tests: virtio-9p: rename PCI configuration test
[qemu.git] / block / vdi.c
1 /*
2 * Block driver for the Virtual Disk Image (VDI) format
3 *
4 * Copyright (c) 2009, 2012 Stefan Weil
5 *
6 * This program is free software: you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation, either version 2 of the License, or
9 * (at your option) version 3 or any later version.
10 *
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
15 *
16 * You should have received a copy of the GNU General Public License
17 * along with this program. If not, see <http://www.gnu.org/licenses/>.
18 *
19 * Reference:
20 * http://forums.virtualbox.org/viewtopic.php?t=8046
21 *
22 * This driver supports create / read / write operations on VDI images.
23 *
24 * Todo (see also TODO in code):
25 *
26 * Some features like snapshots are still missing.
27 *
28 * Deallocation of zero-filled blocks and shrinking images are missing, too
29 * (might be added to common block layer).
30 *
31 * Allocation of blocks could be optimized (less writes to block map and
32 * header).
33 *
34 * Read and write of adjacent blocks could be done in one operation
35 * (current code uses one operation per block (1 MiB).
36 *
37 * The code is not thread safe (missing locks for changes in header and
38 * block table, no problem with current QEMU).
39 *
40 * Hints:
41 *
42 * Blocks (VDI documentation) correspond to clusters (QEMU).
43 * QEMU's backing files could be implemented using VDI snapshot files (TODO).
44 * VDI snapshot files may also contain the complete machine state.
45 * Maybe this machine state can be converted to QEMU PC machine snapshot data.
46 *
47 * The driver keeps a block cache (little endian entries) in memory.
48 * For the standard block size (1 MiB), a 1 TiB disk will use 4 MiB RAM,
49 * so this seems to be reasonable.
50 */
51
52 #include "qemu/osdep.h"
53 #include "qapi/error.h"
54 #include "block/block_int.h"
55 #include "sysemu/block-backend.h"
56 #include "qemu/module.h"
57 #include "qemu/bswap.h"
58 #include "migration/migration.h"
59 #include "qemu/coroutine.h"
60 #include "qemu/cutils.h"
61 #include "qemu/uuid.h"
62
63 /* Code configuration options. */
64
65 /* Enable debug messages. */
66 //~ #define CONFIG_VDI_DEBUG
67
68 /* Support write operations on VDI images. */
69 #define CONFIG_VDI_WRITE
70
71 /* Support non-standard block (cluster) size. This is untested.
72 * Maybe it will be needed for very large images.
73 */
74 //~ #define CONFIG_VDI_BLOCK_SIZE
75
76 /* Support static (fixed, pre-allocated) images. */
77 #define CONFIG_VDI_STATIC_IMAGE
78
79 /* Command line option for static images. */
80 #define BLOCK_OPT_STATIC "static"
81
82 #define KiB 1024
83 #define MiB (KiB * KiB)
84
85 #define SECTOR_SIZE 512
86 #define DEFAULT_CLUSTER_SIZE (1 * MiB)
87
88 #if defined(CONFIG_VDI_DEBUG)
89 #define logout(fmt, ...) \
90 fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__)
91 #else
92 #define logout(fmt, ...) ((void)0)
93 #endif
94
95 /* Image signature. */
96 #define VDI_SIGNATURE 0xbeda107f
97
98 /* Image version. */
99 #define VDI_VERSION_1_1 0x00010001
100
101 /* Image type. */
102 #define VDI_TYPE_DYNAMIC 1
103 #define VDI_TYPE_STATIC 2
104
105 /* Innotek / SUN images use these strings in header.text:
106 * "<<< innotek VirtualBox Disk Image >>>\n"
107 * "<<< Sun xVM VirtualBox Disk Image >>>\n"
108 * "<<< Sun VirtualBox Disk Image >>>\n"
109 * The value does not matter, so QEMU created images use a different text.
110 */
111 #define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n"
112
113 /* A never-allocated block; semantically arbitrary content. */
114 #define VDI_UNALLOCATED 0xffffffffU
115
116 /* A discarded (no longer allocated) block; semantically zero-filled. */
117 #define VDI_DISCARDED 0xfffffffeU
118
119 #define VDI_IS_ALLOCATED(X) ((X) < VDI_DISCARDED)
120
121 /* The bmap will take up VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) bytes; since
122 * the bmap is read and written in a single operation, its size needs to be
123 * limited to INT_MAX; furthermore, when opening an image, the bmap size is
124 * rounded up to be aligned on BDRV_SECTOR_SIZE.
125 * Therefore this should satisfy the following:
126 * VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) + BDRV_SECTOR_SIZE == INT_MAX + 1
127 * (INT_MAX + 1 is the first value not representable as an int)
128 * This guarantees that any value below or equal to the constant will, when
129 * multiplied by sizeof(uint32_t) and rounded up to a BDRV_SECTOR_SIZE boundary,
130 * still be below or equal to INT_MAX. */
131 #define VDI_BLOCKS_IN_IMAGE_MAX \
132 ((unsigned)((INT_MAX + 1u - BDRV_SECTOR_SIZE) / sizeof(uint32_t)))
133 #define VDI_DISK_SIZE_MAX ((uint64_t)VDI_BLOCKS_IN_IMAGE_MAX * \
134 (uint64_t)DEFAULT_CLUSTER_SIZE)
135
136 typedef struct {
137 char text[0x40];
138 uint32_t signature;
139 uint32_t version;
140 uint32_t header_size;
141 uint32_t image_type;
142 uint32_t image_flags;
143 char description[256];
144 uint32_t offset_bmap;
145 uint32_t offset_data;
146 uint32_t cylinders; /* disk geometry, unused here */
147 uint32_t heads; /* disk geometry, unused here */
148 uint32_t sectors; /* disk geometry, unused here */
149 uint32_t sector_size;
150 uint32_t unused1;
151 uint64_t disk_size;
152 uint32_t block_size;
153 uint32_t block_extra; /* unused here */
154 uint32_t blocks_in_image;
155 uint32_t blocks_allocated;
156 QemuUUID uuid_image;
157 QemuUUID uuid_last_snap;
158 QemuUUID uuid_link;
159 QemuUUID uuid_parent;
160 uint64_t unused2[7];
161 } QEMU_PACKED VdiHeader;
162
163 typedef struct {
164 /* The block map entries are little endian (even in memory). */
165 uint32_t *bmap;
166 /* Size of block (bytes). */
167 uint32_t block_size;
168 /* Size of block (sectors). */
169 uint32_t block_sectors;
170 /* First sector of block map. */
171 uint32_t bmap_sector;
172 /* VDI header (converted to host endianness). */
173 VdiHeader header;
174
175 CoMutex write_lock;
176
177 Error *migration_blocker;
178 } BDRVVdiState;
179
180 static void vdi_header_to_cpu(VdiHeader *header)
181 {
182 le32_to_cpus(&header->signature);
183 le32_to_cpus(&header->version);
184 le32_to_cpus(&header->header_size);
185 le32_to_cpus(&header->image_type);
186 le32_to_cpus(&header->image_flags);
187 le32_to_cpus(&header->offset_bmap);
188 le32_to_cpus(&header->offset_data);
189 le32_to_cpus(&header->cylinders);
190 le32_to_cpus(&header->heads);
191 le32_to_cpus(&header->sectors);
192 le32_to_cpus(&header->sector_size);
193 le64_to_cpus(&header->disk_size);
194 le32_to_cpus(&header->block_size);
195 le32_to_cpus(&header->block_extra);
196 le32_to_cpus(&header->blocks_in_image);
197 le32_to_cpus(&header->blocks_allocated);
198 qemu_uuid_bswap(&header->uuid_image);
199 qemu_uuid_bswap(&header->uuid_last_snap);
200 qemu_uuid_bswap(&header->uuid_link);
201 qemu_uuid_bswap(&header->uuid_parent);
202 }
203
204 static void vdi_header_to_le(VdiHeader *header)
205 {
206 cpu_to_le32s(&header->signature);
207 cpu_to_le32s(&header->version);
208 cpu_to_le32s(&header->header_size);
209 cpu_to_le32s(&header->image_type);
210 cpu_to_le32s(&header->image_flags);
211 cpu_to_le32s(&header->offset_bmap);
212 cpu_to_le32s(&header->offset_data);
213 cpu_to_le32s(&header->cylinders);
214 cpu_to_le32s(&header->heads);
215 cpu_to_le32s(&header->sectors);
216 cpu_to_le32s(&header->sector_size);
217 cpu_to_le64s(&header->disk_size);
218 cpu_to_le32s(&header->block_size);
219 cpu_to_le32s(&header->block_extra);
220 cpu_to_le32s(&header->blocks_in_image);
221 cpu_to_le32s(&header->blocks_allocated);
222 qemu_uuid_bswap(&header->uuid_image);
223 qemu_uuid_bswap(&header->uuid_last_snap);
224 qemu_uuid_bswap(&header->uuid_link);
225 qemu_uuid_bswap(&header->uuid_parent);
226 }
227
228 #if defined(CONFIG_VDI_DEBUG)
229 static void vdi_header_print(VdiHeader *header)
230 {
231 char uuid[37];
232 logout("text %s", header->text);
233 logout("signature 0x%08x\n", header->signature);
234 logout("header size 0x%04x\n", header->header_size);
235 logout("image type 0x%04x\n", header->image_type);
236 logout("image flags 0x%04x\n", header->image_flags);
237 logout("description %s\n", header->description);
238 logout("offset bmap 0x%04x\n", header->offset_bmap);
239 logout("offset data 0x%04x\n", header->offset_data);
240 logout("cylinders 0x%04x\n", header->cylinders);
241 logout("heads 0x%04x\n", header->heads);
242 logout("sectors 0x%04x\n", header->sectors);
243 logout("sector size 0x%04x\n", header->sector_size);
244 logout("image size 0x%" PRIx64 " B (%" PRIu64 " MiB)\n",
245 header->disk_size, header->disk_size / MiB);
246 logout("block size 0x%04x\n", header->block_size);
247 logout("block extra 0x%04x\n", header->block_extra);
248 logout("blocks tot. 0x%04x\n", header->blocks_in_image);
249 logout("blocks all. 0x%04x\n", header->blocks_allocated);
250 uuid_unparse(header->uuid_image, uuid);
251 logout("uuid image %s\n", uuid);
252 uuid_unparse(header->uuid_last_snap, uuid);
253 logout("uuid snap %s\n", uuid);
254 uuid_unparse(header->uuid_link, uuid);
255 logout("uuid link %s\n", uuid);
256 uuid_unparse(header->uuid_parent, uuid);
257 logout("uuid parent %s\n", uuid);
258 }
259 #endif
260
261 static int vdi_check(BlockDriverState *bs, BdrvCheckResult *res,
262 BdrvCheckMode fix)
263 {
264 /* TODO: additional checks possible. */
265 BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
266 uint32_t blocks_allocated = 0;
267 uint32_t block;
268 uint32_t *bmap;
269 logout("\n");
270
271 if (fix) {
272 return -ENOTSUP;
273 }
274
275 bmap = g_try_new(uint32_t, s->header.blocks_in_image);
276 if (s->header.blocks_in_image && bmap == NULL) {
277 res->check_errors++;
278 return -ENOMEM;
279 }
280
281 memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t));
282
283 /* Check block map and value of blocks_allocated. */
284 for (block = 0; block < s->header.blocks_in_image; block++) {
285 uint32_t bmap_entry = le32_to_cpu(s->bmap[block]);
286 if (VDI_IS_ALLOCATED(bmap_entry)) {
287 if (bmap_entry < s->header.blocks_in_image) {
288 blocks_allocated++;
289 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) {
290 bmap[bmap_entry] = bmap_entry;
291 } else {
292 fprintf(stderr, "ERROR: block index %" PRIu32
293 " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry);
294 res->corruptions++;
295 }
296 } else {
297 fprintf(stderr, "ERROR: block index %" PRIu32
298 " too large, is %" PRIu32 "\n", block, bmap_entry);
299 res->corruptions++;
300 }
301 }
302 }
303 if (blocks_allocated != s->header.blocks_allocated) {
304 fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32
305 ", should be %" PRIu32 "\n",
306 blocks_allocated, s->header.blocks_allocated);
307 res->corruptions++;
308 }
309
310 g_free(bmap);
311
312 return 0;
313 }
314
315 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
316 {
317 /* TODO: vdi_get_info would be needed for machine snapshots.
318 vm_state_offset is still missing. */
319 BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
320 logout("\n");
321 bdi->cluster_size = s->block_size;
322 bdi->vm_state_offset = 0;
323 bdi->unallocated_blocks_are_zero = true;
324 return 0;
325 }
326
327 static int vdi_make_empty(BlockDriverState *bs)
328 {
329 /* TODO: missing code. */
330 logout("\n");
331 /* The return value for missing code must be 0, see block.c. */
332 return 0;
333 }
334
335 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename)
336 {
337 const VdiHeader *header = (const VdiHeader *)buf;
338 int ret = 0;
339
340 logout("\n");
341
342 if (buf_size < sizeof(*header)) {
343 /* Header too small, no VDI. */
344 } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) {
345 ret = 100;
346 }
347
348 if (ret == 0) {
349 logout("no vdi image\n");
350 } else {
351 logout("%s", header->text);
352 }
353
354 return ret;
355 }
356
357 static int vdi_open(BlockDriverState *bs, QDict *options, int flags,
358 Error **errp)
359 {
360 BDRVVdiState *s = bs->opaque;
361 VdiHeader header;
362 size_t bmap_size;
363 int ret;
364
365 logout("\n");
366
367 ret = bdrv_read(bs->file, 0, (uint8_t *)&header, 1);
368 if (ret < 0) {
369 goto fail;
370 }
371
372 vdi_header_to_cpu(&header);
373 #if defined(CONFIG_VDI_DEBUG)
374 vdi_header_print(&header);
375 #endif
376
377 if (header.disk_size > VDI_DISK_SIZE_MAX) {
378 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
379 ", max supported is 0x%" PRIx64 ")",
380 header.disk_size, VDI_DISK_SIZE_MAX);
381 ret = -ENOTSUP;
382 goto fail;
383 }
384
385 if (header.disk_size % SECTOR_SIZE != 0) {
386 /* 'VBoxManage convertfromraw' can create images with odd disk sizes.
387 We accept them but round the disk size to the next multiple of
388 SECTOR_SIZE. */
389 logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size);
390 header.disk_size = ROUND_UP(header.disk_size, SECTOR_SIZE);
391 }
392
393 if (header.signature != VDI_SIGNATURE) {
394 error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32
395 ")", header.signature);
396 ret = -EINVAL;
397 goto fail;
398 } else if (header.version != VDI_VERSION_1_1) {
399 error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32
400 ")", header.version >> 16, header.version & 0xffff);
401 ret = -ENOTSUP;
402 goto fail;
403 } else if (header.offset_bmap % SECTOR_SIZE != 0) {
404 /* We only support block maps which start on a sector boundary. */
405 error_setg(errp, "unsupported VDI image (unaligned block map offset "
406 "0x%" PRIx32 ")", header.offset_bmap);
407 ret = -ENOTSUP;
408 goto fail;
409 } else if (header.offset_data % SECTOR_SIZE != 0) {
410 /* We only support data blocks which start on a sector boundary. */
411 error_setg(errp, "unsupported VDI image (unaligned data offset 0x%"
412 PRIx32 ")", header.offset_data);
413 ret = -ENOTSUP;
414 goto fail;
415 } else if (header.sector_size != SECTOR_SIZE) {
416 error_setg(errp, "unsupported VDI image (sector size %" PRIu32
417 " is not %u)", header.sector_size, SECTOR_SIZE);
418 ret = -ENOTSUP;
419 goto fail;
420 } else if (header.block_size != DEFAULT_CLUSTER_SIZE) {
421 error_setg(errp, "unsupported VDI image (block size %" PRIu32
422 " is not %u)", header.block_size, DEFAULT_CLUSTER_SIZE);
423 ret = -ENOTSUP;
424 goto fail;
425 } else if (header.disk_size >
426 (uint64_t)header.blocks_in_image * header.block_size) {
427 error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", "
428 "image bitmap has room for %" PRIu64 ")",
429 header.disk_size,
430 (uint64_t)header.blocks_in_image * header.block_size);
431 ret = -ENOTSUP;
432 goto fail;
433 } else if (!qemu_uuid_is_null(&header.uuid_link)) {
434 error_setg(errp, "unsupported VDI image (non-NULL link UUID)");
435 ret = -ENOTSUP;
436 goto fail;
437 } else if (!qemu_uuid_is_null(&header.uuid_parent)) {
438 error_setg(errp, "unsupported VDI image (non-NULL parent UUID)");
439 ret = -ENOTSUP;
440 goto fail;
441 } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) {
442 error_setg(errp, "unsupported VDI image "
443 "(too many blocks %u, max is %u)",
444 header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX);
445 ret = -ENOTSUP;
446 goto fail;
447 }
448
449 bs->total_sectors = header.disk_size / SECTOR_SIZE;
450
451 s->block_size = header.block_size;
452 s->block_sectors = header.block_size / SECTOR_SIZE;
453 s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
454 s->header = header;
455
456 bmap_size = header.blocks_in_image * sizeof(uint32_t);
457 bmap_size = DIV_ROUND_UP(bmap_size, SECTOR_SIZE);
458 s->bmap = qemu_try_blockalign(bs->file->bs, bmap_size * SECTOR_SIZE);
459 if (s->bmap == NULL) {
460 ret = -ENOMEM;
461 goto fail;
462 }
463
464 ret = bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap,
465 bmap_size);
466 if (ret < 0) {
467 goto fail_free_bmap;
468 }
469
470 /* Disable migration when vdi images are used */
471 error_setg(&s->migration_blocker, "The vdi format used by node '%s' "
472 "does not support live migration",
473 bdrv_get_device_or_node_name(bs));
474 migrate_add_blocker(s->migration_blocker);
475
476 qemu_co_mutex_init(&s->write_lock);
477
478 return 0;
479
480 fail_free_bmap:
481 qemu_vfree(s->bmap);
482
483 fail:
484 return ret;
485 }
486
487 static int vdi_reopen_prepare(BDRVReopenState *state,
488 BlockReopenQueue *queue, Error **errp)
489 {
490 return 0;
491 }
492
493 static int64_t coroutine_fn vdi_co_get_block_status(BlockDriverState *bs,
494 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
495 {
496 /* TODO: Check for too large sector_num (in bdrv_is_allocated or here). */
497 BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
498 size_t bmap_index = sector_num / s->block_sectors;
499 size_t sector_in_block = sector_num % s->block_sectors;
500 int n_sectors = s->block_sectors - sector_in_block;
501 uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
502 uint64_t offset;
503 int result;
504
505 logout("%p, %" PRId64 ", %d, %p\n", bs, sector_num, nb_sectors, pnum);
506 if (n_sectors > nb_sectors) {
507 n_sectors = nb_sectors;
508 }
509 *pnum = n_sectors;
510 result = VDI_IS_ALLOCATED(bmap_entry);
511 if (!result) {
512 return 0;
513 }
514
515 offset = s->header.offset_data +
516 (uint64_t)bmap_entry * s->block_size +
517 sector_in_block * SECTOR_SIZE;
518 *file = bs->file->bs;
519 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset;
520 }
521
522 static int coroutine_fn
523 vdi_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
524 QEMUIOVector *qiov, int flags)
525 {
526 BDRVVdiState *s = bs->opaque;
527 QEMUIOVector local_qiov;
528 uint32_t bmap_entry;
529 uint32_t block_index;
530 uint32_t offset_in_block;
531 uint32_t n_bytes;
532 uint64_t bytes_done = 0;
533 int ret = 0;
534
535 logout("\n");
536
537 qemu_iovec_init(&local_qiov, qiov->niov);
538
539 while (ret >= 0 && bytes > 0) {
540 block_index = offset / s->block_size;
541 offset_in_block = offset % s->block_size;
542 n_bytes = MIN(bytes, s->block_size - offset_in_block);
543
544 logout("will read %u bytes starting at offset %" PRIu64 "\n",
545 n_bytes, offset);
546
547 /* prepare next AIO request */
548 bmap_entry = le32_to_cpu(s->bmap[block_index]);
549 if (!VDI_IS_ALLOCATED(bmap_entry)) {
550 /* Block not allocated, return zeros, no need to wait. */
551 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
552 ret = 0;
553 } else {
554 uint64_t data_offset = s->header.offset_data +
555 (uint64_t)bmap_entry * s->block_size +
556 offset_in_block;
557
558 qemu_iovec_reset(&local_qiov);
559 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
560
561 ret = bdrv_co_preadv(bs->file, data_offset, n_bytes,
562 &local_qiov, 0);
563 }
564 logout("%u bytes read\n", n_bytes);
565
566 bytes -= n_bytes;
567 offset += n_bytes;
568 bytes_done += n_bytes;
569 }
570
571 qemu_iovec_destroy(&local_qiov);
572
573 return ret;
574 }
575
576 static int coroutine_fn
577 vdi_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
578 QEMUIOVector *qiov, int flags)
579 {
580 BDRVVdiState *s = bs->opaque;
581 QEMUIOVector local_qiov;
582 uint32_t bmap_entry;
583 uint32_t block_index;
584 uint32_t offset_in_block;
585 uint32_t n_bytes;
586 uint32_t bmap_first = VDI_UNALLOCATED;
587 uint32_t bmap_last = VDI_UNALLOCATED;
588 uint8_t *block = NULL;
589 uint64_t bytes_done = 0;
590 int ret = 0;
591
592 logout("\n");
593
594 qemu_iovec_init(&local_qiov, qiov->niov);
595
596 while (ret >= 0 && bytes > 0) {
597 block_index = offset / s->block_size;
598 offset_in_block = offset % s->block_size;
599 n_bytes = MIN(bytes, s->block_size - offset_in_block);
600
601 logout("will write %u bytes starting at offset %" PRIu64 "\n",
602 n_bytes, offset);
603
604 /* prepare next AIO request */
605 bmap_entry = le32_to_cpu(s->bmap[block_index]);
606 if (!VDI_IS_ALLOCATED(bmap_entry)) {
607 /* Allocate new block and write to it. */
608 uint64_t data_offset;
609 bmap_entry = s->header.blocks_allocated;
610 s->bmap[block_index] = cpu_to_le32(bmap_entry);
611 s->header.blocks_allocated++;
612 data_offset = s->header.offset_data +
613 (uint64_t)bmap_entry * s->block_size;
614 if (block == NULL) {
615 block = g_malloc(s->block_size);
616 bmap_first = block_index;
617 }
618 bmap_last = block_index;
619 /* Copy data to be written to new block and zero unused parts. */
620 memset(block, 0, offset_in_block);
621 qemu_iovec_to_buf(qiov, bytes_done, block + offset_in_block,
622 n_bytes);
623 memset(block + offset_in_block + n_bytes, 0,
624 s->block_size - n_bytes - offset_in_block);
625
626 /* Note that this coroutine does not yield anywhere from reading the
627 * bmap entry until here, so in regards to all the coroutines trying
628 * to write to this cluster, the one doing the allocation will
629 * always be the first to try to acquire the lock.
630 * Therefore, it is also the first that will actually be able to
631 * acquire the lock and thus the padded cluster is written before
632 * the other coroutines can write to the affected area. */
633 qemu_co_mutex_lock(&s->write_lock);
634 ret = bdrv_pwrite(bs->file, data_offset, block, s->block_size);
635 qemu_co_mutex_unlock(&s->write_lock);
636 } else {
637 uint64_t data_offset = s->header.offset_data +
638 (uint64_t)bmap_entry * s->block_size +
639 offset_in_block;
640 qemu_co_mutex_lock(&s->write_lock);
641 /* This lock is only used to make sure the following write operation
642 * is executed after the write issued by the coroutine allocating
643 * this cluster, therefore we do not need to keep it locked.
644 * As stated above, the allocating coroutine will always try to lock
645 * the mutex before all the other concurrent accesses to that
646 * cluster, therefore at this point we can be absolutely certain
647 * that that write operation has returned (there may be other writes
648 * in flight, but they do not concern this very operation). */
649 qemu_co_mutex_unlock(&s->write_lock);
650
651 qemu_iovec_reset(&local_qiov);
652 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
653
654 ret = bdrv_co_pwritev(bs->file, data_offset, n_bytes,
655 &local_qiov, 0);
656 }
657
658 bytes -= n_bytes;
659 offset += n_bytes;
660 bytes_done += n_bytes;
661
662 logout("%u bytes written\n", n_bytes);
663 }
664
665 qemu_iovec_destroy(&local_qiov);
666
667 logout("finished data write\n");
668 if (ret < 0) {
669 return ret;
670 }
671
672 if (block) {
673 /* One or more new blocks were allocated. */
674 VdiHeader *header = (VdiHeader *) block;
675 uint8_t *base;
676 uint64_t offset;
677 uint32_t n_sectors;
678
679 logout("now writing modified header\n");
680 assert(VDI_IS_ALLOCATED(bmap_first));
681 *header = s->header;
682 vdi_header_to_le(header);
683 ret = bdrv_write(bs->file, 0, block, 1);
684 g_free(block);
685 block = NULL;
686
687 if (ret < 0) {
688 return ret;
689 }
690
691 logout("now writing modified block map entry %u...%u\n",
692 bmap_first, bmap_last);
693 /* Write modified sectors from block map. */
694 bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
695 bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
696 n_sectors = bmap_last - bmap_first + 1;
697 offset = s->bmap_sector + bmap_first;
698 base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
699 logout("will write %u block map sectors starting from entry %u\n",
700 n_sectors, bmap_first);
701 ret = bdrv_write(bs->file, offset, base, n_sectors);
702 }
703
704 return ret;
705 }
706
707 static int vdi_create(const char *filename, QemuOpts *opts, Error **errp)
708 {
709 int ret = 0;
710 uint64_t bytes = 0;
711 uint32_t blocks;
712 size_t block_size = DEFAULT_CLUSTER_SIZE;
713 uint32_t image_type = VDI_TYPE_DYNAMIC;
714 VdiHeader header;
715 size_t i;
716 size_t bmap_size;
717 int64_t offset = 0;
718 Error *local_err = NULL;
719 BlockBackend *blk = NULL;
720 uint32_t *bmap = NULL;
721
722 logout("\n");
723
724 /* Read out options. */
725 bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
726 BDRV_SECTOR_SIZE);
727 #if defined(CONFIG_VDI_BLOCK_SIZE)
728 /* TODO: Additional checks (SECTOR_SIZE * 2^n, ...). */
729 block_size = qemu_opt_get_size_del(opts,
730 BLOCK_OPT_CLUSTER_SIZE,
731 DEFAULT_CLUSTER_SIZE);
732 #endif
733 #if defined(CONFIG_VDI_STATIC_IMAGE)
734 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) {
735 image_type = VDI_TYPE_STATIC;
736 }
737 #endif
738
739 if (bytes > VDI_DISK_SIZE_MAX) {
740 ret = -ENOTSUP;
741 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
742 ", max supported is 0x%" PRIx64 ")",
743 bytes, VDI_DISK_SIZE_MAX);
744 goto exit;
745 }
746
747 ret = bdrv_create_file(filename, opts, &local_err);
748 if (ret < 0) {
749 error_propagate(errp, local_err);
750 goto exit;
751 }
752
753 blk = blk_new_open(filename, NULL, NULL,
754 BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
755 if (blk == NULL) {
756 error_propagate(errp, local_err);
757 ret = -EIO;
758 goto exit;
759 }
760
761 blk_set_allow_write_beyond_eof(blk, true);
762
763 /* We need enough blocks to store the given disk size,
764 so always round up. */
765 blocks = DIV_ROUND_UP(bytes, block_size);
766
767 bmap_size = blocks * sizeof(uint32_t);
768 bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE);
769
770 memset(&header, 0, sizeof(header));
771 pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
772 header.signature = VDI_SIGNATURE;
773 header.version = VDI_VERSION_1_1;
774 header.header_size = 0x180;
775 header.image_type = image_type;
776 header.offset_bmap = 0x200;
777 header.offset_data = 0x200 + bmap_size;
778 header.sector_size = SECTOR_SIZE;
779 header.disk_size = bytes;
780 header.block_size = block_size;
781 header.blocks_in_image = blocks;
782 if (image_type == VDI_TYPE_STATIC) {
783 header.blocks_allocated = blocks;
784 }
785 qemu_uuid_generate(&header.uuid_image);
786 qemu_uuid_generate(&header.uuid_last_snap);
787 /* There is no need to set header.uuid_link or header.uuid_parent here. */
788 #if defined(CONFIG_VDI_DEBUG)
789 vdi_header_print(&header);
790 #endif
791 vdi_header_to_le(&header);
792 ret = blk_pwrite(blk, offset, &header, sizeof(header), 0);
793 if (ret < 0) {
794 error_setg(errp, "Error writing header to %s", filename);
795 goto exit;
796 }
797 offset += sizeof(header);
798
799 if (bmap_size > 0) {
800 bmap = g_try_malloc0(bmap_size);
801 if (bmap == NULL) {
802 ret = -ENOMEM;
803 error_setg(errp, "Could not allocate bmap");
804 goto exit;
805 }
806 for (i = 0; i < blocks; i++) {
807 if (image_type == VDI_TYPE_STATIC) {
808 bmap[i] = i;
809 } else {
810 bmap[i] = VDI_UNALLOCATED;
811 }
812 }
813 ret = blk_pwrite(blk, offset, bmap, bmap_size, 0);
814 if (ret < 0) {
815 error_setg(errp, "Error writing bmap to %s", filename);
816 goto exit;
817 }
818 offset += bmap_size;
819 }
820
821 if (image_type == VDI_TYPE_STATIC) {
822 ret = blk_truncate(blk, offset + blocks * block_size);
823 if (ret < 0) {
824 error_setg(errp, "Failed to statically allocate %s", filename);
825 goto exit;
826 }
827 }
828
829 exit:
830 blk_unref(blk);
831 g_free(bmap);
832 return ret;
833 }
834
835 static void vdi_close(BlockDriverState *bs)
836 {
837 BDRVVdiState *s = bs->opaque;
838
839 qemu_vfree(s->bmap);
840
841 migrate_del_blocker(s->migration_blocker);
842 error_free(s->migration_blocker);
843 }
844
845 static QemuOptsList vdi_create_opts = {
846 .name = "vdi-create-opts",
847 .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head),
848 .desc = {
849 {
850 .name = BLOCK_OPT_SIZE,
851 .type = QEMU_OPT_SIZE,
852 .help = "Virtual disk size"
853 },
854 #if defined(CONFIG_VDI_BLOCK_SIZE)
855 {
856 .name = BLOCK_OPT_CLUSTER_SIZE,
857 .type = QEMU_OPT_SIZE,
858 .help = "VDI cluster (block) size",
859 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
860 },
861 #endif
862 #if defined(CONFIG_VDI_STATIC_IMAGE)
863 {
864 .name = BLOCK_OPT_STATIC,
865 .type = QEMU_OPT_BOOL,
866 .help = "VDI static (pre-allocated) image",
867 .def_value_str = "off"
868 },
869 #endif
870 /* TODO: An additional option to set UUID values might be useful. */
871 { /* end of list */ }
872 }
873 };
874
875 static BlockDriver bdrv_vdi = {
876 .format_name = "vdi",
877 .instance_size = sizeof(BDRVVdiState),
878 .bdrv_probe = vdi_probe,
879 .bdrv_open = vdi_open,
880 .bdrv_close = vdi_close,
881 .bdrv_reopen_prepare = vdi_reopen_prepare,
882 .bdrv_create = vdi_create,
883 .bdrv_has_zero_init = bdrv_has_zero_init_1,
884 .bdrv_co_get_block_status = vdi_co_get_block_status,
885 .bdrv_make_empty = vdi_make_empty,
886
887 .bdrv_co_preadv = vdi_co_preadv,
888 #if defined(CONFIG_VDI_WRITE)
889 .bdrv_co_pwritev = vdi_co_pwritev,
890 #endif
891
892 .bdrv_get_info = vdi_get_info,
893
894 .create_opts = &vdi_create_opts,
895 .bdrv_check = vdi_check,
896 };
897
898 static void bdrv_vdi_init(void)
899 {
900 logout("\n");
901 bdrv_register(&bdrv_vdi);
902 }
903
904 block_init(bdrv_vdi_init);