xref: /openbmc/qemu/block/vmdk.c (revision 4b4629d9)
1 /*
2  * Block driver for the VMDK format
3  *
4  * Copyright (c) 2004 Fabrice Bellard
5  * Copyright (c) 2005 Filip Navara
6  *
7  * Permission is hereby granted, free of charge, to any person obtaining a copy
8  * of this software and associated documentation files (the "Software"), to deal
9  * in the Software without restriction, including without limitation the rights
10  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11  * copies of the Software, and to permit persons to whom the Software is
12  * furnished to do so, subject to the following conditions:
13  *
14  * The above copyright notice and this permission notice shall be included in
15  * all copies or substantial portions of the Software.
16  *
17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23  * THE SOFTWARE.
24  */
25 
26 #include "qemu/osdep.h"
27 #include "qapi/error.h"
28 #include "block/block_int.h"
29 #include "sysemu/block-backend.h"
30 #include "qapi/qmp/qerror.h"
31 #include "qemu/error-report.h"
32 #include "qemu/module.h"
33 #include "migration/migration.h"
34 #include "qemu/cutils.h"
35 #include <zlib.h>
36 #include <glib.h>
37 
38 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
39 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
40 #define VMDK4_COMPRESSION_DEFLATE 1
41 #define VMDK4_FLAG_NL_DETECT (1 << 0)
42 #define VMDK4_FLAG_RGD (1 << 1)
43 /* Zeroed-grain enable bit */
44 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
45 #define VMDK4_FLAG_COMPRESS (1 << 16)
46 #define VMDK4_FLAG_MARKER (1 << 17)
47 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
48 
49 #define VMDK_GTE_ZEROED 0x1
50 
51 /* VMDK internal error codes */
52 #define VMDK_OK      0
53 #define VMDK_ERROR   (-1)
54 /* Cluster not allocated */
55 #define VMDK_UNALLOC (-2)
56 #define VMDK_ZEROED  (-3)
57 
58 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
59 
60 typedef struct {
61     uint32_t version;
62     uint32_t flags;
63     uint32_t disk_sectors;
64     uint32_t granularity;
65     uint32_t l1dir_offset;
66     uint32_t l1dir_size;
67     uint32_t file_sectors;
68     uint32_t cylinders;
69     uint32_t heads;
70     uint32_t sectors_per_track;
71 } QEMU_PACKED VMDK3Header;
72 
73 typedef struct {
74     uint32_t version;
75     uint32_t flags;
76     uint64_t capacity;
77     uint64_t granularity;
78     uint64_t desc_offset;
79     uint64_t desc_size;
80     /* Number of GrainTableEntries per GrainTable */
81     uint32_t num_gtes_per_gt;
82     uint64_t rgd_offset;
83     uint64_t gd_offset;
84     uint64_t grain_offset;
85     char filler[1];
86     char check_bytes[4];
87     uint16_t compressAlgorithm;
88 } QEMU_PACKED VMDK4Header;
89 
90 #define L2_CACHE_SIZE 16
91 
92 typedef struct VmdkExtent {
93     BdrvChild *file;
94     bool flat;
95     bool compressed;
96     bool has_marker;
97     bool has_zero_grain;
98     int version;
99     int64_t sectors;
100     int64_t end_sector;
101     int64_t flat_start_offset;
102     int64_t l1_table_offset;
103     int64_t l1_backup_table_offset;
104     uint32_t *l1_table;
105     uint32_t *l1_backup_table;
106     unsigned int l1_size;
107     uint32_t l1_entry_sectors;
108 
109     unsigned int l2_size;
110     uint32_t *l2_cache;
111     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
112     uint32_t l2_cache_counts[L2_CACHE_SIZE];
113 
114     int64_t cluster_sectors;
115     int64_t next_cluster_sector;
116     char *type;
117 } VmdkExtent;
118 
119 typedef struct BDRVVmdkState {
120     CoMutex lock;
121     uint64_t desc_offset;
122     bool cid_updated;
123     bool cid_checked;
124     uint32_t cid;
125     uint32_t parent_cid;
126     int num_extents;
127     /* Extent array with num_extents entries, ascend ordered by address */
128     VmdkExtent *extents;
129     Error *migration_blocker;
130     char *create_type;
131 } BDRVVmdkState;
132 
133 typedef struct VmdkMetaData {
134     unsigned int l1_index;
135     unsigned int l2_index;
136     unsigned int l2_offset;
137     int valid;
138     uint32_t *l2_cache_entry;
139 } VmdkMetaData;
140 
141 typedef struct VmdkGrainMarker {
142     uint64_t lba;
143     uint32_t size;
144     uint8_t  data[0];
145 } QEMU_PACKED VmdkGrainMarker;
146 
147 enum {
148     MARKER_END_OF_STREAM    = 0,
149     MARKER_GRAIN_TABLE      = 1,
150     MARKER_GRAIN_DIRECTORY  = 2,
151     MARKER_FOOTER           = 3,
152 };
153 
154 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
155 {
156     uint32_t magic;
157 
158     if (buf_size < 4) {
159         return 0;
160     }
161     magic = be32_to_cpu(*(uint32_t *)buf);
162     if (magic == VMDK3_MAGIC ||
163         magic == VMDK4_MAGIC) {
164         return 100;
165     } else {
166         const char *p = (const char *)buf;
167         const char *end = p + buf_size;
168         while (p < end) {
169             if (*p == '#') {
170                 /* skip comment line */
171                 while (p < end && *p != '\n') {
172                     p++;
173                 }
174                 p++;
175                 continue;
176             }
177             if (*p == ' ') {
178                 while (p < end && *p == ' ') {
179                     p++;
180                 }
181                 /* skip '\r' if windows line endings used. */
182                 if (p < end && *p == '\r') {
183                     p++;
184                 }
185                 /* only accept blank lines before 'version=' line */
186                 if (p == end || *p != '\n') {
187                     return 0;
188                 }
189                 p++;
190                 continue;
191             }
192             if (end - p >= strlen("version=X\n")) {
193                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
194                     strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
195                     return 100;
196                 }
197             }
198             if (end - p >= strlen("version=X\r\n")) {
199                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
200                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
201                     return 100;
202                 }
203             }
204             return 0;
205         }
206         return 0;
207     }
208 }
209 
210 #define SECTOR_SIZE 512
211 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
212 #define BUF_SIZE 4096
213 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
214 
215 static void vmdk_free_extents(BlockDriverState *bs)
216 {
217     int i;
218     BDRVVmdkState *s = bs->opaque;
219     VmdkExtent *e;
220 
221     for (i = 0; i < s->num_extents; i++) {
222         e = &s->extents[i];
223         g_free(e->l1_table);
224         g_free(e->l2_cache);
225         g_free(e->l1_backup_table);
226         g_free(e->type);
227         if (e->file != bs->file) {
228             bdrv_unref_child(bs, e->file);
229         }
230     }
231     g_free(s->extents);
232 }
233 
234 static void vmdk_free_last_extent(BlockDriverState *bs)
235 {
236     BDRVVmdkState *s = bs->opaque;
237 
238     if (s->num_extents == 0) {
239         return;
240     }
241     s->num_extents--;
242     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents);
243 }
244 
245 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
246 {
247     char *desc;
248     uint32_t cid = 0xffffffff;
249     const char *p_name, *cid_str;
250     size_t cid_str_size;
251     BDRVVmdkState *s = bs->opaque;
252     int ret;
253 
254     desc = g_malloc0(DESC_SIZE);
255     ret = bdrv_pread(bs->file->bs, s->desc_offset, desc, DESC_SIZE);
256     if (ret < 0) {
257         g_free(desc);
258         return 0;
259     }
260 
261     if (parent) {
262         cid_str = "parentCID";
263         cid_str_size = sizeof("parentCID");
264     } else {
265         cid_str = "CID";
266         cid_str_size = sizeof("CID");
267     }
268 
269     desc[DESC_SIZE - 1] = '\0';
270     p_name = strstr(desc, cid_str);
271     if (p_name != NULL) {
272         p_name += cid_str_size;
273         sscanf(p_name, "%" SCNx32, &cid);
274     }
275 
276     g_free(desc);
277     return cid;
278 }
279 
280 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
281 {
282     char *desc, *tmp_desc;
283     char *p_name, *tmp_str;
284     BDRVVmdkState *s = bs->opaque;
285     int ret = 0;
286 
287     desc = g_malloc0(DESC_SIZE);
288     tmp_desc = g_malloc0(DESC_SIZE);
289     ret = bdrv_pread(bs->file->bs, s->desc_offset, desc, DESC_SIZE);
290     if (ret < 0) {
291         goto out;
292     }
293 
294     desc[DESC_SIZE - 1] = '\0';
295     tmp_str = strstr(desc, "parentCID");
296     if (tmp_str == NULL) {
297         ret = -EINVAL;
298         goto out;
299     }
300 
301     pstrcpy(tmp_desc, DESC_SIZE, tmp_str);
302     p_name = strstr(desc, "CID");
303     if (p_name != NULL) {
304         p_name += sizeof("CID");
305         snprintf(p_name, DESC_SIZE - (p_name - desc), "%" PRIx32 "\n", cid);
306         pstrcat(desc, DESC_SIZE, tmp_desc);
307     }
308 
309     ret = bdrv_pwrite_sync(bs->file->bs, s->desc_offset, desc, DESC_SIZE);
310 
311 out:
312     g_free(desc);
313     g_free(tmp_desc);
314     return ret;
315 }
316 
317 static int vmdk_is_cid_valid(BlockDriverState *bs)
318 {
319     BDRVVmdkState *s = bs->opaque;
320     uint32_t cur_pcid;
321 
322     if (!s->cid_checked && bs->backing) {
323         BlockDriverState *p_bs = bs->backing->bs;
324 
325         cur_pcid = vmdk_read_cid(p_bs, 0);
326         if (s->parent_cid != cur_pcid) {
327             /* CID not valid */
328             return 0;
329         }
330     }
331     s->cid_checked = true;
332     /* CID valid */
333     return 1;
334 }
335 
336 /* We have nothing to do for VMDK reopen, stubs just return success */
337 static int vmdk_reopen_prepare(BDRVReopenState *state,
338                                BlockReopenQueue *queue, Error **errp)
339 {
340     assert(state != NULL);
341     assert(state->bs != NULL);
342     return 0;
343 }
344 
345 static int vmdk_parent_open(BlockDriverState *bs)
346 {
347     char *p_name;
348     char *desc;
349     BDRVVmdkState *s = bs->opaque;
350     int ret;
351 
352     desc = g_malloc0(DESC_SIZE + 1);
353     ret = bdrv_pread(bs->file->bs, s->desc_offset, desc, DESC_SIZE);
354     if (ret < 0) {
355         goto out;
356     }
357     ret = 0;
358 
359     p_name = strstr(desc, "parentFileNameHint");
360     if (p_name != NULL) {
361         char *end_name;
362 
363         p_name += sizeof("parentFileNameHint") + 1;
364         end_name = strchr(p_name, '\"');
365         if (end_name == NULL) {
366             ret = -EINVAL;
367             goto out;
368         }
369         if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
370             ret = -EINVAL;
371             goto out;
372         }
373 
374         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
375     }
376 
377 out:
378     g_free(desc);
379     return ret;
380 }
381 
382 /* Create and append extent to the extent array. Return the added VmdkExtent
383  * address. return NULL if allocation failed. */
384 static int vmdk_add_extent(BlockDriverState *bs,
385                            BdrvChild *file, bool flat, int64_t sectors,
386                            int64_t l1_offset, int64_t l1_backup_offset,
387                            uint32_t l1_size,
388                            int l2_size, uint64_t cluster_sectors,
389                            VmdkExtent **new_extent,
390                            Error **errp)
391 {
392     VmdkExtent *extent;
393     BDRVVmdkState *s = bs->opaque;
394     int64_t nb_sectors;
395 
396     if (cluster_sectors > 0x200000) {
397         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
398         error_setg(errp, "Invalid granularity, image may be corrupt");
399         return -EFBIG;
400     }
401     if (l1_size > 512 * 1024 * 1024) {
402         /* Although with big capacity and small l1_entry_sectors, we can get a
403          * big l1_size, we don't want unbounded value to allocate the table.
404          * Limit it to 512M, which is 16PB for default cluster and L2 table
405          * size */
406         error_setg(errp, "L1 size too big");
407         return -EFBIG;
408     }
409 
410     nb_sectors = bdrv_nb_sectors(file->bs);
411     if (nb_sectors < 0) {
412         return nb_sectors;
413     }
414 
415     s->extents = g_renew(VmdkExtent, s->extents, s->num_extents + 1);
416     extent = &s->extents[s->num_extents];
417     s->num_extents++;
418 
419     memset(extent, 0, sizeof(VmdkExtent));
420     extent->file = file;
421     extent->flat = flat;
422     extent->sectors = sectors;
423     extent->l1_table_offset = l1_offset;
424     extent->l1_backup_table_offset = l1_backup_offset;
425     extent->l1_size = l1_size;
426     extent->l1_entry_sectors = l2_size * cluster_sectors;
427     extent->l2_size = l2_size;
428     extent->cluster_sectors = flat ? sectors : cluster_sectors;
429     extent->next_cluster_sector = ROUND_UP(nb_sectors, cluster_sectors);
430 
431     if (s->num_extents > 1) {
432         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
433     } else {
434         extent->end_sector = extent->sectors;
435     }
436     bs->total_sectors = extent->end_sector;
437     if (new_extent) {
438         *new_extent = extent;
439     }
440     return 0;
441 }
442 
443 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
444                             Error **errp)
445 {
446     int ret;
447     size_t l1_size;
448     int i;
449 
450     /* read the L1 table */
451     l1_size = extent->l1_size * sizeof(uint32_t);
452     extent->l1_table = g_try_malloc(l1_size);
453     if (l1_size && extent->l1_table == NULL) {
454         return -ENOMEM;
455     }
456 
457     ret = bdrv_pread(extent->file->bs,
458                      extent->l1_table_offset,
459                      extent->l1_table,
460                      l1_size);
461     if (ret < 0) {
462         error_setg_errno(errp, -ret,
463                          "Could not read l1 table from extent '%s'",
464                          extent->file->bs->filename);
465         goto fail_l1;
466     }
467     for (i = 0; i < extent->l1_size; i++) {
468         le32_to_cpus(&extent->l1_table[i]);
469     }
470 
471     if (extent->l1_backup_table_offset) {
472         extent->l1_backup_table = g_try_malloc(l1_size);
473         if (l1_size && extent->l1_backup_table == NULL) {
474             ret = -ENOMEM;
475             goto fail_l1;
476         }
477         ret = bdrv_pread(extent->file->bs,
478                          extent->l1_backup_table_offset,
479                          extent->l1_backup_table,
480                          l1_size);
481         if (ret < 0) {
482             error_setg_errno(errp, -ret,
483                              "Could not read l1 backup table from extent '%s'",
484                              extent->file->bs->filename);
485             goto fail_l1b;
486         }
487         for (i = 0; i < extent->l1_size; i++) {
488             le32_to_cpus(&extent->l1_backup_table[i]);
489         }
490     }
491 
492     extent->l2_cache =
493         g_new(uint32_t, extent->l2_size * L2_CACHE_SIZE);
494     return 0;
495  fail_l1b:
496     g_free(extent->l1_backup_table);
497  fail_l1:
498     g_free(extent->l1_table);
499     return ret;
500 }
501 
502 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
503                                  BdrvChild *file,
504                                  int flags, Error **errp)
505 {
506     int ret;
507     uint32_t magic;
508     VMDK3Header header;
509     VmdkExtent *extent;
510 
511     ret = bdrv_pread(file->bs, sizeof(magic), &header, sizeof(header));
512     if (ret < 0) {
513         error_setg_errno(errp, -ret,
514                          "Could not read header from file '%s'",
515                          file->bs->filename);
516         return ret;
517     }
518     ret = vmdk_add_extent(bs, file, false,
519                           le32_to_cpu(header.disk_sectors),
520                           (int64_t)le32_to_cpu(header.l1dir_offset) << 9,
521                           0,
522                           le32_to_cpu(header.l1dir_size),
523                           4096,
524                           le32_to_cpu(header.granularity),
525                           &extent,
526                           errp);
527     if (ret < 0) {
528         return ret;
529     }
530     ret = vmdk_init_tables(bs, extent, errp);
531     if (ret) {
532         /* free extent allocated by vmdk_add_extent */
533         vmdk_free_last_extent(bs);
534     }
535     return ret;
536 }
537 
538 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
539                                QDict *options, Error **errp);
540 
541 static char *vmdk_read_desc(BlockDriverState *file, uint64_t desc_offset,
542                             Error **errp)
543 {
544     int64_t size;
545     char *buf;
546     int ret;
547 
548     size = bdrv_getlength(file);
549     if (size < 0) {
550         error_setg_errno(errp, -size, "Could not access file");
551         return NULL;
552     }
553 
554     if (size < 4) {
555         /* Both descriptor file and sparse image must be much larger than 4
556          * bytes, also callers of vmdk_read_desc want to compare the first 4
557          * bytes with VMDK4_MAGIC, let's error out if less is read. */
558         error_setg(errp, "File is too small, not a valid image");
559         return NULL;
560     }
561 
562     size = MIN(size, (1 << 20) - 1);  /* avoid unbounded allocation */
563     buf = g_malloc(size + 1);
564 
565     ret = bdrv_pread(file, desc_offset, buf, size);
566     if (ret < 0) {
567         error_setg_errno(errp, -ret, "Could not read from file");
568         g_free(buf);
569         return NULL;
570     }
571     buf[ret] = 0;
572 
573     return buf;
574 }
575 
576 static int vmdk_open_vmdk4(BlockDriverState *bs,
577                            BdrvChild *file,
578                            int flags, QDict *options, Error **errp)
579 {
580     int ret;
581     uint32_t magic;
582     uint32_t l1_size, l1_entry_sectors;
583     VMDK4Header header;
584     VmdkExtent *extent;
585     BDRVVmdkState *s = bs->opaque;
586     int64_t l1_backup_offset = 0;
587     bool compressed;
588 
589     ret = bdrv_pread(file->bs, sizeof(magic), &header, sizeof(header));
590     if (ret < 0) {
591         error_setg_errno(errp, -ret,
592                          "Could not read header from file '%s'",
593                          file->bs->filename);
594         return -EINVAL;
595     }
596     if (header.capacity == 0) {
597         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
598         if (desc_offset) {
599             char *buf = vmdk_read_desc(file->bs, desc_offset << 9, errp);
600             if (!buf) {
601                 return -EINVAL;
602             }
603             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
604             g_free(buf);
605             return ret;
606         }
607     }
608 
609     if (!s->create_type) {
610         s->create_type = g_strdup("monolithicSparse");
611     }
612 
613     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
614         /*
615          * The footer takes precedence over the header, so read it in. The
616          * footer starts at offset -1024 from the end: One sector for the
617          * footer, and another one for the end-of-stream marker.
618          */
619         struct {
620             struct {
621                 uint64_t val;
622                 uint32_t size;
623                 uint32_t type;
624                 uint8_t pad[512 - 16];
625             } QEMU_PACKED footer_marker;
626 
627             uint32_t magic;
628             VMDK4Header header;
629             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
630 
631             struct {
632                 uint64_t val;
633                 uint32_t size;
634                 uint32_t type;
635                 uint8_t pad[512 - 16];
636             } QEMU_PACKED eos_marker;
637         } QEMU_PACKED footer;
638 
639         ret = bdrv_pread(file->bs,
640             bs->file->bs->total_sectors * 512 - 1536,
641             &footer, sizeof(footer));
642         if (ret < 0) {
643             error_setg_errno(errp, -ret, "Failed to read footer");
644             return ret;
645         }
646 
647         /* Some sanity checks for the footer */
648         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
649             le32_to_cpu(footer.footer_marker.size) != 0  ||
650             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
651             le64_to_cpu(footer.eos_marker.val) != 0  ||
652             le32_to_cpu(footer.eos_marker.size) != 0  ||
653             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
654         {
655             error_setg(errp, "Invalid footer");
656             return -EINVAL;
657         }
658 
659         header = footer.header;
660     }
661 
662     compressed =
663         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
664     if (le32_to_cpu(header.version) > 3) {
665         error_setg(errp, "Unsupported VMDK version %" PRIu32,
666                    le32_to_cpu(header.version));
667         return -ENOTSUP;
668     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR) &&
669                !compressed) {
670         /* VMware KB 2064959 explains that version 3 added support for
671          * persistent changed block tracking (CBT), and backup software can
672          * read it as version=1 if it doesn't care about the changed area
673          * information. So we are safe to enable read only. */
674         error_setg(errp, "VMDK version 3 must be read only");
675         return -EINVAL;
676     }
677 
678     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
679         error_setg(errp, "L2 table size too big");
680         return -EINVAL;
681     }
682 
683     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
684                         * le64_to_cpu(header.granularity);
685     if (l1_entry_sectors == 0) {
686         error_setg(errp, "L1 entry size is invalid");
687         return -EINVAL;
688     }
689     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
690                 / l1_entry_sectors;
691     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
692         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
693     }
694     if (bdrv_nb_sectors(file->bs) < le64_to_cpu(header.grain_offset)) {
695         error_setg(errp, "File truncated, expecting at least %" PRId64 " bytes",
696                    (int64_t)(le64_to_cpu(header.grain_offset)
697                              * BDRV_SECTOR_SIZE));
698         return -EINVAL;
699     }
700 
701     ret = vmdk_add_extent(bs, file, false,
702                           le64_to_cpu(header.capacity),
703                           le64_to_cpu(header.gd_offset) << 9,
704                           l1_backup_offset,
705                           l1_size,
706                           le32_to_cpu(header.num_gtes_per_gt),
707                           le64_to_cpu(header.granularity),
708                           &extent,
709                           errp);
710     if (ret < 0) {
711         return ret;
712     }
713     extent->compressed =
714         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
715     if (extent->compressed) {
716         g_free(s->create_type);
717         s->create_type = g_strdup("streamOptimized");
718     }
719     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
720     extent->version = le32_to_cpu(header.version);
721     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
722     ret = vmdk_init_tables(bs, extent, errp);
723     if (ret) {
724         /* free extent allocated by vmdk_add_extent */
725         vmdk_free_last_extent(bs);
726     }
727     return ret;
728 }
729 
730 /* find an option value out of descriptor file */
731 static int vmdk_parse_description(const char *desc, const char *opt_name,
732         char *buf, int buf_size)
733 {
734     char *opt_pos, *opt_end;
735     const char *end = desc + strlen(desc);
736 
737     opt_pos = strstr(desc, opt_name);
738     if (!opt_pos) {
739         return VMDK_ERROR;
740     }
741     /* Skip "=\"" following opt_name */
742     opt_pos += strlen(opt_name) + 2;
743     if (opt_pos >= end) {
744         return VMDK_ERROR;
745     }
746     opt_end = opt_pos;
747     while (opt_end < end && *opt_end != '"') {
748         opt_end++;
749     }
750     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
751         return VMDK_ERROR;
752     }
753     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
754     return VMDK_OK;
755 }
756 
757 /* Open an extent file and append to bs array */
758 static int vmdk_open_sparse(BlockDriverState *bs, BdrvChild *file, int flags,
759                             char *buf, QDict *options, Error **errp)
760 {
761     uint32_t magic;
762 
763     magic = ldl_be_p(buf);
764     switch (magic) {
765         case VMDK3_MAGIC:
766             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
767             break;
768         case VMDK4_MAGIC:
769             return vmdk_open_vmdk4(bs, file, flags, options, errp);
770             break;
771         default:
772             error_setg(errp, "Image not in VMDK format");
773             return -EINVAL;
774             break;
775     }
776 }
777 
778 static const char *next_line(const char *s)
779 {
780     while (*s) {
781         if (*s == '\n') {
782             return s + 1;
783         }
784         s++;
785     }
786     return s;
787 }
788 
789 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
790                               const char *desc_file_path, QDict *options,
791                               Error **errp)
792 {
793     int ret;
794     int matches;
795     char access[11];
796     char type[11];
797     char fname[512];
798     const char *p, *np;
799     int64_t sectors = 0;
800     int64_t flat_offset;
801     char *extent_path;
802     BdrvChild *extent_file;
803     BDRVVmdkState *s = bs->opaque;
804     VmdkExtent *extent;
805     char extent_opt_prefix[32];
806     Error *local_err = NULL;
807 
808     for (p = desc; *p; p = next_line(p)) {
809         /* parse extent line in one of below formats:
810          *
811          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
812          * RW [size in sectors] SPARSE "file-name.vmdk"
813          * RW [size in sectors] VMFS "file-name.vmdk"
814          * RW [size in sectors] VMFSSPARSE "file-name.vmdk"
815          */
816         flat_offset = -1;
817         matches = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
818                          access, &sectors, type, fname, &flat_offset);
819         if (matches < 4 || strcmp(access, "RW")) {
820             continue;
821         } else if (!strcmp(type, "FLAT")) {
822             if (matches != 5 || flat_offset < 0) {
823                 goto invalid;
824             }
825         } else if (!strcmp(type, "VMFS")) {
826             if (matches == 4) {
827                 flat_offset = 0;
828             } else {
829                 goto invalid;
830             }
831         } else if (matches != 4) {
832             goto invalid;
833         }
834 
835         if (sectors <= 0 ||
836             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
837              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
838             (strcmp(access, "RW"))) {
839             continue;
840         }
841 
842         if (!path_is_absolute(fname) && !path_has_protocol(fname) &&
843             !desc_file_path[0])
844         {
845             error_setg(errp, "Cannot use relative extent paths with VMDK "
846                        "descriptor file '%s'", bs->file->bs->filename);
847             return -EINVAL;
848         }
849 
850         extent_path = g_malloc0(PATH_MAX);
851         path_combine(extent_path, PATH_MAX, desc_file_path, fname);
852 
853         ret = snprintf(extent_opt_prefix, 32, "extents.%d", s->num_extents);
854         assert(ret < 32);
855 
856         extent_file = bdrv_open_child(extent_path, options, extent_opt_prefix,
857                                       bs, &child_file, false, &local_err);
858         g_free(extent_path);
859         if (local_err) {
860             error_propagate(errp, local_err);
861             return -EINVAL;
862         }
863 
864         /* save to extents array */
865         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
866             /* FLAT extent */
867 
868             ret = vmdk_add_extent(bs, extent_file, true, sectors,
869                             0, 0, 0, 0, 0, &extent, errp);
870             if (ret < 0) {
871                 bdrv_unref_child(bs, extent_file);
872                 return ret;
873             }
874             extent->flat_start_offset = flat_offset << 9;
875         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
876             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
877             char *buf = vmdk_read_desc(extent_file->bs, 0, errp);
878             if (!buf) {
879                 ret = -EINVAL;
880             } else {
881                 ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, buf,
882                                        options, errp);
883             }
884             g_free(buf);
885             if (ret) {
886                 bdrv_unref_child(bs, extent_file);
887                 return ret;
888             }
889             extent = &s->extents[s->num_extents - 1];
890         } else {
891             error_setg(errp, "Unsupported extent type '%s'", type);
892             bdrv_unref_child(bs, extent_file);
893             return -ENOTSUP;
894         }
895         extent->type = g_strdup(type);
896     }
897     return 0;
898 
899 invalid:
900     np = next_line(p);
901     assert(np != p);
902     if (np[-1] == '\n') {
903         np--;
904     }
905     error_setg(errp, "Invalid extent line: %.*s", (int)(np - p), p);
906     return -EINVAL;
907 }
908 
909 static int vmdk_open_desc_file(BlockDriverState *bs, int flags, char *buf,
910                                QDict *options, Error **errp)
911 {
912     int ret;
913     char ct[128];
914     BDRVVmdkState *s = bs->opaque;
915 
916     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
917         error_setg(errp, "invalid VMDK image descriptor");
918         ret = -EINVAL;
919         goto exit;
920     }
921     if (strcmp(ct, "monolithicFlat") &&
922         strcmp(ct, "vmfs") &&
923         strcmp(ct, "vmfsSparse") &&
924         strcmp(ct, "twoGbMaxExtentSparse") &&
925         strcmp(ct, "twoGbMaxExtentFlat")) {
926         error_setg(errp, "Unsupported image type '%s'", ct);
927         ret = -ENOTSUP;
928         goto exit;
929     }
930     s->create_type = g_strdup(ct);
931     s->desc_offset = 0;
932     ret = vmdk_parse_extents(buf, bs, bs->file->bs->exact_filename, options,
933                              errp);
934 exit:
935     return ret;
936 }
937 
938 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
939                      Error **errp)
940 {
941     char *buf;
942     int ret;
943     BDRVVmdkState *s = bs->opaque;
944     uint32_t magic;
945 
946     buf = vmdk_read_desc(bs->file->bs, 0, errp);
947     if (!buf) {
948         return -EINVAL;
949     }
950 
951     magic = ldl_be_p(buf);
952     switch (magic) {
953         case VMDK3_MAGIC:
954         case VMDK4_MAGIC:
955             ret = vmdk_open_sparse(bs, bs->file, flags, buf, options,
956                                    errp);
957             s->desc_offset = 0x200;
958             break;
959         default:
960             ret = vmdk_open_desc_file(bs, flags, buf, options, errp);
961             break;
962     }
963     if (ret) {
964         goto fail;
965     }
966 
967     /* try to open parent images, if exist */
968     ret = vmdk_parent_open(bs);
969     if (ret) {
970         goto fail;
971     }
972     s->cid = vmdk_read_cid(bs, 0);
973     s->parent_cid = vmdk_read_cid(bs, 1);
974     qemu_co_mutex_init(&s->lock);
975 
976     /* Disable migration when VMDK images are used */
977     error_setg(&s->migration_blocker, "The vmdk format used by node '%s' "
978                "does not support live migration",
979                bdrv_get_device_or_node_name(bs));
980     migrate_add_blocker(s->migration_blocker);
981     g_free(buf);
982     return 0;
983 
984 fail:
985     g_free(buf);
986     g_free(s->create_type);
987     s->create_type = NULL;
988     vmdk_free_extents(bs);
989     return ret;
990 }
991 
992 
993 static void vmdk_refresh_limits(BlockDriverState *bs, Error **errp)
994 {
995     BDRVVmdkState *s = bs->opaque;
996     int i;
997 
998     for (i = 0; i < s->num_extents; i++) {
999         if (!s->extents[i].flat) {
1000             bs->bl.write_zeroes_alignment =
1001                 MAX(bs->bl.write_zeroes_alignment,
1002                     s->extents[i].cluster_sectors);
1003         }
1004     }
1005 }
1006 
1007 /**
1008  * get_whole_cluster
1009  *
1010  * Copy backing file's cluster that covers @sector_num, otherwise write zero,
1011  * to the cluster at @cluster_sector_num.
1012  *
1013  * If @skip_start_sector < @skip_end_sector, the relative range
1014  * [@skip_start_sector, @skip_end_sector) is not copied or written, and leave
1015  * it for call to write user data in the request.
1016  */
1017 static int get_whole_cluster(BlockDriverState *bs,
1018                              VmdkExtent *extent,
1019                              uint64_t cluster_offset,
1020                              uint64_t offset,
1021                              uint64_t skip_start_bytes,
1022                              uint64_t skip_end_bytes)
1023 {
1024     int ret = VMDK_OK;
1025     int64_t cluster_bytes;
1026     uint8_t *whole_grain;
1027 
1028     /* For COW, align request sector_num to cluster start */
1029     cluster_bytes = extent->cluster_sectors << BDRV_SECTOR_BITS;
1030     offset = QEMU_ALIGN_DOWN(offset, cluster_bytes);
1031     whole_grain = qemu_blockalign(bs, cluster_bytes);
1032 
1033     if (!bs->backing) {
1034         memset(whole_grain, 0, skip_start_bytes);
1035         memset(whole_grain + skip_end_bytes, 0, cluster_bytes - skip_end_bytes);
1036     }
1037 
1038     assert(skip_end_bytes <= cluster_bytes);
1039     /* we will be here if it's first write on non-exist grain(cluster).
1040      * try to read from parent image, if exist */
1041     if (bs->backing && !vmdk_is_cid_valid(bs)) {
1042         ret = VMDK_ERROR;
1043         goto exit;
1044     }
1045 
1046     /* Read backing data before skip range */
1047     if (skip_start_bytes > 0) {
1048         if (bs->backing) {
1049             ret = bdrv_pread(bs->backing->bs, offset, whole_grain,
1050                              skip_start_bytes);
1051             if (ret < 0) {
1052                 ret = VMDK_ERROR;
1053                 goto exit;
1054             }
1055         }
1056         ret = bdrv_pwrite(extent->file->bs, cluster_offset, whole_grain,
1057                           skip_start_bytes);
1058         if (ret < 0) {
1059             ret = VMDK_ERROR;
1060             goto exit;
1061         }
1062     }
1063     /* Read backing data after skip range */
1064     if (skip_end_bytes < cluster_bytes) {
1065         if (bs->backing) {
1066             ret = bdrv_pread(bs->backing->bs, offset + skip_end_bytes,
1067                              whole_grain + skip_end_bytes,
1068                              cluster_bytes - skip_end_bytes);
1069             if (ret < 0) {
1070                 ret = VMDK_ERROR;
1071                 goto exit;
1072             }
1073         }
1074         ret = bdrv_pwrite(extent->file->bs, cluster_offset + skip_end_bytes,
1075                           whole_grain + skip_end_bytes,
1076                           cluster_bytes - skip_end_bytes);
1077         if (ret < 0) {
1078             ret = VMDK_ERROR;
1079             goto exit;
1080         }
1081     }
1082 
1083     ret = VMDK_OK;
1084 exit:
1085     qemu_vfree(whole_grain);
1086     return ret;
1087 }
1088 
1089 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data,
1090                          uint32_t offset)
1091 {
1092     offset = cpu_to_le32(offset);
1093     /* update L2 table */
1094     if (bdrv_pwrite_sync(
1095                 extent->file->bs,
1096                 ((int64_t)m_data->l2_offset * 512)
1097                     + (m_data->l2_index * sizeof(offset)),
1098                 &offset, sizeof(offset)) < 0) {
1099         return VMDK_ERROR;
1100     }
1101     /* update backup L2 table */
1102     if (extent->l1_backup_table_offset != 0) {
1103         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
1104         if (bdrv_pwrite_sync(
1105                     extent->file->bs,
1106                     ((int64_t)m_data->l2_offset * 512)
1107                         + (m_data->l2_index * sizeof(offset)),
1108                     &offset, sizeof(offset)) < 0) {
1109             return VMDK_ERROR;
1110         }
1111     }
1112     if (m_data->l2_cache_entry) {
1113         *m_data->l2_cache_entry = offset;
1114     }
1115 
1116     return VMDK_OK;
1117 }
1118 
1119 /**
1120  * get_cluster_offset
1121  *
1122  * Look up cluster offset in extent file by sector number, and store in
1123  * @cluster_offset.
1124  *
1125  * For flat extents, the start offset as parsed from the description file is
1126  * returned.
1127  *
1128  * For sparse extents, look up in L1, L2 table. If allocate is true, return an
1129  * offset for a new cluster and update L2 cache. If there is a backing file,
1130  * COW is done before returning; otherwise, zeroes are written to the allocated
1131  * cluster. Both COW and zero writing skips the sector range
1132  * [@skip_start_sector, @skip_end_sector) passed in by caller, because caller
1133  * has new data to write there.
1134  *
1135  * Returns: VMDK_OK if cluster exists and mapped in the image.
1136  *          VMDK_UNALLOC if cluster is not mapped and @allocate is false.
1137  *          VMDK_ERROR if failed.
1138  */
1139 static int get_cluster_offset(BlockDriverState *bs,
1140                               VmdkExtent *extent,
1141                               VmdkMetaData *m_data,
1142                               uint64_t offset,
1143                               bool allocate,
1144                               uint64_t *cluster_offset,
1145                               uint64_t skip_start_bytes,
1146                               uint64_t skip_end_bytes)
1147 {
1148     unsigned int l1_index, l2_offset, l2_index;
1149     int min_index, i, j;
1150     uint32_t min_count, *l2_table;
1151     bool zeroed = false;
1152     int64_t ret;
1153     int64_t cluster_sector;
1154 
1155     if (m_data) {
1156         m_data->valid = 0;
1157     }
1158     if (extent->flat) {
1159         *cluster_offset = extent->flat_start_offset;
1160         return VMDK_OK;
1161     }
1162 
1163     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
1164     l1_index = (offset >> 9) / extent->l1_entry_sectors;
1165     if (l1_index >= extent->l1_size) {
1166         return VMDK_ERROR;
1167     }
1168     l2_offset = extent->l1_table[l1_index];
1169     if (!l2_offset) {
1170         return VMDK_UNALLOC;
1171     }
1172     for (i = 0; i < L2_CACHE_SIZE; i++) {
1173         if (l2_offset == extent->l2_cache_offsets[i]) {
1174             /* increment the hit count */
1175             if (++extent->l2_cache_counts[i] == 0xffffffff) {
1176                 for (j = 0; j < L2_CACHE_SIZE; j++) {
1177                     extent->l2_cache_counts[j] >>= 1;
1178                 }
1179             }
1180             l2_table = extent->l2_cache + (i * extent->l2_size);
1181             goto found;
1182         }
1183     }
1184     /* not found: load a new entry in the least used one */
1185     min_index = 0;
1186     min_count = 0xffffffff;
1187     for (i = 0; i < L2_CACHE_SIZE; i++) {
1188         if (extent->l2_cache_counts[i] < min_count) {
1189             min_count = extent->l2_cache_counts[i];
1190             min_index = i;
1191         }
1192     }
1193     l2_table = extent->l2_cache + (min_index * extent->l2_size);
1194     if (bdrv_pread(
1195                 extent->file->bs,
1196                 (int64_t)l2_offset * 512,
1197                 l2_table,
1198                 extent->l2_size * sizeof(uint32_t)
1199             ) != extent->l2_size * sizeof(uint32_t)) {
1200         return VMDK_ERROR;
1201     }
1202 
1203     extent->l2_cache_offsets[min_index] = l2_offset;
1204     extent->l2_cache_counts[min_index] = 1;
1205  found:
1206     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1207     cluster_sector = le32_to_cpu(l2_table[l2_index]);
1208 
1209     if (m_data) {
1210         m_data->valid = 1;
1211         m_data->l1_index = l1_index;
1212         m_data->l2_index = l2_index;
1213         m_data->l2_offset = l2_offset;
1214         m_data->l2_cache_entry = &l2_table[l2_index];
1215     }
1216     if (extent->has_zero_grain && cluster_sector == VMDK_GTE_ZEROED) {
1217         zeroed = true;
1218     }
1219 
1220     if (!cluster_sector || zeroed) {
1221         if (!allocate) {
1222             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1223         }
1224 
1225         cluster_sector = extent->next_cluster_sector;
1226         extent->next_cluster_sector += extent->cluster_sectors;
1227 
1228         /* First of all we write grain itself, to avoid race condition
1229          * that may to corrupt the image.
1230          * This problem may occur because of insufficient space on host disk
1231          * or inappropriate VM shutdown.
1232          */
1233         ret = get_whole_cluster(bs, extent, cluster_sector * BDRV_SECTOR_SIZE,
1234                                 offset, skip_start_bytes, skip_end_bytes);
1235         if (ret) {
1236             return ret;
1237         }
1238     }
1239     *cluster_offset = cluster_sector << BDRV_SECTOR_BITS;
1240     return VMDK_OK;
1241 }
1242 
1243 static VmdkExtent *find_extent(BDRVVmdkState *s,
1244                                 int64_t sector_num, VmdkExtent *start_hint)
1245 {
1246     VmdkExtent *extent = start_hint;
1247 
1248     if (!extent) {
1249         extent = &s->extents[0];
1250     }
1251     while (extent < &s->extents[s->num_extents]) {
1252         if (sector_num < extent->end_sector) {
1253             return extent;
1254         }
1255         extent++;
1256     }
1257     return NULL;
1258 }
1259 
1260 static inline uint64_t vmdk_find_offset_in_cluster(VmdkExtent *extent,
1261                                                    int64_t offset)
1262 {
1263     uint64_t offset_in_cluster, extent_begin_offset, extent_relative_offset;
1264     uint64_t cluster_size = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1265 
1266     extent_begin_offset =
1267         (extent->end_sector - extent->sectors) * BDRV_SECTOR_SIZE;
1268     extent_relative_offset = offset - extent_begin_offset;
1269     offset_in_cluster = extent_relative_offset % cluster_size;
1270 
1271     return offset_in_cluster;
1272 }
1273 
1274 static inline uint64_t vmdk_find_index_in_cluster(VmdkExtent *extent,
1275                                                   int64_t sector_num)
1276 {
1277     uint64_t offset;
1278     offset = vmdk_find_offset_in_cluster(extent, sector_num * BDRV_SECTOR_SIZE);
1279     return offset / BDRV_SECTOR_SIZE;
1280 }
1281 
1282 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1283         int64_t sector_num, int nb_sectors, int *pnum, BlockDriverState **file)
1284 {
1285     BDRVVmdkState *s = bs->opaque;
1286     int64_t index_in_cluster, n, ret;
1287     uint64_t offset;
1288     VmdkExtent *extent;
1289 
1290     extent = find_extent(s, sector_num, NULL);
1291     if (!extent) {
1292         return 0;
1293     }
1294     qemu_co_mutex_lock(&s->lock);
1295     ret = get_cluster_offset(bs, extent, NULL,
1296                              sector_num * 512, false, &offset,
1297                              0, 0);
1298     qemu_co_mutex_unlock(&s->lock);
1299 
1300     index_in_cluster = vmdk_find_index_in_cluster(extent, sector_num);
1301     switch (ret) {
1302     case VMDK_ERROR:
1303         ret = -EIO;
1304         break;
1305     case VMDK_UNALLOC:
1306         ret = 0;
1307         break;
1308     case VMDK_ZEROED:
1309         ret = BDRV_BLOCK_ZERO;
1310         break;
1311     case VMDK_OK:
1312         ret = BDRV_BLOCK_DATA;
1313         if (!extent->compressed) {
1314             ret |= BDRV_BLOCK_OFFSET_VALID;
1315             ret |= (offset + (index_in_cluster << BDRV_SECTOR_BITS))
1316                     & BDRV_BLOCK_OFFSET_MASK;
1317         }
1318         *file = extent->file->bs;
1319         break;
1320     }
1321 
1322     n = extent->cluster_sectors - index_in_cluster;
1323     if (n > nb_sectors) {
1324         n = nb_sectors;
1325     }
1326     *pnum = n;
1327     return ret;
1328 }
1329 
1330 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1331                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1332                             uint64_t qiov_offset, uint64_t n_bytes,
1333                             uint64_t offset)
1334 {
1335     int ret;
1336     VmdkGrainMarker *data = NULL;
1337     uLongf buf_len;
1338     QEMUIOVector local_qiov;
1339     struct iovec iov;
1340     int64_t write_offset;
1341     int64_t write_end_sector;
1342 
1343     if (extent->compressed) {
1344         void *compressed_data;
1345 
1346         if (!extent->has_marker) {
1347             ret = -EINVAL;
1348             goto out;
1349         }
1350         buf_len = (extent->cluster_sectors << 9) * 2;
1351         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1352 
1353         compressed_data = g_malloc(n_bytes);
1354         qemu_iovec_to_buf(qiov, qiov_offset, compressed_data, n_bytes);
1355         ret = compress(data->data, &buf_len, compressed_data, n_bytes);
1356         g_free(compressed_data);
1357 
1358         if (ret != Z_OK || buf_len == 0) {
1359             ret = -EINVAL;
1360             goto out;
1361         }
1362 
1363         data->lba = offset >> BDRV_SECTOR_BITS;
1364         data->size = buf_len;
1365 
1366         n_bytes = buf_len + sizeof(VmdkGrainMarker);
1367         iov = (struct iovec) {
1368             .iov_base   = data,
1369             .iov_len    = n_bytes,
1370         };
1371         qemu_iovec_init_external(&local_qiov, &iov, 1);
1372     } else {
1373         qemu_iovec_init(&local_qiov, qiov->niov);
1374         qemu_iovec_concat(&local_qiov, qiov, qiov_offset, n_bytes);
1375     }
1376 
1377     write_offset = cluster_offset + offset_in_cluster,
1378     ret = bdrv_co_pwritev(extent->file->bs, write_offset, n_bytes,
1379                           &local_qiov, 0);
1380 
1381     write_end_sector = DIV_ROUND_UP(write_offset + n_bytes, BDRV_SECTOR_SIZE);
1382 
1383     if (extent->compressed) {
1384         extent->next_cluster_sector = write_end_sector;
1385     } else {
1386         extent->next_cluster_sector = MAX(extent->next_cluster_sector,
1387                                           write_end_sector);
1388     }
1389 
1390     if (ret < 0) {
1391         goto out;
1392     }
1393     ret = 0;
1394  out:
1395     g_free(data);
1396     if (!extent->compressed) {
1397         qemu_iovec_destroy(&local_qiov);
1398     }
1399     return ret;
1400 }
1401 
1402 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1403                             int64_t offset_in_cluster, QEMUIOVector *qiov,
1404                             int bytes)
1405 {
1406     int ret;
1407     int cluster_bytes, buf_bytes;
1408     uint8_t *cluster_buf, *compressed_data;
1409     uint8_t *uncomp_buf;
1410     uint32_t data_len;
1411     VmdkGrainMarker *marker;
1412     uLongf buf_len;
1413 
1414 
1415     if (!extent->compressed) {
1416         ret = bdrv_co_preadv(extent->file->bs,
1417                              cluster_offset + offset_in_cluster, bytes,
1418                              qiov, 0);
1419         if (ret < 0) {
1420             return ret;
1421         }
1422         return 0;
1423     }
1424     cluster_bytes = extent->cluster_sectors * 512;
1425     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1426     buf_bytes = cluster_bytes * 2;
1427     cluster_buf = g_malloc(buf_bytes);
1428     uncomp_buf = g_malloc(cluster_bytes);
1429     ret = bdrv_pread(extent->file->bs,
1430                 cluster_offset,
1431                 cluster_buf, buf_bytes);
1432     if (ret < 0) {
1433         goto out;
1434     }
1435     compressed_data = cluster_buf;
1436     buf_len = cluster_bytes;
1437     data_len = cluster_bytes;
1438     if (extent->has_marker) {
1439         marker = (VmdkGrainMarker *)cluster_buf;
1440         compressed_data = marker->data;
1441         data_len = le32_to_cpu(marker->size);
1442     }
1443     if (!data_len || data_len > buf_bytes) {
1444         ret = -EINVAL;
1445         goto out;
1446     }
1447     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1448     if (ret != Z_OK) {
1449         ret = -EINVAL;
1450         goto out;
1451 
1452     }
1453     if (offset_in_cluster < 0 ||
1454             offset_in_cluster + bytes > buf_len) {
1455         ret = -EINVAL;
1456         goto out;
1457     }
1458     qemu_iovec_from_buf(qiov, 0, uncomp_buf + offset_in_cluster, bytes);
1459     ret = 0;
1460 
1461  out:
1462     g_free(uncomp_buf);
1463     g_free(cluster_buf);
1464     return ret;
1465 }
1466 
1467 static int coroutine_fn
1468 vmdk_co_preadv(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1469                QEMUIOVector *qiov, int flags)
1470 {
1471     BDRVVmdkState *s = bs->opaque;
1472     int ret;
1473     uint64_t n_bytes, offset_in_cluster;
1474     VmdkExtent *extent = NULL;
1475     QEMUIOVector local_qiov;
1476     uint64_t cluster_offset;
1477     uint64_t bytes_done = 0;
1478 
1479     qemu_iovec_init(&local_qiov, qiov->niov);
1480     qemu_co_mutex_lock(&s->lock);
1481 
1482     while (bytes > 0) {
1483         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1484         if (!extent) {
1485             ret = -EIO;
1486             goto fail;
1487         }
1488         ret = get_cluster_offset(bs, extent, NULL,
1489                                  offset, false, &cluster_offset, 0, 0);
1490         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1491 
1492         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1493                              - offset_in_cluster);
1494 
1495         if (ret != VMDK_OK) {
1496             /* if not allocated, try to read from parent image, if exist */
1497             if (bs->backing && ret != VMDK_ZEROED) {
1498                 if (!vmdk_is_cid_valid(bs)) {
1499                     ret = -EINVAL;
1500                     goto fail;
1501                 }
1502 
1503                 qemu_iovec_reset(&local_qiov);
1504                 qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1505 
1506                 ret = bdrv_co_preadv(bs->backing->bs, offset, n_bytes,
1507                                      &local_qiov, 0);
1508                 if (ret < 0) {
1509                     goto fail;
1510                 }
1511             } else {
1512                 qemu_iovec_memset(qiov, bytes_done, 0, n_bytes);
1513             }
1514         } else {
1515             qemu_iovec_reset(&local_qiov);
1516             qemu_iovec_concat(&local_qiov, qiov, bytes_done, n_bytes);
1517 
1518             ret = vmdk_read_extent(extent, cluster_offset, offset_in_cluster,
1519                                    &local_qiov, n_bytes);
1520             if (ret) {
1521                 goto fail;
1522             }
1523         }
1524         bytes -= n_bytes;
1525         offset += n_bytes;
1526         bytes_done += n_bytes;
1527     }
1528 
1529     ret = 0;
1530 fail:
1531     qemu_co_mutex_unlock(&s->lock);
1532     qemu_iovec_destroy(&local_qiov);
1533 
1534     return ret;
1535 }
1536 
1537 /**
1538  * vmdk_write:
1539  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1540  *                if possible, otherwise return -ENOTSUP.
1541  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1542  *                with each cluster. By dry run we can find if the zero write
1543  *                is possible without modifying image data.
1544  *
1545  * Returns: error code with 0 for success.
1546  */
1547 static int vmdk_pwritev(BlockDriverState *bs, uint64_t offset,
1548                        uint64_t bytes, QEMUIOVector *qiov,
1549                        bool zeroed, bool zero_dry_run)
1550 {
1551     BDRVVmdkState *s = bs->opaque;
1552     VmdkExtent *extent = NULL;
1553     int ret;
1554     int64_t offset_in_cluster, n_bytes;
1555     uint64_t cluster_offset;
1556     uint64_t bytes_done = 0;
1557     VmdkMetaData m_data;
1558 
1559     if (DIV_ROUND_UP(offset, BDRV_SECTOR_SIZE) > bs->total_sectors) {
1560         error_report("Wrong offset: offset=0x%" PRIx64
1561                      " total_sectors=0x%" PRIx64,
1562                      offset, bs->total_sectors);
1563         return -EIO;
1564     }
1565 
1566     while (bytes > 0) {
1567         extent = find_extent(s, offset >> BDRV_SECTOR_BITS, extent);
1568         if (!extent) {
1569             return -EIO;
1570         }
1571         offset_in_cluster = vmdk_find_offset_in_cluster(extent, offset);
1572         n_bytes = MIN(bytes, extent->cluster_sectors * BDRV_SECTOR_SIZE
1573                              - offset_in_cluster);
1574 
1575         ret = get_cluster_offset(bs, extent, &m_data, offset,
1576                                  !(extent->compressed || zeroed),
1577                                  &cluster_offset, offset_in_cluster,
1578                                  offset_in_cluster + n_bytes);
1579         if (extent->compressed) {
1580             if (ret == VMDK_OK) {
1581                 /* Refuse write to allocated cluster for streamOptimized */
1582                 error_report("Could not write to allocated cluster"
1583                               " for streamOptimized");
1584                 return -EIO;
1585             } else {
1586                 /* allocate */
1587                 ret = get_cluster_offset(bs, extent, &m_data, offset,
1588                                          true, &cluster_offset, 0, 0);
1589             }
1590         }
1591         if (ret == VMDK_ERROR) {
1592             return -EINVAL;
1593         }
1594         if (zeroed) {
1595             /* Do zeroed write, buf is ignored */
1596             if (extent->has_zero_grain &&
1597                     offset_in_cluster == 0 &&
1598                     n_bytes >= extent->cluster_sectors * BDRV_SECTOR_SIZE) {
1599                 n_bytes = extent->cluster_sectors * BDRV_SECTOR_SIZE;
1600                 if (!zero_dry_run) {
1601                     /* update L2 tables */
1602                     if (vmdk_L2update(extent, &m_data, VMDK_GTE_ZEROED)
1603                             != VMDK_OK) {
1604                         return -EIO;
1605                     }
1606                 }
1607             } else {
1608                 return -ENOTSUP;
1609             }
1610         } else {
1611             ret = vmdk_write_extent(extent, cluster_offset, offset_in_cluster,
1612                                     qiov, bytes_done, n_bytes, offset);
1613             if (ret) {
1614                 return ret;
1615             }
1616             if (m_data.valid) {
1617                 /* update L2 tables */
1618                 if (vmdk_L2update(extent, &m_data,
1619                                   cluster_offset >> BDRV_SECTOR_BITS)
1620                         != VMDK_OK) {
1621                     return -EIO;
1622                 }
1623             }
1624         }
1625         bytes -= n_bytes;
1626         offset += n_bytes;
1627         bytes_done += n_bytes;
1628 
1629         /* update CID on the first write every time the virtual disk is
1630          * opened */
1631         if (!s->cid_updated) {
1632             ret = vmdk_write_cid(bs, g_random_int());
1633             if (ret < 0) {
1634                 return ret;
1635             }
1636             s->cid_updated = true;
1637         }
1638     }
1639     return 0;
1640 }
1641 
1642 static int coroutine_fn
1643 vmdk_co_pwritev(BlockDriverState *bs, uint64_t offset, uint64_t bytes,
1644                 QEMUIOVector *qiov, int flags)
1645 {
1646     int ret;
1647     BDRVVmdkState *s = bs->opaque;
1648     qemu_co_mutex_lock(&s->lock);
1649     ret = vmdk_pwritev(bs, offset, bytes, qiov, false, false);
1650     qemu_co_mutex_unlock(&s->lock);
1651     return ret;
1652 }
1653 
1654 typedef struct VmdkWriteCompressedCo {
1655     BlockDriverState *bs;
1656     int64_t sector_num;
1657     const uint8_t *buf;
1658     int nb_sectors;
1659     int ret;
1660 } VmdkWriteCompressedCo;
1661 
1662 static void vmdk_co_write_compressed(void *opaque)
1663 {
1664     VmdkWriteCompressedCo *co = opaque;
1665     QEMUIOVector local_qiov;
1666     uint64_t offset = co->sector_num * BDRV_SECTOR_SIZE;
1667     uint64_t bytes = co->nb_sectors * BDRV_SECTOR_SIZE;
1668 
1669     struct iovec iov = (struct iovec) {
1670         .iov_base   = (uint8_t*) co->buf,
1671         .iov_len    = bytes,
1672     };
1673     qemu_iovec_init_external(&local_qiov, &iov, 1);
1674 
1675     co->ret = vmdk_pwritev(co->bs, offset, bytes, &local_qiov, false, false);
1676 }
1677 
1678 static int vmdk_write_compressed(BlockDriverState *bs,
1679                                  int64_t sector_num,
1680                                  const uint8_t *buf,
1681                                  int nb_sectors)
1682 {
1683     BDRVVmdkState *s = bs->opaque;
1684 
1685     if (s->num_extents == 1 && s->extents[0].compressed) {
1686         Coroutine *co;
1687         AioContext *aio_context = bdrv_get_aio_context(bs);
1688         VmdkWriteCompressedCo data = {
1689             .bs         = bs,
1690             .sector_num = sector_num,
1691             .buf        = buf,
1692             .nb_sectors = nb_sectors,
1693             .ret        = -EINPROGRESS,
1694         };
1695         co = qemu_coroutine_create(vmdk_co_write_compressed);
1696         qemu_coroutine_enter(co, &data);
1697         while (data.ret == -EINPROGRESS) {
1698             aio_poll(aio_context, true);
1699         }
1700         return data.ret;
1701     } else {
1702         return -ENOTSUP;
1703     }
1704 }
1705 
1706 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs,
1707                                              int64_t sector_num,
1708                                              int nb_sectors,
1709                                              BdrvRequestFlags flags)
1710 {
1711     int ret;
1712     BDRVVmdkState *s = bs->opaque;
1713     uint64_t offset = sector_num * BDRV_SECTOR_SIZE;
1714     uint64_t bytes = nb_sectors * BDRV_SECTOR_SIZE;
1715 
1716     qemu_co_mutex_lock(&s->lock);
1717     /* write zeroes could fail if sectors not aligned to cluster, test it with
1718      * dry_run == true before really updating image */
1719     ret = vmdk_pwritev(bs, offset, bytes, NULL, true, true);
1720     if (!ret) {
1721         ret = vmdk_pwritev(bs, offset, bytes, NULL, true, false);
1722     }
1723     qemu_co_mutex_unlock(&s->lock);
1724     return ret;
1725 }
1726 
1727 static int vmdk_create_extent(const char *filename, int64_t filesize,
1728                               bool flat, bool compress, bool zeroed_grain,
1729                               QemuOpts *opts, Error **errp)
1730 {
1731     int ret, i;
1732     BlockBackend *blk = NULL;
1733     VMDK4Header header;
1734     Error *local_err = NULL;
1735     uint32_t tmp, magic, grains, gd_sectors, gt_size, gt_count;
1736     uint32_t *gd_buf = NULL;
1737     int gd_buf_size;
1738 
1739     ret = bdrv_create_file(filename, opts, &local_err);
1740     if (ret < 0) {
1741         error_propagate(errp, local_err);
1742         goto exit;
1743     }
1744 
1745     blk = blk_new_open(filename, NULL, NULL,
1746                        BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
1747     if (blk == NULL) {
1748         error_propagate(errp, local_err);
1749         ret = -EIO;
1750         goto exit;
1751     }
1752 
1753     blk_set_allow_write_beyond_eof(blk, true);
1754 
1755     if (flat) {
1756         ret = blk_truncate(blk, filesize);
1757         if (ret < 0) {
1758             error_setg_errno(errp, -ret, "Could not truncate file");
1759         }
1760         goto exit;
1761     }
1762     magic = cpu_to_be32(VMDK4_MAGIC);
1763     memset(&header, 0, sizeof(header));
1764     if (compress) {
1765         header.version = 3;
1766     } else if (zeroed_grain) {
1767         header.version = 2;
1768     } else {
1769         header.version = 1;
1770     }
1771     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1772                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1773                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1774     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1775     header.capacity = filesize / BDRV_SECTOR_SIZE;
1776     header.granularity = 128;
1777     header.num_gtes_per_gt = BDRV_SECTOR_SIZE;
1778 
1779     grains = DIV_ROUND_UP(filesize / BDRV_SECTOR_SIZE, header.granularity);
1780     gt_size = DIV_ROUND_UP(header.num_gtes_per_gt * sizeof(uint32_t),
1781                            BDRV_SECTOR_SIZE);
1782     gt_count = DIV_ROUND_UP(grains, header.num_gtes_per_gt);
1783     gd_sectors = DIV_ROUND_UP(gt_count * sizeof(uint32_t), BDRV_SECTOR_SIZE);
1784 
1785     header.desc_offset = 1;
1786     header.desc_size = 20;
1787     header.rgd_offset = header.desc_offset + header.desc_size;
1788     header.gd_offset = header.rgd_offset + gd_sectors + (gt_size * gt_count);
1789     header.grain_offset =
1790         ROUND_UP(header.gd_offset + gd_sectors + (gt_size * gt_count),
1791                  header.granularity);
1792     /* swap endianness for all header fields */
1793     header.version = cpu_to_le32(header.version);
1794     header.flags = cpu_to_le32(header.flags);
1795     header.capacity = cpu_to_le64(header.capacity);
1796     header.granularity = cpu_to_le64(header.granularity);
1797     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1798     header.desc_offset = cpu_to_le64(header.desc_offset);
1799     header.desc_size = cpu_to_le64(header.desc_size);
1800     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1801     header.gd_offset = cpu_to_le64(header.gd_offset);
1802     header.grain_offset = cpu_to_le64(header.grain_offset);
1803     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1804 
1805     header.check_bytes[0] = 0xa;
1806     header.check_bytes[1] = 0x20;
1807     header.check_bytes[2] = 0xd;
1808     header.check_bytes[3] = 0xa;
1809 
1810     /* write all the data */
1811     ret = blk_pwrite(blk, 0, &magic, sizeof(magic), 0);
1812     if (ret < 0) {
1813         error_setg(errp, QERR_IO_ERROR);
1814         goto exit;
1815     }
1816     ret = blk_pwrite(blk, sizeof(magic), &header, sizeof(header), 0);
1817     if (ret < 0) {
1818         error_setg(errp, QERR_IO_ERROR);
1819         goto exit;
1820     }
1821 
1822     ret = blk_truncate(blk, le64_to_cpu(header.grain_offset) << 9);
1823     if (ret < 0) {
1824         error_setg_errno(errp, -ret, "Could not truncate file");
1825         goto exit;
1826     }
1827 
1828     /* write grain directory */
1829     gd_buf_size = gd_sectors * BDRV_SECTOR_SIZE;
1830     gd_buf = g_malloc0(gd_buf_size);
1831     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_sectors;
1832          i < gt_count; i++, tmp += gt_size) {
1833         gd_buf[i] = cpu_to_le32(tmp);
1834     }
1835     ret = blk_pwrite(blk, le64_to_cpu(header.rgd_offset) * BDRV_SECTOR_SIZE,
1836                      gd_buf, gd_buf_size, 0);
1837     if (ret < 0) {
1838         error_setg(errp, QERR_IO_ERROR);
1839         goto exit;
1840     }
1841 
1842     /* write backup grain directory */
1843     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_sectors;
1844          i < gt_count; i++, tmp += gt_size) {
1845         gd_buf[i] = cpu_to_le32(tmp);
1846     }
1847     ret = blk_pwrite(blk, le64_to_cpu(header.gd_offset) * BDRV_SECTOR_SIZE,
1848                      gd_buf, gd_buf_size, 0);
1849     if (ret < 0) {
1850         error_setg(errp, QERR_IO_ERROR);
1851         goto exit;
1852     }
1853 
1854     ret = 0;
1855 exit:
1856     if (blk) {
1857         blk_unref(blk);
1858     }
1859     g_free(gd_buf);
1860     return ret;
1861 }
1862 
1863 static int filename_decompose(const char *filename, char *path, char *prefix,
1864                               char *postfix, size_t buf_len, Error **errp)
1865 {
1866     const char *p, *q;
1867 
1868     if (filename == NULL || !strlen(filename)) {
1869         error_setg(errp, "No filename provided");
1870         return VMDK_ERROR;
1871     }
1872     p = strrchr(filename, '/');
1873     if (p == NULL) {
1874         p = strrchr(filename, '\\');
1875     }
1876     if (p == NULL) {
1877         p = strrchr(filename, ':');
1878     }
1879     if (p != NULL) {
1880         p++;
1881         if (p - filename >= buf_len) {
1882             return VMDK_ERROR;
1883         }
1884         pstrcpy(path, p - filename + 1, filename);
1885     } else {
1886         p = filename;
1887         path[0] = '\0';
1888     }
1889     q = strrchr(p, '.');
1890     if (q == NULL) {
1891         pstrcpy(prefix, buf_len, p);
1892         postfix[0] = '\0';
1893     } else {
1894         if (q - p >= buf_len) {
1895             return VMDK_ERROR;
1896         }
1897         pstrcpy(prefix, q - p + 1, p);
1898         pstrcpy(postfix, buf_len, q);
1899     }
1900     return VMDK_OK;
1901 }
1902 
1903 static int vmdk_create(const char *filename, QemuOpts *opts, Error **errp)
1904 {
1905     int idx = 0;
1906     BlockBackend *new_blk = NULL;
1907     Error *local_err = NULL;
1908     char *desc = NULL;
1909     int64_t total_size = 0, filesize;
1910     char *adapter_type = NULL;
1911     char *backing_file = NULL;
1912     char *hw_version = NULL;
1913     char *fmt = NULL;
1914     int ret = 0;
1915     bool flat, split, compress;
1916     GString *ext_desc_lines;
1917     char *path = g_malloc0(PATH_MAX);
1918     char *prefix = g_malloc0(PATH_MAX);
1919     char *postfix = g_malloc0(PATH_MAX);
1920     char *desc_line = g_malloc0(BUF_SIZE);
1921     char *ext_filename = g_malloc0(PATH_MAX);
1922     char *desc_filename = g_malloc0(PATH_MAX);
1923     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1924     const char *desc_extent_line;
1925     char *parent_desc_line = g_malloc0(BUF_SIZE);
1926     uint32_t parent_cid = 0xffffffff;
1927     uint32_t number_heads = 16;
1928     bool zeroed_grain = false;
1929     uint32_t desc_offset = 0, desc_len;
1930     const char desc_template[] =
1931         "# Disk DescriptorFile\n"
1932         "version=1\n"
1933         "CID=%" PRIx32 "\n"
1934         "parentCID=%" PRIx32 "\n"
1935         "createType=\"%s\"\n"
1936         "%s"
1937         "\n"
1938         "# Extent description\n"
1939         "%s"
1940         "\n"
1941         "# The Disk Data Base\n"
1942         "#DDB\n"
1943         "\n"
1944         "ddb.virtualHWVersion = \"%s\"\n"
1945         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1946         "ddb.geometry.heads = \"%" PRIu32 "\"\n"
1947         "ddb.geometry.sectors = \"63\"\n"
1948         "ddb.adapterType = \"%s\"\n";
1949 
1950     ext_desc_lines = g_string_new(NULL);
1951 
1952     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1953         ret = -EINVAL;
1954         goto exit;
1955     }
1956     /* Read out options */
1957     total_size = ROUND_UP(qemu_opt_get_size_del(opts, BLOCK_OPT_SIZE, 0),
1958                           BDRV_SECTOR_SIZE);
1959     adapter_type = qemu_opt_get_del(opts, BLOCK_OPT_ADAPTER_TYPE);
1960     backing_file = qemu_opt_get_del(opts, BLOCK_OPT_BACKING_FILE);
1961     hw_version = qemu_opt_get_del(opts, BLOCK_OPT_HWVERSION);
1962     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_COMPAT6, false)) {
1963         if (strcmp(hw_version, "undefined")) {
1964             error_setg(errp,
1965                        "compat6 cannot be enabled with hwversion set");
1966             ret = -EINVAL;
1967             goto exit;
1968         }
1969         g_free(hw_version);
1970         hw_version = g_strdup("6");
1971     }
1972     if (strcmp(hw_version, "undefined") == 0) {
1973         g_free(hw_version);
1974         hw_version = g_strdup("4");
1975     }
1976     fmt = qemu_opt_get_del(opts, BLOCK_OPT_SUBFMT);
1977     if (qemu_opt_get_bool_del(opts, BLOCK_OPT_ZEROED_GRAIN, false)) {
1978         zeroed_grain = true;
1979     }
1980 
1981     if (!adapter_type) {
1982         adapter_type = g_strdup("ide");
1983     } else if (strcmp(adapter_type, "ide") &&
1984                strcmp(adapter_type, "buslogic") &&
1985                strcmp(adapter_type, "lsilogic") &&
1986                strcmp(adapter_type, "legacyESX")) {
1987         error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1988         ret = -EINVAL;
1989         goto exit;
1990     }
1991     if (strcmp(adapter_type, "ide") != 0) {
1992         /* that's the number of heads with which vmware operates when
1993            creating, exporting, etc. vmdk files with a non-ide adapter type */
1994         number_heads = 255;
1995     }
1996     if (!fmt) {
1997         /* Default format to monolithicSparse */
1998         fmt = g_strdup("monolithicSparse");
1999     } else if (strcmp(fmt, "monolithicFlat") &&
2000                strcmp(fmt, "monolithicSparse") &&
2001                strcmp(fmt, "twoGbMaxExtentSparse") &&
2002                strcmp(fmt, "twoGbMaxExtentFlat") &&
2003                strcmp(fmt, "streamOptimized")) {
2004         error_setg(errp, "Unknown subformat: '%s'", fmt);
2005         ret = -EINVAL;
2006         goto exit;
2007     }
2008     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
2009               strcmp(fmt, "twoGbMaxExtentSparse"));
2010     flat = !(strcmp(fmt, "monolithicFlat") &&
2011              strcmp(fmt, "twoGbMaxExtentFlat"));
2012     compress = !strcmp(fmt, "streamOptimized");
2013     if (flat) {
2014         desc_extent_line = "RW %" PRId64 " FLAT \"%s\" 0\n";
2015     } else {
2016         desc_extent_line = "RW %" PRId64 " SPARSE \"%s\"\n";
2017     }
2018     if (flat && backing_file) {
2019         error_setg(errp, "Flat image can't have backing file");
2020         ret = -ENOTSUP;
2021         goto exit;
2022     }
2023     if (flat && zeroed_grain) {
2024         error_setg(errp, "Flat image can't enable zeroed grain");
2025         ret = -ENOTSUP;
2026         goto exit;
2027     }
2028     if (backing_file) {
2029         BlockBackend *blk;
2030         char *full_backing = g_new0(char, PATH_MAX);
2031         bdrv_get_full_backing_filename_from_filename(filename, backing_file,
2032                                                      full_backing, PATH_MAX,
2033                                                      &local_err);
2034         if (local_err) {
2035             g_free(full_backing);
2036             error_propagate(errp, local_err);
2037             ret = -ENOENT;
2038             goto exit;
2039         }
2040 
2041         blk = blk_new_open(full_backing, NULL, NULL,
2042                            BDRV_O_NO_BACKING, errp);
2043         g_free(full_backing);
2044         if (blk == NULL) {
2045             ret = -EIO;
2046             goto exit;
2047         }
2048         if (strcmp(blk_bs(blk)->drv->format_name, "vmdk")) {
2049             blk_unref(blk);
2050             ret = -EINVAL;
2051             goto exit;
2052         }
2053         parent_cid = vmdk_read_cid(blk_bs(blk), 0);
2054         blk_unref(blk);
2055         snprintf(parent_desc_line, BUF_SIZE,
2056                 "parentFileNameHint=\"%s\"", backing_file);
2057     }
2058 
2059     /* Create extents */
2060     filesize = total_size;
2061     while (filesize > 0) {
2062         int64_t size = filesize;
2063 
2064         if (split && size > split_size) {
2065             size = split_size;
2066         }
2067         if (split) {
2068             snprintf(desc_filename, PATH_MAX, "%s-%c%03d%s",
2069                     prefix, flat ? 'f' : 's', ++idx, postfix);
2070         } else if (flat) {
2071             snprintf(desc_filename, PATH_MAX, "%s-flat%s", prefix, postfix);
2072         } else {
2073             snprintf(desc_filename, PATH_MAX, "%s%s", prefix, postfix);
2074         }
2075         snprintf(ext_filename, PATH_MAX, "%s%s", path, desc_filename);
2076 
2077         if (vmdk_create_extent(ext_filename, size,
2078                                flat, compress, zeroed_grain, opts, errp)) {
2079             ret = -EINVAL;
2080             goto exit;
2081         }
2082         filesize -= size;
2083 
2084         /* Format description line */
2085         snprintf(desc_line, BUF_SIZE,
2086                     desc_extent_line, size / BDRV_SECTOR_SIZE, desc_filename);
2087         g_string_append(ext_desc_lines, desc_line);
2088     }
2089     /* generate descriptor file */
2090     desc = g_strdup_printf(desc_template,
2091                            g_random_int(),
2092                            parent_cid,
2093                            fmt,
2094                            parent_desc_line,
2095                            ext_desc_lines->str,
2096                            hw_version,
2097                            total_size /
2098                                (int64_t)(63 * number_heads * BDRV_SECTOR_SIZE),
2099                            number_heads,
2100                            adapter_type);
2101     desc_len = strlen(desc);
2102     /* the descriptor offset = 0x200 */
2103     if (!split && !flat) {
2104         desc_offset = 0x200;
2105     } else {
2106         ret = bdrv_create_file(filename, opts, &local_err);
2107         if (ret < 0) {
2108             error_propagate(errp, local_err);
2109             goto exit;
2110         }
2111     }
2112 
2113     new_blk = blk_new_open(filename, NULL, NULL,
2114                            BDRV_O_RDWR | BDRV_O_PROTOCOL, &local_err);
2115     if (new_blk == NULL) {
2116         error_propagate(errp, local_err);
2117         ret = -EIO;
2118         goto exit;
2119     }
2120 
2121     blk_set_allow_write_beyond_eof(new_blk, true);
2122 
2123     ret = blk_pwrite(new_blk, desc_offset, desc, desc_len, 0);
2124     if (ret < 0) {
2125         error_setg_errno(errp, -ret, "Could not write description");
2126         goto exit;
2127     }
2128     /* bdrv_pwrite write padding zeros to align to sector, we don't need that
2129      * for description file */
2130     if (desc_offset == 0) {
2131         ret = blk_truncate(new_blk, desc_len);
2132         if (ret < 0) {
2133             error_setg_errno(errp, -ret, "Could not truncate file");
2134         }
2135     }
2136 exit:
2137     if (new_blk) {
2138         blk_unref(new_blk);
2139     }
2140     g_free(adapter_type);
2141     g_free(backing_file);
2142     g_free(hw_version);
2143     g_free(fmt);
2144     g_free(desc);
2145     g_free(path);
2146     g_free(prefix);
2147     g_free(postfix);
2148     g_free(desc_line);
2149     g_free(ext_filename);
2150     g_free(desc_filename);
2151     g_free(parent_desc_line);
2152     g_string_free(ext_desc_lines, true);
2153     return ret;
2154 }
2155 
2156 static void vmdk_close(BlockDriverState *bs)
2157 {
2158     BDRVVmdkState *s = bs->opaque;
2159 
2160     vmdk_free_extents(bs);
2161     g_free(s->create_type);
2162 
2163     migrate_del_blocker(s->migration_blocker);
2164     error_free(s->migration_blocker);
2165 }
2166 
2167 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
2168 {
2169     BDRVVmdkState *s = bs->opaque;
2170     int i, err;
2171     int ret = 0;
2172 
2173     for (i = 0; i < s->num_extents; i++) {
2174         err = bdrv_co_flush(s->extents[i].file->bs);
2175         if (err < 0) {
2176             ret = err;
2177         }
2178     }
2179     return ret;
2180 }
2181 
2182 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
2183 {
2184     int i;
2185     int64_t ret = 0;
2186     int64_t r;
2187     BDRVVmdkState *s = bs->opaque;
2188 
2189     ret = bdrv_get_allocated_file_size(bs->file->bs);
2190     if (ret < 0) {
2191         return ret;
2192     }
2193     for (i = 0; i < s->num_extents; i++) {
2194         if (s->extents[i].file == bs->file) {
2195             continue;
2196         }
2197         r = bdrv_get_allocated_file_size(s->extents[i].file->bs);
2198         if (r < 0) {
2199             return r;
2200         }
2201         ret += r;
2202     }
2203     return ret;
2204 }
2205 
2206 static int vmdk_has_zero_init(BlockDriverState *bs)
2207 {
2208     int i;
2209     BDRVVmdkState *s = bs->opaque;
2210 
2211     /* If has a flat extent and its underlying storage doesn't have zero init,
2212      * return 0. */
2213     for (i = 0; i < s->num_extents; i++) {
2214         if (s->extents[i].flat) {
2215             if (!bdrv_has_zero_init(s->extents[i].file->bs)) {
2216                 return 0;
2217             }
2218         }
2219     }
2220     return 1;
2221 }
2222 
2223 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
2224 {
2225     ImageInfo *info = g_new0(ImageInfo, 1);
2226 
2227     *info = (ImageInfo){
2228         .filename         = g_strdup(extent->file->bs->filename),
2229         .format           = g_strdup(extent->type),
2230         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
2231         .compressed       = extent->compressed,
2232         .has_compressed   = extent->compressed,
2233         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
2234         .has_cluster_size = !extent->flat,
2235     };
2236 
2237     return info;
2238 }
2239 
2240 static int vmdk_check(BlockDriverState *bs, BdrvCheckResult *result,
2241                       BdrvCheckMode fix)
2242 {
2243     BDRVVmdkState *s = bs->opaque;
2244     VmdkExtent *extent = NULL;
2245     int64_t sector_num = 0;
2246     int64_t total_sectors = bdrv_nb_sectors(bs);
2247     int ret;
2248     uint64_t cluster_offset;
2249 
2250     if (fix) {
2251         return -ENOTSUP;
2252     }
2253 
2254     for (;;) {
2255         if (sector_num >= total_sectors) {
2256             return 0;
2257         }
2258         extent = find_extent(s, sector_num, extent);
2259         if (!extent) {
2260             fprintf(stderr,
2261                     "ERROR: could not find extent for sector %" PRId64 "\n",
2262                     sector_num);
2263             break;
2264         }
2265         ret = get_cluster_offset(bs, extent, NULL,
2266                                  sector_num << BDRV_SECTOR_BITS,
2267                                  false, &cluster_offset, 0, 0);
2268         if (ret == VMDK_ERROR) {
2269             fprintf(stderr,
2270                     "ERROR: could not get cluster_offset for sector %"
2271                     PRId64 "\n", sector_num);
2272             break;
2273         }
2274         if (ret == VMDK_OK &&
2275             cluster_offset >= bdrv_getlength(extent->file->bs))
2276         {
2277             fprintf(stderr,
2278                     "ERROR: cluster offset for sector %"
2279                     PRId64 " points after EOF\n", sector_num);
2280             break;
2281         }
2282         sector_num += extent->cluster_sectors;
2283     }
2284 
2285     result->corruptions++;
2286     return 0;
2287 }
2288 
2289 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
2290 {
2291     int i;
2292     BDRVVmdkState *s = bs->opaque;
2293     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
2294     ImageInfoList **next;
2295 
2296     *spec_info = (ImageInfoSpecific){
2297         .type = IMAGE_INFO_SPECIFIC_KIND_VMDK,
2298         .u = {
2299             .vmdk.data = g_new0(ImageInfoSpecificVmdk, 1),
2300         },
2301     };
2302 
2303     *spec_info->u.vmdk.data = (ImageInfoSpecificVmdk) {
2304         .create_type = g_strdup(s->create_type),
2305         .cid = s->cid,
2306         .parent_cid = s->parent_cid,
2307     };
2308 
2309     next = &spec_info->u.vmdk.data->extents;
2310     for (i = 0; i < s->num_extents; i++) {
2311         *next = g_new0(ImageInfoList, 1);
2312         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
2313         (*next)->next = NULL;
2314         next = &(*next)->next;
2315     }
2316 
2317     return spec_info;
2318 }
2319 
2320 static bool vmdk_extents_type_eq(const VmdkExtent *a, const VmdkExtent *b)
2321 {
2322     return a->flat == b->flat &&
2323            a->compressed == b->compressed &&
2324            (a->flat || a->cluster_sectors == b->cluster_sectors);
2325 }
2326 
2327 static int vmdk_get_info(BlockDriverState *bs, BlockDriverInfo *bdi)
2328 {
2329     int i;
2330     BDRVVmdkState *s = bs->opaque;
2331     assert(s->num_extents);
2332 
2333     /* See if we have multiple extents but they have different cases */
2334     for (i = 1; i < s->num_extents; i++) {
2335         if (!vmdk_extents_type_eq(&s->extents[0], &s->extents[i])) {
2336             return -ENOTSUP;
2337         }
2338     }
2339     bdi->needs_compressed_writes = s->extents[0].compressed;
2340     if (!s->extents[0].flat) {
2341         bdi->cluster_size = s->extents[0].cluster_sectors << BDRV_SECTOR_BITS;
2342     }
2343     return 0;
2344 }
2345 
2346 static void vmdk_detach_aio_context(BlockDriverState *bs)
2347 {
2348     BDRVVmdkState *s = bs->opaque;
2349     int i;
2350 
2351     for (i = 0; i < s->num_extents; i++) {
2352         bdrv_detach_aio_context(s->extents[i].file->bs);
2353     }
2354 }
2355 
2356 static void vmdk_attach_aio_context(BlockDriverState *bs,
2357                                     AioContext *new_context)
2358 {
2359     BDRVVmdkState *s = bs->opaque;
2360     int i;
2361 
2362     for (i = 0; i < s->num_extents; i++) {
2363         bdrv_attach_aio_context(s->extents[i].file->bs, new_context);
2364     }
2365 }
2366 
2367 static QemuOptsList vmdk_create_opts = {
2368     .name = "vmdk-create-opts",
2369     .head = QTAILQ_HEAD_INITIALIZER(vmdk_create_opts.head),
2370     .desc = {
2371         {
2372             .name = BLOCK_OPT_SIZE,
2373             .type = QEMU_OPT_SIZE,
2374             .help = "Virtual disk size"
2375         },
2376         {
2377             .name = BLOCK_OPT_ADAPTER_TYPE,
2378             .type = QEMU_OPT_STRING,
2379             .help = "Virtual adapter type, can be one of "
2380                     "ide (default), lsilogic, buslogic or legacyESX"
2381         },
2382         {
2383             .name = BLOCK_OPT_BACKING_FILE,
2384             .type = QEMU_OPT_STRING,
2385             .help = "File name of a base image"
2386         },
2387         {
2388             .name = BLOCK_OPT_COMPAT6,
2389             .type = QEMU_OPT_BOOL,
2390             .help = "VMDK version 6 image",
2391             .def_value_str = "off"
2392         },
2393         {
2394             .name = BLOCK_OPT_HWVERSION,
2395             .type = QEMU_OPT_STRING,
2396             .help = "VMDK hardware version",
2397             .def_value_str = "undefined"
2398         },
2399         {
2400             .name = BLOCK_OPT_SUBFMT,
2401             .type = QEMU_OPT_STRING,
2402             .help =
2403                 "VMDK flat extent format, can be one of "
2404                 "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
2405         },
2406         {
2407             .name = BLOCK_OPT_ZEROED_GRAIN,
2408             .type = QEMU_OPT_BOOL,
2409             .help = "Enable efficient zero writes "
2410                     "using the zeroed-grain GTE feature"
2411         },
2412         { /* end of list */ }
2413     }
2414 };
2415 
2416 static BlockDriver bdrv_vmdk = {
2417     .format_name                  = "vmdk",
2418     .instance_size                = sizeof(BDRVVmdkState),
2419     .bdrv_probe                   = vmdk_probe,
2420     .bdrv_open                    = vmdk_open,
2421     .bdrv_check                   = vmdk_check,
2422     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
2423     .bdrv_co_preadv               = vmdk_co_preadv,
2424     .bdrv_co_pwritev              = vmdk_co_pwritev,
2425     .bdrv_write_compressed        = vmdk_write_compressed,
2426     .bdrv_co_write_zeroes         = vmdk_co_write_zeroes,
2427     .bdrv_close                   = vmdk_close,
2428     .bdrv_create                  = vmdk_create,
2429     .bdrv_co_flush_to_disk        = vmdk_co_flush,
2430     .bdrv_co_get_block_status     = vmdk_co_get_block_status,
2431     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
2432     .bdrv_has_zero_init           = vmdk_has_zero_init,
2433     .bdrv_get_specific_info       = vmdk_get_specific_info,
2434     .bdrv_refresh_limits          = vmdk_refresh_limits,
2435     .bdrv_get_info                = vmdk_get_info,
2436     .bdrv_detach_aio_context      = vmdk_detach_aio_context,
2437     .bdrv_attach_aio_context      = vmdk_attach_aio_context,
2438 
2439     .supports_backing             = true,
2440     .create_opts                  = &vmdk_create_opts,
2441 };
2442 
2443 static void bdrv_vmdk_init(void)
2444 {
2445     bdrv_register(&bdrv_vmdk);
2446 }
2447 
2448 block_init(bdrv_vmdk_init);
2449