xref: /openbmc/qemu/block/vdi.c (revision 3d9569b8)
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 
68 /* Code configuration options. */
69 
70 /* Enable debug messages. */
71 //~ #define CONFIG_VDI_DEBUG
72 
73 /* Support write operations on VDI images. */
74 #define CONFIG_VDI_WRITE
75 
76 /* Support non-standard block (cluster) size. This is untested.
77  * Maybe it will be needed for very large images.
78  */
79 //~ #define CONFIG_VDI_BLOCK_SIZE
80 
81 /* Support static (fixed, pre-allocated) images. */
82 #define CONFIG_VDI_STATIC_IMAGE
83 
84 /* Command line option for static images. */
85 #define BLOCK_OPT_STATIC "static"
86 
87 #define SECTOR_SIZE 512
88 #define DEFAULT_CLUSTER_SIZE 1048576
89 /* Note: can't use 1 * MiB, because it's passed to stringify() */
90 
91 #if defined(CONFIG_VDI_DEBUG)
92 #define VDI_DEBUG 1
93 #else
94 #define VDI_DEBUG 0
95 #endif
96 
97 #define logout(fmt, ...) \
98     do {                                                                \
99         if (VDI_DEBUG) {                                                \
100             fprintf(stderr, "vdi\t%-24s" fmt, __func__, ##__VA_ARGS__); \
101         }                                                               \
102     } while (0)
103 
104 /* Image signature. */
105 #define VDI_SIGNATURE 0xbeda107f
106 
107 /* Image version. */
108 #define VDI_VERSION_1_1 0x00010001
109 
110 /* Image type. */
111 #define VDI_TYPE_DYNAMIC 1
112 #define VDI_TYPE_STATIC  2
113 
114 /* Innotek / SUN images use these strings in header.text:
115  * "<<< innotek VirtualBox Disk Image >>>\n"
116  * "<<< Sun xVM VirtualBox Disk Image >>>\n"
117  * "<<< Sun VirtualBox Disk Image >>>\n"
118  * The value does not matter, so QEMU created images use a different text.
119  */
120 #define VDI_TEXT "<<< QEMU VM Virtual Disk Image >>>\n"
121 
122 /* A never-allocated block; semantically arbitrary content. */
123 #define VDI_UNALLOCATED 0xffffffffU
124 
125 /* A discarded (no longer allocated) block; semantically zero-filled. */
126 #define VDI_DISCARDED   0xfffffffeU
127 
128 #define VDI_IS_ALLOCATED(X) ((X) < VDI_DISCARDED)
129 
130 /* The bmap will take up VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) bytes; since
131  * the bmap is read and written in a single operation, its size needs to be
132  * limited to INT_MAX; furthermore, when opening an image, the bmap size is
133  * rounded up to be aligned on BDRV_SECTOR_SIZE.
134  * Therefore this should satisfy the following:
135  * VDI_BLOCKS_IN_IMAGE_MAX * sizeof(uint32_t) + BDRV_SECTOR_SIZE == INT_MAX + 1
136  * (INT_MAX + 1 is the first value not representable as an int)
137  * This guarantees that any value below or equal to the constant will, when
138  * multiplied by sizeof(uint32_t) and rounded up to a BDRV_SECTOR_SIZE boundary,
139  * still be below or equal to INT_MAX. */
140 #define VDI_BLOCKS_IN_IMAGE_MAX \
141     ((unsigned)((INT_MAX + 1u - BDRV_SECTOR_SIZE) / sizeof(uint32_t)))
142 #define VDI_DISK_SIZE_MAX        ((uint64_t)VDI_BLOCKS_IN_IMAGE_MAX * \
143                                   (uint64_t)DEFAULT_CLUSTER_SIZE)
144 
145 static QemuOptsList vdi_create_opts;
146 
147 typedef struct {
148     char text[0x40];
149     uint32_t signature;
150     uint32_t version;
151     uint32_t header_size;
152     uint32_t image_type;
153     uint32_t image_flags;
154     char description[256];
155     uint32_t offset_bmap;
156     uint32_t offset_data;
157     uint32_t cylinders;         /* disk geometry, unused here */
158     uint32_t heads;             /* disk geometry, unused here */
159     uint32_t sectors;           /* disk geometry, unused here */
160     uint32_t sector_size;
161     uint32_t unused1;
162     uint64_t disk_size;
163     uint32_t block_size;
164     uint32_t block_extra;       /* unused here */
165     uint32_t blocks_in_image;
166     uint32_t blocks_allocated;
167     QemuUUID uuid_image;
168     QemuUUID uuid_last_snap;
169     QemuUUID uuid_link;
170     QemuUUID uuid_parent;
171     uint64_t unused2[7];
172 } QEMU_PACKED VdiHeader;
173 
174 QEMU_BUILD_BUG_ON(sizeof(VdiHeader) != 512);
175 
176 typedef struct {
177     /* The block map entries are little endian (even in memory). */
178     uint32_t *bmap;
179     /* Size of block (bytes). */
180     uint32_t block_size;
181     /* First sector of block map. */
182     uint32_t bmap_sector;
183     /* VDI header (converted to host endianness). */
184     VdiHeader header;
185 
186     CoRwlock bmap_lock;
187 
188     Error *migration_blocker;
189 } BDRVVdiState;
190 
191 static void vdi_header_to_cpu(VdiHeader *header)
192 {
193     header->signature = le32_to_cpu(header->signature);
194     header->version = le32_to_cpu(header->version);
195     header->header_size = le32_to_cpu(header->header_size);
196     header->image_type = le32_to_cpu(header->image_type);
197     header->image_flags = le32_to_cpu(header->image_flags);
198     header->offset_bmap = le32_to_cpu(header->offset_bmap);
199     header->offset_data = le32_to_cpu(header->offset_data);
200     header->cylinders = le32_to_cpu(header->cylinders);
201     header->heads = le32_to_cpu(header->heads);
202     header->sectors = le32_to_cpu(header->sectors);
203     header->sector_size = le32_to_cpu(header->sector_size);
204     header->disk_size = le64_to_cpu(header->disk_size);
205     header->block_size = le32_to_cpu(header->block_size);
206     header->block_extra = le32_to_cpu(header->block_extra);
207     header->blocks_in_image = le32_to_cpu(header->blocks_in_image);
208     header->blocks_allocated = le32_to_cpu(header->blocks_allocated);
209     header->uuid_image = qemu_uuid_bswap(header->uuid_image);
210     header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap);
211     header->uuid_link = qemu_uuid_bswap(header->uuid_link);
212     header->uuid_parent = qemu_uuid_bswap(header->uuid_parent);
213 }
214 
215 static void vdi_header_to_le(VdiHeader *header)
216 {
217     header->signature = cpu_to_le32(header->signature);
218     header->version = cpu_to_le32(header->version);
219     header->header_size = cpu_to_le32(header->header_size);
220     header->image_type = cpu_to_le32(header->image_type);
221     header->image_flags = cpu_to_le32(header->image_flags);
222     header->offset_bmap = cpu_to_le32(header->offset_bmap);
223     header->offset_data = cpu_to_le32(header->offset_data);
224     header->cylinders = cpu_to_le32(header->cylinders);
225     header->heads = cpu_to_le32(header->heads);
226     header->sectors = cpu_to_le32(header->sectors);
227     header->sector_size = cpu_to_le32(header->sector_size);
228     header->disk_size = cpu_to_le64(header->disk_size);
229     header->block_size = cpu_to_le32(header->block_size);
230     header->block_extra = cpu_to_le32(header->block_extra);
231     header->blocks_in_image = cpu_to_le32(header->blocks_in_image);
232     header->blocks_allocated = cpu_to_le32(header->blocks_allocated);
233     header->uuid_image = qemu_uuid_bswap(header->uuid_image);
234     header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap);
235     header->uuid_link = qemu_uuid_bswap(header->uuid_link);
236     header->uuid_parent = qemu_uuid_bswap(header->uuid_parent);
237 }
238 
239 static void vdi_header_print(VdiHeader *header)
240 {
241     char uuidstr[37];
242     QemuUUID uuid;
243     logout("text        %s", header->text);
244     logout("signature   0x%08x\n", header->signature);
245     logout("header size 0x%04x\n", header->header_size);
246     logout("image type  0x%04x\n", header->image_type);
247     logout("image flags 0x%04x\n", header->image_flags);
248     logout("description %s\n", header->description);
249     logout("offset bmap 0x%04x\n", header->offset_bmap);
250     logout("offset data 0x%04x\n", header->offset_data);
251     logout("cylinders   0x%04x\n", header->cylinders);
252     logout("heads       0x%04x\n", header->heads);
253     logout("sectors     0x%04x\n", header->sectors);
254     logout("sector size 0x%04x\n", header->sector_size);
255     logout("image size  0x%" PRIx64 " B (%" PRIu64 " MiB)\n",
256            header->disk_size, header->disk_size / MiB);
257     logout("block size  0x%04x\n", header->block_size);
258     logout("block extra 0x%04x\n", header->block_extra);
259     logout("blocks tot. 0x%04x\n", header->blocks_in_image);
260     logout("blocks all. 0x%04x\n", header->blocks_allocated);
261     uuid = header->uuid_image;
262     qemu_uuid_unparse(&uuid, uuidstr);
263     logout("uuid image  %s\n", uuidstr);
264     uuid = header->uuid_last_snap;
265     qemu_uuid_unparse(&uuid, uuidstr);
266     logout("uuid snap   %s\n", uuidstr);
267     uuid = header->uuid_link;
268     qemu_uuid_unparse(&uuid, uuidstr);
269     logout("uuid link   %s\n", uuidstr);
270     uuid = header->uuid_parent;
271     qemu_uuid_unparse(&uuid, uuidstr);
272     logout("uuid parent %s\n", uuidstr);
273 }
274 
275 static int coroutine_fn vdi_co_check(BlockDriverState *bs, BdrvCheckResult *res,
276                                      BdrvCheckMode fix)
277 {
278     /* TODO: additional checks possible. */
279     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
280     uint32_t blocks_allocated = 0;
281     uint32_t block;
282     uint32_t *bmap;
283     logout("\n");
284 
285     if (fix) {
286         return -ENOTSUP;
287     }
288 
289     bmap = g_try_new(uint32_t, s->header.blocks_in_image);
290     if (s->header.blocks_in_image && bmap == NULL) {
291         res->check_errors++;
292         return -ENOMEM;
293     }
294 
295     memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t));
296 
297     /* Check block map and value of blocks_allocated. */
298     for (block = 0; block < s->header.blocks_in_image; block++) {
299         uint32_t bmap_entry = le32_to_cpu(s->bmap[block]);
300         if (VDI_IS_ALLOCATED(bmap_entry)) {
301             if (bmap_entry < s->header.blocks_in_image) {
302                 blocks_allocated++;
303                 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) {
304                     bmap[bmap_entry] = bmap_entry;
305                 } else {
306                     fprintf(stderr, "ERROR: block index %" PRIu32
307                             " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry);
308                     res->corruptions++;
309                 }
310             } else {
311                 fprintf(stderr, "ERROR: block index %" PRIu32
312                         " too large, is %" PRIu32 "\n", block, bmap_entry);
313                 res->corruptions++;
314             }
315         }
316     }
317     if (blocks_allocated != s->header.blocks_allocated) {
318         fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32
319                ", should be %" PRIu32 "\n",
320                blocks_allocated, s->header.blocks_allocated);
321         res->corruptions++;
322     }
323 
324     g_free(bmap);
325 
326     return 0;
327 }
328 
329 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
330 {
331     /* TODO: vdi_get_info would be needed for machine snapshots.
332        vm_state_offset is still missing. */
333     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
334     logout("\n");
335     bdi->cluster_size = s->block_size;
336     bdi->vm_state_offset = 0;
337     bdi->unallocated_blocks_are_zero = true;
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     Error *local_err = NULL;
379     QemuUUID uuid_link, uuid_parent;
380 
381     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
382                                false, errp);
383     if (!bs->file) {
384         return -EINVAL;
385     }
386 
387     logout("\n");
388 
389     ret = bdrv_pread(bs->file, 0, &header, sizeof(header));
390     if (ret < 0) {
391         goto fail;
392     }
393 
394     vdi_header_to_cpu(&header);
395     if (VDI_DEBUG) {
396         vdi_header_print(&header);
397     }
398 
399     if (header.disk_size > VDI_DISK_SIZE_MAX) {
400         error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
401                           ", max supported is 0x%" PRIx64 ")",
402                           header.disk_size, VDI_DISK_SIZE_MAX);
403         ret = -ENOTSUP;
404         goto fail;
405     }
406 
407     uuid_link = header.uuid_link;
408     uuid_parent = header.uuid_parent;
409 
410     if (header.disk_size % SECTOR_SIZE != 0) {
411         /* 'VBoxManage convertfromraw' can create images with odd disk sizes.
412            We accept them but round the disk size to the next multiple of
413            SECTOR_SIZE. */
414         logout("odd disk size %" PRIu64 " B, round up\n", header.disk_size);
415         header.disk_size = ROUND_UP(header.disk_size, SECTOR_SIZE);
416     }
417 
418     if (header.signature != VDI_SIGNATURE) {
419         error_setg(errp, "Image not in VDI format (bad signature %08" PRIx32
420                    ")", header.signature);
421         ret = -EINVAL;
422         goto fail;
423     } else if (header.version != VDI_VERSION_1_1) {
424         error_setg(errp, "unsupported VDI image (version %" PRIu32 ".%" PRIu32
425                    ")", header.version >> 16, header.version & 0xffff);
426         ret = -ENOTSUP;
427         goto fail;
428     } else if (header.offset_bmap % SECTOR_SIZE != 0) {
429         /* We only support block maps which start on a sector boundary. */
430         error_setg(errp, "unsupported VDI image (unaligned block map offset "
431                    "0x%" PRIx32 ")", header.offset_bmap);
432         ret = -ENOTSUP;
433         goto fail;
434     } else if (header.offset_data % SECTOR_SIZE != 0) {
435         /* We only support data blocks which start on a sector boundary. */
436         error_setg(errp, "unsupported VDI image (unaligned data offset 0x%"
437                    PRIx32 ")", header.offset_data);
438         ret = -ENOTSUP;
439         goto fail;
440     } else if (header.sector_size != SECTOR_SIZE) {
441         error_setg(errp, "unsupported VDI image (sector size %" PRIu32
442                    " is not %u)", header.sector_size, SECTOR_SIZE);
443         ret = -ENOTSUP;
444         goto fail;
445     } else if (header.block_size != DEFAULT_CLUSTER_SIZE) {
446         error_setg(errp, "unsupported VDI image (block size %" PRIu32
447                          " is not %" PRIu32 ")",
448                    header.block_size, DEFAULT_CLUSTER_SIZE);
449         ret = -ENOTSUP;
450         goto fail;
451     } else if (header.disk_size >
452                (uint64_t)header.blocks_in_image * header.block_size) {
453         error_setg(errp, "unsupported VDI image (disk size %" PRIu64 ", "
454                    "image bitmap has room for %" PRIu64 ")",
455                    header.disk_size,
456                    (uint64_t)header.blocks_in_image * header.block_size);
457         ret = -ENOTSUP;
458         goto fail;
459     } else if (!qemu_uuid_is_null(&uuid_link)) {
460         error_setg(errp, "unsupported VDI image (non-NULL link UUID)");
461         ret = -ENOTSUP;
462         goto fail;
463     } else if (!qemu_uuid_is_null(&uuid_parent)) {
464         error_setg(errp, "unsupported VDI image (non-NULL parent UUID)");
465         ret = -ENOTSUP;
466         goto fail;
467     } else if (header.blocks_in_image > VDI_BLOCKS_IN_IMAGE_MAX) {
468         error_setg(errp, "unsupported VDI image "
469                          "(too many blocks %u, max is %u)",
470                           header.blocks_in_image, VDI_BLOCKS_IN_IMAGE_MAX);
471         ret = -ENOTSUP;
472         goto fail;
473     }
474 
475     bs->total_sectors = header.disk_size / SECTOR_SIZE;
476 
477     s->block_size = header.block_size;
478     s->bmap_sector = header.offset_bmap / SECTOR_SIZE;
479     s->header = header;
480 
481     bmap_size = header.blocks_in_image * sizeof(uint32_t);
482     bmap_size = DIV_ROUND_UP(bmap_size, SECTOR_SIZE);
483     s->bmap = qemu_try_blockalign(bs->file->bs, bmap_size * SECTOR_SIZE);
484     if (s->bmap == NULL) {
485         ret = -ENOMEM;
486         goto fail;
487     }
488 
489     ret = bdrv_pread(bs->file, header.offset_bmap, s->bmap,
490                      bmap_size * SECTOR_SIZE);
491     if (ret < 0) {
492         goto fail_free_bmap;
493     }
494 
495     /* Disable migration when vdi images are used */
496     error_setg(&s->migration_blocker, "The vdi format used by node '%s' "
497                "does not support live migration",
498                bdrv_get_device_or_node_name(bs));
499     ret = migrate_add_blocker(s->migration_blocker, &local_err);
500     if (local_err) {
501         error_propagate(errp, local_err);
502         error_free(s->migration_blocker);
503         goto fail_free_bmap;
504     }
505 
506     qemu_co_rwlock_init(&s->bmap_lock);
507 
508     return 0;
509 
510  fail_free_bmap:
511     qemu_vfree(s->bmap);
512 
513  fail:
514     return ret;
515 }
516 
517 static int vdi_reopen_prepare(BDRVReopenState *state,
518                               BlockReopenQueue *queue, Error **errp)
519 {
520     return 0;
521 }
522 
523 static int coroutine_fn vdi_co_block_status(BlockDriverState *bs,
524                                             bool want_zero,
525                                             int64_t offset, int64_t bytes,
526                                             int64_t *pnum, int64_t *map,
527                                             BlockDriverState **file)
528 {
529     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
530     size_t bmap_index = offset / s->block_size;
531     size_t index_in_block = offset % s->block_size;
532     uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
533     int result;
534 
535     logout("%p, %" PRId64 ", %" PRId64 ", %p\n", bs, offset, bytes, pnum);
536     *pnum = MIN(s->block_size - index_in_block, bytes);
537     result = VDI_IS_ALLOCATED(bmap_entry);
538     if (!result) {
539         return 0;
540     }
541 
542     *map = s->header.offset_data + (uint64_t)bmap_entry * s->block_size +
543         index_in_block;
544     *file = bs->file->bs;
545     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
546 }
547 
548 static int coroutine_fn
549 vdi_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
550               QEMUIOVector *qiov, int flags)
551 {
552     BDRVVdiState *s = bs->opaque;
553     QEMUIOVector local_qiov;
554     uint32_t bmap_entry;
555     uint32_t block_index;
556     uint32_t offset_in_block;
557     uint32_t n_bytes;
558     uint64_t bytes_done = 0;
559     int ret = 0;
560 
561     logout("\n");
562 
563     qemu_iovec_init(&local_qiov, qiov->niov);
564 
565     while (ret >= 0 && bytes > 0) {
566         block_index = offset / s->block_size;
567         offset_in_block = offset % s->block_size;
568         n_bytes = MIN(bytes, s->block_size - offset_in_block);
569 
570         logout("will read %u bytes starting at offset %" PRIu64 "\n",
571                n_bytes, offset);
572 
573         /* prepare next AIO request */
574         qemu_co_rwlock_rdlock(&s->bmap_lock);
575         bmap_entry = le32_to_cpu(s->bmap[block_index]);
576         qemu_co_rwlock_unlock(&s->bmap_lock);
577         if (!VDI_IS_ALLOCATED(bmap_entry)) {
578             /* Block not allocated, return zeros, no need to wait. */
579             qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
580             ret = 0;
581         } else {
582             uint64_t data_offset = s->header.offset_data +
583                                    (uint64_t)bmap_entry * s->block_size +
584                                    offset_in_block;
585 
586             qemu_iovec_reset(&local_qiov);
587             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
588 
589             ret = bdrv_co_preadv(bs->file, data_offset, n_bytes,
590                                  &local_qiov, 0);
591         }
592         logout("%u bytes read\n", n_bytes);
593 
594         bytes -= n_bytes;
595         offset += n_bytes;
596         bytes_done += n_bytes;
597     }
598 
599     qemu_iovec_destroy(&local_qiov);
600 
601     return ret;
602 }
603 
604 static int coroutine_fn
605 vdi_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
606                QEMUIOVector *qiov, int flags)
607 {
608     BDRVVdiState *s = bs->opaque;
609     QEMUIOVector local_qiov;
610     uint32_t bmap_entry;
611     uint32_t block_index;
612     uint32_t offset_in_block;
613     uint32_t n_bytes;
614     uint64_t data_offset;
615     uint32_t bmap_first = VDI_UNALLOCATED;
616     uint32_t bmap_last = VDI_UNALLOCATED;
617     uint8_t *block = NULL;
618     uint64_t bytes_done = 0;
619     int ret = 0;
620 
621     logout("\n");
622 
623     qemu_iovec_init(&local_qiov, qiov->niov);
624 
625     while (ret >= 0 && bytes > 0) {
626         block_index = offset / s->block_size;
627         offset_in_block = offset % s->block_size;
628         n_bytes = MIN(bytes, s->block_size - offset_in_block);
629 
630         logout("will write %u bytes starting at offset %" PRIu64 "\n",
631                n_bytes, offset);
632 
633         /* prepare next AIO request */
634         qemu_co_rwlock_rdlock(&s->bmap_lock);
635         bmap_entry = le32_to_cpu(s->bmap[block_index]);
636         if (!VDI_IS_ALLOCATED(bmap_entry)) {
637             /* Allocate new block and write to it. */
638             uint64_t data_offset;
639             qemu_co_rwlock_upgrade(&s->bmap_lock);
640             bmap_entry = le32_to_cpu(s->bmap[block_index]);
641             if (VDI_IS_ALLOCATED(bmap_entry)) {
642                 /* A concurrent allocation did the work for us.  */
643                 qemu_co_rwlock_downgrade(&s->bmap_lock);
644                 goto nonallocating_write;
645             }
646 
647             bmap_entry = s->header.blocks_allocated;
648             s->bmap[block_index] = cpu_to_le32(bmap_entry);
649             s->header.blocks_allocated++;
650             data_offset = s->header.offset_data +
651                           (uint64_t)bmap_entry * s->block_size;
652             if (block == NULL) {
653                 block = g_malloc(s->block_size);
654                 bmap_first = block_index;
655             }
656             bmap_last = block_index;
657             /* Copy data to be written to new block and zero unused parts. */
658             memset(block, 0, offset_in_block);
659             qemu_iovec_to_buf(qiov, bytes_done, block + offset_in_block,
660                               n_bytes);
661             memset(block + offset_in_block + n_bytes, 0,
662                    s->block_size - n_bytes - offset_in_block);
663 
664             /* Write the new block under CoRwLock write-side protection,
665              * so this full-cluster write does not overlap a partial write
666              * of the same cluster, issued from the "else" branch.
667              */
668             ret = bdrv_pwrite(bs->file, data_offset, block, s->block_size);
669             qemu_co_rwlock_unlock(&s->bmap_lock);
670         } else {
671 nonallocating_write:
672             data_offset = s->header.offset_data +
673                            (uint64_t)bmap_entry * s->block_size +
674                            offset_in_block;
675             qemu_co_rwlock_unlock(&s->bmap_lock);
676 
677             qemu_iovec_reset(&local_qiov);
678             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
679 
680             ret = bdrv_co_pwritev(bs->file, data_offset, n_bytes,
681                                   &local_qiov, 0);
682         }
683 
684         bytes -= n_bytes;
685         offset += n_bytes;
686         bytes_done += n_bytes;
687 
688         logout("%u bytes written\n", n_bytes);
689     }
690 
691     qemu_iovec_destroy(&local_qiov);
692 
693     logout("finished data write\n");
694     if (ret < 0) {
695         return ret;
696     }
697 
698     if (block) {
699         /* One or more new blocks were allocated. */
700         VdiHeader *header = (VdiHeader *) block;
701         uint8_t *base;
702         uint64_t offset;
703         uint32_t n_sectors;
704 
705         logout("now writing modified header\n");
706         assert(VDI_IS_ALLOCATED(bmap_first));
707         *header = s->header;
708         vdi_header_to_le(header);
709         ret = bdrv_pwrite(bs->file, 0, block, sizeof(VdiHeader));
710         g_free(block);
711         block = NULL;
712 
713         if (ret < 0) {
714             return ret;
715         }
716 
717         logout("now writing modified block map entry %u...%u\n",
718                bmap_first, bmap_last);
719         /* Write modified sectors from block map. */
720         bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
721         bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
722         n_sectors = bmap_last - bmap_first + 1;
723         offset = s->bmap_sector + bmap_first;
724         base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
725         logout("will write %u block map sectors starting from entry %u\n",
726                n_sectors, bmap_first);
727         ret = bdrv_pwrite(bs->file, offset * SECTOR_SIZE, base,
728                           n_sectors * SECTOR_SIZE);
729     }
730 
731     return ret < 0 ? ret : 0;
732 }
733 
734 static int coroutine_fn vdi_co_do_create(BlockdevCreateOptions *create_options,
735                                          size_t block_size, Error **errp)
736 {
737     BlockdevCreateOptionsVdi *vdi_opts;
738     int ret = 0;
739     uint64_t bytes = 0;
740     uint32_t blocks;
741     uint32_t image_type;
742     VdiHeader header;
743     size_t i;
744     size_t bmap_size;
745     int64_t offset = 0;
746     BlockDriverState *bs_file = NULL;
747     BlockBackend *blk = NULL;
748     uint32_t *bmap = NULL;
749     QemuUUID uuid;
750 
751     assert(create_options->driver == BLOCKDEV_DRIVER_VDI);
752     vdi_opts = &create_options->u.vdi;
753 
754     logout("\n");
755 
756     /* Validate options and set default values */
757     bytes = vdi_opts->size;
758 
759     if (!vdi_opts->has_preallocation) {
760         vdi_opts->preallocation = PREALLOC_MODE_OFF;
761     }
762     switch (vdi_opts->preallocation) {
763     case PREALLOC_MODE_OFF:
764         image_type = VDI_TYPE_DYNAMIC;
765         break;
766     case PREALLOC_MODE_METADATA:
767         image_type = VDI_TYPE_STATIC;
768         break;
769     default:
770         error_setg(errp, "Preallocation mode not supported for vdi");
771         return -EINVAL;
772     }
773 
774 #ifndef CONFIG_VDI_STATIC_IMAGE
775     if (image_type == VDI_TYPE_STATIC) {
776         ret = -ENOTSUP;
777         error_setg(errp, "Statically allocated images cannot be created in "
778                    "this build");
779         goto exit;
780     }
781 #endif
782 #ifndef CONFIG_VDI_BLOCK_SIZE
783     if (block_size != DEFAULT_CLUSTER_SIZE) {
784         ret = -ENOTSUP;
785         error_setg(errp,
786                    "A non-default cluster size is not supported in this build");
787         goto exit;
788     }
789 #endif
790 
791     if (bytes > VDI_DISK_SIZE_MAX) {
792         ret = -ENOTSUP;
793         error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
794                           ", max supported is 0x%" PRIx64 ")",
795                           bytes, VDI_DISK_SIZE_MAX);
796         goto exit;
797     }
798 
799     /* Create BlockBackend to write to the image */
800     bs_file = bdrv_open_blockdev_ref(vdi_opts->file, errp);
801     if (!bs_file) {
802         ret = -EIO;
803         goto exit;
804     }
805 
806     blk = blk_new(bdrv_get_aio_context(bs_file),
807                   BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
808     ret = blk_insert_bs(blk, bs_file, errp);
809     if (ret < 0) {
810         goto exit;
811     }
812 
813     blk_set_allow_write_beyond_eof(blk, true);
814 
815     /* We need enough blocks to store the given disk size,
816        so always round up. */
817     blocks = DIV_ROUND_UP(bytes, block_size);
818 
819     bmap_size = blocks * sizeof(uint32_t);
820     bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE);
821 
822     memset(&header, 0, sizeof(header));
823     pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
824     header.signature = VDI_SIGNATURE;
825     header.version = VDI_VERSION_1_1;
826     header.header_size = 0x180;
827     header.image_type = image_type;
828     header.offset_bmap = 0x200;
829     header.offset_data = 0x200 + bmap_size;
830     header.sector_size = SECTOR_SIZE;
831     header.disk_size = bytes;
832     header.block_size = block_size;
833     header.blocks_in_image = blocks;
834     if (image_type == VDI_TYPE_STATIC) {
835         header.blocks_allocated = blocks;
836     }
837     qemu_uuid_generate(&uuid);
838     header.uuid_image = uuid;
839     qemu_uuid_generate(&uuid);
840     header.uuid_last_snap = uuid;
841     /* There is no need to set header.uuid_link or header.uuid_parent here. */
842     if (VDI_DEBUG) {
843         vdi_header_print(&header);
844     }
845     vdi_header_to_le(&header);
846     ret = blk_pwrite(blk, offset, &header, sizeof(header), 0);
847     if (ret < 0) {
848         error_setg(errp, "Error writing header");
849         goto exit;
850     }
851     offset += sizeof(header);
852 
853     if (bmap_size > 0) {
854         bmap = g_try_malloc0(bmap_size);
855         if (bmap == NULL) {
856             ret = -ENOMEM;
857             error_setg(errp, "Could not allocate bmap");
858             goto exit;
859         }
860         for (i = 0; i < blocks; i++) {
861             if (image_type == VDI_TYPE_STATIC) {
862                 bmap[i] = i;
863             } else {
864                 bmap[i] = VDI_UNALLOCATED;
865             }
866         }
867         ret = blk_pwrite(blk, offset, bmap, bmap_size, 0);
868         if (ret < 0) {
869             error_setg(errp, "Error writing bmap");
870             goto exit;
871         }
872         offset += bmap_size;
873     }
874 
875     if (image_type == VDI_TYPE_STATIC) {
876         ret = blk_truncate(blk, offset + blocks * block_size,
877                            PREALLOC_MODE_OFF, errp);
878         if (ret < 0) {
879             error_prepend(errp, "Failed to statically allocate file");
880             goto exit;
881         }
882     }
883 
884     ret = 0;
885 exit:
886     blk_unref(blk);
887     bdrv_unref(bs_file);
888     g_free(bmap);
889     return ret;
890 }
891 
892 static int coroutine_fn vdi_co_create(BlockdevCreateOptions *create_options,
893                                       Error **errp)
894 {
895     return vdi_co_do_create(create_options, DEFAULT_CLUSTER_SIZE, errp);
896 }
897 
898 static int coroutine_fn vdi_co_create_opts(const char *filename, QemuOpts *opts,
899                                            Error **errp)
900 {
901     QDict *qdict = NULL;
902     BlockdevCreateOptions *create_options = NULL;
903     BlockDriverState *bs_file = NULL;
904     uint64_t block_size = DEFAULT_CLUSTER_SIZE;
905     bool is_static = false;
906     Visitor *v;
907     Error *local_err = NULL;
908     int ret;
909 
910     /* Parse options and convert legacy syntax.
911      *
912      * Since CONFIG_VDI_BLOCK_SIZE is disabled by default,
913      * cluster-size is not part of the QAPI schema; therefore we have
914      * to parse it before creating the QAPI object. */
915 #if defined(CONFIG_VDI_BLOCK_SIZE)
916     block_size = qemu_opt_get_size_del(opts,
917                                        BLOCK_OPT_CLUSTER_SIZE,
918                                        DEFAULT_CLUSTER_SIZE);
919     if (block_size < BDRV_SECTOR_SIZE || block_size > UINT32_MAX ||
920         !is_power_of_2(block_size))
921     {
922         error_setg(errp, "Invalid cluster size");
923         ret = -EINVAL;
924         goto done;
925     }
926 #endif
927     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) {
928         is_static = true;
929     }
930 
931     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vdi_create_opts, true);
932 
933     /* Create and open the file (protocol layer) */
934     ret = bdrv_create_file(filename, opts, errp);
935     if (ret < 0) {
936         goto done;
937     }
938 
939     bs_file = bdrv_open(filename, NULL, NULL,
940                         BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
941     if (!bs_file) {
942         ret = -EIO;
943         goto done;
944     }
945 
946     qdict_put_str(qdict, "driver", "vdi");
947     qdict_put_str(qdict, "file", bs_file->node_name);
948     if (is_static) {
949         qdict_put_str(qdict, "preallocation", "metadata");
950     }
951 
952     /* Get the QAPI object */
953     v = qobject_input_visitor_new_flat_confused(qdict, errp);
954     if (!v) {
955         ret = -EINVAL;
956         goto done;
957     }
958     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
959     visit_free(v);
960 
961     if (local_err) {
962         error_propagate(errp, local_err);
963         ret = -EINVAL;
964         goto done;
965     }
966 
967     /* Silently round up size */
968     assert(create_options->driver == BLOCKDEV_DRIVER_VDI);
969     create_options->u.vdi.size = ROUND_UP(create_options->u.vdi.size,
970                                           BDRV_SECTOR_SIZE);
971 
972     /* Create the vdi image (format layer) */
973     ret = vdi_co_do_create(create_options, block_size, errp);
974 done:
975     qobject_unref(qdict);
976     qapi_free_BlockdevCreateOptions(create_options);
977     bdrv_unref(bs_file);
978     return ret;
979 }
980 
981 static void vdi_close(BlockDriverState *bs)
982 {
983     BDRVVdiState *s = bs->opaque;
984 
985     qemu_vfree(s->bmap);
986 
987     migrate_del_blocker(s->migration_blocker);
988     error_free(s->migration_blocker);
989 }
990 
991 static QemuOptsList vdi_create_opts = {
992     .name = "vdi-create-opts",
993     .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head),
994     .desc = {
995         {
996             .name = BLOCK_OPT_SIZE,
997             .type = QEMU_OPT_SIZE,
998             .help = "Virtual disk size"
999         },
1000 #if defined(CONFIG_VDI_BLOCK_SIZE)
1001         {
1002             .name = BLOCK_OPT_CLUSTER_SIZE,
1003             .type = QEMU_OPT_SIZE,
1004             .help = "VDI cluster (block) size",
1005             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
1006         },
1007 #endif
1008 #if defined(CONFIG_VDI_STATIC_IMAGE)
1009         {
1010             .name = BLOCK_OPT_STATIC,
1011             .type = QEMU_OPT_BOOL,
1012             .help = "VDI static (pre-allocated) image",
1013             .def_value_str = "off"
1014         },
1015 #endif
1016         /* TODO: An additional option to set UUID values might be useful. */
1017         { /* end of list */ }
1018     }
1019 };
1020 
1021 static BlockDriver bdrv_vdi = {
1022     .format_name = "vdi",
1023     .instance_size = sizeof(BDRVVdiState),
1024     .bdrv_probe = vdi_probe,
1025     .bdrv_open = vdi_open,
1026     .bdrv_close = vdi_close,
1027     .bdrv_reopen_prepare = vdi_reopen_prepare,
1028     .bdrv_child_perm          = bdrv_format_default_perms,
1029     .bdrv_co_create      = vdi_co_create,
1030     .bdrv_co_create_opts = vdi_co_create_opts,
1031     .bdrv_has_zero_init = bdrv_has_zero_init_1,
1032     .bdrv_co_block_status = vdi_co_block_status,
1033     .bdrv_make_empty = vdi_make_empty,
1034 
1035     .bdrv_co_preadv     = vdi_co_preadv,
1036 #if defined(CONFIG_VDI_WRITE)
1037     .bdrv_co_pwritev    = vdi_co_pwritev,
1038 #endif
1039 
1040     .bdrv_get_info = vdi_get_info,
1041 
1042     .create_opts = &vdi_create_opts,
1043     .bdrv_co_check = vdi_co_check,
1044 };
1045 
1046 static void bdrv_vdi_init(void)
1047 {
1048     logout("\n");
1049     bdrv_register(&bdrv_vdi);
1050 }
1051 
1052 block_init(bdrv_vdi_init);
1053