data = qobject_from_jsonf("{ 'device': %s, 'action': %s, 'operation': %s }", | |
1997 | bdrv->device_name, | |
1998 | action_str, | |
1999 | is_read ? "read" : "write"); | |
2000 | monitor_protocol_event(ev, data); | |
2001 | ||
2002 | qobject_decref(data); | |
2003 | } | |
2004 | ||
2005 | static void bdrv_emit_qmp_eject_event(BlockDriverState *bs, bool ejected) | |
2006 | { | |
2007 | QObject *data; | |
2008 | ||
2009 | data = qobject_from_jsonf("{ 'device': %s, 'tray-open': %i }", | |
2010 | bdrv_get_device_name(bs), ejected); | |
2011 | monitor_protocol_event(QEVENT_DEVICE_TRAY_MOVED, data); | |
2012 | ||
2013 | qobject_decref(data); | |
2014 | } | |
2015 | ||
2016 | static void bdrv_dev_change_media_cb(BlockDriverState *bs, bool load) | |
2017 | { | |
2018 | if (bs->dev_ops && bs->dev_ops->change_media_cb) { | |
2019 | bool tray_was_closed = !bdrv_dev_is_tray_open(bs); | |
2020 | bs->dev_ops->change_media_cb(bs->dev_opaque, load); | |
2021 | if (tray_was_closed) { | |
2022 | /* tray open */ | |
2023 | bdrv_emit_qmp_eject_event(bs, true); | |
2024 | } | |
2025 | if (load) { | |
2026 | /* tray close */ | |
2027 | bdrv_emit_qmp_eject_event(bs, false); | |
2028 | } | |
2029 | } | |
2030 | } | |
2031 | ||
2032 | bool bdrv_dev_has_removable_media(BlockDriverState *bs) | |
2033 | { | |
2034 | return !bs->dev || (bs->dev_ops && bs->dev_ops->change_media_cb); | |
2035 | } | |
2036 | ||
2037 | void bdrv_dev_eject_request(BlockDriverState *bs, bool force) | |
2038 | { | |
2039 | if (bs->dev_ops && bs->dev_ops->eject_request_cb) { | |
2040 | bs->dev_ops->eject_request_cb(bs->dev_opaque, force); | |
2041 | } | |
2042 | } | |
2043 | ||
2044 | bool bdrv_dev_is_tray_open(BlockDriverState *bs) | |
2045 | { | |
2046 | if (bs->dev_ops && bs->dev_ops->is_tray_open) { | |
2047 | return bs->dev_ops->is_tray_open(bs->dev_opaque); | |
2048 | } | |
2049 | return false; | |
2050 | } | |
2051 | ||
2052 | static void bdrv_dev_resize_cb(BlockDriverState *bs) | |
2053 | { | |
2054 | if (bs->dev_ops && bs->dev_ops->resize_cb) { | |
2055 | bs->dev_ops->resize_cb(bs->dev_opaque); | |
2056 | } | |
2057 | } | |
2058 | ||
2059 | bool bdrv_dev_is_medium_locked(BlockDriverState *bs) | |
2060 | { | |
2061 | if (bs->dev_ops && bs->dev_ops->is_medium_locked) { | |
2062 | return bs->dev_ops->is_medium_locked(bs->dev_opaque); | |
2063 | } | |
2064 | return false; | |
2065 | } | |
2066 | ||
2067 | /* | |
2068 | * Run consistency checks on an image | |
2069 | * | |
2070 | * Returns 0 if the check could be completed (it doesn't mean that the image is | |
2071 | * free of errors) or -errno when an internal error occurred. The results of the | |
2072 | * check are stored in res. | |
2073 | */ | |
2074 | int bdrv_check(BlockDriverState *bs, BdrvCheckResult *res, BdrvCheckMode fix) | |
2075 | { | |
2076 | if (bs->drv->bdrv_check == NULL) { | |
2077 | return -ENOTSUP; | |
2078 | } | |
2079 | ||
2080 | memset(res, 0, sizeof(*res)); | |
2081 | return bs->drv->bdrv_check(bs, res, fix); | |
2082 | } | |
2083 | ||
2084 | #define COMMIT_BUF_SECTORS 2048 | |
2085 | ||
2086 | /* commit COW file into the raw image */ | |
2087 | int bdrv_commit(BlockDriverState *bs) | |
2088 | { | |
2089 | BlockDriver *drv = bs->drv; | |
2090 | int64_t sector, total_sectors, length, backing_length; | |
2091 | int n, ro, open_flags; | |
2092 | int ret = 0; | |
2093 | uint8_t *buf = NULL; | |
2094 | char filename[PATH_MAX]; | |
2095 | ||
2096 | if (!drv) | |
2097 | return -ENOMEDIUM; | |
2098 | ||
2099 | if (!bs->backing_hd) { | |
2100 | return -ENOTSUP; | |
2101 | } | |
2102 | ||
2103 | if (bdrv_in_use(bs) || bdrv_in_use(bs->backing_hd)) { | |
2104 | return -EBUSY; | |
2105 | } | |
2106 | ||
2107 | ro = bs->backing_hd->read_only; | |
2108 | /* Use pstrcpy (not strncpy): filename must be NUL-terminated. */ | |
2109 | pstrcpy(filename, sizeof(filename), bs->backing_hd->filename); | |
2110 | open_flags = bs->backing_hd->open_flags; | |
2111 | ||
2112 | if (ro) { | |
2113 | if (bdrv_reopen(bs->backing_hd, open_flags | BDRV_O_RDWR, NULL)) { | |
2114 | return -EACCES; | |
2115 | } | |
2116 | } | |
2117 | ||
2118 | length = bdrv_getlength(bs); | |
2119 | if (length < 0) { | |
2120 | ret = length; | |
2121 | goto ro_cleanup; | |
2122 | } | |
2123 | ||
2124 | backing_length = bdrv_getlength(bs->backing_hd); | |
2125 | if (backing_length < 0) { | |
2126 | ret = backing_length; | |
2127 | goto ro_cleanup; | |
2128 | } | |
2129 | ||
2130 | /* If our top snapshot is larger than the backing file image, | |
2131 | * grow the backing file image if possible. If not possible, | |
2132 | * we must return an error */ | |
2133 | if (length > backing_length) { | |
2134 | ret = bdrv_truncate(bs->backing_hd, length); | |
2135 | if (ret < 0) { | |
2136 | goto ro_cleanup; | |
2137 | } | |
2138 | } | |
2139 | ||
2140 | total_sectors = length >> BDRV_SECTOR_BITS; | |
2141 | buf = g_malloc(COMMIT_BUF_SECTORS * BDRV_SECTOR_SIZE); | |
2142 | ||
2143 | for (sector = 0; sector < total_sectors; sector += n) { | |
2144 | ret = bdrv_is_allocated(bs, sector, COMMIT_BUF_SECTORS, &n); | |
2145 | if (ret < 0) { | |
2146 | goto ro_cleanup; | |
2147 | } | |
2148 | if (ret) { | |
2149 | ret = bdrv_read(bs, sector, buf, n); | |
2150 | if (ret < 0) { | |
2151 | goto ro_cleanup; | |
2152 | } | |
2153 | ||
2154 | ret = bdrv_write(bs->backing_hd, sector, buf, n); | |
2155 | if (ret < 0) { | |
2156 | goto ro_cleanup; | |
2157 | } | |
2158 | } | |
2159 | } | |
2160 | ||
2161 | if (drv->bdrv_make_empty) { | |
2162 | ret = drv->bdrv_make_empty(bs); | |
2163 | if (ret < 0) { | |
2164 | goto ro_cleanup; | |
2165 | } | |
2166 | bdrv_flush(bs); | |
2167 | } | |
2168 | ||
2169 | /* | |
2170 | * Make sure all data we wrote to the backing device is actually | |
2171 | * stable on disk. | |
2172 | */ | |
2173 | if (bs->backing_hd) { | |
2174 | bdrv_flush(bs->backing_hd); | |
2175 | } | |
2176 | ||
2177 | ret = 0; | |
2178 | ro_cleanup: | |
2179 | g_free(buf); | |
2180 | ||
2181 | if (ro) { | |
2182 | /* ignoring error return here */ | |
2183 | bdrv_reopen(bs->backing_hd, open_flags & ~BDRV_O_RDWR, NULL); | |
2184 | } | |
2185 | ||
2186 | return ret; | |
2187 | } | |
2188 | ||
2189 | int bdrv_commit_all(void) | |
2190 | { | |
2191 | BlockDriverState *bs; | |
2192 | ||
2193 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
2194 | if (bs->drv && bs->backing_hd) { | |
2195 | int ret = bdrv_commit(bs); | |
2196 | if (ret < 0) { | |
2197 | return ret; | |
2198 | } | |
2199 | } | |
2200 | } | |
2201 | return 0; | |
2202 | } | |
2203 | ||
2204 | /** | |
2205 | * Remove an active request from the tracked requests list | |
2206 | * | |
2207 | * This function should be called when a tracked request is completing. | |
2208 | */ | |
2209 | static void tracked_request_end(BdrvTrackedRequest *req) | |
2210 | { | |
2211 | QLIST_REMOVE(req, list); | |
2212 | qemu_co_queue_restart_all(&req->wait_queue); | |
2213 | } | |
2214 | ||
2215 | /** | |
2216 | * Add an active request to the tracked requests list | |
2217 | */ | |
2218 | static void tracked_request_begin(BdrvTrackedRequest *req, | |
2219 | BlockDriverState *bs, | |
2220 | int64_t sector_num, | |
2221 | int nb_sectors, bool is_write) | |
2222 | { | |
2223 | *req = (BdrvTrackedRequest){ | |
2224 | .bs = bs, | |
2225 | .sector_num = sector_num, | |
2226 | .nb_sectors = nb_sectors, | |
2227 | .is_write = is_write, | |
2228 | .co = qemu_coroutine_self(), | |
2229 | }; | |
2230 | ||
2231 | qemu_co_queue_init(&req->wait_queue); | |
2232 | ||
2233 | QLIST_INSERT_HEAD(&bs->tracked_requests, req, list); | |
2234 | } | |
2235 | ||
2236 | /** | |
2237 | * Round a region to cluster boundaries | |
2238 | */ | |
2239 | void bdrv_round_to_clusters(BlockDriverState *bs, | |
2240 | int64_t sector_num, int nb_sectors, | |
2241 | int64_t *cluster_sector_num, | |
2242 | int *cluster_nb_sectors) | |
2243 | { | |
2244 | BlockDriverInfo bdi; | |
2245 | ||
2246 | if (bdrv_get_info(bs, &bdi) < 0 || bdi.cluster_size == 0) { | |
2247 | *cluster_sector_num = sector_num; | |
2248 | *cluster_nb_sectors = nb_sectors; | |
2249 | } else { | |
2250 | int64_t c = bdi.cluster_size / BDRV_SECTOR_SIZE; | |
2251 | *cluster_sector_num = QEMU_ALIGN_DOWN(sector_num, c); | |
2252 | *cluster_nb_sectors = QEMU_ALIGN_UP(sector_num - *cluster_sector_num + | |
2253 | nb_sectors, c); | |
2254 | } | |
2255 | } | |
2256 | ||
2257 | static bool tracked_request_overlaps(BdrvTrackedRequest *req, | |
2258 | int64_t sector_num, int nb_sectors) { | |
2259 | /* aaaa bbbb */ | |
2260 | if (sector_num >= req->sector_num + req->nb_sectors) { | |
2261 | return false; | |
2262 | } | |
2263 | /* bbbb aaaa */ | |
2264 | if (req->sector_num >= sector_num + nb_sectors) { | |
2265 | return false; | |
2266 | } | |
2267 | return true; | |
2268 | } | |
2269 | ||
2270 | static void coroutine_fn wait_for_overlapping_requests(BlockDriverState *bs, | |
2271 | int64_t sector_num, int nb_sectors) | |
2272 | { | |
2273 | BdrvTrackedRequest *req; | |
2274 | int64_t cluster_sector_num; | |
2275 | int cluster_nb_sectors; | |
2276 | bool retry; | |
2277 | ||
2278 | /* If we touch the same cluster it counts as an overlap. This guarantees | |
2279 | * that allocating writes will be serialized and not race with each other | |
2280 | * for the same cluster. For example, in copy-on-read it ensures that the | |
2281 | * CoR read and write operations are atomic and guest writes cannot | |
2282 | * interleave between them. | |
2283 | */ | |
2284 | bdrv_round_to_clusters(bs, sector_num, nb_sectors, | |
2285 | &cluster_sector_num, &cluster_nb_sectors); | |
2286 | ||
2287 | do { | |
2288 | retry = false; | |
2289 | QLIST_FOREACH(req, &bs->tracked_requests, list) { | |
2290 | if (tracked_request_overlaps(req, cluster_sector_num, | |
2291 | cluster_nb_sectors)) { | |
2292 | /* Hitting this means there was a reentrant request, for | |
2293 | * example, a block driver issuing nested requests. This must | |
2294 | * never happen since it means deadlock. | |
2295 | */ | |
2296 | assert(qemu_coroutine_self() != req->co); | |
2297 | ||
2298 | qemu_co_queue_wait(&req->wait_queue); | |
2299 | retry = true; | |
2300 | break; | |
2301 | } | |
2302 | } | |
2303 | } while (retry); | |
2304 | } | |
2305 | ||
2306 | /* | |
2307 | * Return values: | |
2308 | * 0 - success | |
2309 | * -EINVAL - backing format specified, but no file | |
2310 | * -ENOSPC - can't update the backing file because no space is left in the | |
2311 | * image file header | |
2312 | * -ENOTSUP - format driver doesn't support changing the backing file | |
2313 | */ | |
2314 | int bdrv_change_backing_file(BlockDriverState *bs, | |
2315 | const char *backing_file, const char *backing_fmt) | |
2316 | { | |
2317 | BlockDriver *drv = bs->drv; | |
2318 | int ret; | |
2319 | ||
2320 | /* Backing file format doesn't make sense without a backing file */ | |
2321 | if (backing_fmt && !backing_file) { | |
2322 | return -EINVAL; | |
2323 | } | |
2324 | ||
2325 | if (drv->bdrv_change_backing_file != NULL) { | |
2326 | ret = drv->bdrv_change_backing_file(bs, backing_file, backing_fmt); | |
2327 | } else { | |
2328 | ret = -ENOTSUP; | |
2329 | } | |
2330 | ||
2331 | if (ret == 0) { | |
2332 | pstrcpy(bs->backing_file, sizeof(bs->backing_file), backing_file ?: ""); | |
2333 | pstrcpy(bs->backing_format, sizeof(bs->backing_format), backing_fmt ?: ""); | |
2334 | } | |
2335 | return ret; | |
2336 | } | |
2337 | ||
2338 | /* | |
2339 | * Finds the image layer in the chain that has 'bs' as its backing file. | |
2340 | * | |
2341 | * active is the current topmost image. | |
2342 | * | |
2343 | * Returns NULL if bs is not found in active's image chain, | |
2344 | * or if active == bs. | |
2345 | */ | |
2346 | BlockDriverState *bdrv_find_overlay(BlockDriverState *active, | |
2347 | BlockDriverState *bs) | |
2348 | { | |
2349 | BlockDriverState *overlay = NULL; | |
2350 | BlockDriverState *intermediate; | |
2351 | ||
2352 | assert(active != NULL); | |
2353 | assert(bs != NULL); | |
2354 | ||
2355 | /* if bs is the same as active, then by definition it has no overlay | |
2356 | */ | |
2357 | if (active == bs) { | |
2358 | return NULL; | |
2359 | } | |
2360 | ||
2361 | intermediate = active; | |
2362 | while (intermediate->backing_hd) { | |
2363 | if (intermediate->backing_hd == bs) { | |
2364 | overlay = intermediate; | |
2365 | break; | |
2366 | } | |
2367 | intermediate = intermediate->backing_hd; | |
2368 | } | |
2369 | ||
2370 | return overlay; | |
2371 | } | |
2372 | ||
2373 | typedef struct BlkIntermediateStates { | |
2374 | BlockDriverState *bs; | |
2375 | QSIMPLEQ_ENTRY(BlkIntermediateStates) entry; | |
2376 | } BlkIntermediateStates; | |
2377 | ||
2378 | ||
2379 | /* | |
2380 | * Drops images above 'base' up to and including 'top', and sets the image | |
2381 | * above 'top' to have base as its backing file. | |
2382 | * | |
2383 | * Requires that the overlay to 'top' is opened r/w, so that the backing file | |
2384 | * information in 'bs' can be properly updated. | |
2385 | * | |
2386 | * E.g., this will convert the following chain: | |
2387 | * bottom <- base <- intermediate <- top <- active | |
2388 | * | |
2389 | * to | |
2390 | * | |
2391 | * bottom <- base <- active | |
2392 | * | |
2393 | * It is allowed for bottom==base, in which case it converts: | |
2394 | * | |
2395 | * base <- intermediate <- top <- active | |
2396 | * | |
2397 | * to | |
2398 | * | |
2399 | * base <- active | |
2400 | * | |
2401 | * Error conditions: | |
2402 | * if active == top, that is considered an error | |
2403 | * | |
2404 | */ | |
2405 | int bdrv_drop_intermediate(BlockDriverState *active, BlockDriverState *top, | |
2406 | BlockDriverState *base) | |
2407 | { | |
2408 | BlockDriverState *intermediate; | |
2409 | BlockDriverState *base_bs = NULL; | |
2410 | BlockDriverState *new_top_bs = NULL; | |
2411 | BlkIntermediateStates *intermediate_state, *next; | |
2412 | int ret = -EIO; | |
2413 | ||
2414 | QSIMPLEQ_HEAD(states_to_delete, BlkIntermediateStates) states_to_delete; | |
2415 | QSIMPLEQ_INIT(&states_to_delete); | |
2416 | ||
2417 | if (!top->drv || !base->drv) { | |
2418 | goto exit; | |
2419 | } | |
2420 | ||
2421 | new_top_bs = bdrv_find_overlay(active, top); | |
2422 | ||
2423 | if (new_top_bs == NULL) { | |
2424 | /* we could not find the image above 'top', this is an error */ | |
2425 | goto exit; | |
2426 | } | |
2427 | ||
2428 | /* special case of new_top_bs->backing_hd already pointing to base - nothing | |
2429 | * to do, no intermediate images */ | |
2430 | if (new_top_bs->backing_hd == base) { | |
2431 | ret = 0; | |
2432 | goto exit; | |
2433 | } | |
2434 | ||
2435 | intermediate = top; | |
2436 | ||
2437 | /* now we will go down through the list, and add each BDS we find | |
2438 | * into our deletion queue, until we hit the 'base' | |
2439 | */ | |
2440 | while (intermediate) { | |
2441 | intermediate_state = g_malloc0(sizeof(BlkIntermediateStates)); | |
2442 | intermediate_state->bs = intermediate; | |
2443 | QSIMPLEQ_INSERT_TAIL(&states_to_delete, intermediate_state, entry); | |
2444 | ||
2445 | if (intermediate->backing_hd == base) { | |
2446 | base_bs = intermediate->backing_hd; | |
2447 | break; | |
2448 | } | |
2449 | intermediate = intermediate->backing_hd; | |
2450 | } | |
2451 | if (base_bs == NULL) { | |
2452 | /* something went wrong, we did not end at the base. safely | |
2453 | * unravel everything, and exit with error */ | |
2454 | goto exit; | |
2455 | } | |
2456 | ||
2457 | /* success - we can delete the intermediate states, and link top->base */ | |
2458 | ret = bdrv_change_backing_file(new_top_bs, base_bs->filename, | |
2459 | base_bs->drv ? base_bs->drv->format_name : ""); | |
2460 | if (ret) { | |
2461 | goto exit; | |
2462 | } | |
2463 | new_top_bs->backing_hd = base_bs; | |
2464 | ||
2465 | bdrv_refresh_limits(new_top_bs); | |
2466 | ||
2467 | QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) { | |
2468 | /* so that bdrv_close() does not recursively close the chain */ | |
2469 | intermediate_state->bs->backing_hd = NULL; | |
2470 | bdrv_unref(intermediate_state->bs); | |
2471 | } | |
2472 | ret = 0; | |
2473 | ||
2474 | exit: | |
2475 | QSIMPLEQ_FOREACH_SAFE(intermediate_state, &states_to_delete, entry, next) { | |
2476 | g_free(intermediate_state); | |
2477 | } | |
2478 | return ret; | |
2479 | } | |
2480 | ||
2481 | ||
2482 | static int bdrv_check_byte_request(BlockDriverState *bs, int64_t offset, | |
2483 | size_t size) | |
2484 | { | |
2485 | int64_t len; | |
2486 | ||
2487 | if (!bdrv_is_inserted(bs)) | |
2488 | return -ENOMEDIUM; | |
2489 | ||
2490 | if (bs->growable) | |
2491 | return 0; | |
2492 | ||
2493 | len = bdrv_getlength(bs); | |
2494 | ||
2495 | if (offset < 0) | |
2496 | return -EIO; | |
2497 | ||
2498 | if ((offset > len) || (len - offset < size)) | |
2499 | return -EIO; | |
2500 | ||
2501 | return 0; | |
2502 | } | |
2503 | ||
2504 | static int bdrv_check_request(BlockDriverState *bs, int64_t sector_num, | |
2505 | int nb_sectors) | |
2506 | { | |
2507 | return bdrv_check_byte_request(bs, sector_num * BDRV_SECTOR_SIZE, | |
2508 | nb_sectors * BDRV_SECTOR_SIZE); | |
2509 | } | |
2510 | ||
2511 | typedef struct RwCo { | |
2512 | BlockDriverState *bs; | |
2513 | int64_t sector_num; | |
2514 | int nb_sectors; | |
2515 | QEMUIOVector *qiov; | |
2516 | bool is_write; | |
2517 | int ret; | |
2518 | BdrvRequestFlags flags; | |
2519 | } RwCo; | |
2520 | ||
2521 | static void coroutine_fn bdrv_rw_co_entry(void *opaque) | |
2522 | { | |
2523 | RwCo *rwco = opaque; | |
2524 | ||
2525 | if (!rwco->is_write) { | |
2526 | rwco->ret = bdrv_co_do_readv(rwco->bs, rwco->sector_num, | |
2527 | rwco->nb_sectors, rwco->qiov, | |
2528 | rwco->flags); | |
2529 | } else { | |
2530 | rwco->ret = bdrv_co_do_writev(rwco->bs, rwco->sector_num, | |
2531 | rwco->nb_sectors, rwco->qiov, | |
2532 | rwco->flags); | |
2533 | } | |
2534 | } | |
2535 | ||
2536 | /* | |
2537 | * Process a vectored synchronous request using coroutines | |
2538 | */ | |
2539 | static int bdrv_rwv_co(BlockDriverState *bs, int64_t sector_num, | |
2540 | QEMUIOVector *qiov, bool is_write, | |
2541 | BdrvRequestFlags flags) | |
2542 | { | |
2543 | Coroutine *co; | |
2544 | RwCo rwco = { | |
2545 | .bs = bs, | |
2546 | .sector_num = sector_num, | |
2547 | .nb_sectors = qiov->size >> BDRV_SECTOR_BITS, | |
2548 | .qiov = qiov, | |
2549 | .is_write = is_write, | |
2550 | .ret = NOT_DONE, | |
2551 | .flags = flags, | |
2552 | }; | |
2553 | assert((qiov->size & (BDRV_SECTOR_SIZE - 1)) == 0); | |
2554 | ||
2555 | /** | |
2556 | * In sync call context, when the vcpu is blocked, this throttling timer | |
2557 | * will not fire; so the I/O throttling function has to be disabled here | |
2558 | * if it has been enabled. | |
2559 | */ | |
2560 | if (bs->io_limits_enabled) { | |
2561 | fprintf(stderr, "Disabling I/O throttling on '%s' due " | |
2562 | "to synchronous I/O.\n", bdrv_get_device_name(bs)); | |
2563 | bdrv_io_limits_disable(bs); | |
2564 | } | |
2565 | ||
2566 | if (qemu_in_coroutine()) { | |
2567 | /* Fast-path if already in coroutine context */ | |
2568 | bdrv_rw_co_entry(&rwco); | |
2569 | } else { | |
2570 | co = qemu_coroutine_create(bdrv_rw_co_entry); | |
2571 | qemu_coroutine_enter(co, &rwco); | |
2572 | while (rwco.ret == NOT_DONE) { | |
2573 | qemu_aio_wait(); | |
2574 | } | |
2575 | } | |
2576 | return rwco.ret; | |
2577 | } | |
2578 | ||
2579 | /* | |
2580 | * Process a synchronous request using coroutines | |
2581 | */ | |
2582 | static int bdrv_rw_co(BlockDriverState *bs, int64_t sector_num, uint8_t *buf, | |
2583 | int nb_sectors, bool is_write, BdrvRequestFlags flags) | |
2584 | { | |
2585 | QEMUIOVector qiov; | |
2586 | struct iovec iov = { | |
2587 | .iov_base = (void *)buf, | |
2588 | .iov_len = nb_sectors * BDRV_SECTOR_SIZE, | |
2589 | }; | |
2590 | ||
2591 | qemu_iovec_init_external(&qiov, &iov, 1); | |
2592 | return bdrv_rwv_co(bs, sector_num, &qiov, is_write, flags); | |
2593 | } | |
2594 | ||
2595 | /* return < 0 if error. See bdrv_write() for the return codes */ | |
2596 | int bdrv_read(BlockDriverState *bs, int64_t sector_num, | |
2597 | uint8_t *buf, int nb_sectors) | |
2598 | { | |
2599 | return bdrv_rw_co(bs, sector_num, buf, nb_sectors, false, 0); | |
2600 | } | |
2601 | ||
2602 | /* Just like bdrv_read(), but with I/O throttling temporarily disabled */ | |
2603 | int bdrv_read_unthrottled(BlockDriverState *bs, int64_t sector_num, | |
2604 | uint8_t *buf, int nb_sectors) | |
2605 | { | |
2606 | bool enabled; | |
2607 | int ret; | |
2608 | ||
2609 | enabled = bs->io_limits_enabled; | |
2610 | bs->io_limits_enabled = false; | |
2611 | ret = bdrv_read(bs, sector_num, buf, nb_sectors); | |
2612 | bs->io_limits_enabled = enabled; | |
2613 | return ret; | |
2614 | } | |
2615 | ||
2616 | /* Return < 0 if error. Important errors are: | |
2617 | -EIO generic I/O error (may happen for all errors) | |
2618 | -ENOMEDIUM No media inserted. | |
2619 | -EINVAL Invalid sector number or nb_sectors | |
2620 | -EACCES Trying to write a read-only device | |
2621 | */ | |
2622 | int bdrv_write(BlockDriverState *bs, int64_t sector_num, | |
2623 | const uint8_t *buf, int nb_sectors) | |
2624 | { | |
2625 | return bdrv_rw_co(bs, sector_num, (uint8_t *)buf, nb_sectors, true, 0); | |
2626 | } | |
2627 | ||
2628 | int bdrv_writev(BlockDriverState *bs, int64_t sector_num, QEMUIOVector *qiov) | |
2629 | { | |
2630 | return bdrv_rwv_co(bs, sector_num, qiov, true, 0); | |
2631 | } | |
2632 | ||
2633 | int bdrv_write_zeroes(BlockDriverState *bs, int64_t sector_num, | |
2634 | int nb_sectors, BdrvRequestFlags flags) | |
2635 | { | |
2636 | return bdrv_rw_co(bs, sector_num, NULL, nb_sectors, true, | |
2637 | BDRV_REQ_ZERO_WRITE | flags); | |
2638 | } | |
2639 | ||
2640 | /* | |
2641 | * Completely zero out a block device with the help of bdrv_write_zeroes. | |
2642 | * The operation is sped up by checking the block status and only writing | |
2643 | * zeroes to the device if they currently do not return zeroes. Optional | |
2644 | * flags are passed through to bdrv_write_zeroes (e.g. BDRV_REQ_MAY_UNMAP). | |
2645 | * | |
2646 | * Returns < 0 on error, 0 on success. For error codes see bdrv_write(). | |
2647 | */ | |
2648 | int bdrv_make_zero(BlockDriverState *bs, BdrvRequestFlags flags) | |
2649 | { | |
2650 | int64_t target_size = bdrv_getlength(bs) / BDRV_SECTOR_SIZE; | |
2651 | int64_t ret, nb_sectors, sector_num = 0; | |
2652 | int n; | |
2653 | ||
2654 | for (;;) { | |
2655 | nb_sectors = target_size - sector_num; | |
2656 | if (nb_sectors <= 0) { | |
2657 | return 0; | |
2658 | } | |
2659 | if (nb_sectors > INT_MAX) { | |
2660 | nb_sectors = INT_MAX; | |
2661 | } | |
2662 | ret = bdrv_get_block_status(bs, sector_num, nb_sectors, &n); | |
2663 | if (ret < 0) { | |
2664 | error_report("error getting block status at sector %" PRId64 ": %s", | |
2665 | sector_num, strerror(-ret)); | |
2666 | return ret; | |
2667 | } | |
2668 | if (ret & BDRV_BLOCK_ZERO) { | |
2669 | sector_num += n; | |
2670 | continue; | |
2671 | } | |
2672 | ret = bdrv_write_zeroes(bs, sector_num, n, flags); | |
2673 | if (ret < 0) { | |
2674 | error_report("error writing zeroes at sector %" PRId64 ": %s", | |
2675 | sector_num, strerror(-ret)); | |
2676 | return ret; | |
2677 | } | |
2678 | sector_num += n; | |
2679 | } | |
2680 | } | |
2681 | ||
2682 | int bdrv_pread(BlockDriverState *bs, int64_t offset, | |
2683 | void *buf, int count1) | |
2684 | { | |
2685 | uint8_t tmp_buf[BDRV_SECTOR_SIZE]; | |
2686 | int len, nb_sectors, count; | |
2687 | int64_t sector_num; | |
2688 | int ret; | |
2689 | ||
2690 | count = count1; | |
2691 | /* first read to align to sector start */ | |
2692 | len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1); | |
2693 | if (len > count) | |
2694 | len = count; | |
2695 | sector_num = offset >> BDRV_SECTOR_BITS; | |
2696 | if (len > 0) { | |
2697 | if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) | |
2698 | return ret; | |
2699 | memcpy(buf, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), len); | |
2700 | count -= len; | |
2701 | if (count == 0) | |
2702 | return count1; | |
2703 | sector_num++; | |
2704 | buf += len; | |
2705 | } | |
2706 | ||
2707 | /* read the sectors "in place" */ | |
2708 | nb_sectors = count >> BDRV_SECTOR_BITS; | |
2709 | if (nb_sectors > 0) { | |
2710 | if ((ret = bdrv_read(bs, sector_num, buf, nb_sectors)) < 0) | |
2711 | return ret; | |
2712 | sector_num += nb_sectors; | |
2713 | len = nb_sectors << BDRV_SECTOR_BITS; | |
2714 | buf += len; | |
2715 | count -= len; | |
2716 | } | |
2717 | ||
2718 | /* add data from the last sector */ | |
2719 | if (count > 0) { | |
2720 | if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) | |
2721 | return ret; | |
2722 | memcpy(buf, tmp_buf, count); | |
2723 | } | |
2724 | return count1; | |
2725 | } | |
2726 | ||
2727 | int bdrv_pwritev(BlockDriverState *bs, int64_t offset, QEMUIOVector *qiov) | |
2728 | { | |
2729 | uint8_t tmp_buf[BDRV_SECTOR_SIZE]; | |
2730 | int len, nb_sectors, count; | |
2731 | int64_t sector_num; | |
2732 | int ret; | |
2733 | ||
2734 | count = qiov->size; | |
2735 | ||
2736 | /* first write to align to sector start */ | |
2737 | len = (BDRV_SECTOR_SIZE - offset) & (BDRV_SECTOR_SIZE - 1); | |
2738 | if (len > count) | |
2739 | len = count; | |
2740 | sector_num = offset >> BDRV_SECTOR_BITS; | |
2741 | if (len > 0) { | |
2742 | if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) | |
2743 | return ret; | |
2744 | qemu_iovec_to_buf(qiov, 0, tmp_buf + (offset & (BDRV_SECTOR_SIZE - 1)), | |
2745 | len); | |
2746 | if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) | |
2747 | return ret; | |
2748 | count -= len; | |
2749 | if (count == 0) | |
2750 | return qiov->size; | |
2751 | sector_num++; | |
2752 | } | |
2753 | ||
2754 | /* write the sectors "in place" */ | |
2755 | nb_sectors = count >> BDRV_SECTOR_BITS; | |
2756 | if (nb_sectors > 0) { | |
2757 | QEMUIOVector qiov_inplace; | |
2758 | ||
2759 | qemu_iovec_init(&qiov_inplace, qiov->niov); | |
2760 | qemu_iovec_concat(&qiov_inplace, qiov, len, | |
2761 | nb_sectors << BDRV_SECTOR_BITS); | |
2762 | ret = bdrv_writev(bs, sector_num, &qiov_inplace); | |
2763 | qemu_iovec_destroy(&qiov_inplace); | |
2764 | if (ret < 0) { | |
2765 | return ret; | |
2766 | } | |
2767 | ||
2768 | sector_num += nb_sectors; | |
2769 | len = nb_sectors << BDRV_SECTOR_BITS; | |
2770 | count -= len; | |
2771 | } | |
2772 | ||
2773 | /* add data from the last sector */ | |
2774 | if (count > 0) { | |
2775 | if ((ret = bdrv_read(bs, sector_num, tmp_buf, 1)) < 0) | |
2776 | return ret; | |
2777 | qemu_iovec_to_buf(qiov, qiov->size - count, tmp_buf, count); | |
2778 | if ((ret = bdrv_write(bs, sector_num, tmp_buf, 1)) < 0) | |
2779 | return ret; | |
2780 | } | |
2781 | return qiov->size; | |
2782 | } | |
2783 | ||
2784 | int bdrv_pwrite(BlockDriverState *bs, int64_t offset, | |
2785 | const void *buf, int count1) | |
2786 | { | |
2787 | QEMUIOVector qiov; | |
2788 | struct iovec iov = { | |
2789 | .iov_base = (void *) buf, | |
2790 | .iov_len = count1, | |
2791 | }; | |
2792 | ||
2793 | qemu_iovec_init_external(&qiov, &iov, 1); | |
2794 | return bdrv_pwritev(bs, offset, &qiov); | |
2795 | } | |
2796 | ||
2797 | /* | |
2798 | * Writes to the file and ensures that no writes are reordered across this | |
2799 | * request (acts as a barrier) | |
2800 | * | |
2801 | * Returns 0 on success, -errno in error cases. | |
2802 | */ | |
2803 | int bdrv_pwrite_sync(BlockDriverState *bs, int64_t offset, | |
2804 | const void *buf, int count) | |
2805 | { | |
2806 | int ret; | |
2807 | ||
2808 | ret = bdrv_pwrite(bs, offset, buf, count); | |
2809 | if (ret < 0) { | |
2810 | return ret; | |
2811 | } | |
2812 | ||
2813 | /* No flush needed for cache modes that already do it */ | |
2814 | if (bs->enable_write_cache) { | |
2815 | bdrv_flush(bs); | |
2816 | } | |
2817 | ||
2818 | return 0; | |
2819 | } | |
2820 | ||
2821 | static int coroutine_fn bdrv_co_do_copy_on_readv(BlockDriverState *bs, | |
2822 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) | |
2823 | { | |
2824 | /* Perform I/O through a temporary buffer so that users who scribble over | |
2825 | * their read buffer while the operation is in progress do not end up | |
2826 | * modifying the image file. This is critical for zero-copy guest I/O | |
2827 | * where anything might happen inside guest memory. | |
2828 | */ | |
2829 | void *bounce_buffer; | |
2830 | ||
2831 | BlockDriver *drv = bs->drv; | |
2832 | struct iovec iov; | |
2833 | QEMUIOVector bounce_qiov; | |
2834 | int64_t cluster_sector_num; | |
2835 | int cluster_nb_sectors; | |
2836 | size_t skip_bytes; | |
2837 | int ret; | |
2838 | ||
2839 | /* Cover entire cluster so no additional backing file I/O is required when | |
2840 | * allocating cluster in the image file. | |
2841 | */ | |
2842 | bdrv_round_to_clusters(bs, sector_num, nb_sectors, | |
2843 | &cluster_sector_num, &cluster_nb_sectors); | |
2844 | ||
2845 | trace_bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, | |
2846 | cluster_sector_num, cluster_nb_sectors); | |
2847 | ||
2848 | iov.iov_len = cluster_nb_sectors * BDRV_SECTOR_SIZE; | |
2849 | iov.iov_base = bounce_buffer = qemu_blockalign(bs, iov.iov_len); | |
2850 | qemu_iovec_init_external(&bounce_qiov, &iov, 1); | |
2851 | ||
2852 | ret = drv->bdrv_co_readv(bs, cluster_sector_num, cluster_nb_sectors, | |
2853 | &bounce_qiov); | |
2854 | if (ret < 0) { | |
2855 | goto err; | |
2856 | } | |
2857 | ||
2858 | if (drv->bdrv_co_write_zeroes && | |
2859 | buffer_is_zero(bounce_buffer, iov.iov_len)) { | |
2860 | ret = bdrv_co_do_write_zeroes(bs, cluster_sector_num, | |
2861 | cluster_nb_sectors, 0); | |
2862 | } else { | |
2863 | /* This does not change the data on the disk, it is not necessary | |
2864 | * to flush even in cache=writethrough mode. | |
2865 | */ | |
2866 | ret = drv->bdrv_co_writev(bs, cluster_sector_num, cluster_nb_sectors, | |
2867 | &bounce_qiov); | |
2868 | } | |
2869 | ||
2870 | if (ret < 0) { | |
2871 | /* It might be okay to ignore write errors for guest requests. If this | |
2872 | * is a deliberate copy-on-read then we don't want to ignore the error. | |
2873 | * Simply report it in all cases. | |
2874 | */ | |
2875 | goto err; | |
2876 | } | |
2877 | ||
2878 | skip_bytes = (sector_num - cluster_sector_num) * BDRV_SECTOR_SIZE; | |
2879 | qemu_iovec_from_buf(qiov, 0, bounce_buffer + skip_bytes, | |
2880 | nb_sectors * BDRV_SECTOR_SIZE); | |
2881 | ||
2882 | err: | |
2883 | qemu_vfree(bounce_buffer); | |
2884 | return ret; | |
2885 | } | |
2886 | ||
2887 | /* | |
2888 | * Forwards an already correctly aligned request to the BlockDriver. This | |
2889 | * handles copy on read and zeroing after EOF; any other features must be | |
2890 | * implemented by the caller. | |
2891 | */ | |
2892 | static int coroutine_fn bdrv_aligned_preadv(BlockDriverState *bs, | |
2893 | int64_t offset, unsigned int bytes, QEMUIOVector *qiov, int flags) | |
2894 | { | |
2895 | BlockDriver *drv = bs->drv; | |
2896 | BdrvTrackedRequest req; | |
2897 | int ret; | |
2898 | ||
2899 | int64_t sector_num = offset >> BDRV_SECTOR_BITS; | |
2900 | unsigned int nb_sectors = bytes >> BDRV_SECTOR_BITS; | |
2901 | ||
2902 | assert((offset & (BDRV_SECTOR_SIZE - 1)) == 0); | |
2903 | assert((bytes & (BDRV_SECTOR_SIZE - 1)) == 0); | |
2904 | ||
2905 | /* Handle Copy on Read and associated serialisation */ | |
2906 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
2907 | bs->copy_on_read_in_flight++; | |
2908 | } | |
2909 | ||
2910 | if (bs->copy_on_read_in_flight) { | |
2911 | wait_for_overlapping_requests(bs, sector_num, nb_sectors); | |
2912 | } | |
2913 | ||
2914 | tracked_request_begin(&req, bs, sector_num, nb_sectors, false); | |
2915 | ||
2916 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
2917 | int pnum; | |
2918 | ||
2919 | ret = bdrv_is_allocated(bs, sector_num, nb_sectors, &pnum); | |
2920 | if (ret < 0) { | |
2921 | goto out; | |
2922 | } | |
2923 | ||
2924 | if (!ret || pnum != nb_sectors) { | |
2925 | ret = bdrv_co_do_copy_on_readv(bs, sector_num, nb_sectors, qiov); | |
2926 | goto out; | |
2927 | } | |
2928 | } | |
2929 | ||
2930 | /* Forward the request to the BlockDriver */ | |
2931 | if (!(bs->zero_beyond_eof && bs->growable)) { | |
2932 | ret = drv->bdrv_co_readv(bs, sector_num, nb_sectors, qiov); | |
2933 | } else { | |
2934 | /* Read zeros after EOF of growable BDSes */ | |
2935 | int64_t len, total_sectors, max_nb_sectors; | |
2936 | ||
2937 | len = bdrv_getlength(bs); | |
2938 | if (len < 0) { | |
2939 | ret = len; | |
2940 | goto out; | |
2941 | } | |
2942 | ||
2943 | total_sectors = DIV_ROUND_UP(len, BDRV_SECTOR_SIZE); | |
2944 | max_nb_sectors = MAX(0, total_sectors - sector_num); | |
2945 | if (max_nb_sectors > 0) { | |
2946 | ret = drv->bdrv_co_readv(bs, sector_num, | |
2947 | MIN(nb_sectors, max_nb_sectors), qiov); | |
2948 | } else { | |
2949 | ret = 0; | |
2950 | } | |
2951 | ||
2952 | /* Reading beyond end of file is supposed to produce zeroes */ | |
2953 | if (ret == 0 && total_sectors < sector_num + nb_sectors) { | |
2954 | uint64_t offset = MAX(0, total_sectors - sector_num); | |
2955 | uint64_t bytes = (sector_num + nb_sectors - offset) * | |
2956 | BDRV_SECTOR_SIZE; | |
2957 | qemu_iovec_memset(qiov, offset * BDRV_SECTOR_SIZE, 0, bytes); | |
2958 | } | |
2959 | } | |
2960 | ||
2961 | out: | |
2962 | tracked_request_end(&req); | |
2963 | ||
2964 | if (flags & BDRV_REQ_COPY_ON_READ) { | |
2965 | bs->copy_on_read_in_flight--; | |
2966 | } | |
2967 | ||
2968 | return ret; | |
2969 | } | |
2970 | ||
2971 | /* | |
2972 | * Handle a read request in coroutine context | |
2973 | */ | |
2974 | static int coroutine_fn bdrv_co_do_readv(BlockDriverState *bs, | |
2975 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
2976 | BdrvRequestFlags flags) | |
2977 | { | |
2978 | BlockDriver *drv = bs->drv; | |
2979 | int ret; | |
2980 | ||
2981 | if (!drv) { | |
2982 | return -ENOMEDIUM; | |
2983 | } | |
2984 | if (bdrv_check_request(bs, sector_num, nb_sectors)) { | |
2985 | return -EIO; | |
2986 | } | |
2987 | ||
2988 | if (bs->copy_on_read) { | |
2989 | flags |= BDRV_REQ_COPY_ON_READ; | |
2990 | } | |
2991 | ||
2992 | /* throttling disk I/O */ | |
2993 | if (bs->io_limits_enabled) { | |
2994 | bdrv_io_limits_intercept(bs, nb_sectors, false); | |
2995 | } | |
2996 | ||
2997 | ret = bdrv_aligned_preadv(bs, sector_num << BDRV_SECTOR_BITS, | |
2998 | nb_sectors << BDRV_SECTOR_BITS, qiov, flags); | |
2999 | return ret; | |
3000 | } | |
3001 | ||
3002 | int coroutine_fn bdrv_co_readv(BlockDriverState *bs, int64_t sector_num, | |
3003 | int nb_sectors, QEMUIOVector *qiov) | |
3004 | { | |
3005 | trace_bdrv_co_readv(bs, sector_num, nb_sectors); | |
3006 | ||
3007 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, 0); | |
3008 | } | |
3009 | ||
3010 | int coroutine_fn bdrv_co_copy_on_readv(BlockDriverState *bs, | |
3011 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov) | |
3012 | { | |
3013 | trace_bdrv_co_copy_on_readv(bs, sector_num, nb_sectors); | |
3014 | ||
3015 | return bdrv_co_do_readv(bs, sector_num, nb_sectors, qiov, | |
3016 | BDRV_REQ_COPY_ON_READ); | |
3017 | } | |
3018 | ||
3019 | /* if no limit is specified in the BlockLimits use a default | |
3020 | * of 32768 512-byte sectors (16 MiB) per request. | |
3021 | */ | |
3022 | #define MAX_WRITE_ZEROES_DEFAULT 32768 | |
3023 | ||
3024 | static int coroutine_fn bdrv_co_do_write_zeroes(BlockDriverState *bs, | |
3025 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags) | |
3026 | { | |
3027 | BlockDriver *drv = bs->drv; | |
3028 | QEMUIOVector qiov; | |
3029 | struct iovec iov = {0}; | |
3030 | int ret = 0; | |
3031 | ||
3032 | int max_write_zeroes = bs->bl.max_write_zeroes ? | |
3033 | bs->bl.max_write_zeroes : MAX_WRITE_ZEROES_DEFAULT; | |
3034 | ||
3035 | while (nb_sectors > 0 && !ret) { | |
3036 | int num = nb_sectors; | |
3037 | ||
3038 | /* Align request. Block drivers can expect the "bulk" of the request | |
3039 | * to be aligned. | |
3040 | */ | |
3041 | if (bs->bl.write_zeroes_alignment | |
3042 | && num > bs->bl.write_zeroes_alignment) { | |
3043 | if (sector_num % bs->bl.write_zeroes_alignment != 0) { | |
3044 | /* Make a small request up to the first aligned sector. */ | |
3045 | num = bs->bl.write_zeroes_alignment; | |
3046 | num -= sector_num % bs->bl.write_zeroes_alignment; | |
3047 | } else if ((sector_num + num) % bs->bl.write_zeroes_alignment != 0) { | |
3048 | /* Shorten the request to the last aligned sector. num cannot | |
3049 | * underflow because num > bs->bl.write_zeroes_alignment. | |
3050 | */ | |
3051 | num -= (sector_num + num) % bs->bl.write_zeroes_alignment; | |
3052 | } | |
3053 | } | |
3054 | ||
3055 | /* limit request size */ | |
3056 | if (num > max_write_zeroes) { | |
3057 | num = max_write_zeroes; | |
3058 | } | |
3059 | ||
3060 | ret = -ENOTSUP; | |
3061 | /* First try the efficient write zeroes operation */ | |
3062 | if (drv->bdrv_co_write_zeroes) { | |
3063 | ret = drv->bdrv_co_write_zeroes(bs, sector_num, num, flags); | |
3064 | } | |
3065 | ||
3066 | if (ret == -ENOTSUP) { | |
3067 | /* Fall back to bounce buffer if write zeroes is unsupported */ | |
3068 | iov.iov_len = num * BDRV_SECTOR_SIZE; | |
3069 | if (iov.iov_base == NULL) { | |
3070 | iov.iov_base = qemu_blockalign(bs, num * BDRV_SECTOR_SIZE); | |
3071 | memset(iov.iov_base, 0, num * BDRV_SECTOR_SIZE); | |
3072 | } | |
3073 | qemu_iovec_init_external(&qiov, &iov, 1); | |
3074 | ||
3075 | ret = drv->bdrv_co_writev(bs, sector_num, num, &qiov); | |
3076 | ||
3077 | /* Keep bounce buffer around if it is big enough for all | |
3078 | * all future requests. | |
3079 | */ | |
3080 | if (num < max_write_zeroes) { | |
3081 | qemu_vfree(iov.iov_base); | |
3082 | iov.iov_base = NULL; | |
3083 | } | |
3084 | } | |
3085 | ||
3086 | sector_num += num; | |
3087 | nb_sectors -= num; | |
3088 | } | |
3089 | ||
3090 | qemu_vfree(iov.iov_base); | |
3091 | return ret; | |
3092 | } | |
3093 | ||
3094 | /* | |
3095 | * Handle a write request in coroutine context | |
3096 | */ | |
3097 | static int coroutine_fn bdrv_co_do_writev(BlockDriverState *bs, | |
3098 | int64_t sector_num, int nb_sectors, QEMUIOVector *qiov, | |
3099 | BdrvRequestFlags flags) | |
3100 | { | |
3101 | BlockDriver *drv = bs->drv; | |
3102 | BdrvTrackedRequest req; | |
3103 | int ret; | |
3104 | ||
3105 | if (!bs->drv) { | |
3106 | return -ENOMEDIUM; | |
3107 | } | |
3108 | if (bs->read_only) { | |
3109 | return -EACCES; | |
3110 | } | |
3111 | if (bdrv_check_request(bs, sector_num, nb_sectors)) { | |
3112 | return -EIO; | |
3113 | } | |
3114 | ||
3115 | if (bs->copy_on_read_in_flight) { | |
3116 | wait_for_overlapping_requests(bs, sector_num, nb_sectors); | |
3117 | } | |
3118 | ||
3119 | /* throttling disk I/O */ | |
3120 | if (bs->io_limits_enabled) { | |
3121 | bdrv_io_limits_intercept(bs, nb_sectors, true); | |
3122 | } | |
3123 | ||
3124 | tracked_request_begin(&req, bs, sector_num, nb_sectors, true); | |
3125 | ||
3126 | ret = notifier_with_return_list_notify(&bs->before_write_notifiers, &req); | |
3127 | ||
3128 | if (ret < 0) { | |
3129 | /* Do nothing, write notifier decided to fail this request */ | |
3130 | } else if (flags & BDRV_REQ_ZERO_WRITE) { | |
3131 | ret = bdrv_co_do_write_zeroes(bs, sector_num, nb_sectors, flags); | |
3132 | } else { | |
3133 | ret = drv->bdrv_co_writev(bs, sector_num, nb_sectors, qiov); | |
3134 | } | |
3135 | ||
3136 | if (ret == 0 && !bs->enable_write_cache) { | |
3137 | ret = bdrv_co_flush(bs); | |
3138 | } | |
3139 | ||
3140 | bdrv_set_dirty(bs, sector_num, nb_sectors); | |
3141 | ||
3142 | if (bs->wr_highest_sector < sector_num + nb_sectors - 1) { | |
3143 | bs->wr_highest_sector = sector_num + nb_sectors - 1; | |
3144 | } | |
3145 | if (bs->growable && ret >= 0) { | |
3146 | bs->total_sectors = MAX(bs->total_sectors, sector_num + nb_sectors); | |
3147 | } | |
3148 | ||
3149 | tracked_request_end(&req); | |
3150 | ||
3151 | return ret; | |
3152 | } | |
3153 | ||
3154 | int coroutine_fn bdrv_co_writev(BlockDriverState *bs, int64_t sector_num, | |
3155 | int nb_sectors, QEMUIOVector *qiov) | |
3156 | { | |
3157 | trace_bdrv_co_writev(bs, sector_num, nb_sectors); | |
3158 | ||
3159 | return bdrv_co_do_writev(bs, sector_num, nb_sectors, qiov, 0); | |
3160 | } | |
3161 | ||
3162 | int coroutine_fn bdrv_co_write_zeroes(BlockDriverState *bs, | |
3163 | int64_t sector_num, int nb_sectors, | |
3164 | BdrvRequestFlags flags) | |
3165 | { | |
3166 | trace_bdrv_co_write_zeroes(bs, sector_num, nb_sectors, flags); | |
3167 | ||
3168 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
3169 | flags &= ~BDRV_REQ_MAY_UNMAP; | |
3170 | } | |
3171 | ||
3172 | return bdrv_co_do_writev(bs, sector_num, nb_sectors, NULL, | |
3173 | BDRV_REQ_ZERO_WRITE | flags); | |
3174 | } | |
3175 | ||
3176 | /** | |
3177 | * Truncate file to 'offset' bytes (needed only for file protocols) | |
3178 | */ | |
3179 | int bdrv_truncate(BlockDriverState *bs, int64_t offset) | |
3180 | { | |
3181 | BlockDriver *drv = bs->drv; | |
3182 | int ret; | |
3183 | if (!drv) | |
3184 | return -ENOMEDIUM; | |
3185 | if (!drv->bdrv_truncate) | |
3186 | return -ENOTSUP; | |
3187 | if (bs->read_only) | |
3188 | return -EACCES; | |
3189 | if (bdrv_in_use(bs)) | |
3190 | return -EBUSY; | |
3191 | ret = drv->bdrv_truncate(bs, offset); | |
3192 | if (ret == 0) { | |
3193 | ret = refresh_total_sectors(bs, offset >> BDRV_SECTOR_BITS); | |
3194 | bdrv_dev_resize_cb(bs); | |
3195 | } | |
3196 | return ret; | |
3197 | } | |
3198 | ||
3199 | /** | |
3200 | * Length of a allocated file in bytes. Sparse files are counted by actual | |
3201 | * allocated space. Return < 0 if error or unknown. | |
3202 | */ | |
3203 | int64_t bdrv_get_allocated_file_size(BlockDriverState *bs) | |
3204 | { | |
3205 | BlockDriver *drv = bs->drv; | |
3206 | if (!drv) { | |
3207 | return -ENOMEDIUM; | |
3208 | } | |
3209 | if (drv->bdrv_get_allocated_file_size) { | |
3210 | return drv->bdrv_get_allocated_file_size(bs); | |
3211 | } | |
3212 | if (bs->file) { | |
3213 | return bdrv_get_allocated_file_size(bs->file); | |
3214 | } | |
3215 | return -ENOTSUP; | |
3216 | } | |
3217 | ||
3218 | /** | |
3219 | * Length of a file in bytes. Return < 0 if error or unknown. | |
3220 | */ | |
3221 | int64_t bdrv_getlength(BlockDriverState *bs) | |
3222 | { | |
3223 | BlockDriver *drv = bs->drv; | |
3224 | if (!drv) | |
3225 | return -ENOMEDIUM; | |
3226 | ||
3227 | if (drv->has_variable_length) { | |
3228 | int ret = refresh_total_sectors(bs, bs->total_sectors); | |
3229 | if (ret < 0) { | |
3230 | return ret; | |
3231 | } | |
3232 | } | |
3233 | return bs->total_sectors * BDRV_SECTOR_SIZE; | |
3234 | } | |
3235 | ||
3236 | /* return 0 as number of sectors if no device present or error */ | |
3237 | void bdrv_get_geometry(BlockDriverState *bs, uint64_t *nb_sectors_ptr) | |
3238 | { | |
3239 | int64_t length; | |
3240 | length = bdrv_getlength(bs); | |
3241 | if (length < 0) | |
3242 | length = 0; | |
3243 | else | |
3244 | length = length >> BDRV_SECTOR_BITS; | |
3245 | *nb_sectors_ptr = length; | |
3246 | } | |
3247 | ||
3248 | void bdrv_set_on_error(BlockDriverState *bs, BlockdevOnError on_read_error, | |
3249 | BlockdevOnError on_write_error) | |
3250 | { | |
3251 | bs->on_read_error = on_read_error; | |
3252 | bs->on_write_error = on_write_error; | |
3253 | } | |
3254 | ||
3255 | BlockdevOnError bdrv_get_on_error(BlockDriverState *bs, bool is_read) | |
3256 | { | |
3257 | return is_read ? bs->on_read_error : bs->on_write_error; | |
3258 | } | |
3259 | ||
3260 | BlockErrorAction bdrv_get_error_action(BlockDriverState *bs, bool is_read, int error) | |
3261 | { | |
3262 | BlockdevOnError on_err = is_read ? bs->on_read_error : bs->on_write_error; | |
3263 | ||
3264 | switch (on_err) { | |
3265 | case BLOCKDEV_ON_ERROR_ENOSPC: | |
3266 | return (error == ENOSPC) ? BDRV_ACTION_STOP : BDRV_ACTION_REPORT; | |
3267 | case BLOCKDEV_ON_ERROR_STOP: | |
3268 | return BDRV_ACTION_STOP; | |
3269 | case BLOCKDEV_ON_ERROR_REPORT: | |
3270 | return BDRV_ACTION_REPORT; | |
3271 | case BLOCKDEV_ON_ERROR_IGNORE: | |
3272 | return BDRV_ACTION_IGNORE; | |
3273 | default: | |
3274 | abort(); | |
3275 | } | |
3276 | } | |
3277 | ||
3278 | /* This is done by device models because, while the block layer knows | |
3279 | * about the error, it does not know whether an operation comes from | |
3280 | * the device or the block layer (from a job, for example). | |
3281 | */ | |
3282 | void bdrv_error_action(BlockDriverState *bs, BlockErrorAction action, | |
3283 | bool is_read, int error) | |
3284 | { | |
3285 | assert(error >= 0); | |
3286 | bdrv_emit_qmp_error_event(bs, QEVENT_BLOCK_IO_ERROR, action, is_read); | |
3287 | if (action == BDRV_ACTION_STOP) { | |
3288 | vm_stop(RUN_STATE_IO_ERROR); | |
3289 | bdrv_iostatus_set_err(bs, error); | |
3290 | } | |
3291 | } | |
3292 | ||
3293 | int bdrv_is_read_only(BlockDriverState *bs) | |
3294 | { | |
3295 | return bs->read_only; | |
3296 | } | |
3297 | ||
3298 | int bdrv_is_sg(BlockDriverState *bs) | |
3299 | { | |
3300 | return bs->sg; | |
3301 | } | |
3302 | ||
3303 | int bdrv_enable_write_cache(BlockDriverState *bs) | |
3304 | { | |
3305 | return bs->enable_write_cache; | |
3306 | } | |
3307 | ||
3308 | void bdrv_set_enable_write_cache(BlockDriverState *bs, bool wce) | |
3309 | { | |
3310 | bs->enable_write_cache = wce; | |
3311 | ||
3312 | /* so a reopen() will preserve wce */ | |
3313 | if (wce) { | |
3314 | bs->open_flags |= BDRV_O_CACHE_WB; | |
3315 | } else { | |
3316 | bs->open_flags &= ~BDRV_O_CACHE_WB; | |
3317 | } | |
3318 | } | |
3319 | ||
3320 | int bdrv_is_encrypted(BlockDriverState *bs) | |
3321 | { | |
3322 | if (bs->backing_hd && bs->backing_hd->encrypted) | |
3323 | return 1; | |
3324 | return bs->encrypted; | |
3325 | } | |
3326 | ||
3327 | int bdrv_key_required(BlockDriverState *bs) | |
3328 | { | |
3329 | BlockDriverState *backing_hd = bs->backing_hd; | |
3330 | ||
3331 | if (backing_hd && backing_hd->encrypted && !backing_hd->valid_key) | |
3332 | return 1; | |
3333 | return (bs->encrypted && !bs->valid_key); | |
3334 | } | |
3335 | ||
3336 | int bdrv_set_key(BlockDriverState *bs, const char *key) | |
3337 | { | |
3338 | int ret; | |
3339 | if (bs->backing_hd && bs->backing_hd->encrypted) { | |
3340 | ret = bdrv_set_key(bs->backing_hd, key); | |
3341 | if (ret < 0) | |
3342 | return ret; | |
3343 | if (!bs->encrypted) | |
3344 | return 0; | |
3345 | } | |
3346 | if (!bs->encrypted) { | |
3347 | return -EINVAL; | |
3348 | } else if (!bs->drv || !bs->drv->bdrv_set_key) { | |
3349 | return -ENOMEDIUM; | |
3350 | } | |
3351 | ret = bs->drv->bdrv_set_key(bs, key); | |
3352 | if (ret < 0) { | |
3353 | bs->valid_key = 0; | |
3354 | } else if (!bs->valid_key) { | |
3355 | bs->valid_key = 1; | |
3356 | /* call the change callback now, we skipped it on open */ | |
3357 | bdrv_dev_change_media_cb(bs, true); | |
3358 | } | |
3359 | return ret; | |
3360 | } | |
3361 | ||
3362 | const char *bdrv_get_format_name(BlockDriverState *bs) | |
3363 | { | |
3364 | return bs->drv ? bs->drv->format_name : NULL; | |
3365 | } | |
3366 | ||
3367 | void bdrv_iterate_format(void (*it)(void *opaque, const char *name), | |
3368 | void *opaque) | |
3369 | { | |
3370 | BlockDriver *drv; | |
3371 | ||
3372 | QLIST_FOREACH(drv, &bdrv_drivers, list) { | |
3373 | it(opaque, drv->format_name); | |
3374 | } | |
3375 | } | |
3376 | ||
3377 | /* This function is to find block backend bs */ | |
3378 | BlockDriverState *bdrv_find(const char *name) | |
3379 | { | |
3380 | BlockDriverState *bs; | |
3381 | ||
3382 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
3383 | if (!strcmp(name, bs->device_name)) { | |
3384 | return bs; | |
3385 | } | |
3386 | } | |
3387 | return NULL; | |
3388 | } | |
3389 | ||
3390 | /* This function is to find a node in the bs graph */ | |
3391 | BlockDriverState *bdrv_find_node(const char *node_name) | |
3392 | { | |
3393 | BlockDriverState *bs; | |
3394 | ||
3395 | assert(node_name); | |
3396 | ||
3397 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
3398 | if (!strcmp(node_name, bs->node_name)) { | |
3399 | return bs; | |
3400 | } | |
3401 | } | |
3402 | return NULL; | |
3403 | } | |
3404 | ||
3405 | /* Put this QMP function here so it can access the static graph_bdrv_states. */ | |
3406 | BlockDeviceInfoList *bdrv_named_nodes_list(void) | |
3407 | { | |
3408 | BlockDeviceInfoList *list, *entry; | |
3409 | BlockDriverState *bs; | |
3410 | ||
3411 | list = NULL; | |
3412 | QTAILQ_FOREACH(bs, &graph_bdrv_states, node_list) { | |
3413 | entry = g_malloc0(sizeof(*entry)); | |
3414 | entry->value = bdrv_block_device_info(bs); | |
3415 | entry->next = list; | |
3416 | list = entry; | |
3417 | } | |
3418 | ||
3419 | return list; | |
3420 | } | |
3421 | ||
3422 | BlockDriverState *bdrv_lookup_bs(const char *device, | |
3423 | const char *node_name, | |
3424 | Error **errp) | |
3425 | { | |
3426 | BlockDriverState *bs = NULL; | |
3427 | ||
3428 | if ((!device && !node_name) || (device && node_name)) { | |
3429 | error_setg(errp, "Use either device or node-name but not both"); | |
3430 | return NULL; | |
3431 | } | |
3432 | ||
3433 | if (device) { | |
3434 | bs = bdrv_find(device); | |
3435 | ||
3436 | if (!bs) { | |
3437 | error_set(errp, QERR_DEVICE_NOT_FOUND, device); | |
3438 | return NULL; | |
3439 | } | |
3440 | ||
3441 | return bs; | |
3442 | } | |
3443 | ||
3444 | bs = bdrv_find_node(node_name); | |
3445 | ||
3446 | if (!bs) { | |
3447 | error_set(errp, QERR_DEVICE_NOT_FOUND, node_name); | |
3448 | return NULL; | |
3449 | } | |
3450 | ||
3451 | return bs; | |
3452 | } | |
3453 | ||
3454 | BlockDriverState *bdrv_next(BlockDriverState *bs) | |
3455 | { | |
3456 | if (!bs) { | |
3457 | return QTAILQ_FIRST(&bdrv_states); | |
3458 | } | |
3459 | return QTAILQ_NEXT(bs, device_list); | |
3460 | } | |
3461 | ||
3462 | void bdrv_iterate(void (*it)(void *opaque, BlockDriverState *bs), void *opaque) | |
3463 | { | |
3464 | BlockDriverState *bs; | |
3465 | ||
3466 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
3467 | it(opaque, bs); | |
3468 | } | |
3469 | } | |
3470 | ||
3471 | const char *bdrv_get_device_name(BlockDriverState *bs) | |
3472 | { | |
3473 | return bs->device_name; | |
3474 | } | |
3475 | ||
3476 | int bdrv_get_flags(BlockDriverState *bs) | |
3477 | { | |
3478 | return bs->open_flags; | |
3479 | } | |
3480 | ||
3481 | int bdrv_flush_all(void) | |
3482 | { | |
3483 | BlockDriverState *bs; | |
3484 | int result = 0; | |
3485 | ||
3486 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
3487 | int ret = bdrv_flush(bs); | |
3488 | if (ret < 0 && !result) { | |
3489 | result = ret; | |
3490 | } | |
3491 | } | |
3492 | ||
3493 | return result; | |
3494 | } | |
3495 | ||
3496 | int bdrv_has_zero_init_1(BlockDriverState *bs) | |
3497 | { | |
3498 | return 1; | |
3499 | } | |
3500 | ||
3501 | int bdrv_has_zero_init(BlockDriverState *bs) | |
3502 | { | |
3503 | assert(bs->drv); | |
3504 | ||
3505 | /* If BS is a copy on write image, it is initialized to | |
3506 | the contents of the base image, which may not be zeroes. */ | |
3507 | if (bs->backing_hd) { | |
3508 | return 0; | |
3509 | } | |
3510 | if (bs->drv->bdrv_has_zero_init) { | |
3511 | return bs->drv->bdrv_has_zero_init(bs); | |
3512 | } | |
3513 | ||
3514 | /* safe default */ | |
3515 | return 0; | |
3516 | } | |
3517 | ||
3518 | bool bdrv_unallocated_blocks_are_zero(BlockDriverState *bs) | |
3519 | { | |
3520 | BlockDriverInfo bdi; | |
3521 | ||
3522 | if (bs->backing_hd) { | |
3523 | return false; | |
3524 | } | |
3525 | ||
3526 | if (bdrv_get_info(bs, &bdi) == 0) { | |
3527 | return bdi.unallocated_blocks_are_zero; | |
3528 | } | |
3529 | ||
3530 | return false; | |
3531 | } | |
3532 | ||
3533 | bool bdrv_can_write_zeroes_with_unmap(BlockDriverState *bs) | |
3534 | { | |
3535 | BlockDriverInfo bdi; | |
3536 | ||
3537 | if (bs->backing_hd || !(bs->open_flags & BDRV_O_UNMAP)) { | |
3538 | return false; | |
3539 | } | |
3540 | ||
3541 | if (bdrv_get_info(bs, &bdi) == 0) { | |
3542 | return bdi.can_write_zeroes_with_unmap; | |
3543 | } | |
3544 | ||
3545 | return false; | |
3546 | } | |
3547 | ||
3548 | typedef struct BdrvCoGetBlockStatusData { | |
3549 | BlockDriverState *bs; | |
3550 | BlockDriverState *base; | |
3551 | int64_t sector_num; | |
3552 | int nb_sectors; | |
3553 | int *pnum; | |
3554 | int64_t ret; | |
3555 | bool done; | |
3556 | } BdrvCoGetBlockStatusData; | |
3557 | ||
3558 | /* | |
3559 | * Returns true iff the specified sector is present in the disk image. Drivers | |
3560 | * not implementing the functionality are assumed to not support backing files, | |
3561 | * hence all their sectors are reported as allocated. | |
3562 | * | |
3563 | * If 'sector_num' is beyond the end of the disk image the return value is 0 | |
3564 | * and 'pnum' is set to 0. | |
3565 | * | |
3566 | * 'pnum' is set to the number of sectors (including and immediately following | |
3567 | * the specified sector) that are known to be in the same | |
3568 | * allocated/unallocated state. | |
3569 | * | |
3570 | * 'nb_sectors' is the max value 'pnum' should be set to. If nb_sectors goes | |
3571 | * beyond the end of the disk image it will be clamped. | |
3572 | */ | |
3573 | static int64_t coroutine_fn bdrv_co_get_block_status(BlockDriverState *bs, | |
3574 | int64_t sector_num, | |
3575 | int nb_sectors, int *pnum) | |
3576 | { | |
3577 | int64_t length; | |
3578 | int64_t n; | |
3579 | int64_t ret, ret2; | |
3580 | ||
3581 | length = bdrv_getlength(bs); | |
3582 | if (length < 0) { | |
3583 | return length; | |
3584 | } | |
3585 | ||
3586 | if (sector_num >= (length >> BDRV_SECTOR_BITS)) { | |
3587 | *pnum = 0; | |
3588 | return 0; | |
3589 | } | |
3590 | ||
3591 | n = bs->total_sectors - sector_num; | |
3592 | if (n < nb_sectors) { | |
3593 | nb_sectors = n; | |
3594 | } | |
3595 | ||
3596 | if (!bs->drv->bdrv_co_get_block_status) { | |
3597 | *pnum = nb_sectors; | |
3598 | ret = BDRV_BLOCK_DATA; | |
3599 | if (bs->drv->protocol_name) { | |
3600 | ret |= BDRV_BLOCK_OFFSET_VALID | (sector_num * BDRV_SECTOR_SIZE); | |
3601 | } | |
3602 | return ret; | |
3603 | } | |
3604 | ||
3605 | ret = bs->drv->bdrv_co_get_block_status(bs, sector_num, nb_sectors, pnum); | |
3606 | if (ret < 0) { | |
3607 | *pnum = 0; | |
3608 | return ret; | |
3609 | } | |
3610 | ||
3611 | if (ret & BDRV_BLOCK_RAW) { | |
3612 | assert(ret & BDRV_BLOCK_OFFSET_VALID); | |
3613 | return bdrv_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS, | |
3614 | *pnum, pnum); | |
3615 | } | |
3616 | ||
3617 | if (!(ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO)) { | |
3618 | if (bdrv_unallocated_blocks_are_zero(bs)) { | |
3619 | ret |= BDRV_BLOCK_ZERO; | |
3620 | } else if (bs->backing_hd) { | |
3621 | BlockDriverState *bs2 = bs->backing_hd; | |
3622 | int64_t length2 = bdrv_getlength(bs2); | |
3623 | if (length2 >= 0 && sector_num >= (length2 >> BDRV_SECTOR_BITS)) { | |
3624 | ret |= BDRV_BLOCK_ZERO; | |
3625 | } | |
3626 | } | |
3627 | } | |
3628 | ||
3629 | if (bs->file && | |
3630 | (ret & BDRV_BLOCK_DATA) && !(ret & BDRV_BLOCK_ZERO) && | |
3631 | (ret & BDRV_BLOCK_OFFSET_VALID)) { | |
3632 | ret2 = bdrv_co_get_block_status(bs->file, ret >> BDRV_SECTOR_BITS, | |
3633 | *pnum, pnum); | |
3634 | if (ret2 >= 0) { | |
3635 | /* Ignore errors. This is just providing extra information, it | |
3636 | * is useful but not necessary. | |
3637 | */ | |
3638 | ret |= (ret2 & BDRV_BLOCK_ZERO); | |
3639 | } | |
3640 | } | |
3641 | ||
3642 | return ret; | |
3643 | } | |
3644 | ||
3645 | /* Coroutine wrapper for bdrv_get_block_status() */ | |
3646 | static void coroutine_fn bdrv_get_block_status_co_entry(void *opaque) | |
3647 | { | |
3648 | BdrvCoGetBlockStatusData *data = opaque; | |
3649 | BlockDriverState *bs = data->bs; | |
3650 | ||
3651 | data->ret = bdrv_co_get_block_status(bs, data->sector_num, data->nb_sectors, | |
3652 | data->pnum); | |
3653 | data->done = true; | |
3654 | } | |
3655 | ||
3656 | /* | |
3657 | * Synchronous wrapper around bdrv_co_get_block_status(). | |
3658 | * | |
3659 | * See bdrv_co_get_block_status() for details. | |
3660 | */ | |
3661 | int64_t bdrv_get_block_status(BlockDriverState *bs, int64_t sector_num, | |
3662 | int nb_sectors, int *pnum) | |
3663 | { | |
3664 | Coroutine *co; | |
3665 | BdrvCoGetBlockStatusData data = { | |
3666 | .bs = bs, | |
3667 | .sector_num = sector_num, | |
3668 | .nb_sectors = nb_sectors, | |
3669 | .pnum = pnum, | |
3670 | .done = false, | |
3671 | }; | |
3672 | ||
3673 | if (qemu_in_coroutine()) { | |
3674 | /* Fast-path if already in coroutine context */ | |
3675 | bdrv_get_block_status_co_entry(&data); | |
3676 | } else { | |
3677 | co = qemu_coroutine_create(bdrv_get_block_status_co_entry); | |
3678 | qemu_coroutine_enter(co, &data); | |
3679 | while (!data.done) { | |
3680 | qemu_aio_wait(); | |
3681 | } | |
3682 | } | |
3683 | return data.ret; | |
3684 | } | |
3685 | ||
3686 | int coroutine_fn bdrv_is_allocated(BlockDriverState *bs, int64_t sector_num, | |
3687 | int nb_sectors, int *pnum) | |
3688 | { | |
3689 | int64_t ret = bdrv_get_block_status(bs, sector_num, nb_sectors, pnum); | |
3690 | if (ret < 0) { | |
3691 | return ret; | |
3692 | } | |
3693 | return | |
3694 | (ret & BDRV_BLOCK_DATA) || | |
3695 | ((ret & BDRV_BLOCK_ZERO) && !bdrv_has_zero_init(bs)); | |
3696 | } | |
3697 | ||
3698 | /* | |
3699 | * Given an image chain: ... -> [BASE] -> [INTER1] -> [INTER2] -> [TOP] | |
3700 | * | |
3701 | * Return true if the given sector is allocated in any image between | |
3702 | * BASE and TOP (inclusive). BASE can be NULL to check if the given | |
3703 | * sector is allocated in any image of the chain. Return false otherwise. | |
3704 | * | |
3705 | * 'pnum' is set to the number of sectors (including and immediately following | |
3706 | * the specified sector) that are known to be in the same | |
3707 | * allocated/unallocated state. | |
3708 | * | |
3709 | */ | |
3710 | int bdrv_is_allocated_above(BlockDriverState *top, | |
3711 | BlockDriverState *base, | |
3712 | int64_t sector_num, | |
3713 | int nb_sectors, int *pnum) | |
3714 | { | |
3715 | BlockDriverState *intermediate; | |
3716 | int ret, n = nb_sectors; | |
3717 | ||
3718 | intermediate = top; | |
3719 | while (intermediate && intermediate != base) { | |
3720 | int pnum_inter; | |
3721 | ret = bdrv_is_allocated(intermediate, sector_num, nb_sectors, | |
3722 | &pnum_inter); | |
3723 | if (ret < 0) { | |
3724 | return ret; | |
3725 | } else if (ret) { | |
3726 | *pnum = pnum_inter; | |
3727 | return 1; | |
3728 | } | |
3729 | ||
3730 | /* | |
3731 | * [sector_num, nb_sectors] is unallocated on top but intermediate | |
3732 | * might have | |
3733 | * | |
3734 | * [sector_num+x, nr_sectors] allocated. | |
3735 | */ | |
3736 | if (n > pnum_inter && | |
3737 | (intermediate == top || | |
3738 | sector_num + pnum_inter < intermediate->total_sectors)) { | |
3739 | n = pnum_inter; | |
3740 | } | |
3741 | ||
3742 | intermediate = intermediate->backing_hd; | |
3743 | } | |
3744 | ||
3745 | *pnum = n; | |
3746 | return 0; | |
3747 | } | |
3748 | ||
3749 | const char *bdrv_get_encrypted_filename(BlockDriverState *bs) | |
3750 | { | |
3751 | if (bs->backing_hd && bs->backing_hd->encrypted) | |
3752 | return bs->backing_file; | |
3753 | else if (bs->encrypted) | |
3754 | return bs->filename; | |
3755 | else | |
3756 | return NULL; | |
3757 | } | |
3758 | ||
3759 | void bdrv_get_backing_filename(BlockDriverState *bs, | |
3760 | char *filename, int filename_size) | |
3761 | { | |
3762 | pstrcpy(filename, filename_size, bs->backing_file); | |
3763 | } | |
3764 | ||
3765 | int bdrv_write_compressed(BlockDriverState *bs, int64_t sector_num, | |
3766 | const uint8_t *buf, int nb_sectors) | |
3767 | { | |
3768 | BlockDriver *drv = bs->drv; | |
3769 | if (!drv) | |
3770 | return -ENOMEDIUM; | |
3771 | if (!drv->bdrv_write_compressed) | |
3772 | return -ENOTSUP; | |
3773 | if (bdrv_check_request(bs, sector_num, nb_sectors)) | |
3774 | return -EIO; | |
3775 | ||
3776 | assert(QLIST_EMPTY(&bs->dirty_bitmaps)); | |
3777 | ||
3778 | return drv->bdrv_write_compressed(bs, sector_num, buf, nb_sectors); | |
3779 | } | |
3780 | ||
3781 | int bdrv_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) | |
3782 | { | |
3783 | BlockDriver *drv = bs->drv; | |
3784 | if (!drv) | |
3785 | return -ENOMEDIUM; | |
3786 | if (!drv->bdrv_get_info) | |
3787 | return -ENOTSUP; | |
3788 | memset(bdi, 0, sizeof(*bdi)); | |
3789 | return drv->bdrv_get_info(bs, bdi); | |
3790 | } | |
3791 | ||
3792 | ImageInfoSpecific *bdrv_get_specific_info(BlockDriverState *bs) | |
3793 | { | |
3794 | BlockDriver *drv = bs->drv; | |
3795 | if (drv && drv->bdrv_get_specific_info) { | |
3796 | return drv->bdrv_get_specific_info(bs); | |
3797 | } | |
3798 | return NULL; | |
3799 | } | |
3800 | ||
3801 | int bdrv_save_vmstate(BlockDriverState *bs, const uint8_t *buf, | |
3802 | int64_t pos, int size) | |
3803 | { | |
3804 | QEMUIOVector qiov; | |
3805 | struct iovec iov = { | |
3806 | .iov_base = (void *) buf, | |
3807 | .iov_len = size, | |
3808 | }; | |
3809 | ||
3810 | qemu_iovec_init_external(&qiov, &iov, 1); | |
3811 | return bdrv_writev_vmstate(bs, &qiov, pos); | |
3812 | } | |
3813 | ||
3814 | int bdrv_writev_vmstate(BlockDriverState *bs, QEMUIOVector *qiov, int64_t pos) | |
3815 | { | |
3816 | BlockDriver *drv = bs->drv; | |
3817 | ||
3818 | if (!drv) { | |
3819 | return -ENOMEDIUM; | |
3820 | } else if (drv->bdrv_save_vmstate) { | |
3821 | return drv->bdrv_save_vmstate(bs, qiov, pos); | |
3822 | } else if (bs->file) { | |
3823 | return bdrv_writev_vmstate(bs->file, qiov, pos); | |
3824 | } | |
3825 | ||
3826 | return -ENOTSUP; | |
3827 | } | |
3828 | ||
3829 | int bdrv_load_vmstate(BlockDriverState *bs, uint8_t *buf, | |
3830 | int64_t pos, int size) | |
3831 | { | |
3832 | BlockDriver *drv = bs->drv; | |
3833 | if (!drv) | |
3834 | return -ENOMEDIUM; | |
3835 | if (drv->bdrv_load_vmstate) | |
3836 | return drv->bdrv_load_vmstate(bs, buf, pos, size); | |
3837 | if (bs->file) | |
3838 | return bdrv_load_vmstate(bs->file, buf, pos, size); | |
3839 | return -ENOTSUP; | |
3840 | } | |
3841 | ||
3842 | void bdrv_debug_event(BlockDriverState *bs, BlkDebugEvent event) | |
3843 | { | |
3844 | if (!bs || !bs->drv || !bs->drv->bdrv_debug_event) { | |
3845 | return; | |
3846 | } | |
3847 | ||
3848 | bs->drv->bdrv_debug_event(bs, event); | |
3849 | } | |
3850 | ||
3851 | int bdrv_debug_breakpoint(BlockDriverState *bs, const char *event, | |
3852 | const char *tag) | |
3853 | { | |
3854 | while (bs && bs->drv && !bs->drv->bdrv_debug_breakpoint) { | |
3855 | bs = bs->file; | |
3856 | } | |
3857 | ||
3858 | if (bs && bs->drv && bs->drv->bdrv_debug_breakpoint) { | |
3859 | return bs->drv->bdrv_debug_breakpoint(bs, event, tag); | |
3860 | } | |
3861 | ||
3862 | return -ENOTSUP; | |
3863 | } | |
3864 | ||
3865 | int bdrv_debug_remove_breakpoint(BlockDriverState *bs, const char *tag) | |
3866 | { | |
3867 | while (bs && bs->drv && !bs->drv->bdrv_debug_remove_breakpoint) { | |
3868 | bs = bs->file; | |
3869 | } | |
3870 | ||
3871 | if (bs && bs->drv && bs->drv->bdrv_debug_remove_breakpoint) { | |
3872 | return bs->drv->bdrv_debug_remove_breakpoint(bs, tag); | |
3873 | } | |
3874 | ||
3875 | return -ENOTSUP; | |
3876 | } | |
3877 | ||
3878 | int bdrv_debug_resume(BlockDriverState *bs, const char *tag) | |
3879 | { | |
3880 | while (bs && bs->drv && !bs->drv->bdrv_debug_resume) { | |
3881 | bs = bs->file; | |
3882 | } | |
3883 | ||
3884 | if (bs && bs->drv && bs->drv->bdrv_debug_resume) { | |
3885 | return bs->drv->bdrv_debug_resume(bs, tag); | |
3886 | } | |
3887 | ||
3888 | return -ENOTSUP; | |
3889 | } | |
3890 | ||
3891 | bool bdrv_debug_is_suspended(BlockDriverState *bs, const char *tag) | |
3892 | { | |
3893 | while (bs && bs->drv && !bs->drv->bdrv_debug_is_suspended) { | |
3894 | bs = bs->file; | |
3895 | } | |
3896 | ||
3897 | if (bs && bs->drv && bs->drv->bdrv_debug_is_suspended) { | |
3898 | return bs->drv->bdrv_debug_is_suspended(bs, tag); | |
3899 | } | |
3900 | ||
3901 | return false; | |
3902 | } | |
3903 | ||
3904 | int bdrv_is_snapshot(BlockDriverState *bs) | |
3905 | { | |
3906 | return !!(bs->open_flags & BDRV_O_SNAPSHOT); | |
3907 | } | |
3908 | ||
3909 | /* backing_file can either be relative, or absolute, or a protocol. If it is | |
3910 | * relative, it must be relative to the chain. So, passing in bs->filename | |
3911 | * from a BDS as backing_file should not be done, as that may be relative to | |
3912 | * the CWD rather than the chain. */ | |
3913 | BlockDriverState *bdrv_find_backing_image(BlockDriverState *bs, | |
3914 | const char *backing_file) | |
3915 | { | |
3916 | char *filename_full = NULL; | |
3917 | char *backing_file_full = NULL; | |
3918 | char *filename_tmp = NULL; | |
3919 | int is_protocol = 0; | |
3920 | BlockDriverState *curr_bs = NULL; | |
3921 | BlockDriverState *retval = NULL; | |
3922 | ||
3923 | if (!bs || !bs->drv || !backing_file) { | |
3924 | return NULL; | |
3925 | } | |
3926 | ||
3927 | filename_full = g_malloc(PATH_MAX); | |
3928 | backing_file_full = g_malloc(PATH_MAX); | |
3929 | filename_tmp = g_malloc(PATH_MAX); | |
3930 | ||
3931 | is_protocol = path_has_protocol(backing_file); | |
3932 | ||
3933 | for (curr_bs = bs; curr_bs->backing_hd; curr_bs = curr_bs->backing_hd) { | |
3934 | ||
3935 | /* If either of the filename paths is actually a protocol, then | |
3936 | * compare unmodified paths; otherwise make paths relative */ | |
3937 | if (is_protocol || path_has_protocol(curr_bs->backing_file)) { | |
3938 | if (strcmp(backing_file, curr_bs->backing_file) == 0) { | |
3939 | retval = curr_bs->backing_hd; | |
3940 | break; | |
3941 | } | |
3942 | } else { | |
3943 | /* If not an absolute filename path, make it relative to the current | |
3944 | * image's filename path */ | |
3945 | path_combine(filename_tmp, PATH_MAX, curr_bs->filename, | |
3946 | backing_file); | |
3947 | ||
3948 | /* We are going to compare absolute pathnames */ | |
3949 | if (!realpath(filename_tmp, filename_full)) { | |
3950 | continue; | |
3951 | } | |
3952 | ||
3953 | /* We need to make sure the backing filename we are comparing against | |
3954 | * is relative to the current image filename (or absolute) */ | |
3955 | path_combine(filename_tmp, PATH_MAX, curr_bs->filename, | |
3956 | curr_bs->backing_file); | |
3957 | ||
3958 | if (!realpath(filename_tmp, backing_file_full)) { | |
3959 | continue; | |
3960 | } | |
3961 | ||
3962 | if (strcmp(backing_file_full, filename_full) == 0) { | |
3963 | retval = curr_bs->backing_hd; | |
3964 | break; | |
3965 | } | |
3966 | } | |
3967 | } | |
3968 | ||
3969 | g_free(filename_full); | |
3970 | g_free(backing_file_full); | |
3971 | g_free(filename_tmp); | |
3972 | return retval; | |
3973 | } | |
3974 | ||
3975 | int bdrv_get_backing_file_depth(BlockDriverState *bs) | |
3976 | { | |
3977 | if (!bs->drv) { | |
3978 | return 0; | |
3979 | } | |
3980 | ||
3981 | if (!bs->backing_hd) { | |
3982 | return 0; | |
3983 | } | |
3984 | ||
3985 | return 1 + bdrv_get_backing_file_depth(bs->backing_hd); | |
3986 | } | |
3987 | ||
3988 | BlockDriverState *bdrv_find_base(BlockDriverState *bs) | |
3989 | { | |
3990 | BlockDriverState *curr_bs = NULL; | |
3991 | ||
3992 | if (!bs) { | |
3993 | return NULL; | |
3994 | } | |
3995 | ||
3996 | curr_bs = bs; | |
3997 | ||
3998 | while (curr_bs->backing_hd) { | |
3999 | curr_bs = curr_bs->backing_hd; | |
4000 | } | |
4001 | return curr_bs; | |
4002 | } | |
4003 | ||
4004 | /**************************************************************/ | |
4005 | /* async I/Os */ | |
4006 | ||
4007 | BlockDriverAIOCB *bdrv_aio_readv(BlockDriverState *bs, int64_t sector_num, | |
4008 | QEMUIOVector *qiov, int nb_sectors, | |
4009 | BlockDriverCompletionFunc *cb, void *opaque) | |
4010 | { | |
4011 | trace_bdrv_aio_readv(bs, sector_num, nb_sectors, opaque); | |
4012 | ||
4013 | return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0, | |
4014 | cb, opaque, false); | |
4015 | } | |
4016 | ||
4017 | BlockDriverAIOCB *bdrv_aio_writev(BlockDriverState *bs, int64_t sector_num, | |
4018 | QEMUIOVector *qiov, int nb_sectors, | |
4019 | BlockDriverCompletionFunc *cb, void *opaque) | |
4020 | { | |
4021 | trace_bdrv_aio_writev(bs, sector_num, nb_sectors, opaque); | |
4022 | ||
4023 | return bdrv_co_aio_rw_vector(bs, sector_num, qiov, nb_sectors, 0, | |
4024 | cb, opaque, true); | |
4025 | } | |
4026 | ||
4027 | BlockDriverAIOCB *bdrv_aio_write_zeroes(BlockDriverState *bs, | |
4028 | int64_t sector_num, int nb_sectors, BdrvRequestFlags flags, | |
4029 | BlockDriverCompletionFunc *cb, void *opaque) | |
4030 | { | |
4031 | trace_bdrv_aio_write_zeroes(bs, sector_num, nb_sectors, flags, opaque); | |
4032 | ||
4033 | return bdrv_co_aio_rw_vector(bs, sector_num, NULL, nb_sectors, | |
4034 | BDRV_REQ_ZERO_WRITE | flags, | |
4035 | cb, opaque, true); | |
4036 | } | |
4037 | ||
4038 | ||
4039 | typedef struct MultiwriteCB { | |
4040 | int error; | |
4041 | int num_requests; | |
4042 | int num_callbacks; | |
4043 | struct { | |
4044 | BlockDriverCompletionFunc *cb; | |
4045 | void *opaque; | |
4046 | QEMUIOVector *free_qiov; | |
4047 | } callbacks[]; | |
4048 | } MultiwriteCB; | |
4049 | ||
4050 | static void multiwrite_user_cb(MultiwriteCB *mcb) | |
4051 | { | |
4052 | int i; | |
4053 | ||
4054 | for (i = 0; i < mcb->num_callbacks; i++) { | |
4055 | mcb->callbacks[i].cb(mcb->callbacks[i].opaque, mcb->error); | |
4056 | if (mcb->callbacks[i].free_qiov) { | |
4057 | qemu_iovec_destroy(mcb->callbacks[i].free_qiov); | |
4058 | } | |
4059 | g_free(mcb->callbacks[i].free_qiov); | |
4060 | } | |
4061 | } | |
4062 | ||
4063 | static void multiwrite_cb(void *opaque, int ret) | |
4064 | { | |
4065 | MultiwriteCB *mcb = opaque; | |
4066 | ||
4067 | trace_multiwrite_cb(mcb, ret); | |
4068 | ||
4069 | if (ret < 0 && !mcb->error) { | |
4070 | mcb->error = ret; | |
4071 | } | |
4072 | ||
4073 | mcb->num_requests--; | |
4074 | if (mcb->num_requests == 0) { | |
4075 | multiwrite_user_cb(mcb); | |
4076 | g_free(mcb); | |
4077 | } | |
4078 | } | |
4079 | ||
4080 | static int multiwrite_req_compare(const void *a, const void *b) | |
4081 | { | |
4082 | const BlockRequest *req1 = a, *req2 = b; | |
4083 | ||
4084 | /* | |
4085 | * Note that we can't simply subtract req2->sector from req1->sector | |
4086 | * here as that could overflow the return value. | |
4087 | */ | |
4088 | if (req1->sector > req2->sector) { | |
4089 | return 1; | |
4090 | } else if (req1->sector < req2->sector) { | |
4091 | return -1; | |
4092 | } else { | |
4093 | return 0; | |
4094 | } | |
4095 | } | |
4096 | ||
4097 | /* | |
4098 | * Takes a bunch of requests and tries to merge them. Returns the number of | |
4099 | * requests that remain after merging. | |
4100 | */ | |
4101 | static int multiwrite_merge(BlockDriverState *bs, BlockRequest *reqs, | |
4102 | int num_reqs, MultiwriteCB *mcb) | |
4103 | { | |
4104 | int i, outidx; | |
4105 | ||
4106 | // Sort requests by start sector | |
4107 | qsort(reqs, num_reqs, sizeof(*reqs), &multiwrite_req_compare); | |
4108 | ||
4109 | // Check if adjacent requests touch the same clusters. If so, combine them, | |
4110 | // filling up gaps with zero sectors. | |
4111 | outidx = 0; | |
4112 | for (i = 1; i < num_reqs; i++) { | |
4113 | int merge = 0; | |
4114 | int64_t oldreq_last = reqs[outidx].sector + reqs[outidx].nb_sectors; | |
4115 | ||
4116 | // Handle exactly sequential writes and overlapping writes. | |
4117 | if (reqs[i].sector <= oldreq_last) { | |
4118 | merge = 1; | |
4119 | } | |
4120 | ||
4121 | if (reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1 > IOV_MAX) { | |
4122 | merge = 0; | |
4123 | } | |
4124 | ||
4125 | if (merge) { | |
4126 | size_t size; | |
4127 | QEMUIOVector *qiov = g_malloc0(sizeof(*qiov)); | |
4128 | qemu_iovec_init(qiov, | |
4129 | reqs[outidx].qiov->niov + reqs[i].qiov->niov + 1); | |
4130 | ||
4131 | // Add the first request to the merged one. If the requests are | |
4132 | // overlapping, drop the last sectors of the first request. | |
4133 | size = (reqs[i].sector - reqs[outidx].sector) << 9; | |
4134 | qemu_iovec_concat(qiov, reqs[outidx].qiov, 0, size); | |
4135 | ||
4136 | // We should need to add any zeros between the two requests | |
4137 | assert (reqs[i].sector <= oldreq_last); | |
4138 | ||
4139 | // Add the second request | |
4140 | qemu_iovec_concat(qiov, reqs[i].qiov, 0, reqs[i].qiov->size); | |
4141 | ||
4142 | reqs[outidx].nb_sectors = qiov->size >> 9; | |
4143 | reqs[outidx].qiov = qiov; | |
4144 | ||
4145 | mcb->callbacks[i].free_qiov = reqs[outidx].qiov; | |
4146 | } else { | |
4147 | outidx++; | |
4148 | reqs[outidx].sector = reqs[i].sector; | |
4149 | reqs[outidx].nb_sectors = reqs[i].nb_sectors; | |
4150 | reqs[outidx].qiov = reqs[i].qiov; | |
4151 | } | |
4152 | } | |
4153 | ||
4154 | return outidx + 1; | |
4155 | } | |
4156 | ||
4157 | /* | |
4158 | * Submit multiple AIO write requests at once. | |
4159 | * | |
4160 | * On success, the function returns 0 and all requests in the reqs array have | |
4161 | * been submitted. In error case this function returns -1, and any of the | |
4162 | * requests may or may not be submitted yet. In particular, this means that the | |
4163 | * callback will be called for some of the requests, for others it won't. The | |
4164 | * caller must check the error field of the BlockRequest to wait for the right | |
4165 | * callbacks (if error != 0, no callback will be called). | |
4166 | * | |
4167 | * The implementation may modify the contents of the reqs array, e.g. to merge | |
4168 | * requests. However, the fields opaque and error are left unmodified as they | |
4169 | * are used to signal failure for a single request to the caller. | |
4170 | */ | |
4171 | int bdrv_aio_multiwrite(BlockDriverState *bs, BlockRequest *reqs, int num_reqs) | |
4172 | { | |
4173 | MultiwriteCB *mcb; | |
4174 | int i; | |
4175 | ||
4176 | /* don't submit writes if we don't have a medium */ | |
4177 | if (bs->drv == NULL) { | |
4178 | for (i = 0; i < num_reqs; i++) { | |
4179 | reqs[i].error = -ENOMEDIUM; | |
4180 | } | |
4181 | return -1; | |
4182 | } | |
4183 | ||
4184 | if (num_reqs == 0) { | |
4185 | return 0; | |
4186 | } | |
4187 | ||
4188 | // Create MultiwriteCB structure | |
4189 | mcb = g_malloc0(sizeof(*mcb) + num_reqs * sizeof(*mcb->callbacks)); | |
4190 | mcb->num_requests = 0; | |
4191 | mcb->num_callbacks = num_reqs; | |
4192 | ||
4193 | for (i = 0; i < num_reqs; i++) { | |
4194 | mcb->callbacks[i].cb = reqs[i].cb; | |
4195 | mcb->callbacks[i].opaque = reqs[i].opaque; | |
4196 | } | |
4197 | ||
4198 | // Check for mergable requests | |
4199 | num_reqs = multiwrite_merge(bs, reqs, num_reqs, mcb); | |
4200 | ||
4201 | trace_bdrv_aio_multiwrite(mcb, mcb->num_callbacks, num_reqs); | |
4202 | ||
4203 | /* Run the aio requests. */ | |
4204 | mcb->num_requests = num_reqs; | |
4205 | for (i = 0; i < num_reqs; i++) { | |
4206 | bdrv_co_aio_rw_vector(bs, reqs[i].sector, reqs[i].qiov, | |
4207 | reqs[i].nb_sectors, reqs[i].flags, | |
4208 | multiwrite_cb, mcb, | |
4209 | true); | |
4210 | } | |
4211 | ||
4212 | return 0; | |
4213 | } | |
4214 | ||
4215 | void bdrv_aio_cancel(BlockDriverAIOCB *acb) | |
4216 | { | |
4217 | acb->aiocb_info->cancel(acb); | |
4218 | } | |
4219 | ||
4220 | /**************************************************************/ | |
4221 | /* async block device emulation */ | |
4222 | ||
4223 | typedef struct BlockDriverAIOCBSync { | |
4224 | BlockDriverAIOCB common; | |
4225 | QEMUBH *bh; | |
4226 | int ret; | |
4227 | /* vector translation state */ | |
4228 | QEMUIOVector *qiov; | |
4229 | uint8_t *bounce; | |
4230 | int is_write; | |
4231 | } BlockDriverAIOCBSync; | |
4232 | ||
4233 | static void bdrv_aio_cancel_em(BlockDriverAIOCB *blockacb) | |
4234 | { | |
4235 | BlockDriverAIOCBSync *acb = | |
4236 | container_of(blockacb, BlockDriverAIOCBSync, common); | |
4237 | qemu_bh_delete(acb->bh); | |
4238 | acb->bh = NULL; | |
4239 | qemu_aio_release(acb); | |
4240 | } | |
4241 | ||
4242 | static const AIOCBInfo bdrv_em_aiocb_info = { | |
4243 | .aiocb_size = sizeof(BlockDriverAIOCBSync), | |
4244 | .cancel = bdrv_aio_cancel_em, | |
4245 | }; | |
4246 | ||
4247 | static void bdrv_aio_bh_cb(void *opaque) | |
4248 | { | |
4249 | BlockDriverAIOCBSync *acb = opaque; | |
4250 | ||
4251 | if (!acb->is_write) | |
4252 | qemu_iovec_from_buf(acb->qiov, 0, acb->bounce, acb->qiov->size); | |
4253 | qemu_vfree(acb->bounce); | |
4254 | acb->common.cb(acb->common.opaque, acb->ret); | |
4255 | qemu_bh_delete(acb->bh); | |
4256 | acb->bh = NULL; | |
4257 | qemu_aio_release(acb); | |
4258 | } | |
4259 | ||
4260 | static BlockDriverAIOCB *bdrv_aio_rw_vector(BlockDriverState *bs, | |
4261 | int64_t sector_num, | |
4262 | QEMUIOVector *qiov, | |
4263 | int nb_sectors, | |
4264 | BlockDriverCompletionFunc *cb, | |
4265 | void *opaque, | |
4266 | int is_write) | |
4267 | ||
4268 | { | |
4269 | BlockDriverAIOCBSync *acb; | |
4270 | ||
4271 | acb = qemu_aio_get(&bdrv_em_aiocb_info, bs, cb, opaque); | |
4272 | acb->is_write = is_write; | |
4273 | acb->qiov = qiov; | |
4274 | acb->bounce = qemu_blockalign(bs, qiov->size); | |
4275 | acb->bh = qemu_bh_new(bdrv_aio_bh_cb, acb); | |
4276 | ||
4277 | if (is_write) { | |
4278 | qemu_iovec_to_buf(acb->qiov, 0, acb->bounce, qiov->size); | |
4279 | acb->ret = bs->drv->bdrv_write(bs, sector_num, acb->bounce, nb_sectors); | |
4280 | } else { | |
4281 | acb->ret = bs->drv->bdrv_read(bs, sector_num, acb->bounce, nb_sectors); | |
4282 | } | |
4283 | ||
4284 | qemu_bh_schedule(acb->bh); | |
4285 | ||
4286 | return &acb->common; | |
4287 | } | |
4288 | ||
4289 | static BlockDriverAIOCB *bdrv_aio_readv_em(BlockDriverState *bs, | |
4290 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
4291 | BlockDriverCompletionFunc *cb, void *opaque) | |
4292 | { | |
4293 | return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 0); | |
4294 | } | |
4295 | ||
4296 | static BlockDriverAIOCB *bdrv_aio_writev_em(BlockDriverState *bs, | |
4297 | int64_t sector_num, QEMUIOVector *qiov, int nb_sectors, | |
4298 | BlockDriverCompletionFunc *cb, void *opaque) | |
4299 | { | |
4300 | return bdrv_aio_rw_vector(bs, sector_num, qiov, nb_sectors, cb, opaque, 1); | |
4301 | } | |
4302 | ||
4303 | ||
4304 | typedef struct BlockDriverAIOCBCoroutine { | |
4305 | BlockDriverAIOCB common; | |
4306 | BlockRequest req; | |
4307 | bool is_write; | |
4308 | bool *done; | |
4309 | QEMUBH* bh; | |
4310 | } BlockDriverAIOCBCoroutine; | |
4311 | ||
4312 | static void bdrv_aio_co_cancel_em(BlockDriverAIOCB *blockacb) | |
4313 | { | |
4314 | BlockDriverAIOCBCoroutine *acb = | |
4315 | container_of(blockacb, BlockDriverAIOCBCoroutine, common); | |
4316 | bool done = false; | |
4317 | ||
4318 | acb->done = &done; | |
4319 | while (!done) { | |
4320 | qemu_aio_wait(); | |
4321 | } | |
4322 | } | |
4323 | ||
4324 | static const AIOCBInfo bdrv_em_co_aiocb_info = { | |
4325 | .aiocb_size = sizeof(BlockDriverAIOCBCoroutine), | |
4326 | .cancel = bdrv_aio_co_cancel_em, | |
4327 | }; | |
4328 | ||
4329 | static void bdrv_co_em_bh(void *opaque) | |
4330 | { | |
4331 | BlockDriverAIOCBCoroutine *acb = opaque; | |
4332 | ||
4333 | acb->common.cb(acb->common.opaque, acb->req.error); | |
4334 | ||
4335 | if (acb->done) { | |
4336 | *acb->done = true; | |
4337 | } | |
4338 | ||
4339 | qemu_bh_delete(acb->bh); | |
4340 | qemu_aio_release(acb); | |
4341 | } | |
4342 | ||
4343 | /* Invoke bdrv_co_do_readv/bdrv_co_do_writev */ | |
4344 | static void coroutine_fn bdrv_co_do_rw(void *opaque) | |
4345 | { | |
4346 | BlockDriverAIOCBCoroutine *acb = opaque; | |
4347 | BlockDriverState *bs = acb->common.bs; | |
4348 | ||
4349 | if (!acb->is_write) { | |
4350 | acb->req.error = bdrv_co_do_readv(bs, acb->req.sector, | |
4351 | acb->req.nb_sectors, acb->req.qiov, acb->req.flags); | |
4352 | } else { | |
4353 | acb->req.error = bdrv_co_do_writev(bs, acb->req.sector, | |
4354 | acb->req.nb_sectors, acb->req.qiov, acb->req.flags); | |
4355 | } | |
4356 | ||
4357 | acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); | |
4358 | qemu_bh_schedule(acb->bh); | |
4359 | } | |
4360 | ||
4361 | static BlockDriverAIOCB *bdrv_co_aio_rw_vector(BlockDriverState *bs, | |
4362 | int64_t sector_num, | |
4363 | QEMUIOVector *qiov, | |
4364 | int nb_sectors, | |
4365 | BdrvRequestFlags flags, | |
4366 | BlockDriverCompletionFunc *cb, | |
4367 | void *opaque, | |
4368 | bool is_write) | |
4369 | { | |
4370 | Coroutine *co; | |
4371 | BlockDriverAIOCBCoroutine *acb; | |
4372 | ||
4373 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
4374 | acb->req.sector = sector_num; | |
4375 | acb->req.nb_sectors = nb_sectors; | |
4376 | acb->req.qiov = qiov; | |
4377 | acb->req.flags = flags; | |
4378 | acb->is_write = is_write; | |
4379 | acb->done = NULL; | |
4380 | ||
4381 | co = qemu_coroutine_create(bdrv_co_do_rw); | |
4382 | qemu_coroutine_enter(co, acb); | |
4383 | ||
4384 | return &acb->common; | |
4385 | } | |
4386 | ||
4387 | static void coroutine_fn bdrv_aio_flush_co_entry(void *opaque) | |
4388 | { | |
4389 | BlockDriverAIOCBCoroutine *acb = opaque; | |
4390 | BlockDriverState *bs = acb->common.bs; | |
4391 | ||
4392 | acb->req.error = bdrv_co_flush(bs); | |
4393 | acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); | |
4394 | qemu_bh_schedule(acb->bh); | |
4395 | } | |
4396 | ||
4397 | BlockDriverAIOCB *bdrv_aio_flush(BlockDriverState *bs, | |
4398 | BlockDriverCompletionFunc *cb, void *opaque) | |
4399 | { | |
4400 | trace_bdrv_aio_flush(bs, opaque); | |
4401 | ||
4402 | Coroutine *co; | |
4403 | BlockDriverAIOCBCoroutine *acb; | |
4404 | ||
4405 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
4406 | acb->done = NULL; | |
4407 | ||
4408 | co = qemu_coroutine_create(bdrv_aio_flush_co_entry); | |
4409 | qemu_coroutine_enter(co, acb); | |
4410 | ||
4411 | return &acb->common; | |
4412 | } | |
4413 | ||
4414 | static void coroutine_fn bdrv_aio_discard_co_entry(void *opaque) | |
4415 | { | |
4416 | BlockDriverAIOCBCoroutine *acb = opaque; | |
4417 | BlockDriverState *bs = acb->common.bs; | |
4418 | ||
4419 | acb->req.error = bdrv_co_discard(bs, acb->req.sector, acb->req.nb_sectors); | |
4420 | acb->bh = qemu_bh_new(bdrv_co_em_bh, acb); | |
4421 | qemu_bh_schedule(acb->bh); | |
4422 | } | |
4423 | ||
4424 | BlockDriverAIOCB *bdrv_aio_discard(BlockDriverState *bs, | |
4425 | int64_t sector_num, int nb_sectors, | |
4426 | BlockDriverCompletionFunc *cb, void *opaque) | |
4427 | { | |
4428 | Coroutine *co; | |
4429 | BlockDriverAIOCBCoroutine *acb; | |
4430 | ||
4431 | trace_bdrv_aio_discard(bs, sector_num, nb_sectors, opaque); | |
4432 | ||
4433 | acb = qemu_aio_get(&bdrv_em_co_aiocb_info, bs, cb, opaque); | |
4434 | acb->req.sector = sector_num; | |
4435 | acb->req.nb_sectors = nb_sectors; | |
4436 | acb->done = NULL; | |
4437 | co = qemu_coroutine_create(bdrv_aio_discard_co_entry); | |
4438 | qemu_coroutine_enter(co, acb); | |
4439 | ||
4440 | return &acb->common; | |
4441 | } | |
4442 | ||
4443 | void bdrv_init(void) | |
4444 | { | |
4445 | module_call_init(MODULE_INIT_BLOCK); | |
4446 | } | |
4447 | ||
4448 | void bdrv_init_with_whitelist(void) | |
4449 | { | |
4450 | use_bdrv_whitelist = 1; | |
4451 | bdrv_init(); | |
4452 | } | |
4453 | ||
4454 | void *qemu_aio_get(const AIOCBInfo *aiocb_info, BlockDriverState *bs, | |
4455 | BlockDriverCompletionFunc *cb, void *opaque) | |
4456 | { | |
4457 | BlockDriverAIOCB *acb; | |
4458 | ||
4459 | acb = g_slice_alloc(aiocb_info->aiocb_size); | |
4460 | acb->aiocb_info = aiocb_info; | |
4461 | acb->bs = bs; | |
4462 | acb->cb = cb; | |
4463 | acb->opaque = opaque; | |
4464 | return acb; | |
4465 | } | |
4466 | ||
4467 | void qemu_aio_release(void *p) | |
4468 | { | |
4469 | BlockDriverAIOCB *acb = p; | |
4470 | g_slice_free1(acb->aiocb_info->aiocb_size, acb); | |
4471 | } | |
4472 | ||
4473 | /**************************************************************/ | |
4474 | /* Coroutine block device emulation */ | |
4475 | ||
4476 | typedef struct CoroutineIOCompletion { | |
4477 | Coroutine *coroutine; | |
4478 | int ret; | |
4479 | } CoroutineIOCompletion; | |
4480 | ||
4481 | static void bdrv_co_io_em_complete(void *opaque, int ret) | |
4482 | { | |
4483 | CoroutineIOCompletion *co = opaque; | |
4484 | ||
4485 | co->ret = ret; | |
4486 | qemu_coroutine_enter(co->coroutine, NULL); | |
4487 | } | |
4488 | ||
4489 | static int coroutine_fn bdrv_co_io_em(BlockDriverState *bs, int64_t sector_num, | |
4490 | int nb_sectors, QEMUIOVector *iov, | |
4491 | bool is_write) | |
4492 | { | |
4493 | CoroutineIOCompletion co = { | |
4494 | .coroutine = qemu_coroutine_self(), | |
4495 | }; | |
4496 | BlockDriverAIOCB *acb; | |
4497 | ||
4498 | if (is_write) { | |
4499 | acb = bs->drv->bdrv_aio_writev(bs, sector_num, iov, nb_sectors, | |
4500 | bdrv_co_io_em_complete, &co); | |
4501 | } else { | |
4502 | acb = bs->drv->bdrv_aio_readv(bs, sector_num, iov, nb_sectors, | |
4503 | bdrv_co_io_em_complete, &co); | |
4504 | } | |
4505 | ||
4506 | trace_bdrv_co_io_em(bs, sector_num, nb_sectors, is_write, acb); | |
4507 | if (!acb) { | |
4508 | return -EIO; | |
4509 | } | |
4510 | qemu_coroutine_yield(); | |
4511 | ||
4512 | return co.ret; | |
4513 | } | |
4514 | ||
4515 | static int coroutine_fn bdrv_co_readv_em(BlockDriverState *bs, | |
4516 | int64_t sector_num, int nb_sectors, | |
4517 | QEMUIOVector *iov) | |
4518 | { | |
4519 | return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, false); | |
4520 | } | |
4521 | ||
4522 | static int coroutine_fn bdrv_co_writev_em(BlockDriverState *bs, | |
4523 | int64_t sector_num, int nb_sectors, | |
4524 | QEMUIOVector *iov) | |
4525 | { | |
4526 | return bdrv_co_io_em(bs, sector_num, nb_sectors, iov, true); | |
4527 | } | |
4528 | ||
4529 | static void coroutine_fn bdrv_flush_co_entry(void *opaque) | |
4530 | { | |
4531 | RwCo *rwco = opaque; | |
4532 | ||
4533 | rwco->ret = bdrv_co_flush(rwco->bs); | |
4534 | } | |
4535 | ||
4536 | int coroutine_fn bdrv_co_flush(BlockDriverState *bs) | |
4537 | { | |
4538 | int ret; | |
4539 | ||
4540 | if (!bs || !bdrv_is_inserted(bs) || bdrv_is_read_only(bs)) { | |
4541 | return 0; | |
4542 | } | |
4543 | ||
4544 | /* Write back cached data to the OS even with cache=unsafe */ | |
4545 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_OS); | |
4546 | if (bs->drv->bdrv_co_flush_to_os) { | |
4547 | ret = bs->drv->bdrv_co_flush_to_os(bs); | |
4548 | if (ret < 0) { | |
4549 | return ret; | |
4550 | } | |
4551 | } | |
4552 | ||
4553 | /* But don't actually force it to the disk with cache=unsafe */ | |
4554 | if (bs->open_flags & BDRV_O_NO_FLUSH) { | |
4555 | goto flush_parent; | |
4556 | } | |
4557 | ||
4558 | BLKDBG_EVENT(bs->file, BLKDBG_FLUSH_TO_DISK); | |
4559 | if (bs->drv->bdrv_co_flush_to_disk) { | |
4560 | ret = bs->drv->bdrv_co_flush_to_disk(bs); | |
4561 | } else if (bs->drv->bdrv_aio_flush) { | |
4562 | BlockDriverAIOCB *acb; | |
4563 | CoroutineIOCompletion co = { | |
4564 | .coroutine = qemu_coroutine_self(), | |
4565 | }; | |
4566 | ||
4567 | acb = bs->drv->bdrv_aio_flush(bs, bdrv_co_io_em_complete, &co); | |
4568 | if (acb == NULL) { | |
4569 | ret = -EIO; | |
4570 | } else { | |
4571 | qemu_coroutine_yield(); | |
4572 | ret = co.ret; | |
4573 | } | |
4574 | } else { | |
4575 | /* | |
4576 | * Some block drivers always operate in either writethrough or unsafe | |
4577 | * mode and don't support bdrv_flush therefore. Usually qemu doesn't | |
4578 | * know how the server works (because the behaviour is hardcoded or | |
4579 | * depends on server-side configuration), so we can't ensure that | |
4580 | * everything is safe on disk. Returning an error doesn't work because | |
4581 | * that would break guests even if the server operates in writethrough | |
4582 | * mode. | |
4583 | * | |
4584 | * Let's hope the user knows what he's doing. | |
4585 | */ | |
4586 | ret = 0; | |
4587 | } | |
4588 | if (ret < 0) { | |
4589 | return ret; | |
4590 | } | |
4591 | ||
4592 | /* Now flush the underlying protocol. It will also have BDRV_O_NO_FLUSH | |
4593 | * in the case of cache=unsafe, so there are no useless flushes. | |
4594 | */ | |
4595 | flush_parent: | |
4596 | return bdrv_co_flush(bs->file); | |
4597 | } | |
4598 | ||
4599 | void bdrv_invalidate_cache(BlockDriverState *bs) | |
4600 | { | |
4601 | if (bs->drv && bs->drv->bdrv_invalidate_cache) { | |
4602 | bs->drv->bdrv_invalidate_cache(bs); | |
4603 | } | |
4604 | } | |
4605 | ||
4606 | void bdrv_invalidate_cache_all(void) | |
4607 | { | |
4608 | BlockDriverState *bs; | |
4609 | ||
4610 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
4611 | bdrv_invalidate_cache(bs); | |
4612 | } | |
4613 | } | |
4614 | ||
4615 | void bdrv_clear_incoming_migration_all(void) | |
4616 | { | |
4617 | BlockDriverState *bs; | |
4618 | ||
4619 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
4620 | bs->open_flags = bs->open_flags & ~(BDRV_O_INCOMING); | |
4621 | } | |
4622 | } | |
4623 | ||
4624 | int bdrv_flush(BlockDriverState *bs) | |
4625 | { | |
4626 | Coroutine *co; | |
4627 | RwCo rwco = { | |
4628 | .bs = bs, | |
4629 | .ret = NOT_DONE, | |
4630 | }; | |
4631 | ||
4632 | if (qemu_in_coroutine()) { | |
4633 | /* Fast-path if already in coroutine context */ | |
4634 | bdrv_flush_co_entry(&rwco); | |
4635 | } else { | |
4636 | co = qemu_coroutine_create(bdrv_flush_co_entry); | |
4637 | qemu_coroutine_enter(co, &rwco); | |
4638 | while (rwco.ret == NOT_DONE) { | |
4639 | qemu_aio_wait(); | |
4640 | } | |
4641 | } | |
4642 | ||
4643 | return rwco.ret; | |
4644 | } | |
4645 | ||
4646 | static void coroutine_fn bdrv_discard_co_entry(void *opaque) | |
4647 | { | |
4648 | RwCo *rwco = opaque; | |
4649 | ||
4650 | rwco->ret = bdrv_co_discard(rwco->bs, rwco->sector_num, rwco->nb_sectors); | |
4651 | } | |
4652 | ||
4653 | /* if no limit is specified in the BlockLimits use a default | |
4654 | * of 32768 512-byte sectors (16 MiB) per request. | |
4655 | */ | |
4656 | #define MAX_DISCARD_DEFAULT 32768 | |
4657 | ||
4658 | int coroutine_fn bdrv_co_discard(BlockDriverState *bs, int64_t sector_num, | |
4659 | int nb_sectors) | |
4660 | { | |
4661 | int max_discard; | |
4662 | ||
4663 | if (!bs->drv) { | |
4664 | return -ENOMEDIUM; | |
4665 | } else if (bdrv_check_request(bs, sector_num, nb_sectors)) { | |
4666 | return -EIO; | |
4667 | } else if (bs->read_only) { | |
4668 | return -EROFS; | |
4669 | } | |
4670 | ||
4671 | bdrv_reset_dirty(bs, sector_num, nb_sectors); | |
4672 | ||
4673 | /* Do nothing if disabled. */ | |
4674 | if (!(bs->open_flags & BDRV_O_UNMAP)) { | |
4675 | return 0; | |
4676 | } | |
4677 | ||
4678 | if (!bs->drv->bdrv_co_discard && !bs->drv->bdrv_aio_discard) { | |
4679 | return 0; | |
4680 | } | |
4681 | ||
4682 | max_discard = bs->bl.max_discard ? bs->bl.max_discard : MAX_DISCARD_DEFAULT; | |
4683 | while (nb_sectors > 0) { | |
4684 | int ret; | |
4685 | int num = nb_sectors; | |
4686 | ||
4687 | /* align request */ | |
4688 | if (bs->bl.discard_alignment && | |
4689 | num >= bs->bl.discard_alignment && | |
4690 | sector_num % bs->bl.discard_alignment) { | |
4691 | if (num > bs->bl.discard_alignment) { | |
4692 | num = bs->bl.discard_alignment; | |
4693 | } | |
4694 | num -= sector_num % bs->bl.discard_alignment; | |
4695 | } | |
4696 | ||
4697 | /* limit request size */ | |
4698 | if (num > max_discard) { | |
4699 | num = max_discard; | |
4700 | } | |
4701 | ||
4702 | if (bs->drv->bdrv_co_discard) { | |
4703 | ret = bs->drv->bdrv_co_discard(bs, sector_num, num); | |
4704 | } else { | |
4705 | BlockDriverAIOCB *acb; | |
4706 | CoroutineIOCompletion co = { | |
4707 | .coroutine = qemu_coroutine_self(), | |
4708 | }; | |
4709 | ||
4710 | acb = bs->drv->bdrv_aio_discard(bs, sector_num, nb_sectors, | |
4711 | bdrv_co_io_em_complete, &co); | |
4712 | if (acb == NULL) { | |
4713 | return -EIO; | |
4714 | } else { | |
4715 | qemu_coroutine_yield(); | |
4716 | ret = co.ret; | |
4717 | } | |
4718 | } | |
4719 | if (ret && ret != -ENOTSUP) { | |
4720 | return ret; | |
4721 | } | |
4722 | ||
4723 | sector_num += num; | |
4724 | nb_sectors -= num; | |
4725 | } | |
4726 | return 0; | |
4727 | } | |
4728 | ||
4729 | int bdrv_discard(BlockDriverState *bs, int64_t sector_num, int nb_sectors) | |
4730 | { | |
4731 | Coroutine *co; | |
4732 | RwCo rwco = { | |
4733 | .bs = bs, | |
4734 | .sector_num = sector_num, | |
4735 | .nb_sectors = nb_sectors, | |
4736 | .ret = NOT_DONE, | |
4737 | }; | |
4738 | ||
4739 | if (qemu_in_coroutine()) { | |
4740 | /* Fast-path if already in coroutine context */ | |
4741 | bdrv_discard_co_entry(&rwco); | |
4742 | } else { | |
4743 | co = qemu_coroutine_create(bdrv_discard_co_entry); | |
4744 | qemu_coroutine_enter(co, &rwco); | |
4745 | while (rwco.ret == NOT_DONE) { | |
4746 | qemu_aio_wait(); | |
4747 | } | |
4748 | } | |
4749 | ||
4750 | return rwco.ret; | |
4751 | } | |
4752 | ||
4753 | /**************************************************************/ | |
4754 | /* removable device support */ | |
4755 | ||
4756 | /** | |
4757 | * Return TRUE if the media is present | |
4758 | */ | |
4759 | int bdrv_is_inserted(BlockDriverState *bs) | |
4760 | { | |
4761 | BlockDriver *drv = bs->drv; | |
4762 | ||
4763 | if (!drv) | |
4764 | return 0; | |
4765 | if (!drv->bdrv_is_inserted) | |
4766 | return 1; | |
4767 | return drv->bdrv_is_inserted(bs); | |
4768 | } | |
4769 | ||
4770 | /** | |
4771 | * Return whether the media changed since the last call to this | |
4772 | * function, or -ENOTSUP if we don't know. Most drivers don't know. | |
4773 | */ | |
4774 | int bdrv_media_changed(BlockDriverState *bs) | |
4775 | { | |
4776 | BlockDriver *drv = bs->drv; | |
4777 | ||
4778 | if (drv && drv->bdrv_media_changed) { | |
4779 | return drv->bdrv_media_changed(bs); | |
4780 | } | |
4781 | return -ENOTSUP; | |
4782 | } | |
4783 | ||
4784 | /** | |
4785 | * If eject_flag is TRUE, eject the media. Otherwise, close the tray | |
4786 | */ | |
4787 | void bdrv_eject(BlockDriverState *bs, bool eject_flag) | |
4788 | { | |
4789 | BlockDriver *drv = bs->drv; | |
4790 | ||
4791 | if (drv && drv->bdrv_eject) { | |
4792 | drv->bdrv_eject(bs, eject_flag); | |
4793 | } | |
4794 | ||
4795 | if (bs->device_name[0] != '\0') { | |
4796 | bdrv_emit_qmp_eject_event(bs, eject_flag); | |
4797 | } | |
4798 | } | |
4799 | ||
4800 | /** | |
4801 | * Lock or unlock the media (if it is locked, the user won't be able | |
4802 | * to eject it manually). | |
4803 | */ | |
4804 | void bdrv_lock_medium(BlockDriverState *bs, bool locked) | |
4805 | { | |
4806 | BlockDriver *drv = bs->drv; | |
4807 | ||
4808 | trace_bdrv_lock_medium(bs, locked); | |
4809 | ||
4810 | if (drv && drv->bdrv_lock_medium) { | |
4811 | drv->bdrv_lock_medium(bs, locked); | |
4812 | } | |
4813 | } | |
4814 | ||
4815 | /* needed for generic scsi interface */ | |
4816 | ||
4817 | int bdrv_ioctl(BlockDriverState *bs, unsigned long int req, void *buf) | |
4818 | { | |
4819 | BlockDriver *drv = bs->drv; | |
4820 | ||
4821 | if (drv && drv->bdrv_ioctl) | |
4822 | return drv->bdrv_ioctl(bs, req, buf); | |
4823 | return -ENOTSUP; | |
4824 | } | |
4825 | ||
4826 | BlockDriverAIOCB *bdrv_aio_ioctl(BlockDriverState *bs, | |
4827 | unsigned long int req, void *buf, | |
4828 | BlockDriverCompletionFunc *cb, void *opaque) | |
4829 | { | |
4830 | BlockDriver *drv = bs->drv; | |
4831 | ||
4832 | if (drv && drv->bdrv_aio_ioctl) | |
4833 | return drv->bdrv_aio_ioctl(bs, req, buf, cb, opaque); | |
4834 | return NULL; | |
4835 | } | |
4836 | ||
4837 | void bdrv_set_guest_block_size(BlockDriverState *bs, int align) | |
4838 | { | |
4839 | bs->guest_block_size = align; | |
4840 | } | |
4841 | ||
4842 | void *qemu_blockalign(BlockDriverState *bs, size_t size) | |
4843 | { | |
4844 | return qemu_memalign(bdrv_opt_mem_align(bs), size); | |
4845 | } | |
4846 | ||
4847 | /* | |
4848 | * Check if all memory in this vector is sector aligned. | |
4849 | */ | |
4850 | bool bdrv_qiov_is_aligned(BlockDriverState *bs, QEMUIOVector *qiov) | |
4851 | { | |
4852 | int i; | |
4853 | size_t alignment = bdrv_opt_mem_align(bs); | |
4854 | ||
4855 | for (i = 0; i < qiov->niov; i++) { | |
4856 | if ((uintptr_t) qiov->iov[i].iov_base % alignment) { | |
4857 | return false; | |
4858 | } | |
4859 | if (qiov->iov[i].iov_len % alignment) { | |
4860 | return false; | |
4861 | } | |
4862 | } | |
4863 | ||
4864 | return true; | |
4865 | } | |
4866 | ||
4867 | BdrvDirtyBitmap *bdrv_create_dirty_bitmap(BlockDriverState *bs, int granularity) | |
4868 | { | |
4869 | int64_t bitmap_size; | |
4870 | BdrvDirtyBitmap *bitmap; | |
4871 | ||
4872 | assert((granularity & (granularity - 1)) == 0); | |
4873 | ||
4874 | granularity >>= BDRV_SECTOR_BITS; | |
4875 | assert(granularity); | |
4876 | bitmap_size = (bdrv_getlength(bs) >> BDRV_SECTOR_BITS); | |
4877 | bitmap = g_malloc0(sizeof(BdrvDirtyBitmap)); | |
4878 | bitmap->bitmap = hbitmap_alloc(bitmap_size, ffs(granularity) - 1); | |
4879 | QLIST_INSERT_HEAD(&bs->dirty_bitmaps, bitmap, list); | |
4880 | return bitmap; | |
4881 | } | |
4882 | ||
4883 | void bdrv_release_dirty_bitmap(BlockDriverState *bs, BdrvDirtyBitmap *bitmap) | |
4884 | { | |
4885 | BdrvDirtyBitmap *bm, *next; | |
4886 | QLIST_FOREACH_SAFE(bm, &bs->dirty_bitmaps, list, next) { | |
4887 | if (bm == bitmap) { | |
4888 | QLIST_REMOVE(bitmap, list); | |
4889 | hbitmap_free(bitmap->bitmap); | |
4890 | g_free(bitmap); | |
4891 | return; | |
4892 | } | |
4893 | } | |
4894 | } | |
4895 | ||
4896 | BlockDirtyInfoList *bdrv_query_dirty_bitmaps(BlockDriverState *bs) | |
4897 | { | |
4898 | BdrvDirtyBitmap *bm; | |
4899 | BlockDirtyInfoList *list = NULL; | |
4900 | BlockDirtyInfoList **plist = &list; | |
4901 | ||
4902 | QLIST_FOREACH(bm, &bs->dirty_bitmaps, list) { | |
4903 | BlockDirtyInfo *info = g_malloc0(sizeof(BlockDirtyInfo)); | |
4904 | BlockDirtyInfoList *entry = g_malloc0(sizeof(BlockDirtyInfoList)); | |
4905 | info->count = bdrv_get_dirty_count(bs, bm); | |
4906 | info->granularity = | |
4907 | ((int64_t) BDRV_SECTOR_SIZE << hbitmap_granularity(bm->bitmap)); | |
4908 | entry->value = info; | |
4909 | *plist = entry; | |
4910 | plist = &entry->next; | |
4911 | } | |
4912 | ||
4913 | return list; | |
4914 | } | |
4915 | ||
4916 | int bdrv_get_dirty(BlockDriverState *bs, BdrvDirtyBitmap *bitmap, int64_t sector) | |
4917 | { | |
4918 | if (bitmap) { | |
4919 | return hbitmap_get(bitmap->bitmap, sector); | |
4920 | } else { | |
4921 | return 0; | |
4922 | } | |
4923 | } | |
4924 | ||
4925 | void bdrv_dirty_iter_init(BlockDriverState *bs, | |
4926 | BdrvDirtyBitmap *bitmap, HBitmapIter *hbi) | |
4927 | { | |
4928 | hbitmap_iter_init(hbi, bitmap->bitmap, 0); | |
4929 | } | |
4930 | ||
4931 | void bdrv_set_dirty(BlockDriverState *bs, int64_t cur_sector, | |
4932 | int nr_sectors) | |
4933 | { | |
4934 | BdrvDirtyBitmap *bitmap; | |
4935 | QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) { | |
4936 | hbitmap_set(bitmap->bitmap, cur_sector, nr_sectors); | |
4937 | } | |
4938 | } | |
4939 | ||
4940 | void bdrv_reset_dirty(BlockDriverState *bs, int64_t cur_sector, int nr_sectors) | |
4941 | { | |
4942 | BdrvDirtyBitmap *bitmap; | |
4943 | QLIST_FOREACH(bitmap, &bs->dirty_bitmaps, list) { | |
4944 | hbitmap_reset(bitmap->bitmap, cur_sector, nr_sectors); | |
4945 | } | |
4946 | } | |
4947 | ||
4948 | int64_t bdrv_get_dirty_count(BlockDriverState *bs, BdrvDirtyBitmap *bitmap) | |
4949 | { | |
4950 | return hbitmap_count(bitmap->bitmap); | |
4951 | } | |
4952 | ||
4953 | /* Get a reference to bs */ | |
4954 | void bdrv_ref(BlockDriverState *bs) | |
4955 | { | |
4956 | bs->refcnt++; | |
4957 | } | |
4958 | ||
4959 | /* Release a previously grabbed reference to bs. | |
4960 | * If after releasing, reference count is zero, the BlockDriverState is | |
4961 | * deleted. */ | |
4962 | void bdrv_unref(BlockDriverState *bs) | |
4963 | { | |
4964 | assert(bs->refcnt > 0); | |
4965 | if (--bs->refcnt == 0) { | |
4966 | bdrv_delete(bs); | |
4967 | } | |
4968 | } | |
4969 | ||
4970 | void bdrv_set_in_use(BlockDriverState *bs, int in_use) | |
4971 | { | |
4972 | assert(bs->in_use != in_use); | |
4973 | bs->in_use = in_use; | |
4974 | } | |
4975 | ||
4976 | int bdrv_in_use(BlockDriverState *bs) | |
4977 | { | |
4978 | return bs->in_use; | |
4979 | } | |
4980 | ||
4981 | void bdrv_iostatus_enable(BlockDriverState *bs) | |
4982 | { | |
4983 | bs->iostatus_enabled = true; | |
4984 | bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; | |
4985 | } | |
4986 | ||
4987 | /* The I/O status is only enabled if the drive explicitly | |
4988 | * enables it _and_ the VM is configured to stop on errors */ | |
4989 | bool bdrv_iostatus_is_enabled(const BlockDriverState *bs) | |
4990 | { | |
4991 | return (bs->iostatus_enabled && | |
4992 | (bs->on_write_error == BLOCKDEV_ON_ERROR_ENOSPC || | |
4993 | bs->on_write_error == BLOCKDEV_ON_ERROR_STOP || | |
4994 | bs->on_read_error == BLOCKDEV_ON_ERROR_STOP)); | |
4995 | } | |
4996 | ||
4997 | void bdrv_iostatus_disable(BlockDriverState *bs) | |
4998 | { | |
4999 | bs->iostatus_enabled = false; | |
5000 | } | |
5001 | ||
5002 | void bdrv_iostatus_reset(BlockDriverState *bs) | |
5003 | { | |
5004 | if (bdrv_iostatus_is_enabled(bs)) { | |
5005 | bs->iostatus = BLOCK_DEVICE_IO_STATUS_OK; | |
5006 | if (bs->job) { | |
5007 | block_job_iostatus_reset(bs->job); | |
5008 | } | |
5009 | } | |
5010 | } | |
5011 | ||
5012 | void bdrv_iostatus_set_err(BlockDriverState *bs, int error) | |
5013 | { | |
5014 | assert(bdrv_iostatus_is_enabled(bs)); | |
5015 | if (bs->iostatus == BLOCK_DEVICE_IO_STATUS_OK) { | |
5016 | bs->iostatus = error == ENOSPC ? BLOCK_DEVICE_IO_STATUS_NOSPACE : | |
5017 | BLOCK_DEVICE_IO_STATUS_FAILED; | |
5018 | } | |
5019 | } | |
5020 | ||
5021 | void | |
5022 | bdrv_acct_start(BlockDriverState *bs, BlockAcctCookie *cookie, int64_t bytes, | |
5023 | enum BlockAcctType type) | |
5024 | { | |
5025 | assert(type < BDRV_MAX_IOTYPE); | |
5026 | ||
5027 | cookie->bytes = bytes; | |
5028 | cookie->start_time_ns = get_clock(); | |
5029 | cookie->type = type; | |
5030 | } | |
5031 | ||
5032 | void | |
5033 | bdrv_acct_done(BlockDriverState *bs, BlockAcctCookie *cookie) | |
5034 | { | |
5035 | assert(cookie->type < BDRV_MAX_IOTYPE); | |
5036 | ||
5037 | bs->nr_bytes[cookie->type] += cookie->bytes; | |
5038 | bs->nr_ops[cookie->type]++; | |
5039 | bs->total_time_ns[cookie->type] += get_clock() - cookie->start_time_ns; | |
5040 | } | |
5041 | ||
5042 | void bdrv_img_create(const char *filename, const char *fmt, | |
5043 | const char *base_filename, const char *base_fmt, | |
5044 | char *options, uint64_t img_size, int flags, | |
5045 | Error **errp, bool quiet) | |
5046 | { | |
5047 | QEMUOptionParameter *param = NULL, *create_options = NULL; | |
5048 | QEMUOptionParameter *backing_fmt, *backing_file, *size; | |
5049 | BlockDriver *drv, *proto_drv; | |
5050 | BlockDriver *backing_drv = NULL; | |
5051 | Error *local_err = NULL; | |
5052 | int ret = 0; | |
5053 | ||
5054 | /* Find driver and parse its options */ | |
5055 | drv = bdrv_find_format(fmt); | |
5056 | if (!drv) { | |
5057 | error_setg(errp, "Unknown file format '%s'", fmt); | |
5058 | return; | |
5059 | } | |
5060 | ||
5061 | proto_drv = bdrv_find_protocol(filename, true); | |
5062 | if (!proto_drv) { | |
5063 | error_setg(errp, "Unknown protocol '%s'", filename); | |
5064 | return; | |
5065 | } | |
5066 | ||
5067 | create_options = append_option_parameters(create_options, | |
5068 | drv->create_options); | |
5069 | create_options = append_option_parameters(create_options, | |
5070 | proto_drv->create_options); | |
5071 | ||
5072 | /* Create parameter list with default values */ | |
5073 | param = parse_option_parameters("", create_options, param); | |
5074 | ||
5075 | set_option_parameter_int(param, BLOCK_OPT_SIZE, img_size); | |
5076 | ||
5077 | /* Parse -o options */ | |
5078 | if (options) { | |
5079 | param = parse_option_parameters(options, create_options, param); | |
5080 | if (param == NULL) { | |
5081 | error_setg(errp, "Invalid options for file format '%s'.", fmt); | |
5082 | goto out; | |
5083 | } | |
5084 | } | |
5085 | ||
5086 | if (base_filename) { | |
5087 | if (set_option_parameter(param, BLOCK_OPT_BACKING_FILE, | |
5088 | base_filename)) { | |
5089 | error_setg(errp, "Backing file not supported for file format '%s'", | |
5090 | fmt); | |
5091 | goto out; | |
5092 | } | |
5093 | } | |
5094 | ||
5095 | if (base_fmt) { | |
5096 | if (set_option_parameter(param, BLOCK_OPT_BACKING_FMT, base_fmt)) { | |
5097 | error_setg(errp, "Backing file format not supported for file " | |
5098 | "format '%s'", fmt); | |
5099 | goto out; | |
5100 | } | |
5101 | } | |
5102 | ||
5103 | backing_file = get_option_parameter(param, BLOCK_OPT_BACKING_FILE); | |
5104 | if (backing_file && backing_file->value.s) { | |
5105 | if (!strcmp(filename, backing_file->value.s)) { | |
5106 | error_setg(errp, "Error: Trying to create an image with the " | |
5107 | "same filename as the backing file"); | |
5108 | goto out; | |
5109 | } | |
5110 | } | |
5111 | ||
5112 | backing_fmt = get_option_parameter(param, BLOCK_OPT_BACKING_FMT); | |
5113 | if (backing_fmt && backing_fmt->value.s) { | |
5114 | backing_drv = bdrv_find_format(backing_fmt->value.s); | |
5115 | if (!backing_drv) { | |
5116 | error_setg(errp, "Unknown backing file format '%s'", | |
5117 | backing_fmt->value.s); | |
5118 | goto out; | |
5119 | } | |
5120 | } | |
5121 | ||
5122 | // The size for the image must always be specified, with one exception: | |
5123 | // If we are using a backing file, we can obtain the size from there | |
5124 | size = get_option_parameter(param, BLOCK_OPT_SIZE); | |
5125 | if (size && size->value.n == -1) { | |
5126 | if (backing_file && backing_file->value.s) { | |
5127 | BlockDriverState *bs; | |
5128 | uint64_t size; | |
5129 | char buf[32]; | |
5130 | int back_flags; | |
5131 | ||
5132 | /* backing files always opened read-only */ | |
5133 | back_flags = | |
5134 | flags & ~(BDRV_O_RDWR | BDRV_O_SNAPSHOT | BDRV_O_NO_BACKING); | |
5135 | ||
5136 | bs = bdrv_new(""); | |
5137 | ||
5138 | ret = bdrv_open(bs, backing_file->value.s, NULL, back_flags, | |
5139 | backing_drv, &local_err); | |
5140 | if (ret < 0) { | |
5141 | error_setg_errno(errp, -ret, "Could not open '%s': %s", | |
5142 | backing_file->value.s, | |
5143 | error_get_pretty(local_err)); | |
5144 | error_free(local_err); | |
5145 | local_err = NULL; | |
5146 | bdrv_unref(bs); | |
5147 | goto out; | |
5148 | } | |
5149 | bdrv_get_geometry(bs, &size); | |
5150 | size *= 512; | |
5151 | ||
5152 | snprintf(buf, sizeof(buf), "%" PRId64, size); | |
5153 | set_option_parameter(param, BLOCK_OPT_SIZE, buf); | |
5154 | ||
5155 | bdrv_unref(bs); | |
5156 | } else { | |
5157 | error_setg(errp, "Image creation needs a size parameter"); | |
5158 | goto out; | |
5159 | } | |
5160 | } | |
5161 | ||
5162 | if (!quiet) { | |
5163 | printf("Formatting '%s', fmt=%s ", filename, fmt); | |
5164 | print_option_parameters(param); | |
5165 | puts(""); | |
5166 | } | |
5167 | ret = bdrv_create(drv, filename, param, &local_err); | |
5168 | if (ret == -EFBIG) { | |
5169 | /* This is generally a better message than whatever the driver would | |
5170 | * deliver (especially because of the cluster_size_hint), since that | |
5171 | * is most probably not much different from "image too large". */ | |
5172 | const char *cluster_size_hint = ""; | |
5173 | if (get_option_parameter(create_options, BLOCK_OPT_CLUSTER_SIZE)) { | |
5174 | cluster_size_hint = " (try using a larger cluster size)"; | |
5175 | } | |
5176 | error_setg(errp, "The image size is too large for file format '%s'" | |
5177 | "%s", fmt, cluster_size_hint); | |
5178 | error_free(local_err); | |
5179 | local_err = NULL; | |
5180 | } | |
5181 | ||
5182 | out: | |
5183 | free_option_parameters(create_options); | |
5184 | free_option_parameters(param); | |
5185 | ||
5186 | if (error_is_set(&local_err)) { | |
5187 | error_propagate(errp, local_err); | |
5188 | } | |
5189 | } | |
5190 | ||
5191 | AioContext *bdrv_get_aio_context(BlockDriverState *bs) | |
5192 | { | |
5193 | /* Currently BlockDriverState always uses the main loop AioContext */ | |
5194 | return qemu_get_aio_context(); | |
5195 | } | |
5196 | ||
5197 | void bdrv_add_before_write_notifier(BlockDriverState *bs, | |
5198 | NotifierWithReturn *notifier) | |
5199 | { | |
5200 | notifier_with_return_list_add(&bs->before_write_notifiers, notifier); | |
5201 | } | |
5202 | ||
5203 | int bdrv_amend_options(BlockDriverState *bs, QEMUOptionParameter *options) | |
5204 | { | |
5205 | if (bs->drv->bdrv_amend_options == NULL) { | |
5206 | return -ENOTSUP; | |
5207 | } | |
5208 | return bs->drv->bdrv_amend_options(bs, options); | |
5209 | } | |
5210 | ||
5211 | /* Used to recurse on single child block filters. | |
5212 | * Single child block filter will store their child in bs->file. | |
5213 | */ | |
5214 | bool bdrv_generic_is_first_non_filter(BlockDriverState *bs, | |
5215 | BlockDriverState *candidate) | |
5216 | { | |
5217 | if (!bs->drv) { | |
5218 | return false; | |
5219 | } | |
5220 | ||
5221 | if (!bs->drv->authorizations[BS_IS_A_FILTER]) { | |
5222 | if (bs == candidate) { | |
5223 | return true; | |
5224 | } else { | |
5225 | return false; | |
5226 | } | |
5227 | } | |
5228 | ||
5229 | if (!bs->drv->authorizations[BS_FILTER_PASS_DOWN]) { | |
5230 | return false; | |
5231 | } | |
5232 | ||
5233 | if (!bs->file) { | |
5234 | return false; | |
5235 | } | |
5236 | ||
5237 | return bdrv_recurse_is_first_non_filter(bs->file, candidate); | |
5238 | } | |
5239 | ||
5240 | bool bdrv_recurse_is_first_non_filter(BlockDriverState *bs, | |
5241 | BlockDriverState *candidate) | |
5242 | { | |
5243 | if (bs->drv && bs->drv->bdrv_recurse_is_first_non_filter) { | |
5244 | return bs->drv->bdrv_recurse_is_first_non_filter(bs, candidate); | |
5245 | } | |
5246 | ||
5247 | return bdrv_generic_is_first_non_filter(bs, candidate); | |
5248 | } | |
5249 | ||
5250 | /* This function checks if the candidate is the first non filter bs down it's | |
5251 | * bs chain. Since we don't have pointers to parents it explore all bs chains | |
5252 | * from the top. Some filters can choose not to pass down the recursion. | |
5253 | */ | |
5254 | bool bdrv_is_first_non_filter(BlockDriverState *candidate) | |
5255 | { | |
5256 | BlockDriverState *bs; | |
5257 | ||
5258 | /* walk down the bs forest recursively */ | |
5259 | QTAILQ_FOREACH(bs, &bdrv_states, device_list) { | |
5260 | bool perm; | |
5261 | ||
5262 | if (!bs->file) { | |
5263 | continue; | |
5264 | } | |
5265 | ||
5266 | perm = bdrv_recurse_is_first_non_filter(bs->file, candidate); | |
5267 | ||
5268 | /* candidate is the first non filter */ | |
5269 | if (perm) { | |
5270 | return true; | |
5271 | } | |
5272 | } | |
5273 | ||
5274 | return false; | |
5275 | } |