xref: /openbmc/qemu/block/vdi.c (revision 848c7092)
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 typedef struct {
175     /* The block map entries are little endian (even in memory). */
176     uint32_t *bmap;
177     /* Size of block (bytes). */
178     uint32_t block_size;
179     /* First sector of block map. */
180     uint32_t bmap_sector;
181     /* VDI header (converted to host endianness). */
182     VdiHeader header;
183 
184     CoRwlock bmap_lock;
185 
186     Error *migration_blocker;
187 } BDRVVdiState;
188 
189 static void vdi_header_to_cpu(VdiHeader *header)
190 {
191     header->signature = le32_to_cpu(header->signature);
192     header->version = le32_to_cpu(header->version);
193     header->header_size = le32_to_cpu(header->header_size);
194     header->image_type = le32_to_cpu(header->image_type);
195     header->image_flags = le32_to_cpu(header->image_flags);
196     header->offset_bmap = le32_to_cpu(header->offset_bmap);
197     header->offset_data = le32_to_cpu(header->offset_data);
198     header->cylinders = le32_to_cpu(header->cylinders);
199     header->heads = le32_to_cpu(header->heads);
200     header->sectors = le32_to_cpu(header->sectors);
201     header->sector_size = le32_to_cpu(header->sector_size);
202     header->disk_size = le64_to_cpu(header->disk_size);
203     header->block_size = le32_to_cpu(header->block_size);
204     header->block_extra = le32_to_cpu(header->block_extra);
205     header->blocks_in_image = le32_to_cpu(header->blocks_in_image);
206     header->blocks_allocated = le32_to_cpu(header->blocks_allocated);
207     header->uuid_image = qemu_uuid_bswap(header->uuid_image);
208     header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap);
209     header->uuid_link = qemu_uuid_bswap(header->uuid_link);
210     header->uuid_parent = qemu_uuid_bswap(header->uuid_parent);
211 }
212 
213 static void vdi_header_to_le(VdiHeader *header)
214 {
215     header->signature = cpu_to_le32(header->signature);
216     header->version = cpu_to_le32(header->version);
217     header->header_size = cpu_to_le32(header->header_size);
218     header->image_type = cpu_to_le32(header->image_type);
219     header->image_flags = cpu_to_le32(header->image_flags);
220     header->offset_bmap = cpu_to_le32(header->offset_bmap);
221     header->offset_data = cpu_to_le32(header->offset_data);
222     header->cylinders = cpu_to_le32(header->cylinders);
223     header->heads = cpu_to_le32(header->heads);
224     header->sectors = cpu_to_le32(header->sectors);
225     header->sector_size = cpu_to_le32(header->sector_size);
226     header->disk_size = cpu_to_le64(header->disk_size);
227     header->block_size = cpu_to_le32(header->block_size);
228     header->block_extra = cpu_to_le32(header->block_extra);
229     header->blocks_in_image = cpu_to_le32(header->blocks_in_image);
230     header->blocks_allocated = cpu_to_le32(header->blocks_allocated);
231     header->uuid_image = qemu_uuid_bswap(header->uuid_image);
232     header->uuid_last_snap = qemu_uuid_bswap(header->uuid_last_snap);
233     header->uuid_link = qemu_uuid_bswap(header->uuid_link);
234     header->uuid_parent = qemu_uuid_bswap(header->uuid_parent);
235 }
236 
237 static void vdi_header_print(VdiHeader *header)
238 {
239     char uuidstr[37];
240     QemuUUID uuid;
241     logout("text        %s", header->text);
242     logout("signature   0x%08x\n", header->signature);
243     logout("header size 0x%04x\n", header->header_size);
244     logout("image type  0x%04x\n", header->image_type);
245     logout("image flags 0x%04x\n", header->image_flags);
246     logout("description %s\n", header->description);
247     logout("offset bmap 0x%04x\n", header->offset_bmap);
248     logout("offset data 0x%04x\n", header->offset_data);
249     logout("cylinders   0x%04x\n", header->cylinders);
250     logout("heads       0x%04x\n", header->heads);
251     logout("sectors     0x%04x\n", header->sectors);
252     logout("sector size 0x%04x\n", header->sector_size);
253     logout("image size  0x%" PRIx64 " B (%" PRIu64 " MiB)\n",
254            header->disk_size, header->disk_size / MiB);
255     logout("block size  0x%04x\n", header->block_size);
256     logout("block extra 0x%04x\n", header->block_extra);
257     logout("blocks tot. 0x%04x\n", header->blocks_in_image);
258     logout("blocks all. 0x%04x\n", header->blocks_allocated);
259     uuid = header->uuid_image;
260     qemu_uuid_unparse(&uuid, uuidstr);
261     logout("uuid image  %s\n", uuidstr);
262     uuid = header->uuid_last_snap;
263     qemu_uuid_unparse(&uuid, uuidstr);
264     logout("uuid snap   %s\n", uuidstr);
265     uuid = header->uuid_link;
266     qemu_uuid_unparse(&uuid, uuidstr);
267     logout("uuid link   %s\n", uuidstr);
268     uuid = header->uuid_parent;
269     qemu_uuid_unparse(&uuid, uuidstr);
270     logout("uuid parent %s\n", uuidstr);
271 }
272 
273 static int coroutine_fn vdi_co_check(BlockDriverState *bs, BdrvCheckResult *res,
274                                      BdrvCheckMode fix)
275 {
276     /* TODO: additional checks possible. */
277     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
278     uint32_t blocks_allocated = 0;
279     uint32_t block;
280     uint32_t *bmap;
281     logout("\n");
282 
283     if (fix) {
284         return -ENOTSUP;
285     }
286 
287     bmap = g_try_new(uint32_t, s->header.blocks_in_image);
288     if (s->header.blocks_in_image && bmap == NULL) {
289         res->check_errors++;
290         return -ENOMEM;
291     }
292 
293     memset(bmap, 0xff, s->header.blocks_in_image * sizeof(uint32_t));
294 
295     /* Check block map and value of blocks_allocated. */
296     for (block = 0; block < s->header.blocks_in_image; block++) {
297         uint32_t bmap_entry = le32_to_cpu(s->bmap[block]);
298         if (VDI_IS_ALLOCATED(bmap_entry)) {
299             if (bmap_entry < s->header.blocks_in_image) {
300                 blocks_allocated++;
301                 if (!VDI_IS_ALLOCATED(bmap[bmap_entry])) {
302                     bmap[bmap_entry] = bmap_entry;
303                 } else {
304                     fprintf(stderr, "ERROR: block index %" PRIu32
305                             " also used by %" PRIu32 "\n", bmap[bmap_entry], bmap_entry);
306                     res->corruptions++;
307                 }
308             } else {
309                 fprintf(stderr, "ERROR: block index %" PRIu32
310                         " too large, is %" PRIu32 "\n", block, bmap_entry);
311                 res->corruptions++;
312             }
313         }
314     }
315     if (blocks_allocated != s->header.blocks_allocated) {
316         fprintf(stderr, "ERROR: allocated blocks mismatch, is %" PRIu32
317                ", should be %" PRIu32 "\n",
318                blocks_allocated, s->header.blocks_allocated);
319         res->corruptions++;
320     }
321 
322     g_free(bmap);
323 
324     return 0;
325 }
326 
327 static int vdi_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
328 {
329     /* TODO: vdi_get_info would be needed for machine snapshots.
330        vm_state_offset is still missing. */
331     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
332     logout("\n");
333     bdi->cluster_size = s->block_size;
334     bdi->vm_state_offset = 0;
335     bdi->unallocated_blocks_are_zero = true;
336     return 0;
337 }
338 
339 static int vdi_make_empty(BlockDriverState *bs)
340 {
341     /* TODO: missing code. */
342     logout("\n");
343     /* The return value for missing code must be 0, see block.c. */
344     return 0;
345 }
346 
347 static int vdi_probe(const uint8_t *buf, int buf_size, const char *filename)
348 {
349     const VdiHeader *header = (const VdiHeader *)buf;
350     int ret = 0;
351 
352     logout("\n");
353 
354     if (buf_size < sizeof(*header)) {
355         /* Header too small, no VDI. */
356     } else if (le32_to_cpu(header->signature) == VDI_SIGNATURE) {
357         ret = 100;
358     }
359 
360     if (ret == 0) {
361         logout("no vdi image\n");
362     } else {
363         logout("%s", header->text);
364     }
365 
366     return ret;
367 }
368 
369 static int vdi_open(BlockDriverState *bs, QDict *options, int flags,
370                     Error **errp)
371 {
372     BDRVVdiState *s = bs->opaque;
373     VdiHeader header;
374     size_t bmap_size;
375     int ret;
376     Error *local_err = NULL;
377     QemuUUID uuid_link, uuid_parent;
378 
379     bs->file = bdrv_open_child(NULL, options, "file", bs, &child_file,
380                                false, errp);
381     if (!bs->file) {
382         return -EINVAL;
383     }
384 
385     logout("\n");
386 
387     ret = bdrv_read(bs->file, 0, (uint8_t *)&header, 1);
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_read(bs->file, s->bmap_sector, (uint8_t *)s->bmap,
488                     bmap_size);
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, &local_err);
498     if (local_err) {
499         error_propagate(errp, local_err);
500         error_free(s->migration_blocker);
501         goto fail_free_bmap;
502     }
503 
504     qemu_co_rwlock_init(&s->bmap_lock);
505 
506     return 0;
507 
508  fail_free_bmap:
509     qemu_vfree(s->bmap);
510 
511  fail:
512     return ret;
513 }
514 
515 static int vdi_reopen_prepare(BDRVReopenState *state,
516                               BlockReopenQueue *queue, Error **errp)
517 {
518     return 0;
519 }
520 
521 static int coroutine_fn vdi_co_block_status(BlockDriverState *bs,
522                                             bool want_zero,
523                                             int64_t offset, int64_t bytes,
524                                             int64_t *pnum, int64_t *map,
525                                             BlockDriverState **file)
526 {
527     BDRVVdiState *s = (BDRVVdiState *)bs->opaque;
528     size_t bmap_index = offset / s->block_size;
529     size_t index_in_block = offset % s->block_size;
530     uint32_t bmap_entry = le32_to_cpu(s->bmap[bmap_index]);
531     int result;
532 
533     logout("%p, %" PRId64 ", %" PRId64 ", %p\n", bs, offset, bytes, pnum);
534     *pnum = MIN(s->block_size - index_in_block, bytes);
535     result = VDI_IS_ALLOCATED(bmap_entry);
536     if (!result) {
537         return 0;
538     }
539 
540     *map = s->header.offset_data + (uint64_t)bmap_entry * s->block_size +
541         index_in_block;
542     *file = bs->file->bs;
543     return BDRV_BLOCK_DATA | BDRV_BLOCK_OFFSET_VALID;
544 }
545 
546 static int coroutine_fn
547 vdi_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
548               QEMUIOVector *qiov, int 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, uint64_t offset, uint64_t bytes,
604                QEMUIOVector *qiov, int 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_pwrite(bs->file, data_offset, block, s->block_size);
667             qemu_co_rwlock_unlock(&s->bmap_lock);
668         } else {
669 nonallocating_write:
670             data_offset = s->header.offset_data +
671                            (uint64_t)bmap_entry * s->block_size +
672                            offset_in_block;
673             qemu_co_rwlock_unlock(&s->bmap_lock);
674 
675             qemu_iovec_reset(&local_qiov);
676             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
677 
678             ret = bdrv_co_pwritev(bs->file, data_offset, n_bytes,
679                                   &local_qiov, 0);
680         }
681 
682         bytes -= n_bytes;
683         offset += n_bytes;
684         bytes_done += n_bytes;
685 
686         logout("%u bytes written\n", n_bytes);
687     }
688 
689     qemu_iovec_destroy(&local_qiov);
690 
691     logout("finished data write\n");
692     if (ret < 0) {
693         return ret;
694     }
695 
696     if (block) {
697         /* One or more new blocks were allocated. */
698         VdiHeader *header = (VdiHeader *) block;
699         uint8_t *base;
700         uint64_t offset;
701         uint32_t n_sectors;
702 
703         logout("now writing modified header\n");
704         assert(VDI_IS_ALLOCATED(bmap_first));
705         *header = s->header;
706         vdi_header_to_le(header);
707         ret = bdrv_write(bs->file, 0, block, 1);
708         g_free(block);
709         block = NULL;
710 
711         if (ret < 0) {
712             return ret;
713         }
714 
715         logout("now writing modified block map entry %u...%u\n",
716                bmap_first, bmap_last);
717         /* Write modified sectors from block map. */
718         bmap_first /= (SECTOR_SIZE / sizeof(uint32_t));
719         bmap_last /= (SECTOR_SIZE / sizeof(uint32_t));
720         n_sectors = bmap_last - bmap_first + 1;
721         offset = s->bmap_sector + bmap_first;
722         base = ((uint8_t *)&s->bmap[0]) + bmap_first * SECTOR_SIZE;
723         logout("will write %u block map sectors starting from entry %u\n",
724                n_sectors, bmap_first);
725         ret = bdrv_write(bs->file, offset, base, n_sectors);
726     }
727 
728     return ret;
729 }
730 
731 static int coroutine_fn vdi_co_do_create(BlockdevCreateOptions *create_options,
732                                          size_t block_size, Error **errp)
733 {
734     BlockdevCreateOptionsVdi *vdi_opts;
735     int ret = 0;
736     uint64_t bytes = 0;
737     uint32_t blocks;
738     uint32_t image_type;
739     VdiHeader header;
740     size_t i;
741     size_t bmap_size;
742     int64_t offset = 0;
743     BlockDriverState *bs_file = NULL;
744     BlockBackend *blk = NULL;
745     uint32_t *bmap = NULL;
746     QemuUUID uuid;
747 
748     assert(create_options->driver == BLOCKDEV_DRIVER_VDI);
749     vdi_opts = &create_options->u.vdi;
750 
751     logout("\n");
752 
753     /* Validate options and set default values */
754     bytes = vdi_opts->size;
755 
756     if (!vdi_opts->has_preallocation) {
757         vdi_opts->preallocation = PREALLOC_MODE_OFF;
758     }
759     switch (vdi_opts->preallocation) {
760     case PREALLOC_MODE_OFF:
761         image_type = VDI_TYPE_DYNAMIC;
762         break;
763     case PREALLOC_MODE_METADATA:
764         image_type = VDI_TYPE_STATIC;
765         break;
766     default:
767         error_setg(errp, "Preallocation mode not supported for vdi");
768         return -EINVAL;
769     }
770 
771 #ifndef CONFIG_VDI_STATIC_IMAGE
772     if (image_type == VDI_TYPE_STATIC) {
773         ret = -ENOTSUP;
774         error_setg(errp, "Statically allocated images cannot be created in "
775                    "this build");
776         goto exit;
777     }
778 #endif
779 #ifndef CONFIG_VDI_BLOCK_SIZE
780     if (block_size != DEFAULT_CLUSTER_SIZE) {
781         ret = -ENOTSUP;
782         error_setg(errp,
783                    "A non-default cluster size is not supported in this build");
784         goto exit;
785     }
786 #endif
787 
788     if (bytes > VDI_DISK_SIZE_MAX) {
789         ret = -ENOTSUP;
790         error_setg(errp, "Unsupported VDI image size (size is 0x%" PRIx64
791                           ", max supported is 0x%" PRIx64 ")",
792                           bytes, VDI_DISK_SIZE_MAX);
793         goto exit;
794     }
795 
796     /* Create BlockBackend to write to the image */
797     bs_file = bdrv_open_blockdev_ref(vdi_opts->file, errp);
798     if (!bs_file) {
799         ret = -EIO;
800         goto exit;
801     }
802 
803     blk = blk_new(BLK_PERM_WRITE | BLK_PERM_RESIZE, BLK_PERM_ALL);
804     ret = blk_insert_bs(blk, bs_file, errp);
805     if (ret < 0) {
806         goto exit;
807     }
808 
809     blk_set_allow_write_beyond_eof(blk, true);
810 
811     /* We need enough blocks to store the given disk size,
812        so always round up. */
813     blocks = DIV_ROUND_UP(bytes, block_size);
814 
815     bmap_size = blocks * sizeof(uint32_t);
816     bmap_size = ROUND_UP(bmap_size, SECTOR_SIZE);
817 
818     memset(&header, 0, sizeof(header));
819     pstrcpy(header.text, sizeof(header.text), VDI_TEXT);
820     header.signature = VDI_SIGNATURE;
821     header.version = VDI_VERSION_1_1;
822     header.header_size = 0x180;
823     header.image_type = image_type;
824     header.offset_bmap = 0x200;
825     header.offset_data = 0x200 + bmap_size;
826     header.sector_size = SECTOR_SIZE;
827     header.disk_size = bytes;
828     header.block_size = block_size;
829     header.blocks_in_image = blocks;
830     if (image_type == VDI_TYPE_STATIC) {
831         header.blocks_allocated = blocks;
832     }
833     qemu_uuid_generate(&uuid);
834     header.uuid_image = uuid;
835     qemu_uuid_generate(&uuid);
836     header.uuid_last_snap = uuid;
837     /* There is no need to set header.uuid_link or header.uuid_parent here. */
838     if (VDI_DEBUG) {
839         vdi_header_print(&header);
840     }
841     vdi_header_to_le(&header);
842     ret = blk_pwrite(blk, offset, &header, sizeof(header), 0);
843     if (ret < 0) {
844         error_setg(errp, "Error writing header");
845         goto exit;
846     }
847     offset += sizeof(header);
848 
849     if (bmap_size > 0) {
850         bmap = g_try_malloc0(bmap_size);
851         if (bmap == NULL) {
852             ret = -ENOMEM;
853             error_setg(errp, "Could not allocate bmap");
854             goto exit;
855         }
856         for (i = 0; i < blocks; i++) {
857             if (image_type == VDI_TYPE_STATIC) {
858                 bmap[i] = i;
859             } else {
860                 bmap[i] = VDI_UNALLOCATED;
861             }
862         }
863         ret = blk_pwrite(blk, offset, bmap, bmap_size, 0);
864         if (ret < 0) {
865             error_setg(errp, "Error writing bmap");
866             goto exit;
867         }
868         offset += bmap_size;
869     }
870 
871     if (image_type == VDI_TYPE_STATIC) {
872         ret = blk_truncate(blk, offset + blocks * block_size,
873                            PREALLOC_MODE_OFF, errp);
874         if (ret < 0) {
875             error_prepend(errp, "Failed to statically allocate file");
876             goto exit;
877         }
878     }
879 
880     ret = 0;
881 exit:
882     blk_unref(blk);
883     bdrv_unref(bs_file);
884     g_free(bmap);
885     return ret;
886 }
887 
888 static int coroutine_fn vdi_co_create(BlockdevCreateOptions *create_options,
889                                       Error **errp)
890 {
891     return vdi_co_do_create(create_options, DEFAULT_CLUSTER_SIZE, errp);
892 }
893 
894 static int coroutine_fn vdi_co_create_opts(const char *filename, QemuOpts *opts,
895                                            Error **errp)
896 {
897     QDict *qdict = NULL;
898     BlockdevCreateOptions *create_options = NULL;
899     BlockDriverState *bs_file = NULL;
900     uint64_t block_size = DEFAULT_CLUSTER_SIZE;
901     bool is_static = false;
902     Visitor *v;
903     Error *local_err = NULL;
904     int ret;
905 
906     /* Parse options and convert legacy syntax.
907      *
908      * Since CONFIG_VDI_BLOCK_SIZE is disabled by default,
909      * cluster-size is not part of the QAPI schema; therefore we have
910      * to parse it before creating the QAPI object. */
911 #if defined(CONFIG_VDI_BLOCK_SIZE)
912     block_size = qemu_opt_get_size_del(opts,
913                                        BLOCK_OPT_CLUSTER_SIZE,
914                                        DEFAULT_CLUSTER_SIZE);
915     if (block_size < BDRV_SECTOR_SIZE || block_size > UINT32_MAX ||
916         !is_power_of_2(block_size))
917     {
918         error_setg(errp, "Invalid cluster size");
919         ret = -EINVAL;
920         goto done;
921     }
922 #endif
923     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_STATIC, false)) {
924         is_static = true;
925     }
926 
927     qdict = qemu_opts_to_qdict_filtered(opts, NULL, &vdi_create_opts, true);
928 
929     /* Create and open the file (protocol layer) */
930     ret = bdrv_create_file(filename, opts, errp);
931     if (ret < 0) {
932         goto done;
933     }
934 
935     bs_file = bdrv_open(filename, NULL, NULL,
936                         BDRV_O_RDWR | BDRV_O_RESIZE | BDRV_O_PROTOCOL, errp);
937     if (!bs_file) {
938         ret = -EIO;
939         goto done;
940     }
941 
942     qdict_put_str(qdict, "driver", "vdi");
943     qdict_put_str(qdict, "file", bs_file->node_name);
944     if (is_static) {
945         qdict_put_str(qdict, "preallocation", "metadata");
946     }
947 
948     /* Get the QAPI object */
949     v = qobject_input_visitor_new_flat_confused(qdict, errp);
950     if (!v) {
951         ret = -EINVAL;
952         goto done;
953     }
954     visit_type_BlockdevCreateOptions(v, NULL, &create_options, &local_err);
955     visit_free(v);
956 
957     if (local_err) {
958         error_propagate(errp, local_err);
959         ret = -EINVAL;
960         goto done;
961     }
962 
963     /* Silently round up size */
964     assert(create_options->driver == BLOCKDEV_DRIVER_VDI);
965     create_options->u.vdi.size = ROUND_UP(create_options->u.vdi.size,
966                                           BDRV_SECTOR_SIZE);
967 
968     /* Create the vdi image (format layer) */
969     ret = vdi_co_do_create(create_options, block_size, errp);
970 done:
971     qobject_unref(qdict);
972     qapi_free_BlockdevCreateOptions(create_options);
973     bdrv_unref(bs_file);
974     return ret;
975 }
976 
977 static void vdi_close(BlockDriverState *bs)
978 {
979     BDRVVdiState *s = bs->opaque;
980 
981     qemu_vfree(s->bmap);
982 
983     migrate_del_blocker(s->migration_blocker);
984     error_free(s->migration_blocker);
985 }
986 
987 static QemuOptsList vdi_create_opts = {
988     .name = "vdi-create-opts",
989     .head = QTAILQ_HEAD_INITIALIZER(vdi_create_opts.head),
990     .desc = {
991         {
992             .name = BLOCK_OPT_SIZE,
993             .type = QEMU_OPT_SIZE,
994             .help = "Virtual disk size"
995         },
996 #if defined(CONFIG_VDI_BLOCK_SIZE)
997         {
998             .name = BLOCK_OPT_CLUSTER_SIZE,
999             .type = QEMU_OPT_SIZE,
1000             .help = "VDI cluster (block) size",
1001             .def_value_str = stringify(DEFAULT_CLUSTER_SIZE)
1002         },
1003 #endif
1004 #if defined(CONFIG_VDI_STATIC_IMAGE)
1005         {
1006             .name = BLOCK_OPT_STATIC,
1007             .type = QEMU_OPT_BOOL,
1008             .help = "VDI static (pre-allocated) image",
1009             .def_value_str = "off"
1010         },
1011 #endif
1012         /* TODO: An additional option to set UUID values might be useful. */
1013         { /* end of list */ }
1014     }
1015 };
1016 
1017 static BlockDriver bdrv_vdi = {
1018     .format_name = "vdi",
1019     .instance_size = sizeof(BDRVVdiState),
1020     .bdrv_probe = vdi_probe,
1021     .bdrv_open = vdi_open,
1022     .bdrv_close = vdi_close,
1023     .bdrv_reopen_prepare = vdi_reopen_prepare,
1024     .bdrv_child_perm          = bdrv_format_default_perms,
1025     .bdrv_co_create      = vdi_co_create,
1026     .bdrv_co_create_opts = vdi_co_create_opts,
1027     .bdrv_has_zero_init = bdrv_has_zero_init_1,
1028     .bdrv_co_block_status = vdi_co_block_status,
1029     .bdrv_make_empty = vdi_make_empty,
1030 
1031     .bdrv_co_preadv     = vdi_co_preadv,
1032 #if defined(CONFIG_VDI_WRITE)
1033     .bdrv_co_pwritev    = vdi_co_pwritev,
1034 #endif
1035 
1036     .bdrv_get_info = vdi_get_info,
1037 
1038     .create_opts = &vdi_create_opts,
1039     .bdrv_co_check = vdi_co_check,
1040 };
1041 
1042 static void bdrv_vdi_init(void)
1043 {
1044     logout("\n");
1045     bdrv_register(&bdrv_vdi);
1046 }
1047 
1048 block_init(bdrv_vdi_init);
1049