spapr: Workaround for broken radix guests
[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 Error *local_err = NULL;
365
366 bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
367 false, errp);
368 if (!bs->file) {
369 return -EINVAL;
370 }
371
372 logout("\n");
373
374 ret = bdrv_read(bs->file, 0, (uint8_t *)&header, 1);
375 if (ret < 0) {
376 goto fail;
377 }
378
379 vdi_header_to_cpu(&header);
380 #if defined(CONFIG_VDI_DEBUG)
381 vdi_header_print(&header);
382 #endif
383
384 if (header.disk_size > VDI_DISK_SIZE_MAX) {
385 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
386 ", max supported is 0x%" PRIx64 ")",
387 header.disk_size, VDI_DISK_SIZE_MAX);
388 ret = -ENOTSUP;
389 goto fail;
390 }
391
392 if (header.disk_size % SECTOR_SIZE != 0) {
393 /* 'VBoxManage convertfromraw' can create images with odd disk sizes.
394 We accept them but round the disk size to the next multiple of
395 SECTOR_SIZE. */
396 logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size);
397 header.disk_size = ROUND_UP(header.disk_size, SECTOR_SIZE);
398 }
399
400 if (header.signature != VDI_SIGNATURE) {
401 error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32
402 ")", header.signature);
403 ret = -EINVAL;
404 goto fail;
405 } else if (header.version != VDI_VERSION_1_1) {
406 error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32
407 ")", header.version >> 16, header.version & 0xffff);
408 ret = -ENOTSUP;
409 goto fail;
410 } else if (header.offset_bmap % SECTOR_SIZE != 0) {
411 /* We only support block maps which start on a sector boundary. */
412 error_setg(errp, "unsupported VDI image (unaligned block map offset "
413 "0x%" PRIx32 ")", header.offset_bmap);
414 ret = -ENOTSUP;
415 goto fail;
416 } else if (header.offset_data % SECTOR_SIZE != 0) {
417 /* We only support data blocks which start on a sector boundary. */
418 error_setg(errp, "unsupported VDI image (unaligned data offset 0x%"
419 PRIx32 ")", header.offset_data);
420 ret = -ENOTSUP;
421 goto fail;
422 } else if (header.sector_size != SECTOR_SIZE) {
423 error_setg(errp, "unsupported VDI image (sector size %" PRIu32
424 " is not %u)", header.sector_size, SECTOR_SIZE);
425 ret = -ENOTSUP;
426 goto fail;
427 } else if (header.block_size != DEFAULT_CLUSTER_SIZE) {
428 error_setg(errp, "unsupported VDI image (block size %" PRIu32
429 " is not %u)", header.block_size, DEFAULT_CLUSTER_SIZE);
430 ret = -ENOTSUP;
431 goto fail;
432 } else if (header.disk_size >
433 (uint64_t)header.blocks_in_image * header.block_size) {
434 error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", "
435 "image bitmap has room for %" PRIu64 ")",
436 header.disk_size,
437 (uint64_t)header.blocks_in_image * header.block_size);
438 ret = -ENOTSUP;
439 goto fail;
440 } else if (!qemu_uuid_is_null(&header.uuid_link)) {
441 error_setg(errp, "unsupported VDI image (non-NULL link UUID)");
442 ret = -ENOTSUP;
443 goto fail;
444 } else if (!qemu_uuid_is_null(&header.uuid_parent)) {
445 error_setg(errp, "unsupported VDI image (non-NULL parent UUID)");
446 ret = -ENOTSUP;
447 goto fail;
448 } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) {
449 error_setg(errp, "unsupported VDI image "
450 "(too many blocks %u, max is %u)",
451 header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX);
452 ret = -ENOTSUP;
453 goto fail;
454 }
455
456 bs->total_sectors = header.disk_size / SECTOR_SIZE;
457
458 s->block_size = header.block_size;
459 s->block_sectors = header.block_size / SECTOR_SIZE;
460 s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
461 s->header = header;
462
463 bmap_size = header.blocks_in_image * sizeof(uint32_t);
464 bmap_size = DIV_ROUND_UP(bmap_size, SECTOR_SIZE);
465 s->bmap = qemu_try_blockalign(bs->file->bs, bmap_size * SECTOR_SIZE);
466 if (s->bmap == NULL) {
467 ret = -ENOMEM;
468 goto fail;
469 }
470
471 ret = bdrv_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap,
472 bmap_size);
473 if (ret < 0) {
474 goto fail_free_bmap;
475 }
476
477 /* Disable migration when vdi images are used */
478 error_setg(&s->migration_blocker, "The vdi format used by node '%s' "
479 "does not support live migration",
480 bdrv_get_device_or_node_name(bs));
481 ret = migrate_add_blocker(s->migration_blocker, &local_err);
482 if (local_err) {
483 error_propagate(errp, local_err);
484 error_free(s->migration_blocker);
485 goto fail_free_bmap;
486 }
487
488 qemu_co_mutex_init(&s->write_lock);
489
490 return 0;
491
492 fail_free_bmap:
493 qemu_vfree(s->bmap);
494
495 fail:
496 return ret;
497 }
498
499 static int vdi_reopen_prepare(BDRVReopenState *state,
500 BlockReopenQueue *queue, Error **errp)
501 {
502 return 0;
503 }
504
505 static int64_t coroutine_fn vdi_co_get_block_status(BlockDriverState *bs,
506 int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
507 {
508 /* TODO: Check for too large sector_num (in bdrv_is_allocated or here). */
509 BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
510 size_t bmap_index = sector_num / s->block_sectors;
511 size_t sector_in_block = sector_num % s->block_sectors;
512 int n_sectors = s->block_sectors - sector_in_block;
513 uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
514 uint64_t offset;
515 int result;
516
517 logout("%p, %" PRId64 ", %d, %p\n", bs, sector_num, nb_sectors, pnum);
518 if (n_sectors > nb_sectors) {
519 n_sectors = nb_sectors;
520 }
521 *pnum = n_sectors;
522 result = VDI_IS_ALLOCATED(bmap_entry);
523 if (!result) {
524 return 0;
525 }
526
527 offset = s->header.offset_data +
528 (uint64_t)bmap_entry * s->block_size +
529 sector_in_block * SECTOR_SIZE;
530 *file = bs->file->bs;
531 return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID | offset;
532 }
533
534 static int coroutine_fn
535 vdi_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
536 QEMUIOVector *qiov, int flags)
537 {
538 BDRVVdiState *s = bs->opaque;
539 QEMUIOVector local_qiov;
540 uint32_t bmap_entry;
541 uint32_t block_index;
542 uint32_t offset_in_block;
543 uint32_t n_bytes;
544 uint64_t bytes_done = 0;
545 int ret = 0;
546
547 logout("\n");
548
549 qemu_iovec_init(&local_qiov, qiov->niov);
550
551 while (ret >= 0 && bytes > 0) {
552 block_index = offset / s->block_size;
553 offset_in_block = offset % s->block_size;
554 n_bytes = MIN(bytes, s->block_size - offset_in_block);
555
556 logout("will read %u bytes starting at offset %" PRIu64 "\n",
557 n_bytes, offset);
558
559 /* prepare next AIO request */
560 bmap_entry = le32_to_cpu(s->bmap[block_index]);
561 if (!VDI_IS_ALLOCATED(bmap_entry)) {
562 /* Block not allocated, return zeros, no need to wait. */
563 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
564 ret = 0;
565 } else {
566 uint64_t data_offset = s->header.offset_data +
567 (uint64_t)bmap_entry * s->block_size +
568 offset_in_block;
569
570 qemu_iovec_reset(&local_qiov);
571 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
572
573 ret = bdrv_co_preadv(bs->file, data_offset, n_bytes,
574 &local_qiov, 0);
575 }
576 logout("%u bytes read\n", n_bytes);
577
578 bytes -= n_bytes;
579 offset += n_bytes;
580 bytes_done += n_bytes;
581 }
582
583 qemu_iovec_destroy(&local_qiov);
584
585 return ret;
586 }
587
588 static int coroutine_fn
589 vdi_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
590 QEMUIOVector *qiov, int flags)
591 {
592 BDRVVdiState *s = bs->opaque;
593 QEMUIOVector local_qiov;
594 uint32_t bmap_entry;
595 uint32_t block_index;
596 uint32_t offset_in_block;
597 uint32_t n_bytes;
598 uint32_t bmap_first = VDI_UNALLOCATED;
599 uint32_t bmap_last = VDI_UNALLOCATED;
600 uint8_t *block = NULL;
601 uint64_t bytes_done = 0;
602 int ret = 0;
603
604 logout("\n");
605
606 qemu_iovec_init(&local_qiov, qiov->niov);
607
608 while (ret >= 0 && bytes > 0) {
609 block_index = offset / s->block_size;
610 offset_in_block = offset % s->block_size;
611 n_bytes = MIN(bytes, s->block_size - offset_in_block);
612
613 logout("will write %u bytes starting at offset %" PRIu64 "\n",
614 n_bytes, offset);
615
616 /* prepare next AIO request */
617 bmap_entry = le32_to_cpu(s->bmap[block_index]);
618 if (!VDI_IS_ALLOCATED(bmap_entry)) {
619 /* Allocate new block and write to it. */
620 uint64_t data_offset;
621 bmap_entry = s->header.blocks_allocated;
622 s->bmap[block_index] = cpu_to_le32(bmap_entry);
623 s->header.blocks_allocated++;
624 data_offset = s->header.offset_data +
625 (uint64_t)bmap_entry * s->block_size;
626 if (block == NULL) {
627 block = g_malloc(s->block_size);
628 bmap_first = block_index;
629 }
630 bmap_last = block_index;
631 /* Copy data to be written to new block and zero unused parts. */
632 memset(block, 0, offset_in_block);
633 qemu_iovec_to_buf(qiov, bytes_done, block + offset_in_block,
634 n_bytes);
635 memset(block + offset_in_block + n_bytes, 0,
636 s->block_size - n_bytes - offset_in_block);
637
638 /* Note that this coroutine does not yield anywhere from reading the
639 * bmap entry until here, so in regards to all the coroutines trying
640 * to write to this cluster, the one doing the allocation will
641 * always be the first to try to acquire the lock.
642 * Therefore, it is also the first that will actually be able to
643 * acquire the lock and thus the padded cluster is written before
644 * the other coroutines can write to the affected area. */
645 qemu_co_mutex_lock(&s->write_lock);
646 ret = bdrv_pwrite(bs->file, data_offset, block, s->block_size);
647 qemu_co_mutex_unlock(&s->write_lock);
648 } else {
649 uint64_t data_offset = s->header.offset_data +
650 (uint64_t)bmap_entry * s->block_size +
651 offset_in_block;
652 qemu_co_mutex_lock(&s->write_lock);
653 /* This lock is only used to make sure the following write operation
654 * is executed after the write issued by the coroutine allocating
655 * this cluster, therefore we do not need to keep it locked.
656 * As stated above, the allocating coroutine will always try to lock
657 * the mutex before all the other concurrent accesses to that
658 * cluster, therefore at this point we can be absolutely certain
659 * that that write operation has returned (there may be other writes
660 * in flight, but they do not concern this very operation). */
661 qemu_co_mutex_unlock(&s->write_lock);
662
663 qemu_iovec_reset(&local_qiov);
664 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
665
666 ret = bdrv_co_pwritev(bs->file, data_offset, n_bytes,
667 &local_qiov, 0);
668 }
669
670 bytes -= n_bytes;
671 offset += n_bytes;
672 bytes_done += n_bytes;
673
674 logout("%u bytes written\n", n_bytes);
675 }
676
677 qemu_iovec_destroy(&local_qiov);
678
679 logout("finished data write\n");
680 if (ret < 0) {
681 return ret;
682 }
683
684 if (block) {
685 /* One or more new blocks were allocated. */
686 VdiHeader *header = (VdiHeader *) block;
687 uint8_t *base;
688 uint64_t offset;
689 uint32_t n_sectors;
690
691 logout("now writing modified header\n");
692 assert(VDI_IS_ALLOCATED(bmap_first));
693 *header = s->header;
694 vdi_header_to_le(header);
695 ret = bdrv_write(bs->file, 0, block, 1);
696 g_free(block);
697 block = NULL;
698
699 if (ret < 0) {
700 return ret;
701 }
702
703 logout("now writing modified block map entry %u...%u\n",
704 bmap_first, bmap_last);
705 /* Write modified sectors from block map. */
706 bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
707 bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
708 n_sectors = bmap_last - bmap_first + 1;
709 offset = s->bmap_sector + bmap_first;
710 base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
711 logout("will write %u block map sectors starting from entry %u\n",
712 n_sectors, bmap_first);
713 ret = bdrv_write(bs->file, offset, base, n_sectors);
714 }
715
716 return ret;
717 }
718
719 static int vdi_create(const char *filename, QemuOpts *opts, Error **errp)
720 {
721 int ret = 0;
722 uint64_t bytes = 0;
723 uint32_t blocks;
724 size_t block_size = DEFAULT_CLUSTER_SIZE;
725 uint32_t image_type = VDI_TYPE_DYNAMIC;
726 VdiHeader header;
727 size_t i;
728 size_t bmap_size;
729 int64_t offset = 0;
730 Error *local_err = NULL;
731 BlockBackend *blk = NULL;
732 uint32_t *bmap = NULL;
733
734 logout("\n");
735
736 /* Read out options. */
737 bytes = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
738 BDRV_SECTOR_SIZE);
739 #if defined(CONFIG_VDI_BLOCK_SIZE)
740 /* TODO: Additional checks (SECTOR_SIZE * 2^n, ...). */
741 block_size = qemu_opt_get_size_del(opts,
742 BLOCK_OPT_CLUSTER_SIZE,
743 DEFAULT_CLUSTER_SIZE);
744 #endif
745 #if defined(CONFIG_VDI_STATIC_IMAGE)
746 if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) {
747 image_type = VDI_TYPE_STATIC;
748 }
749 #endif
750
751 if (bytes > VDI_DISK_SIZE_MAX) {
752 ret = -ENOTSUP;
753 error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
754 ", max supported is 0x%" PRIx64 ")",
755 bytes, VDI_DISK_SIZE_MAX);
756 goto exit;
757 }
758
759 ret = bdrv_create_file(filename, opts, &local_err);
760 if (ret < 0) {
761 error_propagate(errp, local_err);
762 goto exit;
763 }
764
765 blk = blk_new_open(filename, NULL, NULL,
766 BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL,
767 &local_err);
768 if (blk == NULL) {
769 error_propagate(errp, local_err);
770 ret = -EIO;
771 goto exit;
772 }
773
774 blk_set_allow_write_beyond_eof(blk, true);
775
776 /* We need enough blocks to store the given disk size,
777 so always round up. */
778 blocks = DIV_ROUND_UP(bytes, block_size);
779
780 bmap_size = blocks * sizeof(uint32_t);
781 bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE);
782
783 memset(&header, 0, sizeof(header));
784 pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
785 header.signature = VDI_SIGNATURE;
786 header.version = VDI_VERSION_1_1;
787 header.header_size = 0x180;
788 header.image_type = image_type;
789 header.offset_bmap = 0x200;
790 header.offset_data = 0x200 + bmap_size;
791 header.sector_size = SECTOR_SIZE;
792 header.disk_size = bytes;
793 header.block_size = block_size;
794 header.blocks_in_image = blocks;
795 if (image_type == VDI_TYPE_STATIC) {
796 header.blocks_allocated = blocks;
797 }
798 qemu_uuid_generate(&header.uuid_image);
799 qemu_uuid_generate(&header.uuid_last_snap);
800 /* There is no need to set header.uuid_link or header.uuid_parent here. */
801 #if defined(CONFIG_VDI_DEBUG)
802 vdi_header_print(&header);
803 #endif
804 vdi_header_to_le(&header);
805 ret = blk_pwrite(blk, offset, &header, sizeof(header), 0);
806 if (ret < 0) {
807 error_setg(errp, "Error writing header to %s", filename);
808 goto exit;
809 }
810 offset += sizeof(header);
811
812 if (bmap_size > 0) {
813 bmap = g_try_malloc0(bmap_size);
814 if (bmap == NULL) {
815 ret = -ENOMEM;
816 error_setg(errp, "Could not allocate bmap");
817 goto exit;
818 }
819 for (i = 0; i < blocks; i++) {
820 if (image_type == VDI_TYPE_STATIC) {
821 bmap[i] = i;
822 } else {
823 bmap[i] = VDI_UNALLOCATED;
824 }
825 }
826 ret = blk_pwrite(blk, offset, bmap, bmap_size, 0);
827 if (ret < 0) {
828 error_setg(errp, "Error writing bmap to %s", filename);
829 goto exit;
830 }
831 offset += bmap_size;
832 }
833
834 if (image_type == VDI_TYPE_STATIC) {
835 ret = blk_truncate(blk, offset + blocks * block_size);
836 if (ret < 0) {
837 error_setg(errp, "Failed to statically allocate %s", filename);
838 goto exit;
839 }
840 }
841
842 exit:
843 blk_unref(blk);
844 g_free(bmap);
845 return ret;
846 }
847
848 static void vdi_close(BlockDriverState *bs)
849 {
850 BDRVVdiState *s = bs->opaque;
851
852 qemu_vfree(s->bmap);
853
854 migrate_del_blocker(s->migration_blocker);
855 error_free(s->migration_blocker);
856 }
857
858 static QemuOptsList vdi_create_opts = {
859 .name = "vdi-create-opts",
860 .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head),
861 .desc = {
862 {
863 .name = BLOCK_OPT_SIZE,
864 .type = QEMU_OPT_SIZE,
865 .help = "Virtual disk size"
866 },
867 #if defined(CONFIG_VDI_BLOCK_SIZE)
868 {
869 .name = BLOCK_OPT_CLUSTER_SIZE,
870 .type = QEMU_OPT_SIZE,
871 .help = "VDI cluster (block) size",
872 .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
873 },
874 #endif
875 #if defined(CONFIG_VDI_STATIC_IMAGE)
876 {
877 .name = BLOCK_OPT_STATIC,
878 .type = QEMU_OPT_BOOL,
879 .help = "VDI static (pre-allocated) image",
880 .def_value_str = "off"
881 },
882 #endif
883 /* TODO: An additional option to set UUID values might be useful. */
884 { /* end of list */ }
885 }
886 };
887
888 static BlockDriver bdrv_vdi = {
889 .format_name = "vdi",
890 .instance_size = sizeof(BDRVVdiState),
891 .bdrv_probe = vdi_probe,
892 .bdrv_open = vdi_open,
893 .bdrv_close = vdi_close,
894 .bdrv_reopen_prepare = vdi_reopen_prepare,
895 .bdrv_child_perm = bdrv_format_default_perms,
896 .bdrv_create = vdi_create,
897 .bdrv_has_zero_init = bdrv_has_zero_init_1,
898 .bdrv_co_get_block_status = vdi_co_get_block_status,
899 .bdrv_make_empty = vdi_make_empty,
900
901 .bdrv_co_preadv = vdi_co_preadv,
902 #if defined(CONFIG_VDI_WRITE)
903 .bdrv_co_pwritev = vdi_co_pwritev,
904 #endif
905
906 .bdrv_get_info = vdi_get_info,
907
908 .create_opts = &vdi_create_opts,
909 .bdrv_check = vdi_check,
910 };
911
912 static void bdrv_vdi_init(void)
913 {
914 logout("\n");
915 bdrv_register(&bdrv_vdi);
916 }
917
918 block_init(bdrv_vdi_init);