xref: /openbmc/qemu/block/vmdk.c (revision 1b111dc1)
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-common.h"
27 #include "block/block_int.h"
28 #include "qemu/module.h"
29 #include "migration/migration.h"
30 #include <zlib.h>
31 
32 #define VMDK3_MAGIC (('C' << 24) | ('O' << 16) | ('W' << 8) | 'D')
33 #define VMDK4_MAGIC (('K' << 24) | ('D' << 16) | ('M' << 8) | 'V')
34 #define VMDK4_COMPRESSION_DEFLATE 1
35 #define VMDK4_FLAG_NL_DETECT (1 << 0)
36 #define VMDK4_FLAG_RGD (1 << 1)
37 /* Zeroed-grain enable bit */
38 #define VMDK4_FLAG_ZERO_GRAIN   (1 << 2)
39 #define VMDK4_FLAG_COMPRESS (1 << 16)
40 #define VMDK4_FLAG_MARKER (1 << 17)
41 #define VMDK4_GD_AT_END 0xffffffffffffffffULL
42 
43 #define VMDK_GTE_ZEROED 0x1
44 
45 /* VMDK internal error codes */
46 #define VMDK_OK      0
47 #define VMDK_ERROR   (-1)
48 /* Cluster not allocated */
49 #define VMDK_UNALLOC (-2)
50 #define VMDK_ZEROED  (-3)
51 
52 #define BLOCK_OPT_ZEROED_GRAIN "zeroed_grain"
53 
54 typedef struct {
55     uint32_t version;
56     uint32_t flags;
57     uint32_t disk_sectors;
58     uint32_t granularity;
59     uint32_t l1dir_offset;
60     uint32_t l1dir_size;
61     uint32_t file_sectors;
62     uint32_t cylinders;
63     uint32_t heads;
64     uint32_t sectors_per_track;
65 } QEMU_PACKED VMDK3Header;
66 
67 typedef struct {
68     uint32_t version;
69     uint32_t flags;
70     uint64_t capacity;
71     uint64_t granularity;
72     uint64_t desc_offset;
73     uint64_t desc_size;
74     /* Number of GrainTableEntries per GrainTable */
75     uint32_t num_gtes_per_gt;
76     uint64_t rgd_offset;
77     uint64_t gd_offset;
78     uint64_t grain_offset;
79     char filler[1];
80     char check_bytes[4];
81     uint16_t compressAlgorithm;
82 } QEMU_PACKED VMDK4Header;
83 
84 #define L2_CACHE_SIZE 16
85 
86 typedef struct VmdkExtent {
87     BlockDriverState *file;
88     bool flat;
89     bool compressed;
90     bool has_marker;
91     bool has_zero_grain;
92     int version;
93     int64_t sectors;
94     int64_t end_sector;
95     int64_t flat_start_offset;
96     int64_t l1_table_offset;
97     int64_t l1_backup_table_offset;
98     uint32_t *l1_table;
99     uint32_t *l1_backup_table;
100     unsigned int l1_size;
101     uint32_t l1_entry_sectors;
102 
103     unsigned int l2_size;
104     uint32_t *l2_cache;
105     uint32_t l2_cache_offsets[L2_CACHE_SIZE];
106     uint32_t l2_cache_counts[L2_CACHE_SIZE];
107 
108     int64_t cluster_sectors;
109     char *type;
110 } VmdkExtent;
111 
112 typedef struct BDRVVmdkState {
113     CoMutex lock;
114     uint64_t desc_offset;
115     bool cid_updated;
116     bool cid_checked;
117     uint32_t cid;
118     uint32_t parent_cid;
119     int num_extents;
120     /* Extent array with num_extents entries, ascend ordered by address */
121     VmdkExtent *extents;
122     Error *migration_blocker;
123     char *create_type;
124 } BDRVVmdkState;
125 
126 typedef struct VmdkMetaData {
127     uint32_t offset;
128     unsigned int l1_index;
129     unsigned int l2_index;
130     unsigned int l2_offset;
131     int valid;
132     uint32_t *l2_cache_entry;
133 } VmdkMetaData;
134 
135 typedef struct VmdkGrainMarker {
136     uint64_t lba;
137     uint32_t size;
138     uint8_t  data[0];
139 } QEMU_PACKED VmdkGrainMarker;
140 
141 enum {
142     MARKER_END_OF_STREAM    = 0,
143     MARKER_GRAIN_TABLE      = 1,
144     MARKER_GRAIN_DIRECTORY  = 2,
145     MARKER_FOOTER           = 3,
146 };
147 
148 static int vmdk_probe(const uint8_t *buf, int buf_size, const char *filename)
149 {
150     uint32_t magic;
151 
152     if (buf_size < 4) {
153         return 0;
154     }
155     magic = be32_to_cpu(*(uint32_t *)buf);
156     if (magic == VMDK3_MAGIC ||
157         magic == VMDK4_MAGIC) {
158         return 100;
159     } else {
160         const char *p = (const char *)buf;
161         const char *end = p + buf_size;
162         while (p < end) {
163             if (*p == '#') {
164                 /* skip comment line */
165                 while (p < end && *p != '\n') {
166                     p++;
167                 }
168                 p++;
169                 continue;
170             }
171             if (*p == ' ') {
172                 while (p < end && *p == ' ') {
173                     p++;
174                 }
175                 /* skip '\r' if windows line endings used. */
176                 if (p < end && *p == '\r') {
177                     p++;
178                 }
179                 /* only accept blank lines before 'version=' line */
180                 if (p == end || *p != '\n') {
181                     return 0;
182                 }
183                 p++;
184                 continue;
185             }
186             if (end - p >= strlen("version=X\n")) {
187                 if (strncmp("version=1\n", p, strlen("version=1\n")) == 0 ||
188                     strncmp("version=2\n", p, strlen("version=2\n")) == 0) {
189                     return 100;
190                 }
191             }
192             if (end - p >= strlen("version=X\r\n")) {
193                 if (strncmp("version=1\r\n", p, strlen("version=1\r\n")) == 0 ||
194                     strncmp("version=2\r\n", p, strlen("version=2\r\n")) == 0) {
195                     return 100;
196                 }
197             }
198             return 0;
199         }
200         return 0;
201     }
202 }
203 
204 #define SECTOR_SIZE 512
205 #define DESC_SIZE (20 * SECTOR_SIZE)    /* 20 sectors of 512 bytes each */
206 #define BUF_SIZE 4096
207 #define HEADER_SIZE 512                 /* first sector of 512 bytes */
208 
209 static void vmdk_free_extents(BlockDriverState *bs)
210 {
211     int i;
212     BDRVVmdkState *s = bs->opaque;
213     VmdkExtent *e;
214 
215     for (i = 0; i < s->num_extents; i++) {
216         e = &s->extents[i];
217         g_free(e->l1_table);
218         g_free(e->l2_cache);
219         g_free(e->l1_backup_table);
220         g_free(e->type);
221         if (e->file != bs->file) {
222             bdrv_unref(e->file);
223         }
224     }
225     g_free(s->extents);
226 }
227 
228 static void vmdk_free_last_extent(BlockDriverState *bs)
229 {
230     BDRVVmdkState *s = bs->opaque;
231 
232     if (s->num_extents == 0) {
233         return;
234     }
235     s->num_extents--;
236     s->extents = g_realloc(s->extents, s->num_extents * sizeof(VmdkExtent));
237 }
238 
239 static uint32_t vmdk_read_cid(BlockDriverState *bs, int parent)
240 {
241     char desc[DESC_SIZE];
242     uint32_t cid = 0xffffffff;
243     const char *p_name, *cid_str;
244     size_t cid_str_size;
245     BDRVVmdkState *s = bs->opaque;
246     int ret;
247 
248     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
249     if (ret < 0) {
250         return 0;
251     }
252 
253     if (parent) {
254         cid_str = "parentCID";
255         cid_str_size = sizeof("parentCID");
256     } else {
257         cid_str = "CID";
258         cid_str_size = sizeof("CID");
259     }
260 
261     desc[DESC_SIZE - 1] = '\0';
262     p_name = strstr(desc, cid_str);
263     if (p_name != NULL) {
264         p_name += cid_str_size;
265         sscanf(p_name, "%x", &cid);
266     }
267 
268     return cid;
269 }
270 
271 static int vmdk_write_cid(BlockDriverState *bs, uint32_t cid)
272 {
273     char desc[DESC_SIZE], tmp_desc[DESC_SIZE];
274     char *p_name, *tmp_str;
275     BDRVVmdkState *s = bs->opaque;
276     int ret;
277 
278     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
279     if (ret < 0) {
280         return ret;
281     }
282 
283     desc[DESC_SIZE - 1] = '\0';
284     tmp_str = strstr(desc, "parentCID");
285     if (tmp_str == NULL) {
286         return -EINVAL;
287     }
288 
289     pstrcpy(tmp_desc, sizeof(tmp_desc), tmp_str);
290     p_name = strstr(desc, "CID");
291     if (p_name != NULL) {
292         p_name += sizeof("CID");
293         snprintf(p_name, sizeof(desc) - (p_name - desc), "%x\n", cid);
294         pstrcat(desc, sizeof(desc), tmp_desc);
295     }
296 
297     ret = bdrv_pwrite_sync(bs->file, s->desc_offset, desc, DESC_SIZE);
298     if (ret < 0) {
299         return ret;
300     }
301 
302     return 0;
303 }
304 
305 static int vmdk_is_cid_valid(BlockDriverState *bs)
306 {
307     BDRVVmdkState *s = bs->opaque;
308     BlockDriverState *p_bs = bs->backing_hd;
309     uint32_t cur_pcid;
310 
311     if (!s->cid_checked && p_bs) {
312         cur_pcid = vmdk_read_cid(p_bs, 0);
313         if (s->parent_cid != cur_pcid) {
314             /* CID not valid */
315             return 0;
316         }
317     }
318     s->cid_checked = true;
319     /* CID valid */
320     return 1;
321 }
322 
323 /* Queue extents, if any, for reopen() */
324 static int vmdk_reopen_prepare(BDRVReopenState *state,
325                                BlockReopenQueue *queue, Error **errp)
326 {
327     BDRVVmdkState *s;
328     int ret = -1;
329     int i;
330     VmdkExtent *e;
331 
332     assert(state != NULL);
333     assert(state->bs != NULL);
334 
335     if (queue == NULL) {
336         error_setg(errp, "No reopen queue for VMDK extents");
337         goto exit;
338     }
339 
340     s = state->bs->opaque;
341 
342     assert(s != NULL);
343 
344     for (i = 0; i < s->num_extents; i++) {
345         e = &s->extents[i];
346         if (e->file != state->bs->file) {
347             bdrv_reopen_queue(queue, e->file, state->flags);
348         }
349     }
350     ret = 0;
351 
352 exit:
353     return ret;
354 }
355 
356 static int vmdk_parent_open(BlockDriverState *bs)
357 {
358     char *p_name;
359     char desc[DESC_SIZE + 1];
360     BDRVVmdkState *s = bs->opaque;
361     int ret;
362 
363     desc[DESC_SIZE] = '\0';
364     ret = bdrv_pread(bs->file, s->desc_offset, desc, DESC_SIZE);
365     if (ret < 0) {
366         return ret;
367     }
368 
369     p_name = strstr(desc, "parentFileNameHint");
370     if (p_name != NULL) {
371         char *end_name;
372 
373         p_name += sizeof("parentFileNameHint") + 1;
374         end_name = strchr(p_name, '\"');
375         if (end_name == NULL) {
376             return -EINVAL;
377         }
378         if ((end_name - p_name) > sizeof(bs->backing_file) - 1) {
379             return -EINVAL;
380         }
381 
382         pstrcpy(bs->backing_file, end_name - p_name + 1, p_name);
383     }
384 
385     return 0;
386 }
387 
388 /* Create and append extent to the extent array. Return the added VmdkExtent
389  * address. return NULL if allocation failed. */
390 static int vmdk_add_extent(BlockDriverState *bs,
391                            BlockDriverState *file, bool flat, int64_t sectors,
392                            int64_t l1_offset, int64_t l1_backup_offset,
393                            uint32_t l1_size,
394                            int l2_size, uint64_t cluster_sectors,
395                            VmdkExtent **new_extent,
396                            Error **errp)
397 {
398     VmdkExtent *extent;
399     BDRVVmdkState *s = bs->opaque;
400 
401     if (cluster_sectors > 0x200000) {
402         /* 0x200000 * 512Bytes = 1GB for one cluster is unrealistic */
403         error_setg(errp, "Invalid granularity, image may be corrupt");
404         return -EFBIG;
405     }
406     if (l1_size > 512 * 1024 * 1024) {
407         /* Although with big capacity and small l1_entry_sectors, we can get a
408          * big l1_size, we don't want unbounded value to allocate the table.
409          * Limit it to 512M, which is 16PB for default cluster and L2 table
410          * size */
411         error_setg(errp, "L1 size too big");
412         return -EFBIG;
413     }
414 
415     s->extents = g_realloc(s->extents,
416                               (s->num_extents + 1) * sizeof(VmdkExtent));
417     extent = &s->extents[s->num_extents];
418     s->num_extents++;
419 
420     memset(extent, 0, sizeof(VmdkExtent));
421     extent->file = file;
422     extent->flat = flat;
423     extent->sectors = sectors;
424     extent->l1_table_offset = l1_offset;
425     extent->l1_backup_table_offset = l1_backup_offset;
426     extent->l1_size = l1_size;
427     extent->l1_entry_sectors = l2_size * cluster_sectors;
428     extent->l2_size = l2_size;
429     extent->cluster_sectors = flat ? sectors : cluster_sectors;
430 
431     if (!flat) {
432         bs->bl.write_zeroes_alignment =
433             MAX(bs->bl.write_zeroes_alignment, cluster_sectors);
434     }
435     if (s->num_extents > 1) {
436         extent->end_sector = (*(extent - 1)).end_sector + extent->sectors;
437     } else {
438         extent->end_sector = extent->sectors;
439     }
440     bs->total_sectors = extent->end_sector;
441     if (new_extent) {
442         *new_extent = extent;
443     }
444     return 0;
445 }
446 
447 static int vmdk_init_tables(BlockDriverState *bs, VmdkExtent *extent,
448                             Error **errp)
449 {
450     int ret;
451     int l1_size, i;
452 
453     /* read the L1 table */
454     l1_size = extent->l1_size * sizeof(uint32_t);
455     extent->l1_table = g_malloc(l1_size);
456     ret = bdrv_pread(extent->file,
457                      extent->l1_table_offset,
458                      extent->l1_table,
459                      l1_size);
460     if (ret < 0) {
461         error_setg_errno(errp, -ret,
462                          "Could not read l1 table from extent '%s'",
463                          extent->file->filename);
464         goto fail_l1;
465     }
466     for (i = 0; i < extent->l1_size; i++) {
467         le32_to_cpus(&extent->l1_table[i]);
468     }
469 
470     if (extent->l1_backup_table_offset) {
471         extent->l1_backup_table = g_malloc(l1_size);
472         ret = bdrv_pread(extent->file,
473                          extent->l1_backup_table_offset,
474                          extent->l1_backup_table,
475                          l1_size);
476         if (ret < 0) {
477             error_setg_errno(errp, -ret,
478                              "Could not read l1 backup table from extent '%s'",
479                              extent->file->filename);
480             goto fail_l1b;
481         }
482         for (i = 0; i < extent->l1_size; i++) {
483             le32_to_cpus(&extent->l1_backup_table[i]);
484         }
485     }
486 
487     extent->l2_cache =
488         g_malloc(extent->l2_size * L2_CACHE_SIZE * sizeof(uint32_t));
489     return 0;
490  fail_l1b:
491     g_free(extent->l1_backup_table);
492  fail_l1:
493     g_free(extent->l1_table);
494     return ret;
495 }
496 
497 static int vmdk_open_vmfs_sparse(BlockDriverState *bs,
498                                  BlockDriverState *file,
499                                  int flags, Error **errp)
500 {
501     int ret;
502     uint32_t magic;
503     VMDK3Header header;
504     VmdkExtent *extent;
505 
506     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
507     if (ret < 0) {
508         error_setg_errno(errp, -ret,
509                          "Could not read header from file '%s'",
510                          file->filename);
511         return ret;
512     }
513     ret = vmdk_add_extent(bs, file, false,
514                           le32_to_cpu(header.disk_sectors),
515                           le32_to_cpu(header.l1dir_offset) << 9,
516                           0,
517                           le32_to_cpu(header.l1dir_size),
518                           4096,
519                           le32_to_cpu(header.granularity),
520                           &extent,
521                           errp);
522     if (ret < 0) {
523         return ret;
524     }
525     ret = vmdk_init_tables(bs, extent, errp);
526     if (ret) {
527         /* free extent allocated by vmdk_add_extent */
528         vmdk_free_last_extent(bs);
529     }
530     return ret;
531 }
532 
533 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
534                                uint64_t desc_offset, Error **errp);
535 
536 static int vmdk_open_vmdk4(BlockDriverState *bs,
537                            BlockDriverState *file,
538                            int flags, Error **errp)
539 {
540     int ret;
541     uint32_t magic;
542     uint32_t l1_size, l1_entry_sectors;
543     VMDK4Header header;
544     VmdkExtent *extent;
545     BDRVVmdkState *s = bs->opaque;
546     int64_t l1_backup_offset = 0;
547 
548     ret = bdrv_pread(file, sizeof(magic), &header, sizeof(header));
549     if (ret < 0) {
550         error_setg_errno(errp, -ret,
551                          "Could not read header from file '%s'",
552                          file->filename);
553     }
554     if (header.capacity == 0) {
555         uint64_t desc_offset = le64_to_cpu(header.desc_offset);
556         if (desc_offset) {
557             return vmdk_open_desc_file(bs, flags, desc_offset << 9, errp);
558         }
559     }
560 
561     if (!s->create_type) {
562         s->create_type = g_strdup("monolithicSparse");
563     }
564 
565     if (le64_to_cpu(header.gd_offset) == VMDK4_GD_AT_END) {
566         /*
567          * The footer takes precedence over the header, so read it in. The
568          * footer starts at offset -1024 from the end: One sector for the
569          * footer, and another one for the end-of-stream marker.
570          */
571         struct {
572             struct {
573                 uint64_t val;
574                 uint32_t size;
575                 uint32_t type;
576                 uint8_t pad[512 - 16];
577             } QEMU_PACKED footer_marker;
578 
579             uint32_t magic;
580             VMDK4Header header;
581             uint8_t pad[512 - 4 - sizeof(VMDK4Header)];
582 
583             struct {
584                 uint64_t val;
585                 uint32_t size;
586                 uint32_t type;
587                 uint8_t pad[512 - 16];
588             } QEMU_PACKED eos_marker;
589         } QEMU_PACKED footer;
590 
591         ret = bdrv_pread(file,
592             bs->file->total_sectors * 512 - 1536,
593             &footer, sizeof(footer));
594         if (ret < 0) {
595             return ret;
596         }
597 
598         /* Some sanity checks for the footer */
599         if (be32_to_cpu(footer.magic) != VMDK4_MAGIC ||
600             le32_to_cpu(footer.footer_marker.size) != 0  ||
601             le32_to_cpu(footer.footer_marker.type) != MARKER_FOOTER ||
602             le64_to_cpu(footer.eos_marker.val) != 0  ||
603             le32_to_cpu(footer.eos_marker.size) != 0  ||
604             le32_to_cpu(footer.eos_marker.type) != MARKER_END_OF_STREAM)
605         {
606             return -EINVAL;
607         }
608 
609         header = footer.header;
610     }
611 
612     if (le32_to_cpu(header.version) > 3) {
613         char buf[64];
614         snprintf(buf, sizeof(buf), "VMDK version %d",
615                  le32_to_cpu(header.version));
616         qerror_report(QERR_UNKNOWN_BLOCK_FORMAT_FEATURE,
617                 bs->device_name, "vmdk", buf);
618         return -ENOTSUP;
619     } else if (le32_to_cpu(header.version) == 3 && (flags & BDRV_O_RDWR)) {
620         /* VMware KB 2064959 explains that version 3 added support for
621          * persistent changed block tracking (CBT), and backup software can
622          * read it as version=1 if it doesn't care about the changed area
623          * information. So we are safe to enable read only. */
624         error_setg(errp, "VMDK version 3 must be read only");
625         return -EINVAL;
626     }
627 
628     if (le32_to_cpu(header.num_gtes_per_gt) > 512) {
629         error_report("L2 table size too big");
630         return -EINVAL;
631     }
632 
633     l1_entry_sectors = le32_to_cpu(header.num_gtes_per_gt)
634                         * le64_to_cpu(header.granularity);
635     if (l1_entry_sectors == 0) {
636         return -EINVAL;
637     }
638     l1_size = (le64_to_cpu(header.capacity) + l1_entry_sectors - 1)
639                 / l1_entry_sectors;
640     if (le32_to_cpu(header.flags) & VMDK4_FLAG_RGD) {
641         l1_backup_offset = le64_to_cpu(header.rgd_offset) << 9;
642     }
643     ret = vmdk_add_extent(bs, file, false,
644                           le64_to_cpu(header.capacity),
645                           le64_to_cpu(header.gd_offset) << 9,
646                           l1_backup_offset,
647                           l1_size,
648                           le32_to_cpu(header.num_gtes_per_gt),
649                           le64_to_cpu(header.granularity),
650                           &extent,
651                           errp);
652     if (ret < 0) {
653         return ret;
654     }
655     extent->compressed =
656         le16_to_cpu(header.compressAlgorithm) == VMDK4_COMPRESSION_DEFLATE;
657     extent->has_marker = le32_to_cpu(header.flags) & VMDK4_FLAG_MARKER;
658     extent->version = le32_to_cpu(header.version);
659     extent->has_zero_grain = le32_to_cpu(header.flags) & VMDK4_FLAG_ZERO_GRAIN;
660     ret = vmdk_init_tables(bs, extent, errp);
661     if (ret) {
662         /* free extent allocated by vmdk_add_extent */
663         vmdk_free_last_extent(bs);
664     }
665     return ret;
666 }
667 
668 /* find an option value out of descriptor file */
669 static int vmdk_parse_description(const char *desc, const char *opt_name,
670         char *buf, int buf_size)
671 {
672     char *opt_pos, *opt_end;
673     const char *end = desc + strlen(desc);
674 
675     opt_pos = strstr(desc, opt_name);
676     if (!opt_pos) {
677         return VMDK_ERROR;
678     }
679     /* Skip "=\"" following opt_name */
680     opt_pos += strlen(opt_name) + 2;
681     if (opt_pos >= end) {
682         return VMDK_ERROR;
683     }
684     opt_end = opt_pos;
685     while (opt_end < end && *opt_end != '"') {
686         opt_end++;
687     }
688     if (opt_end == end || buf_size < opt_end - opt_pos + 1) {
689         return VMDK_ERROR;
690     }
691     pstrcpy(buf, opt_end - opt_pos + 1, opt_pos);
692     return VMDK_OK;
693 }
694 
695 /* Open an extent file and append to bs array */
696 static int vmdk_open_sparse(BlockDriverState *bs,
697                             BlockDriverState *file,
698                             int flags, Error **errp)
699 {
700     uint32_t magic;
701 
702     if (bdrv_pread(file, 0, &magic, sizeof(magic)) != sizeof(magic)) {
703         return -EIO;
704     }
705 
706     magic = be32_to_cpu(magic);
707     switch (magic) {
708         case VMDK3_MAGIC:
709             return vmdk_open_vmfs_sparse(bs, file, flags, errp);
710             break;
711         case VMDK4_MAGIC:
712             return vmdk_open_vmdk4(bs, file, flags, errp);
713             break;
714         default:
715             return -EMEDIUMTYPE;
716             break;
717     }
718 }
719 
720 static int vmdk_parse_extents(const char *desc, BlockDriverState *bs,
721                               const char *desc_file_path, Error **errp)
722 {
723     int ret;
724     char access[11];
725     char type[11];
726     char fname[512];
727     const char *p = desc;
728     int64_t sectors = 0;
729     int64_t flat_offset;
730     char extent_path[PATH_MAX];
731     BlockDriverState *extent_file;
732     BDRVVmdkState *s = bs->opaque;
733     VmdkExtent *extent;
734 
735     while (*p) {
736         /* parse extent line:
737          * RW [size in sectors] FLAT "file-name.vmdk" OFFSET
738          * or
739          * RW [size in sectors] SPARSE "file-name.vmdk"
740          */
741         flat_offset = -1;
742         ret = sscanf(p, "%10s %" SCNd64 " %10s \"%511[^\n\r\"]\" %" SCNd64,
743                 access, &sectors, type, fname, &flat_offset);
744         if (ret < 4 || strcmp(access, "RW")) {
745             goto next_line;
746         } else if (!strcmp(type, "FLAT")) {
747             if (ret != 5 || flat_offset < 0) {
748                 error_setg(errp, "Invalid extent lines: \n%s", p);
749                 return -EINVAL;
750             }
751         } else if (!strcmp(type, "VMFS")) {
752             flat_offset = 0;
753         } else if (ret != 4) {
754             error_setg(errp, "Invalid extent lines: \n%s", p);
755             return -EINVAL;
756         }
757 
758         if (sectors <= 0 ||
759             (strcmp(type, "FLAT") && strcmp(type, "SPARSE") &&
760              strcmp(type, "VMFS") && strcmp(type, "VMFSSPARSE")) ||
761             (strcmp(access, "RW"))) {
762             goto next_line;
763         }
764 
765         path_combine(extent_path, sizeof(extent_path),
766                 desc_file_path, fname);
767         ret = bdrv_file_open(&extent_file, extent_path, NULL, bs->open_flags,
768                              errp);
769         if (ret) {
770             return ret;
771         }
772 
773         /* save to extents array */
774         if (!strcmp(type, "FLAT") || !strcmp(type, "VMFS")) {
775             /* FLAT extent */
776 
777             ret = vmdk_add_extent(bs, extent_file, true, sectors,
778                             0, 0, 0, 0, 0, &extent, errp);
779             if (ret < 0) {
780                 return ret;
781             }
782             extent->flat_start_offset = flat_offset << 9;
783         } else if (!strcmp(type, "SPARSE") || !strcmp(type, "VMFSSPARSE")) {
784             /* SPARSE extent and VMFSSPARSE extent are both "COWD" sparse file*/
785             ret = vmdk_open_sparse(bs, extent_file, bs->open_flags, errp);
786             if (ret) {
787                 bdrv_unref(extent_file);
788                 return ret;
789             }
790             extent = &s->extents[s->num_extents - 1];
791         } else {
792             error_setg(errp, "Unsupported extent type '%s'", type);
793             return -ENOTSUP;
794         }
795         extent->type = g_strdup(type);
796 next_line:
797         /* move to next line */
798         while (*p) {
799             if (*p == '\n') {
800                 p++;
801                 break;
802             }
803             p++;
804         }
805     }
806     return 0;
807 }
808 
809 static int vmdk_open_desc_file(BlockDriverState *bs, int flags,
810                                uint64_t desc_offset, Error **errp)
811 {
812     int ret;
813     char *buf = NULL;
814     char ct[128];
815     BDRVVmdkState *s = bs->opaque;
816     int64_t size;
817 
818     size = bdrv_getlength(bs->file);
819     if (size < 0) {
820         return -EINVAL;
821     }
822 
823     size = MIN(size, 1 << 20);  /* avoid unbounded allocation */
824     buf = g_malloc0(size + 1);
825 
826     ret = bdrv_pread(bs->file, desc_offset, buf, size);
827     if (ret < 0) {
828         goto exit;
829     }
830     if (vmdk_parse_description(buf, "createType", ct, sizeof(ct))) {
831         ret = -EMEDIUMTYPE;
832         goto exit;
833     }
834     if (strcmp(ct, "monolithicFlat") &&
835         strcmp(ct, "vmfs") &&
836         strcmp(ct, "vmfsSparse") &&
837         strcmp(ct, "twoGbMaxExtentSparse") &&
838         strcmp(ct, "twoGbMaxExtentFlat")) {
839         error_setg(errp, "Unsupported image type '%s'", ct);
840         ret = -ENOTSUP;
841         goto exit;
842     }
843     s->create_type = g_strdup(ct);
844     s->desc_offset = 0;
845     ret = vmdk_parse_extents(buf, bs, bs->file->filename, errp);
846 exit:
847     g_free(buf);
848     return ret;
849 }
850 
851 static int vmdk_open(BlockDriverState *bs, QDict *options, int flags,
852                      Error **errp)
853 {
854     int ret;
855     BDRVVmdkState *s = bs->opaque;
856 
857     if (vmdk_open_sparse(bs, bs->file, flags, errp) == 0) {
858         s->desc_offset = 0x200;
859     } else {
860         ret = vmdk_open_desc_file(bs, flags, 0, errp);
861         if (ret) {
862             goto fail;
863         }
864     }
865     /* try to open parent images, if exist */
866     ret = vmdk_parent_open(bs);
867     if (ret) {
868         goto fail;
869     }
870     s->cid = vmdk_read_cid(bs, 0);
871     s->parent_cid = vmdk_read_cid(bs, 1);
872     qemu_co_mutex_init(&s->lock);
873 
874     /* Disable migration when VMDK images are used */
875     error_set(&s->migration_blocker,
876               QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED,
877               "vmdk", bs->device_name, "live migration");
878     migrate_add_blocker(s->migration_blocker);
879 
880     return 0;
881 
882 fail:
883     g_free(s->create_type);
884     s->create_type = NULL;
885     vmdk_free_extents(bs);
886     return ret;
887 }
888 
889 static int get_whole_cluster(BlockDriverState *bs,
890                 VmdkExtent *extent,
891                 uint64_t cluster_offset,
892                 uint64_t offset,
893                 bool allocate)
894 {
895     int ret = VMDK_OK;
896     uint8_t *whole_grain = NULL;
897 
898     /* we will be here if it's first write on non-exist grain(cluster).
899      * try to read from parent image, if exist */
900     if (bs->backing_hd) {
901         whole_grain =
902             qemu_blockalign(bs, extent->cluster_sectors << BDRV_SECTOR_BITS);
903         if (!vmdk_is_cid_valid(bs)) {
904             ret = VMDK_ERROR;
905             goto exit;
906         }
907 
908         /* floor offset to cluster */
909         offset -= offset % (extent->cluster_sectors * 512);
910         ret = bdrv_read(bs->backing_hd, offset >> 9, whole_grain,
911                 extent->cluster_sectors);
912         if (ret < 0) {
913             ret = VMDK_ERROR;
914             goto exit;
915         }
916 
917         /* Write grain only into the active image */
918         ret = bdrv_write(extent->file, cluster_offset, whole_grain,
919                 extent->cluster_sectors);
920         if (ret < 0) {
921             ret = VMDK_ERROR;
922             goto exit;
923         }
924     }
925 exit:
926     qemu_vfree(whole_grain);
927     return ret;
928 }
929 
930 static int vmdk_L2update(VmdkExtent *extent, VmdkMetaData *m_data)
931 {
932     uint32_t offset;
933     QEMU_BUILD_BUG_ON(sizeof(offset) != sizeof(m_data->offset));
934     offset = cpu_to_le32(m_data->offset);
935     /* update L2 table */
936     if (bdrv_pwrite_sync(
937                 extent->file,
938                 ((int64_t)m_data->l2_offset * 512)
939                     + (m_data->l2_index * sizeof(m_data->offset)),
940                 &offset, sizeof(offset)) < 0) {
941         return VMDK_ERROR;
942     }
943     /* update backup L2 table */
944     if (extent->l1_backup_table_offset != 0) {
945         m_data->l2_offset = extent->l1_backup_table[m_data->l1_index];
946         if (bdrv_pwrite_sync(
947                     extent->file,
948                     ((int64_t)m_data->l2_offset * 512)
949                         + (m_data->l2_index * sizeof(m_data->offset)),
950                     &offset, sizeof(offset)) < 0) {
951             return VMDK_ERROR;
952         }
953     }
954     if (m_data->l2_cache_entry) {
955         *m_data->l2_cache_entry = offset;
956     }
957 
958     return VMDK_OK;
959 }
960 
961 static int get_cluster_offset(BlockDriverState *bs,
962                                     VmdkExtent *extent,
963                                     VmdkMetaData *m_data,
964                                     uint64_t offset,
965                                     int allocate,
966                                     uint64_t *cluster_offset)
967 {
968     unsigned int l1_index, l2_offset, l2_index;
969     int min_index, i, j;
970     uint32_t min_count, *l2_table;
971     bool zeroed = false;
972 
973     if (m_data) {
974         m_data->valid = 0;
975     }
976     if (extent->flat) {
977         *cluster_offset = extent->flat_start_offset;
978         return VMDK_OK;
979     }
980 
981     offset -= (extent->end_sector - extent->sectors) * SECTOR_SIZE;
982     l1_index = (offset >> 9) / extent->l1_entry_sectors;
983     if (l1_index >= extent->l1_size) {
984         return VMDK_ERROR;
985     }
986     l2_offset = extent->l1_table[l1_index];
987     if (!l2_offset) {
988         return VMDK_UNALLOC;
989     }
990     for (i = 0; i < L2_CACHE_SIZE; i++) {
991         if (l2_offset == extent->l2_cache_offsets[i]) {
992             /* increment the hit count */
993             if (++extent->l2_cache_counts[i] == 0xffffffff) {
994                 for (j = 0; j < L2_CACHE_SIZE; j++) {
995                     extent->l2_cache_counts[j] >>= 1;
996                 }
997             }
998             l2_table = extent->l2_cache + (i * extent->l2_size);
999             goto found;
1000         }
1001     }
1002     /* not found: load a new entry in the least used one */
1003     min_index = 0;
1004     min_count = 0xffffffff;
1005     for (i = 0; i < L2_CACHE_SIZE; i++) {
1006         if (extent->l2_cache_counts[i] < min_count) {
1007             min_count = extent->l2_cache_counts[i];
1008             min_index = i;
1009         }
1010     }
1011     l2_table = extent->l2_cache + (min_index * extent->l2_size);
1012     if (bdrv_pread(
1013                 extent->file,
1014                 (int64_t)l2_offset * 512,
1015                 l2_table,
1016                 extent->l2_size * sizeof(uint32_t)
1017             ) != extent->l2_size * sizeof(uint32_t)) {
1018         return VMDK_ERROR;
1019     }
1020 
1021     extent->l2_cache_offsets[min_index] = l2_offset;
1022     extent->l2_cache_counts[min_index] = 1;
1023  found:
1024     l2_index = ((offset >> 9) / extent->cluster_sectors) % extent->l2_size;
1025     *cluster_offset = le32_to_cpu(l2_table[l2_index]);
1026 
1027     if (m_data) {
1028         m_data->valid = 1;
1029         m_data->l1_index = l1_index;
1030         m_data->l2_index = l2_index;
1031         m_data->offset = *cluster_offset;
1032         m_data->l2_offset = l2_offset;
1033         m_data->l2_cache_entry = &l2_table[l2_index];
1034     }
1035     if (extent->has_zero_grain && *cluster_offset == VMDK_GTE_ZEROED) {
1036         zeroed = true;
1037     }
1038 
1039     if (!*cluster_offset || zeroed) {
1040         if (!allocate) {
1041             return zeroed ? VMDK_ZEROED : VMDK_UNALLOC;
1042         }
1043 
1044         /* Avoid the L2 tables update for the images that have snapshots. */
1045         *cluster_offset = bdrv_getlength(extent->file);
1046         if (!extent->compressed) {
1047             bdrv_truncate(
1048                 extent->file,
1049                 *cluster_offset + (extent->cluster_sectors << 9)
1050             );
1051         }
1052 
1053         *cluster_offset >>= 9;
1054         l2_table[l2_index] = cpu_to_le32(*cluster_offset);
1055 
1056         /* First of all we write grain itself, to avoid race condition
1057          * that may to corrupt the image.
1058          * This problem may occur because of insufficient space on host disk
1059          * or inappropriate VM shutdown.
1060          */
1061         if (get_whole_cluster(
1062                 bs, extent, *cluster_offset, offset, allocate) == -1) {
1063             return VMDK_ERROR;
1064         }
1065 
1066         if (m_data) {
1067             m_data->offset = *cluster_offset;
1068         }
1069     }
1070     *cluster_offset <<= 9;
1071     return VMDK_OK;
1072 }
1073 
1074 static VmdkExtent *find_extent(BDRVVmdkState *s,
1075                                 int64_t sector_num, VmdkExtent *start_hint)
1076 {
1077     VmdkExtent *extent = start_hint;
1078 
1079     if (!extent) {
1080         extent = &s->extents[0];
1081     }
1082     while (extent < &s->extents[s->num_extents]) {
1083         if (sector_num < extent->end_sector) {
1084             return extent;
1085         }
1086         extent++;
1087     }
1088     return NULL;
1089 }
1090 
1091 static int64_t coroutine_fn vmdk_co_get_block_status(BlockDriverState *bs,
1092         int64_t sector_num, int nb_sectors, int *pnum)
1093 {
1094     BDRVVmdkState *s = bs->opaque;
1095     int64_t index_in_cluster, n, ret;
1096     uint64_t offset;
1097     VmdkExtent *extent;
1098 
1099     extent = find_extent(s, sector_num, NULL);
1100     if (!extent) {
1101         return 0;
1102     }
1103     qemu_co_mutex_lock(&s->lock);
1104     ret = get_cluster_offset(bs, extent, NULL,
1105                             sector_num * 512, 0, &offset);
1106     qemu_co_mutex_unlock(&s->lock);
1107 
1108     switch (ret) {
1109     case VMDK_ERROR:
1110         ret = -EIO;
1111         break;
1112     case VMDK_UNALLOC:
1113         ret = 0;
1114         break;
1115     case VMDK_ZEROED:
1116         ret = BDRV_BLOCK_ZERO;
1117         break;
1118     case VMDK_OK:
1119         ret = BDRV_BLOCK_DATA;
1120         if (extent->file == bs->file) {
1121             ret |= BDRV_BLOCK_OFFSET_VALID | offset;
1122         }
1123 
1124         break;
1125     }
1126 
1127     index_in_cluster = sector_num % extent->cluster_sectors;
1128     n = extent->cluster_sectors - index_in_cluster;
1129     if (n > nb_sectors) {
1130         n = nb_sectors;
1131     }
1132     *pnum = n;
1133     return ret;
1134 }
1135 
1136 static int vmdk_write_extent(VmdkExtent *extent, int64_t cluster_offset,
1137                             int64_t offset_in_cluster, const uint8_t *buf,
1138                             int nb_sectors, int64_t sector_num)
1139 {
1140     int ret;
1141     VmdkGrainMarker *data = NULL;
1142     uLongf buf_len;
1143     const uint8_t *write_buf = buf;
1144     int write_len = nb_sectors * 512;
1145 
1146     if (extent->compressed) {
1147         if (!extent->has_marker) {
1148             ret = -EINVAL;
1149             goto out;
1150         }
1151         buf_len = (extent->cluster_sectors << 9) * 2;
1152         data = g_malloc(buf_len + sizeof(VmdkGrainMarker));
1153         if (compress(data->data, &buf_len, buf, nb_sectors << 9) != Z_OK ||
1154                 buf_len == 0) {
1155             ret = -EINVAL;
1156             goto out;
1157         }
1158         data->lba = sector_num;
1159         data->size = buf_len;
1160         write_buf = (uint8_t *)data;
1161         write_len = buf_len + sizeof(VmdkGrainMarker);
1162     }
1163     ret = bdrv_pwrite(extent->file,
1164                         cluster_offset + offset_in_cluster,
1165                         write_buf,
1166                         write_len);
1167     if (ret != write_len) {
1168         ret = ret < 0 ? ret : -EIO;
1169         goto out;
1170     }
1171     ret = 0;
1172  out:
1173     g_free(data);
1174     return ret;
1175 }
1176 
1177 static int vmdk_read_extent(VmdkExtent *extent, int64_t cluster_offset,
1178                             int64_t offset_in_cluster, uint8_t *buf,
1179                             int nb_sectors)
1180 {
1181     int ret;
1182     int cluster_bytes, buf_bytes;
1183     uint8_t *cluster_buf, *compressed_data;
1184     uint8_t *uncomp_buf;
1185     uint32_t data_len;
1186     VmdkGrainMarker *marker;
1187     uLongf buf_len;
1188 
1189 
1190     if (!extent->compressed) {
1191         ret = bdrv_pread(extent->file,
1192                           cluster_offset + offset_in_cluster,
1193                           buf, nb_sectors * 512);
1194         if (ret == nb_sectors * 512) {
1195             return 0;
1196         } else {
1197             return -EIO;
1198         }
1199     }
1200     cluster_bytes = extent->cluster_sectors * 512;
1201     /* Read two clusters in case GrainMarker + compressed data > one cluster */
1202     buf_bytes = cluster_bytes * 2;
1203     cluster_buf = g_malloc(buf_bytes);
1204     uncomp_buf = g_malloc(cluster_bytes);
1205     ret = bdrv_pread(extent->file,
1206                 cluster_offset,
1207                 cluster_buf, buf_bytes);
1208     if (ret < 0) {
1209         goto out;
1210     }
1211     compressed_data = cluster_buf;
1212     buf_len = cluster_bytes;
1213     data_len = cluster_bytes;
1214     if (extent->has_marker) {
1215         marker = (VmdkGrainMarker *)cluster_buf;
1216         compressed_data = marker->data;
1217         data_len = le32_to_cpu(marker->size);
1218     }
1219     if (!data_len || data_len > buf_bytes) {
1220         ret = -EINVAL;
1221         goto out;
1222     }
1223     ret = uncompress(uncomp_buf, &buf_len, compressed_data, data_len);
1224     if (ret != Z_OK) {
1225         ret = -EINVAL;
1226         goto out;
1227 
1228     }
1229     if (offset_in_cluster < 0 ||
1230             offset_in_cluster + nb_sectors * 512 > buf_len) {
1231         ret = -EINVAL;
1232         goto out;
1233     }
1234     memcpy(buf, uncomp_buf + offset_in_cluster, nb_sectors * 512);
1235     ret = 0;
1236 
1237  out:
1238     g_free(uncomp_buf);
1239     g_free(cluster_buf);
1240     return ret;
1241 }
1242 
1243 static int vmdk_read(BlockDriverState *bs, int64_t sector_num,
1244                     uint8_t *buf, int nb_sectors)
1245 {
1246     BDRVVmdkState *s = bs->opaque;
1247     int ret;
1248     uint64_t n, index_in_cluster;
1249     uint64_t extent_begin_sector, extent_relative_sector_num;
1250     VmdkExtent *extent = NULL;
1251     uint64_t cluster_offset;
1252 
1253     while (nb_sectors > 0) {
1254         extent = find_extent(s, sector_num, extent);
1255         if (!extent) {
1256             return -EIO;
1257         }
1258         ret = get_cluster_offset(
1259                             bs, extent, NULL,
1260                             sector_num << 9, 0, &cluster_offset);
1261         extent_begin_sector = extent->end_sector - extent->sectors;
1262         extent_relative_sector_num = sector_num - extent_begin_sector;
1263         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1264         n = extent->cluster_sectors - index_in_cluster;
1265         if (n > nb_sectors) {
1266             n = nb_sectors;
1267         }
1268         if (ret != VMDK_OK) {
1269             /* if not allocated, try to read from parent image, if exist */
1270             if (bs->backing_hd && ret != VMDK_ZEROED) {
1271                 if (!vmdk_is_cid_valid(bs)) {
1272                     return -EINVAL;
1273                 }
1274                 ret = bdrv_read(bs->backing_hd, sector_num, buf, n);
1275                 if (ret < 0) {
1276                     return ret;
1277                 }
1278             } else {
1279                 memset(buf, 0, 512 * n);
1280             }
1281         } else {
1282             ret = vmdk_read_extent(extent,
1283                             cluster_offset, index_in_cluster * 512,
1284                             buf, n);
1285             if (ret) {
1286                 return ret;
1287             }
1288         }
1289         nb_sectors -= n;
1290         sector_num += n;
1291         buf += n * 512;
1292     }
1293     return 0;
1294 }
1295 
1296 static coroutine_fn int vmdk_co_read(BlockDriverState *bs, int64_t sector_num,
1297                                      uint8_t *buf, int nb_sectors)
1298 {
1299     int ret;
1300     BDRVVmdkState *s = bs->opaque;
1301     qemu_co_mutex_lock(&s->lock);
1302     ret = vmdk_read(bs, sector_num, buf, nb_sectors);
1303     qemu_co_mutex_unlock(&s->lock);
1304     return ret;
1305 }
1306 
1307 /**
1308  * vmdk_write:
1309  * @zeroed:       buf is ignored (data is zero), use zeroed_grain GTE feature
1310  *                if possible, otherwise return -ENOTSUP.
1311  * @zero_dry_run: used for zeroed == true only, don't update L2 table, just try
1312  *                with each cluster. By dry run we can find if the zero write
1313  *                is possible without modifying image data.
1314  *
1315  * Returns: error code with 0 for success.
1316  */
1317 static int vmdk_write(BlockDriverState *bs, int64_t sector_num,
1318                       const uint8_t *buf, int nb_sectors,
1319                       bool zeroed, bool zero_dry_run)
1320 {
1321     BDRVVmdkState *s = bs->opaque;
1322     VmdkExtent *extent = NULL;
1323     int n, ret;
1324     int64_t index_in_cluster;
1325     uint64_t extent_begin_sector, extent_relative_sector_num;
1326     uint64_t cluster_offset;
1327     VmdkMetaData m_data;
1328 
1329     if (sector_num > bs->total_sectors) {
1330         error_report("Wrong offset: sector_num=0x%" PRIx64
1331                 " total_sectors=0x%" PRIx64 "\n",
1332                 sector_num, bs->total_sectors);
1333         return -EIO;
1334     }
1335 
1336     while (nb_sectors > 0) {
1337         extent = find_extent(s, sector_num, extent);
1338         if (!extent) {
1339             return -EIO;
1340         }
1341         ret = get_cluster_offset(
1342                                 bs,
1343                                 extent,
1344                                 &m_data,
1345                                 sector_num << 9, !extent->compressed,
1346                                 &cluster_offset);
1347         if (extent->compressed) {
1348             if (ret == VMDK_OK) {
1349                 /* Refuse write to allocated cluster for streamOptimized */
1350                 error_report("Could not write to allocated cluster"
1351                               " for streamOptimized");
1352                 return -EIO;
1353             } else {
1354                 /* allocate */
1355                 ret = get_cluster_offset(
1356                                         bs,
1357                                         extent,
1358                                         &m_data,
1359                                         sector_num << 9, 1,
1360                                         &cluster_offset);
1361             }
1362         }
1363         if (ret == VMDK_ERROR) {
1364             return -EINVAL;
1365         }
1366         extent_begin_sector = extent->end_sector - extent->sectors;
1367         extent_relative_sector_num = sector_num - extent_begin_sector;
1368         index_in_cluster = extent_relative_sector_num % extent->cluster_sectors;
1369         n = extent->cluster_sectors - index_in_cluster;
1370         if (n > nb_sectors) {
1371             n = nb_sectors;
1372         }
1373         if (zeroed) {
1374             /* Do zeroed write, buf is ignored */
1375             if (extent->has_zero_grain &&
1376                     index_in_cluster == 0 &&
1377                     n >= extent->cluster_sectors) {
1378                 n = extent->cluster_sectors;
1379                 if (!zero_dry_run) {
1380                     m_data.offset = VMDK_GTE_ZEROED;
1381                     /* update L2 tables */
1382                     if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1383                         return -EIO;
1384                     }
1385                 }
1386             } else {
1387                 return -ENOTSUP;
1388             }
1389         } else {
1390             ret = vmdk_write_extent(extent,
1391                             cluster_offset, index_in_cluster * 512,
1392                             buf, n, sector_num);
1393             if (ret) {
1394                 return ret;
1395             }
1396             if (m_data.valid) {
1397                 /* update L2 tables */
1398                 if (vmdk_L2update(extent, &m_data) != VMDK_OK) {
1399                     return -EIO;
1400                 }
1401             }
1402         }
1403         nb_sectors -= n;
1404         sector_num += n;
1405         buf += n * 512;
1406 
1407         /* update CID on the first write every time the virtual disk is
1408          * opened */
1409         if (!s->cid_updated) {
1410             ret = vmdk_write_cid(bs, time(NULL));
1411             if (ret < 0) {
1412                 return ret;
1413             }
1414             s->cid_updated = true;
1415         }
1416     }
1417     return 0;
1418 }
1419 
1420 static coroutine_fn int vmdk_co_write(BlockDriverState *bs, int64_t sector_num,
1421                                       const uint8_t *buf, int nb_sectors)
1422 {
1423     int ret;
1424     BDRVVmdkState *s = bs->opaque;
1425     qemu_co_mutex_lock(&s->lock);
1426     ret = vmdk_write(bs, sector_num, buf, nb_sectors, false, false);
1427     qemu_co_mutex_unlock(&s->lock);
1428     return ret;
1429 }
1430 
1431 static int coroutine_fn vmdk_co_write_zeroes(BlockDriverState *bs,
1432                                              int64_t sector_num,
1433                                              int nb_sectors,
1434                                              BdrvRequestFlags flags)
1435 {
1436     int ret;
1437     BDRVVmdkState *s = bs->opaque;
1438     qemu_co_mutex_lock(&s->lock);
1439     /* write zeroes could fail if sectors not aligned to cluster, test it with
1440      * dry_run == true before really updating image */
1441     ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, true);
1442     if (!ret) {
1443         ret = vmdk_write(bs, sector_num, NULL, nb_sectors, true, false);
1444     }
1445     qemu_co_mutex_unlock(&s->lock);
1446     return ret;
1447 }
1448 
1449 static int vmdk_create_extent(const char *filename, int64_t filesize,
1450                               bool flat, bool compress, bool zeroed_grain)
1451 {
1452     int ret, i;
1453     int fd = 0;
1454     VMDK4Header header;
1455     uint32_t tmp, magic, grains, gd_size, gt_size, gt_count;
1456 
1457     fd = qemu_open(filename,
1458                    O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1459                    0644);
1460     if (fd < 0) {
1461         return -errno;
1462     }
1463     if (flat) {
1464         ret = ftruncate(fd, filesize);
1465         if (ret < 0) {
1466             ret = -errno;
1467         }
1468         goto exit;
1469     }
1470     magic = cpu_to_be32(VMDK4_MAGIC);
1471     memset(&header, 0, sizeof(header));
1472     header.version = zeroed_grain ? 2 : 1;
1473     header.flags = VMDK4_FLAG_RGD | VMDK4_FLAG_NL_DETECT
1474                    | (compress ? VMDK4_FLAG_COMPRESS | VMDK4_FLAG_MARKER : 0)
1475                    | (zeroed_grain ? VMDK4_FLAG_ZERO_GRAIN : 0);
1476     header.compressAlgorithm = compress ? VMDK4_COMPRESSION_DEFLATE : 0;
1477     header.capacity = filesize / 512;
1478     header.granularity = 128;
1479     header.num_gtes_per_gt = 512;
1480 
1481     grains = (filesize / 512 + header.granularity - 1) / header.granularity;
1482     gt_size = ((header.num_gtes_per_gt * sizeof(uint32_t)) + 511) >> 9;
1483     gt_count =
1484         (grains + header.num_gtes_per_gt - 1) / header.num_gtes_per_gt;
1485     gd_size = (gt_count * sizeof(uint32_t) + 511) >> 9;
1486 
1487     header.desc_offset = 1;
1488     header.desc_size = 20;
1489     header.rgd_offset = header.desc_offset + header.desc_size;
1490     header.gd_offset = header.rgd_offset + gd_size + (gt_size * gt_count);
1491     header.grain_offset =
1492        ((header.gd_offset + gd_size + (gt_size * gt_count) +
1493          header.granularity - 1) / header.granularity) *
1494         header.granularity;
1495     /* swap endianness for all header fields */
1496     header.version = cpu_to_le32(header.version);
1497     header.flags = cpu_to_le32(header.flags);
1498     header.capacity = cpu_to_le64(header.capacity);
1499     header.granularity = cpu_to_le64(header.granularity);
1500     header.num_gtes_per_gt = cpu_to_le32(header.num_gtes_per_gt);
1501     header.desc_offset = cpu_to_le64(header.desc_offset);
1502     header.desc_size = cpu_to_le64(header.desc_size);
1503     header.rgd_offset = cpu_to_le64(header.rgd_offset);
1504     header.gd_offset = cpu_to_le64(header.gd_offset);
1505     header.grain_offset = cpu_to_le64(header.grain_offset);
1506     header.compressAlgorithm = cpu_to_le16(header.compressAlgorithm);
1507 
1508     header.check_bytes[0] = 0xa;
1509     header.check_bytes[1] = 0x20;
1510     header.check_bytes[2] = 0xd;
1511     header.check_bytes[3] = 0xa;
1512 
1513     /* write all the data */
1514     ret = qemu_write_full(fd, &magic, sizeof(magic));
1515     if (ret != sizeof(magic)) {
1516         ret = -errno;
1517         goto exit;
1518     }
1519     ret = qemu_write_full(fd, &header, sizeof(header));
1520     if (ret != sizeof(header)) {
1521         ret = -errno;
1522         goto exit;
1523     }
1524 
1525     ret = ftruncate(fd, le64_to_cpu(header.grain_offset) << 9);
1526     if (ret < 0) {
1527         ret = -errno;
1528         goto exit;
1529     }
1530 
1531     /* write grain directory */
1532     lseek(fd, le64_to_cpu(header.rgd_offset) << 9, SEEK_SET);
1533     for (i = 0, tmp = le64_to_cpu(header.rgd_offset) + gd_size;
1534          i < gt_count; i++, tmp += gt_size) {
1535         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1536         if (ret != sizeof(tmp)) {
1537             ret = -errno;
1538             goto exit;
1539         }
1540     }
1541 
1542     /* write backup grain directory */
1543     lseek(fd, le64_to_cpu(header.gd_offset) << 9, SEEK_SET);
1544     for (i = 0, tmp = le64_to_cpu(header.gd_offset) + gd_size;
1545          i < gt_count; i++, tmp += gt_size) {
1546         ret = qemu_write_full(fd, &tmp, sizeof(tmp));
1547         if (ret != sizeof(tmp)) {
1548             ret = -errno;
1549             goto exit;
1550         }
1551     }
1552 
1553     ret = 0;
1554  exit:
1555     qemu_close(fd);
1556     return ret;
1557 }
1558 
1559 static int filename_decompose(const char *filename, char *path, char *prefix,
1560                               char *postfix, size_t buf_len, Error **errp)
1561 {
1562     const char *p, *q;
1563 
1564     if (filename == NULL || !strlen(filename)) {
1565         error_setg(errp, "No filename provided");
1566         return VMDK_ERROR;
1567     }
1568     p = strrchr(filename, '/');
1569     if (p == NULL) {
1570         p = strrchr(filename, '\\');
1571     }
1572     if (p == NULL) {
1573         p = strrchr(filename, ':');
1574     }
1575     if (p != NULL) {
1576         p++;
1577         if (p - filename >= buf_len) {
1578             return VMDK_ERROR;
1579         }
1580         pstrcpy(path, p - filename + 1, filename);
1581     } else {
1582         p = filename;
1583         path[0] = '\0';
1584     }
1585     q = strrchr(p, '.');
1586     if (q == NULL) {
1587         pstrcpy(prefix, buf_len, p);
1588         postfix[0] = '\0';
1589     } else {
1590         if (q - p >= buf_len) {
1591             return VMDK_ERROR;
1592         }
1593         pstrcpy(prefix, q - p + 1, p);
1594         pstrcpy(postfix, buf_len, q);
1595     }
1596     return VMDK_OK;
1597 }
1598 
1599 static int vmdk_create(const char *filename, QEMUOptionParameter *options,
1600                        Error **errp)
1601 {
1602     int fd, idx = 0;
1603     char *desc = NULL;
1604     int64_t total_size = 0, filesize;
1605     const char *adapter_type = NULL;
1606     const char *backing_file = NULL;
1607     const char *fmt = NULL;
1608     int flags = 0;
1609     int ret = 0;
1610     bool flat, split, compress;
1611     GString *ext_desc_lines;
1612     char path[PATH_MAX], prefix[PATH_MAX], postfix[PATH_MAX];
1613     const int64_t split_size = 0x80000000;  /* VMDK has constant split size */
1614     const char *desc_extent_line;
1615     char parent_desc_line[BUF_SIZE] = "";
1616     uint32_t parent_cid = 0xffffffff;
1617     uint32_t number_heads = 16;
1618     bool zeroed_grain = false;
1619     const char desc_template[] =
1620         "# Disk DescriptorFile\n"
1621         "version=1\n"
1622         "CID=%x\n"
1623         "parentCID=%x\n"
1624         "createType=\"%s\"\n"
1625         "%s"
1626         "\n"
1627         "# Extent description\n"
1628         "%s"
1629         "\n"
1630         "# The Disk Data Base\n"
1631         "#DDB\n"
1632         "\n"
1633         "ddb.virtualHWVersion = \"%d\"\n"
1634         "ddb.geometry.cylinders = \"%" PRId64 "\"\n"
1635         "ddb.geometry.heads = \"%d\"\n"
1636         "ddb.geometry.sectors = \"63\"\n"
1637         "ddb.adapterType = \"%s\"\n";
1638 
1639     ext_desc_lines = g_string_new(NULL);
1640 
1641     if (filename_decompose(filename, path, prefix, postfix, PATH_MAX, errp)) {
1642         ret = -EINVAL;
1643         goto exit;
1644     }
1645     /* Read out options */
1646     while (options && options->name) {
1647         if (!strcmp(options->name, BLOCK_OPT_SIZE)) {
1648             total_size = options->value.n;
1649         } else if (!strcmp(options->name, BLOCK_OPT_ADAPTER_TYPE)) {
1650             adapter_type = options->value.s;
1651         } else if (!strcmp(options->name, BLOCK_OPT_BACKING_FILE)) {
1652             backing_file = options->value.s;
1653         } else if (!strcmp(options->name, BLOCK_OPT_COMPAT6)) {
1654             flags |= options->value.n ? BLOCK_FLAG_COMPAT6 : 0;
1655         } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) {
1656             fmt = options->value.s;
1657         } else if (!strcmp(options->name, BLOCK_OPT_ZEROED_GRAIN)) {
1658             zeroed_grain |= options->value.n;
1659         }
1660         options++;
1661     }
1662     if (!adapter_type) {
1663         adapter_type = "ide";
1664     } else if (strcmp(adapter_type, "ide") &&
1665                strcmp(adapter_type, "buslogic") &&
1666                strcmp(adapter_type, "lsilogic") &&
1667                strcmp(adapter_type, "legacyESX")) {
1668         error_setg(errp, "Unknown adapter type: '%s'", adapter_type);
1669         ret = -EINVAL;
1670         goto exit;
1671     }
1672     if (strcmp(adapter_type, "ide") != 0) {
1673         /* that's the number of heads with which vmware operates when
1674            creating, exporting, etc. vmdk files with a non-ide adapter type */
1675         number_heads = 255;
1676     }
1677     if (!fmt) {
1678         /* Default format to monolithicSparse */
1679         fmt = "monolithicSparse";
1680     } else if (strcmp(fmt, "monolithicFlat") &&
1681                strcmp(fmt, "monolithicSparse") &&
1682                strcmp(fmt, "twoGbMaxExtentSparse") &&
1683                strcmp(fmt, "twoGbMaxExtentFlat") &&
1684                strcmp(fmt, "streamOptimized")) {
1685         error_setg(errp, "Unknown subformat: '%s'", fmt);
1686         ret = -EINVAL;
1687         goto exit;
1688     }
1689     split = !(strcmp(fmt, "twoGbMaxExtentFlat") &&
1690               strcmp(fmt, "twoGbMaxExtentSparse"));
1691     flat = !(strcmp(fmt, "monolithicFlat") &&
1692              strcmp(fmt, "twoGbMaxExtentFlat"));
1693     compress = !strcmp(fmt, "streamOptimized");
1694     if (flat) {
1695         desc_extent_line = "RW %lld FLAT \"%s\" 0\n";
1696     } else {
1697         desc_extent_line = "RW %lld SPARSE \"%s\"\n";
1698     }
1699     if (flat && backing_file) {
1700         error_setg(errp, "Flat image can't have backing file");
1701         ret = -ENOTSUP;
1702         goto exit;
1703     }
1704     if (flat && zeroed_grain) {
1705         error_setg(errp, "Flat image can't enable zeroed grain");
1706         ret = -ENOTSUP;
1707         goto exit;
1708     }
1709     if (backing_file) {
1710         BlockDriverState *bs = bdrv_new("");
1711         ret = bdrv_open(bs, backing_file, NULL, BDRV_O_NO_BACKING, NULL, errp);
1712         if (ret != 0) {
1713             bdrv_unref(bs);
1714             goto exit;
1715         }
1716         if (strcmp(bs->drv->format_name, "vmdk")) {
1717             bdrv_unref(bs);
1718             ret = -EINVAL;
1719             goto exit;
1720         }
1721         parent_cid = vmdk_read_cid(bs, 0);
1722         bdrv_unref(bs);
1723         snprintf(parent_desc_line, sizeof(parent_desc_line),
1724                 "parentFileNameHint=\"%s\"", backing_file);
1725     }
1726 
1727     /* Create extents */
1728     filesize = total_size;
1729     while (filesize > 0) {
1730         char desc_line[BUF_SIZE];
1731         char ext_filename[PATH_MAX];
1732         char desc_filename[PATH_MAX];
1733         int64_t size = filesize;
1734 
1735         if (split && size > split_size) {
1736             size = split_size;
1737         }
1738         if (split) {
1739             snprintf(desc_filename, sizeof(desc_filename), "%s-%c%03d%s",
1740                     prefix, flat ? 'f' : 's', ++idx, postfix);
1741         } else if (flat) {
1742             snprintf(desc_filename, sizeof(desc_filename), "%s-flat%s",
1743                     prefix, postfix);
1744         } else {
1745             snprintf(desc_filename, sizeof(desc_filename), "%s%s",
1746                     prefix, postfix);
1747         }
1748         snprintf(ext_filename, sizeof(ext_filename), "%s%s",
1749                 path, desc_filename);
1750 
1751         if (vmdk_create_extent(ext_filename, size,
1752                                flat, compress, zeroed_grain)) {
1753             ret = -EINVAL;
1754             goto exit;
1755         }
1756         filesize -= size;
1757 
1758         /* Format description line */
1759         snprintf(desc_line, sizeof(desc_line),
1760                     desc_extent_line, size / 512, desc_filename);
1761         g_string_append(ext_desc_lines, desc_line);
1762     }
1763     /* generate descriptor file */
1764     desc = g_strdup_printf(desc_template,
1765                            (unsigned int)time(NULL),
1766                            parent_cid,
1767                            fmt,
1768                            parent_desc_line,
1769                            ext_desc_lines->str,
1770                            (flags & BLOCK_FLAG_COMPAT6 ? 6 : 4),
1771                            total_size / (int64_t)(63 * number_heads * 512),
1772                            number_heads,
1773                            adapter_type);
1774     if (split || flat) {
1775         fd = qemu_open(filename,
1776                        O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
1777                        0644);
1778     } else {
1779         fd = qemu_open(filename,
1780                        O_WRONLY | O_BINARY | O_LARGEFILE,
1781                        0644);
1782     }
1783     if (fd < 0) {
1784         ret = -errno;
1785         goto exit;
1786     }
1787     /* the descriptor offset = 0x200 */
1788     if (!split && !flat && 0x200 != lseek(fd, 0x200, SEEK_SET)) {
1789         ret = -errno;
1790         goto close_exit;
1791     }
1792     ret = qemu_write_full(fd, desc, strlen(desc));
1793     if (ret != strlen(desc)) {
1794         ret = -errno;
1795         goto close_exit;
1796     }
1797     ret = 0;
1798 close_exit:
1799     qemu_close(fd);
1800 exit:
1801     g_free(desc);
1802     g_string_free(ext_desc_lines, true);
1803     return ret;
1804 }
1805 
1806 static void vmdk_close(BlockDriverState *bs)
1807 {
1808     BDRVVmdkState *s = bs->opaque;
1809 
1810     vmdk_free_extents(bs);
1811     g_free(s->create_type);
1812 
1813     migrate_del_blocker(s->migration_blocker);
1814     error_free(s->migration_blocker);
1815 }
1816 
1817 static coroutine_fn int vmdk_co_flush(BlockDriverState *bs)
1818 {
1819     BDRVVmdkState *s = bs->opaque;
1820     int i, err;
1821     int ret = 0;
1822 
1823     for (i = 0; i < s->num_extents; i++) {
1824         err = bdrv_co_flush(s->extents[i].file);
1825         if (err < 0) {
1826             ret = err;
1827         }
1828     }
1829     return ret;
1830 }
1831 
1832 static int64_t vmdk_get_allocated_file_size(BlockDriverState *bs)
1833 {
1834     int i;
1835     int64_t ret = 0;
1836     int64_t r;
1837     BDRVVmdkState *s = bs->opaque;
1838 
1839     ret = bdrv_get_allocated_file_size(bs->file);
1840     if (ret < 0) {
1841         return ret;
1842     }
1843     for (i = 0; i < s->num_extents; i++) {
1844         if (s->extents[i].file == bs->file) {
1845             continue;
1846         }
1847         r = bdrv_get_allocated_file_size(s->extents[i].file);
1848         if (r < 0) {
1849             return r;
1850         }
1851         ret += r;
1852     }
1853     return ret;
1854 }
1855 
1856 static int vmdk_has_zero_init(BlockDriverState *bs)
1857 {
1858     int i;
1859     BDRVVmdkState *s = bs->opaque;
1860 
1861     /* If has a flat extent and its underlying storage doesn't have zero init,
1862      * return 0. */
1863     for (i = 0; i < s->num_extents; i++) {
1864         if (s->extents[i].flat) {
1865             if (!bdrv_has_zero_init(s->extents[i].file)) {
1866                 return 0;
1867             }
1868         }
1869     }
1870     return 1;
1871 }
1872 
1873 static ImageInfo *vmdk_get_extent_info(VmdkExtent *extent)
1874 {
1875     ImageInfo *info = g_new0(ImageInfo, 1);
1876 
1877     *info = (ImageInfo){
1878         .filename         = g_strdup(extent->file->filename),
1879         .format           = g_strdup(extent->type),
1880         .virtual_size     = extent->sectors * BDRV_SECTOR_SIZE,
1881         .compressed       = extent->compressed,
1882         .has_compressed   = extent->compressed,
1883         .cluster_size     = extent->cluster_sectors * BDRV_SECTOR_SIZE,
1884         .has_cluster_size = !extent->flat,
1885     };
1886 
1887     return info;
1888 }
1889 
1890 static ImageInfoSpecific *vmdk_get_specific_info(BlockDriverState *bs)
1891 {
1892     int i;
1893     BDRVVmdkState *s = bs->opaque;
1894     ImageInfoSpecific *spec_info = g_new0(ImageInfoSpecific, 1);
1895     ImageInfoList **next;
1896 
1897     *spec_info = (ImageInfoSpecific){
1898         .kind = IMAGE_INFO_SPECIFIC_KIND_VMDK,
1899         {
1900             .vmdk = g_new0(ImageInfoSpecificVmdk, 1),
1901         },
1902     };
1903 
1904     *spec_info->vmdk = (ImageInfoSpecificVmdk) {
1905         .create_type = g_strdup(s->create_type),
1906         .cid = s->cid,
1907         .parent_cid = s->parent_cid,
1908     };
1909 
1910     next = &spec_info->vmdk->extents;
1911     for (i = 0; i < s->num_extents; i++) {
1912         *next = g_new0(ImageInfoList, 1);
1913         (*next)->value = vmdk_get_extent_info(&s->extents[i]);
1914         (*next)->next = NULL;
1915         next = &(*next)->next;
1916     }
1917 
1918     return spec_info;
1919 }
1920 
1921 static QEMUOptionParameter vmdk_create_options[] = {
1922     {
1923         .name = BLOCK_OPT_SIZE,
1924         .type = OPT_SIZE,
1925         .help = "Virtual disk size"
1926     },
1927     {
1928         .name = BLOCK_OPT_ADAPTER_TYPE,
1929         .type = OPT_STRING,
1930         .help = "Virtual adapter type, can be one of "
1931                 "ide (default), lsilogic, buslogic or legacyESX"
1932     },
1933     {
1934         .name = BLOCK_OPT_BACKING_FILE,
1935         .type = OPT_STRING,
1936         .help = "File name of a base image"
1937     },
1938     {
1939         .name = BLOCK_OPT_COMPAT6,
1940         .type = OPT_FLAG,
1941         .help = "VMDK version 6 image"
1942     },
1943     {
1944         .name = BLOCK_OPT_SUBFMT,
1945         .type = OPT_STRING,
1946         .help =
1947             "VMDK flat extent format, can be one of "
1948             "{monolithicSparse (default) | monolithicFlat | twoGbMaxExtentSparse | twoGbMaxExtentFlat | streamOptimized} "
1949     },
1950     {
1951         .name = BLOCK_OPT_ZEROED_GRAIN,
1952         .type = OPT_FLAG,
1953         .help = "Enable efficient zero writes using the zeroed-grain GTE feature"
1954     },
1955     { NULL }
1956 };
1957 
1958 static BlockDriver bdrv_vmdk = {
1959     .format_name                  = "vmdk",
1960     .instance_size                = sizeof(BDRVVmdkState),
1961     .bdrv_probe                   = vmdk_probe,
1962     .bdrv_open                    = vmdk_open,
1963     .bdrv_reopen_prepare          = vmdk_reopen_prepare,
1964     .bdrv_read                    = vmdk_co_read,
1965     .bdrv_write                   = vmdk_co_write,
1966     .bdrv_co_write_zeroes         = vmdk_co_write_zeroes,
1967     .bdrv_close                   = vmdk_close,
1968     .bdrv_create                  = vmdk_create,
1969     .bdrv_co_flush_to_disk        = vmdk_co_flush,
1970     .bdrv_co_get_block_status     = vmdk_co_get_block_status,
1971     .bdrv_get_allocated_file_size = vmdk_get_allocated_file_size,
1972     .bdrv_has_zero_init           = vmdk_has_zero_init,
1973     .bdrv_get_specific_info       = vmdk_get_specific_info,
1974 
1975     .create_options               = vmdk_create_options,
1976 };
1977 
1978 static void bdrv_vmdk_init(void)
1979 {
1980     bdrv_register(&bdrv_vmdk);
1981 }
1982 
1983 block_init(bdrv_vmdk_init);
1984