xref: /openbmc/qemu/block/vdi.c (revision 2475a0d0)
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 "qemu/units.h"
54 #include "qapi/error.h"
55 #include "qapi/qobject-input-visitor.h"
56 #include "qapi/qapi-visit-block-core.h"
57 #include "block/block_int.h"
58 #include "block/qdict.h"
59 #include "sysemu/block-backend.h"
60 #include "qemu/module.h"
61 #include "qemu/option.h"
62 #include "qemu/bswap.h"
63 #include "migration/blocker.h"
64 #include "qemu/coroutine.h"
65 #include "qemu/cutils.h"
66 #include "qemu/uuid.h"
67 #include "qemu/memalign.h"
68 
69 /* Code configuration options. */
70 
71 /* Enable debug messages. */
72 //~ #define CONFIG_VDI_DEBUG
73 
74 /* Support write operations on VDI images. */
75 #define CONFIG_VDI_WRITE
76 
77 /* Support non-standard block (cluster) size. This is untested.
78  * Maybe it will be needed for very large images.
79  */
80 //~ #define CONFIG_VDI_BLOCK_SIZE
81 
82 /* Support static (fixed, pre-allocated) images. */
83 #define CONFIG_VDI_STATIC_IMAGE
84 
85 /* Command line option for static images. */
86 #define BLOCK_OPT_STATIC "static"
87 
88 #define SECTOR_SIZE 512
89 #define DEFAULT_CLUSTER_SIZE 1048576
90 /* Note: can't use 1 * MiB, because it's passed to stringify() */
91 
92 #if defined(CONFIG_VDI_DEBUG)
93 #define VDI_DEBUG 1
94 #else
95 #define VDI_DEBUG 0
96 #endif
97 
98 #define logout(fmt, ...) \
99     do {                                                                \
100         if (VDI_DEBUG) {                                                \
101             fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__); \
102         }                                                               \
103     } while (0)
104 
105 /* Image signature. */
106 #define VDI_SIGNATURE 0xbeda107f
107 
108 /* Image version. */
109 #define VDI_VERSION_1_1 0x00010001
110 
111 /* Image type. */
112 #define VDI_TYPE_DYNAMIC 1
113 #define VDI_TYPE_STATIC  2
114 
115 /* Innotek / SUN images use these strings in header.text:
116  * "<<< innotek VirtualBox Disk Image >>>\n"
117  * "<<< Sun xVM VirtualBox Disk Image >>>\n"
118  * "<<< Sun VirtualBox Disk Image >>>\n"
119  * The value does not matter, so QEMU created images use a different text.
120  */
121 #define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n"
122 
123 /* A never-allocated block; semantically arbitrary content. */
124 #define VDI_UNALLOCATED 0xffffffffU
125 
126 /* A discarded (no longer allocated) block; semantically zero-filled. */
127 #define VDI_DISCARDED   0xfffffffeU
128 
129 #define VDI_IS_ALLOCATED(X) ((X) < VDI_DISCARDED)
130 
131 /* The bmap will take up VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) bytes; since
132  * the bmap is read and written in a single operation, its size needs to be
133  * limited to INT_MAX; furthermore, when opening an image, the bmap size is
134  * rounded up to be aligned on BDRV_SECTOR_SIZE.
135  * Therefore this should satisfy the following:
136  * VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) + BDRV_SECTOR_SIZE == INT_MAX + 1
137  * (INT_MAX + 1 is the first value not representable as an int)
138  * This guarantees that any value below or equal to the constant will, when
139  * multiplied by sizeof(uint32_t) and rounded up to a BDRV_SECTOR_SIZE boundary,
140  * still be below or equal to INT_MAX. */
141 #define VDI_BLOCKS_IN_IMAGE_MAX \
142     ((unsigned)((INT_MAX + 1u - BDRV_SECTOR_SIZE) / sizeof(uint32_t)))
143 #define VDI_DISK_SIZE_MAX        ((uint64_t)VDI_BLOCKS_IN_IMAGE_MAX * \
144                                   (uint64_t)DEFAULT_CLUSTER_SIZE)
145 
146 static QemuOptsList vdi_create_opts;
147 
148 typedef struct {
149     char text[0x40];
150     uint32_t signature;
151     uint32_t version;
152     uint32_t header_size;
153     uint32_t image_type;
154     uint32_t image_flags;
155     char description[256];
156     uint32_t offset_bmap;
157     uint32_t offset_data;
158     uint32_t cylinders;         /* disk geometry, unused here */
159     uint32_t heads;             /* disk geometry, unused here */
160     uint32_t sectors;           /* disk geometry, unused here */
161     uint32_t sector_size;
162     uint32_t unused1;
163     uint64_t disk_size;
164     uint32_t block_size;
165     uint32_t block_extra;       /* unused here */
166     uint32_t blocks_in_image;
167     uint32_t blocks_allocated;
168     QemuUUID uuid_image;
169     QemuUUID uuid_last_snap;
170     QemuUUID uuid_link;
171     QemuUUID uuid_parent;
172     uint64_t unused2[7];
173 } QEMU_PACKED VdiHeader;
174 
175 QEMU_BUILD_BUG_ON(sizeof(VdiHeader) != 512);
176 
177 typedef struct {
178     /* The block map entries are little endian (even in memory). */
179     uint32_t *bmap;
180     /* Size of block (bytes). */
181     uint32_t block_size;
182     /* First sector of block map. */
183     uint32_t bmap_sector;
184     /* VDI header (converted to host endianness). */
185     VdiHeader header;
186 
187     CoRwlock bmap_lock;
188 
189     Error *migration_blocker;
190 } BDRVVdiState;
191 
192 static void vdi_header_to_cpu(VdiHeader *header)
193 {
194     header->signature = le32_to_cpu(header->signature);
195     header->version = le32_to_cpu(header->version);
196     header->header_size = le32_to_cpu(header->header_size);
197     header->image_type = le32_to_cpu(header->image_type);
198     header->image_flags = le32_to_cpu(header->image_flags);
199     header->offset_bmap = le32_to_cpu(header->offset_bmap);
200     header->offset_data = le32_to_cpu(header->offset_data);
201     header->cylinders = le32_to_cpu(header->cylinders);
202     header->heads = le32_to_cpu(header->heads);
203     header->sectors = le32_to_cpu(header->sectors);
204     header->sector_size = le32_to_cpu(header->sector_size);
205     header->disk_size = le64_to_cpu(header->disk_size);
206     header->block_size = le32_to_cpu(header->block_size);
207     header->block_extra = le32_to_cpu(header->block_extra);
208     header->blocks_in_image = le32_to_cpu(header->blocks_in_image);
209     header->blocks_allocated = le32_to_cpu(header->blocks_allocated);
210     header->uuid_image = qemu_uuid_bswap(header->uuid_image);
211     header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap);
212     header->uuid_link = qemu_uuid_bswap(header->uuid_link);
213     header->uuid_parent = qemu_uuid_bswap(header->uuid_parent);
214 }
215 
216 static void vdi_header_to_le(VdiHeader *header)
217 {
218     header->signature = cpu_to_le32(header->signature);
219     header->version = cpu_to_le32(header->version);
220     header->header_size = cpu_to_le32(header->header_size);
221     header->image_type = cpu_to_le32(header->image_type);
222     header->image_flags = cpu_to_le32(header->image_flags);
223     header->offset_bmap = cpu_to_le32(header->offset_bmap);
224     header->offset_data = cpu_to_le32(header->offset_data);
225     header->cylinders = cpu_to_le32(header->cylinders);
226     header->heads = cpu_to_le32(header->heads);
227     header->sectors = cpu_to_le32(header->sectors);
228     header->sector_size = cpu_to_le32(header->sector_size);
229     header->disk_size = cpu_to_le64(header->disk_size);
230     header->block_size = cpu_to_le32(header->block_size);
231     header->block_extra = cpu_to_le32(header->block_extra);
232     header->blocks_in_image = cpu_to_le32(header->blocks_in_image);
233     header->blocks_allocated = cpu_to_le32(header->blocks_allocated);
234     header->uuid_image = qemu_uuid_bswap(header->uuid_image);
235     header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap);
236     header->uuid_link = qemu_uuid_bswap(header->uuid_link);
237     header->uuid_parent = qemu_uuid_bswap(header->uuid_parent);
238 }
239 
240 static void vdi_header_print(VdiHeader *header)
241 {
242     char uuidstr[37];
243     QemuUUID uuid;
244     logout("text        %s", header->text);
245     logout("signature   0x%08x\n", header->signature);
246     logout("header size 0x%04x\n", header->header_size);
247     logout("image type  0x%04x\n", header->image_type);
248     logout("image flags 0x%04x\n", header->image_flags);
249     logout("description %s\n", header->description);
250     logout("offset bmap 0x%04x\n", header->offset_bmap);
251     logout("offset data 0x%04x\n", header->offset_data);
252     logout("cylinders   0x%04x\n", header->cylinders);
253     logout("heads       0x%04x\n", header->heads);
254     logout("sectors     0x%04x\n", header->sectors);
255     logout("sector size 0x%04x\n", header->sector_size);
256     logout("image size  0x%" PRIx64 " B (%" PRIu64 " MiB)\n",
257            header->disk_size, header->disk_size / MiB);
258     logout("block size  0x%04x\n", header->block_size);
259     logout("block extra 0x%04x\n", header->block_extra);
260     logout("blocks tot. 0x%04x\n", header->blocks_in_image);
261     logout("blocks all. 0x%04x\n", header->blocks_allocated);
262     uuid = header->uuid_image;
263     qemu_uuid_unparse(&uuid, uuidstr);
264     logout("uuid image  %s\n", uuidstr);
265     uuid = header->uuid_last_snap;
266     qemu_uuid_unparse(&uuid, uuidstr);
267     logout("uuid snap   %s\n", uuidstr);
268     uuid = header->uuid_link;
269     qemu_uuid_unparse(&uuid, uuidstr);
270     logout("uuid link   %s\n", uuidstr);
271     uuid = header->uuid_parent;
272     qemu_uuid_unparse(&uuid, uuidstr);
273     logout("uuid parent %s\n", uuidstr);
274 }
275 
276 static int coroutine_fn vdi_co_check(BlockDriverState *bs, BdrvCheckResult *res,
277                                      BdrvCheckMode fix)
278 {
279     /* TODO: additional checks possible. */
280     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
281     uint32_t blocks_allocated = 0;
282     uint32_t block;
283     uint32_t *bmap;
284     logout("\n");
285 
286     if (fix) {
287         return -ENOTSUP;
288     }
289 
290     bmap = g_try_new(uint32_t, s->header.blocks_in_image);
291     if (s->header.blocks_in_image && bmap == NULL) {
292         res->check_errors++;
293         return -ENOMEM;
294     }
295 
296     memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t));
297 
298     /* Check block map and value of blocks_allocated. */
299     for (block = 0; block < s->header.blocks_in_image; block++) {
300         uint32_t bmap_entry = le32_to_cpu(s->bmap[block]);
301         if (VDI_IS_ALLOCATED(bmap_entry)) {
302             if (bmap_entry < s->header.blocks_in_image) {
303                 blocks_allocated++;
304                 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) {
305                     bmap[bmap_entry] = bmap_entry;
306                 } else {
307                     fprintf(stderr, "ERROR: block index %" PRIu32
308                             " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry);
309                     res->corruptions++;
310                 }
311             } else {
312                 fprintf(stderr, "ERROR: block index %" PRIu32
313                         " too large, is %" PRIu32 "\n", block, bmap_entry);
314                 res->corruptions++;
315             }
316         }
317     }
318     if (blocks_allocated != s->header.blocks_allocated) {
319         fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32
320                ", should be %" PRIu32 "\n",
321                blocks_allocated, s->header.blocks_allocated);
322         res->corruptions++;
323     }
324 
325     g_free(bmap);
326 
327     return 0;
328 }
329 
330 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
331 {
332     /* TODO: vdi_get_info would be needed for machine snapshots.
333        vm_state_offset is still missing. */
334     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
335     logout("\n");
336     bdi->cluster_size = s->block_size;
337     bdi->vm_state_offset = 0;
338     return 0;
339 }
340 
341 static int vdi_make_empty(BlockDriverState *bs)
342 {
343     /* TODO: missing code. */
344     logout("\n");
345     /* The return value for missing code must be 0, see block.c. */
346     return 0;
347 }
348 
349 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename)
350 {
351     const VdiHeader *header = (const VdiHeader *)buf;
352     int ret = 0;
353 
354     logout("\n");
355 
356     if (buf_size < sizeof(*header)) {
357         /* Header too small, no VDI. */
358     } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) {
359         ret = 100;
360     }
361 
362     if (ret == 0) {
363         logout("no vdi image\n");
364     } else {
365         logout("%s", header->text);
366     }
367 
368     return ret;
369 }
370 
371 static int vdi_open(BlockDriverState *bs, QDict *options, int flags,
372                     Error **errp)
373 {
374     BDRVVdiState *s = bs->opaque;
375     VdiHeader header;
376     size_t bmap_size;
377     int ret;
378     QemuUUID uuid_link, uuid_parent;
379 
380     ret = bdrv_open_file_child(NULL, options, "file", bs, errp);
381     if (ret < 0) {
382         return ret;
383     }
384 
385     logout("\n");
386 
387     ret = bdrv_pread(bs->file, 0, sizeof(header), &header, 0);
388     if (ret < 0) {
389         goto fail;
390     }
391 
392     vdi_header_to_cpu(&header);
393     if (VDI_DEBUG) {
394         vdi_header_print(&header);
395     }
396 
397     if (header.disk_size > VDI_DISK_SIZE_MAX) {
398         error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
399                           ", max supported is 0x%" PRIx64 ")",
400                           header.disk_size, VDI_DISK_SIZE_MAX);
401         ret = -ENOTSUP;
402         goto fail;
403     }
404 
405     uuid_link = header.uuid_link;
406     uuid_parent = header.uuid_parent;
407 
408     if (header.disk_size % SECTOR_SIZE != 0) {
409         /* 'VBoxManage convertfromraw' can create images with odd disk sizes.
410            We accept them but round the disk size to the next multiple of
411            SECTOR_SIZE. */
412         logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size);
413         header.disk_size = ROUND_UP(header.disk_size, SECTOR_SIZE);
414     }
415 
416     if (header.signature != VDI_SIGNATURE) {
417         error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32
418                    ")", header.signature);
419         ret = -EINVAL;
420         goto fail;
421     } else if (header.version != VDI_VERSION_1_1) {
422         error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32
423                    ")", header.version >> 16, header.version & 0xffff);
424         ret = -ENOTSUP;
425         goto fail;
426     } else if (header.offset_bmap % SECTOR_SIZE != 0) {
427         /* We only support block maps which start on a sector boundary. */
428         error_setg(errp, "unsupported VDI image (unaligned block map offset "
429                    "0x%" PRIx32 ")", header.offset_bmap);
430         ret = -ENOTSUP;
431         goto fail;
432     } else if (header.offset_data % SECTOR_SIZE != 0) {
433         /* We only support data blocks which start on a sector boundary. */
434         error_setg(errp, "unsupported VDI image (unaligned data offset 0x%"
435                    PRIx32 ")", header.offset_data);
436         ret = -ENOTSUP;
437         goto fail;
438     } else if (header.sector_size != SECTOR_SIZE) {
439         error_setg(errp, "unsupported VDI image (sector size %" PRIu32
440                    " is not %u)", header.sector_size, SECTOR_SIZE);
441         ret = -ENOTSUP;
442         goto fail;
443     } else if (header.block_size != DEFAULT_CLUSTER_SIZE) {
444         error_setg(errp, "unsupported VDI image (block size %" PRIu32
445                          " is not %" PRIu32 ")",
446                    header.block_size, DEFAULT_CLUSTER_SIZE);
447         ret = -ENOTSUP;
448         goto fail;
449     } else if (header.disk_size >
450                (uint64_t)header.blocks_in_image * header.block_size) {
451         error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", "
452                    "image bitmap has room for %" PRIu64 ")",
453                    header.disk_size,
454                    (uint64_t)header.blocks_in_image * header.block_size);
455         ret = -ENOTSUP;
456         goto fail;
457     } else if (!qemu_uuid_is_null(&uuid_link)) {
458         error_setg(errp, "unsupported VDI image (non-NULL link UUID)");
459         ret = -ENOTSUP;
460         goto fail;
461     } else if (!qemu_uuid_is_null(&uuid_parent)) {
462         error_setg(errp, "unsupported VDI image (non-NULL parent UUID)");
463         ret = -ENOTSUP;
464         goto fail;
465     } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) {
466         error_setg(errp, "unsupported VDI image "
467                          "(too many blocks %u, max is %u)",
468                           header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX);
469         ret = -ENOTSUP;
470         goto fail;
471     }
472 
473     bs->total_sectors = header.disk_size / SECTOR_SIZE;
474 
475     s->block_size = header.block_size;
476     s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
477     s->header = header;
478 
479     bmap_size = header.blocks_in_image * sizeof(uint32_t);
480     bmap_size = DIV_ROUND_UP(bmap_size, SECTOR_SIZE);
481     s->bmap = qemu_try_blockalign(bs->file->bs, bmap_size * SECTOR_SIZE);
482     if (s->bmap == NULL) {
483         ret = -ENOMEM;
484         goto fail;
485     }
486 
487     ret = bdrv_pread(bs->file, header.offset_bmap, bmap_size * SECTOR_SIZE,
488                      s->bmap, 0);
489     if (ret < 0) {
490         goto fail_free_bmap;
491     }
492 
493     /* Disable migration when vdi images are used */
494     error_setg(&s->migration_blocker, "The vdi format used by node '%s' "
495                "does not support live migration",
496                bdrv_get_device_or_node_name(bs));
497     ret = migrate_add_blocker(s->migration_blocker, errp);
498     if (ret < 0) {
499         error_free(s->migration_blocker);
500         goto fail_free_bmap;
501     }
502 
503     qemu_co_rwlock_init(&s->bmap_lock);
504 
505     return 0;
506 
507  fail_free_bmap:
508     qemu_vfree(s->bmap);
509 
510  fail:
511     return ret;
512 }
513 
514 static int vdi_reopen_prepare(BDRVReopenState *state,
515                               BlockReopenQueue *queue, Error **errp)
516 {
517     return 0;
518 }
519 
520 static int coroutine_fn vdi_co_block_status(BlockDriverState *bs,
521                                             bool want_zero,
522                                             int64_t offset, int64_t bytes,
523                                             int64_t *pnum, int64_t *map,
524                                             BlockDriverState **file)
525 {
526     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
527     size_t bmap_index = offset / s->block_size;
528     size_t index_in_block = offset % s->block_size;
529     uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
530     int result;
531 
532     logout("%p, %" PRId64 ", %" PRId64 ", %p\n", bs, offset, bytes, pnum);
533     *pnum = MIN(s->block_size - index_in_block, bytes);
534     result = VDI_IS_ALLOCATED(bmap_entry);
535     if (!result) {
536         return BDRV_BLOCK_ZERO;
537     }
538 
539     *map = s->header.offset_data + (uint64_t)bmap_entry * s->block_size +
540         index_in_block;
541     *file = bs->file->bs;
542     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID |
543         (s->header.image_type == VDI_TYPE_STATIC ? BDRV_BLOCK_RECURSE : 0);
544 }
545 
546 static int coroutine_fn
547 vdi_co_preadv(BlockDriverState *bs, int64_t offset, int64_t bytes,
548               QEMUIOVector *qiov, BdrvRequestFlags flags)
549 {
550     BDRVVdiState *s = bs->opaque;
551     QEMUIOVector local_qiov;
552     uint32_t bmap_entry;
553     uint32_t block_index;
554     uint32_t offset_in_block;
555     uint32_t n_bytes;
556     uint64_t bytes_done = 0;
557     int ret = 0;
558 
559     logout("\n");
560 
561     qemu_iovec_init(&local_qiov, qiov->niov);
562 
563     while (ret >= 0 && bytes > 0) {
564         block_index = offset / s->block_size;
565         offset_in_block = offset % s->block_size;
566         n_bytes = MIN(bytes, s->block_size - offset_in_block);
567 
568         logout("will read %u bytes starting at offset %" PRIu64 "\n",
569                n_bytes, offset);
570 
571         /* prepare next AIO request */
572         qemu_co_rwlock_rdlock(&s->bmap_lock);
573         bmap_entry = le32_to_cpu(s->bmap[block_index]);
574         qemu_co_rwlock_unlock(&s->bmap_lock);
575         if (!VDI_IS_ALLOCATED(bmap_entry)) {
576             /* Block not allocated, return zeros, no need to wait. */
577             qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
578             ret = 0;
579         } else {
580             uint64_t data_offset = s->header.offset_data +
581                                    (uint64_t)bmap_entry * s->block_size +
582                                    offset_in_block;
583 
584             qemu_iovec_reset(&local_qiov);
585             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
586 
587             ret = bdrv_co_preadv(bs->file, data_offset, n_bytes,
588                                  &local_qiov, 0);
589         }
590         logout("%u bytes read\n", n_bytes);
591 
592         bytes -= n_bytes;
593         offset += n_bytes;
594         bytes_done += n_bytes;
595     }
596 
597     qemu_iovec_destroy(&local_qiov);
598 
599     return ret;
600 }
601 
602 static int coroutine_fn
603 vdi_co_pwritev(BlockDriverState *bs, int64_t offset, int64_t bytes,
604                QEMUIOVector *qiov, BdrvRequestFlags flags)
605 {
606     BDRVVdiState *s = bs->opaque;
607     QEMUIOVector local_qiov;
608     uint32_t bmap_entry;
609     uint32_t block_index;
610     uint32_t offset_in_block;
611     uint32_t n_bytes;
612     uint64_t data_offset;
613     uint32_t bmap_first = VDI_UNALLOCATED;
614     uint32_t bmap_last = VDI_UNALLOCATED;
615     uint8_t *block = NULL;
616     uint64_t bytes_done = 0;
617     int ret = 0;
618 
619     logout("\n");
620 
621     qemu_iovec_init(&local_qiov, qiov->niov);
622 
623     while (ret >= 0 && bytes > 0) {
624         block_index = offset / s->block_size;
625         offset_in_block = offset % s->block_size;
626         n_bytes = MIN(bytes, s->block_size - offset_in_block);
627 
628         logout("will write %u bytes starting at offset %" PRIu64 "\n",
629                n_bytes, offset);
630 
631         /* prepare next AIO request */
632         qemu_co_rwlock_rdlock(&s->bmap_lock);
633         bmap_entry = le32_to_cpu(s->bmap[block_index]);
634         if (!VDI_IS_ALLOCATED(bmap_entry)) {
635             /* Allocate new block and write to it. */
636             uint64_t data_offset;
637             qemu_co_rwlock_upgrade(&s->bmap_lock);
638             bmap_entry = le32_to_cpu(s->bmap[block_index]);
639             if (VDI_IS_ALLOCATED(bmap_entry)) {
640                 /* A concurrent allocation did the work for us.  */
641                 qemu_co_rwlock_downgrade(&s->bmap_lock);
642                 goto nonallocating_write;
643             }
644 
645             bmap_entry = s->header.blocks_allocated;
646             s->bmap[block_index] = cpu_to_le32(bmap_entry);
647             s->header.blocks_allocated++;
648             data_offset = s->header.offset_data +
649                           (uint64_t)bmap_entry * s->block_size;
650             if (block == NULL) {
651                 block = g_malloc(s->block_size);
652                 bmap_first = block_index;
653             }
654             bmap_last = block_index;
655             /* Copy data to be written to new block and zero unused parts. */
656             memset(block, 0, offset_in_block);
657             qemu_iovec_to_buf(qiov, bytes_done, block + offset_in_block,
658                               n_bytes);
659             memset(block + offset_in_block + n_bytes, 0,
660                    s->block_size - n_bytes - offset_in_block);
661 
662             /* Write the new block under CoRwLock write-side protection,
663              * so this full-cluster write does not overlap a partial write
664              * of the same cluster, issued from the "else" branch.
665              */
666             ret = bdrv_co_pwrite(bs->file, data_offset, s->block_size, block,
667                                  0);
668             qemu_co_rwlock_unlock(&s->bmap_lock);
669         } else {
670 nonallocating_write:
671             data_offset = s->header.offset_data +
672                            (uint64_t)bmap_entry * s->block_size +
673                            offset_in_block;
674             qemu_co_rwlock_unlock(&s->bmap_lock);
675 
676             qemu_iovec_reset(&local_qiov);
677             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
678 
679             ret = bdrv_co_pwritev(bs->file, data_offset, n_bytes,
680                                   &local_qiov, 0);
681         }
682 
683         bytes -= n_bytes;
684         offset += n_bytes;
685         bytes_done += n_bytes;
686 
687         logout("%u bytes written\n", n_bytes);
688     }
689 
690     qemu_iovec_destroy(&local_qiov);
691 
692     logout("finished data write\n");
693     if (ret < 0) {
694         g_free(block);
695         return ret;
696     }
697 
698     if (block) {
699         /* One or more new blocks were allocated. */
700         VdiHeader *header;
701         uint8_t *base;
702         uint64_t offset;
703         uint32_t n_sectors;
704 
705         g_free(block);
706         header = g_malloc(sizeof(*header));
707 
708         logout("now writing modified header\n");
709         assert(VDI_IS_ALLOCATED(bmap_first));
710         *header = s->header;
711         vdi_header_to_le(header);
712         ret = bdrv_co_pwrite(bs->file, 0, sizeof(*header), header, 0);
713         g_free(header);
714 
715         if (ret < 0) {
716             return ret;
717         }
718 
719         logout("now writing modified block map entry %u...%u\n",
720                bmap_first, bmap_last);
721         /* Write modified sectors from block map. */
722         bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
723         bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
724         n_sectors = bmap_last - bmap_first + 1;
725         offset = s->bmap_sector + bmap_first;
726         base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
727         logout("will write %u block map sectors starting from entry %u\n",
728                n_sectors, bmap_first);
729         ret = bdrv_co_pwrite(bs->file, offset * SECTOR_SIZE,
730                              n_sectors * SECTOR_SIZE, base, 0);
731     }
732 
733     return ret;
734 }
735 
736 static int coroutine_fn vdi_co_do_create(BlockdevCreateOptions *create_options,
737                                          size_t block_size, Error **errp)
738 {
739     BlockdevCreateOptionsVdi *vdi_opts;
740     int ret = 0;
741     uint64_t bytes = 0;
742     uint32_t blocks;
743     uint32_t image_type;
744     VdiHeader header;
745     size_t i;
746     size_t bmap_size;
747     int64_t offset = 0;
748     BlockDriverState *bs_file = NULL;
749     BlockBackend *blk = NULL;
750     uint32_t *bmap = NULL;
751     QemuUUID uuid;
752 
753     assert(create_options->driver == BLOCKDEV_DRIVER_VDI);
754     vdi_opts = &create_options->u.vdi;
755 
756     logout("\n");
757 
758     /* Validate options and set default values */
759     bytes = vdi_opts->size;
760 
761     if (!vdi_opts->has_preallocation) {
762         vdi_opts->preallocation = PREALLOC_MODE_OFF;
763     }
764     switch (vdi_opts->preallocation) {
765     case PREALLOC_MODE_OFF:
766         image_type = VDI_TYPE_DYNAMIC;
767         break;
768     case PREALLOC_MODE_METADATA:
769         image_type = VDI_TYPE_STATIC;
770         break;
771     default:
772         error_setg(errp, "Preallocation mode not supported for vdi");
773         return -EINVAL;
774     }
775 
776 #ifndef CONFIG_VDI_STATIC_IMAGE
777     if (image_type == VDI_TYPE_STATIC) {
778         ret = -ENOTSUP;
779         error_setg(errp, "Statically allocated images cannot be created in "
780                    "this build");
781         goto exit;
782     }
783 #endif
784 #ifndef CONFIG_VDI_BLOCK_SIZE
785     if (block_size != DEFAULT_CLUSTER_SIZE) {
786         ret = -ENOTSUP;
787         error_setg(errp,
788                    "A non-default cluster size is not supported in this build");
789         goto exit;
790     }
791 #endif
792 
793     if (bytes > VDI_DISK_SIZE_MAX) {
794         ret = -ENOTSUP;
795         error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
796                           ", max supported is 0x%" PRIx64 ")",
797                           bytes, VDI_DISK_SIZE_MAX);
798         goto exit;
799     }
800 
801     /* Create BlockBackend to write to the image */
802     bs_file = bdrv_open_blockdev_ref(vdi_opts->file, errp);
803     if (!bs_file) {
804         ret = -EIO;
805         goto exit;
806     }
807 
808     blk = blk_new_with_bs(bs_file, BLK_PERM_WRITE | BLK_PERM_RESIZE,
809                           BLK_PERM_ALL, errp);
810     if (!blk) {
811         ret = -EPERM;
812         goto exit;
813     }
814 
815     blk_set_allow_write_beyond_eof(blk, true);
816 
817     /* We need enough blocks to store the given disk size,
818        so always round up. */
819     blocks = DIV_ROUND_UP(bytes, block_size);
820 
821     bmap_size = blocks * sizeof(uint32_t);
822     bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE);
823 
824     memset(&header, 0, sizeof(header));
825     pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
826     header.signature = VDI_SIGNATURE;
827     header.version = VDI_VERSION_1_1;
828     header.header_size = 0x180;
829     header.image_type = image_type;
830     header.offset_bmap = 0x200;
831     header.offset_data = 0x200 + bmap_size;
832     header.sector_size = SECTOR_SIZE;
833     header.disk_size = bytes;
834     header.block_size = block_size;
835     header.blocks_in_image = blocks;
836     if (image_type == VDI_TYPE_STATIC) {
837         header.blocks_allocated = blocks;
838     }
839     qemu_uuid_generate(&uuid);
840     header.uuid_image = uuid;
841     qemu_uuid_generate(&uuid);
842     header.uuid_last_snap = uuid;
843     /* There is no need to set header.uuid_link or header.uuid_parent here. */
844     if (VDI_DEBUG) {
845         vdi_header_print(&header);
846     }
847     vdi_header_to_le(&header);
848     ret = blk_co_pwrite(blk, offset, sizeof(header), &header, 0);
849     if (ret < 0) {
850         error_setg(errp, "Error writing header");
851         goto exit;
852     }
853     offset += sizeof(header);
854 
855     if (bmap_size > 0) {
856         bmap = g_try_malloc0(bmap_size);
857         if (bmap == NULL) {
858             ret = -ENOMEM;
859             error_setg(errp, "Could not allocate bmap");
860             goto exit;
861         }
862         for (i = 0; i < blocks; i++) {
863             if (image_type == VDI_TYPE_STATIC) {
864                 bmap[i] = i;
865             } else {
866                 bmap[i] = VDI_UNALLOCATED;
867             }
868         }
869         ret = blk_co_pwrite(blk, offset, bmap_size, bmap, 0);
870         if (ret < 0) {
871             error_setg(errp, "Error writing bmap");
872             goto exit;
873         }
874         offset += bmap_size;
875     }
876 
877     if (image_type == VDI_TYPE_STATIC) {
878         ret = blk_co_truncate(blk, offset + blocks * block_size, false,
879                               PREALLOC_MODE_OFF, 0, errp);
880         if (ret < 0) {
881             error_prepend(errp, "Failed to statically allocate file");
882             goto exit;
883         }
884     }
885 
886     ret = 0;
887 exit:
888     blk_unref(blk);
889     bdrv_unref(bs_file);
890     g_free(bmap);
891     return ret;
892 }
893 
894 static int coroutine_fn vdi_co_create(BlockdevCreateOptions *create_options,
895                                       Error **errp)
896 {
897     return vdi_co_do_create(create_options, DEFAULT_CLUSTER_SIZE, errp);
898 }
899 
900 static int coroutine_fn vdi_co_create_opts(BlockDriver *drv,
901                                            const char *filename,
902                                            QemuOpts *opts,
903                                            Error **errp)
904 {
905     QDict *qdict = NULL;
906     BlockdevCreateOptions *create_options = NULL;
907     BlockDriverState *bs_file = NULL;
908     uint64_t block_size = DEFAULT_CLUSTER_SIZE;
909     bool is_static = false;
910     Visitor *v;
911     int ret;
912 
913     /* Parse options and convert legacy syntax.
914      *
915      * Since CONFIG_VDI_BLOCK_SIZE is disabled by default,
916      * cluster-size is not part of the QAPI schema; therefore we have
917      * to parse it before creating the QAPI object. */
918 #if defined(CONFIG_VDI_BLOCK_SIZE)
919     block_size = qemu_opt_get_size_del(opts,
920                                        BLOCK_OPT_CLUSTER_SIZE,
921                                        DEFAULT_CLUSTER_SIZE);
922     if (block_size < BDRV_SECTOR_SIZE || block_size > UINT32_MAX ||
923         !is_power_of_2(block_size))
924     {
925         error_setg(errp, "Invalid cluster size");
926         ret = -EINVAL;
927         goto done;
928     }
929 #endif
930     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) {
931         is_static = true;
932     }
933 
934     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vdi_create_opts, true);
935 
936     /* Create and open the file (protocol layer) */
937     ret = bdrv_co_create_file(filename, opts, errp);
938     if (ret < 0) {
939         goto done;
940     }
941 
942     bs_file = bdrv_open(filename, NULL, NULL,
943                         BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
944     if (!bs_file) {
945         ret = -EIO;
946         goto done;
947     }
948 
949     qdict_put_str(qdict, "driver", "vdi");
950     qdict_put_str(qdict, "file", bs_file->node_name);
951     if (is_static) {
952         qdict_put_str(qdict, "preallocation", "metadata");
953     }
954 
955     /* Get the QAPI object */
956     v = qobject_input_visitor_new_flat_confused(qdict, errp);
957     if (!v) {
958         ret = -EINVAL;
959         goto done;
960     }
961     visit_type_BlockdevCreateOptions(v, NULL, &create_options, errp);
962     visit_free(v);
963     if (!create_options) {
964         ret = -EINVAL;
965         goto done;
966     }
967 
968     /* Silently round up size */
969     assert(create_options->driver == BLOCKDEV_DRIVER_VDI);
970     create_options->u.vdi.size = ROUND_UP(create_options->u.vdi.size,
971                                           BDRV_SECTOR_SIZE);
972 
973     /* Create the vdi image (format layer) */
974     ret = vdi_co_do_create(create_options, block_size, errp);
975 done:
976     qobject_unref(qdict);
977     qapi_free_BlockdevCreateOptions(create_options);
978     bdrv_unref(bs_file);
979     return ret;
980 }
981 
982 static void vdi_close(BlockDriverState *bs)
983 {
984     BDRVVdiState *s = bs->opaque;
985 
986     qemu_vfree(s->bmap);
987 
988     migrate_del_blocker(s->migration_blocker);
989     error_free(s->migration_blocker);
990 }
991 
992 static int vdi_has_zero_init(BlockDriverState *bs)
993 {
994     BDRVVdiState *s = bs->opaque;
995 
996     if (s->header.image_type == VDI_TYPE_STATIC) {
997         return bdrv_has_zero_init(bs->file->bs);
998     } else {
999         return 1;
1000     }
1001 }
1002 
1003 static QemuOptsList vdi_create_opts = {
1004     .name = "vdi-create-opts",
1005     .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head),
1006     .desc = {
1007         {
1008             .name = BLOCK_OPT_SIZE,
1009             .type = QEMU_OPT_SIZE,
1010             .help = "Virtual disk size"
1011         },
1012 #if defined(CONFIG_VDI_BLOCK_SIZE)
1013         {
1014             .name = BLOCK_OPT_CLUSTER_SIZE,
1015             .type = QEMU_OPT_SIZE,
1016             .help = "VDI cluster (block) size",
1017             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
1018         },
1019 #endif
1020 #if defined(CONFIG_VDI_STATIC_IMAGE)
1021         {
1022             .name = BLOCK_OPT_STATIC,
1023             .type = QEMU_OPT_BOOL,
1024             .help = "VDI static (pre-allocated) image",
1025             .def_value_str = "off"
1026         },
1027 #endif
1028         /* TODO: An additional option to set UUID values might be useful. */
1029         { /* end of list */ }
1030     }
1031 };
1032 
1033 static BlockDriver bdrv_vdi = {
1034     .format_name = "vdi",
1035     .instance_size = sizeof(BDRVVdiState),
1036     .bdrv_probe = vdi_probe,
1037     .bdrv_open = vdi_open,
1038     .bdrv_close = vdi_close,
1039     .bdrv_reopen_prepare = vdi_reopen_prepare,
1040     .bdrv_child_perm          = bdrv_default_perms,
1041     .bdrv_co_create      = vdi_co_create,
1042     .bdrv_co_create_opts = vdi_co_create_opts,
1043     .bdrv_has_zero_init  = vdi_has_zero_init,
1044     .bdrv_co_block_status = vdi_co_block_status,
1045     .bdrv_make_empty = vdi_make_empty,
1046 
1047     .bdrv_co_preadv     = vdi_co_preadv,
1048 #if defined(CONFIG_VDI_WRITE)
1049     .bdrv_co_pwritev    = vdi_co_pwritev,
1050 #endif
1051 
1052     .bdrv_get_info = vdi_get_info,
1053 
1054     .is_format = true,
1055     .create_opts = &vdi_create_opts,
1056     .bdrv_co_check = vdi_co_check,
1057 };
1058 
1059 static void bdrv_vdi_init(void)
1060 {
1061     logout("\n");
1062     bdrv_register(&bdrv_vdi);
1063 }
1064 
1065 block_init(bdrv_vdi_init);
1066