]>
Commit | Line | Data |
---|---|---|
e8d4e5ff JC |
1 | /* |
2 | * Block driver for Hyper-V VHDX Images | |
3 | * | |
4 | * Copyright (c) 2013 Red Hat, Inc., | |
5 | * | |
6 | * Authors: | |
7 | * Jeff Cody <jcody@redhat.com> | |
8 | * | |
6e9d290b | 9 | * This is based on the "VHDX Format Specification v1.00", published 8/25/2012 |
e8d4e5ff | 10 | * by Microsoft: |
6e9d290b | 11 | * https://www.microsoft.com/en-us/download/details.aspx?id=34750 |
e8d4e5ff JC |
12 | * |
13 | * This work is licensed under the terms of the GNU LGPL, version 2 or later. | |
14 | * See the COPYING.LIB file in the top-level directory. | |
15 | * | |
16 | */ | |
17 | ||
18 | #include "qemu-common.h" | |
19 | #include "block/block_int.h" | |
20 | #include "qemu/module.h" | |
21 | #include "qemu/crc32c.h" | |
22 | #include "block/vhdx.h" | |
5641bf40 | 23 | #include "migration/migration.h" |
e8d4e5ff | 24 | |
4f18b782 | 25 | #include <uuid/uuid.h> |
3412f7b1 JC |
26 | #include <glib.h> |
27 | ||
28 | /* Options for VHDX creation */ | |
29 | ||
30 | #define VHDX_BLOCK_OPT_LOG_SIZE "log_size" | |
31 | #define VHDX_BLOCK_OPT_BLOCK_SIZE "block_size" | |
32 | #define VHDX_BLOCK_OPT_ZERO "block_state_zero" | |
33 | ||
34 | typedef enum VHDXImageType { | |
35 | VHDX_TYPE_DYNAMIC = 0, | |
36 | VHDX_TYPE_FIXED, | |
37 | VHDX_TYPE_DIFFERENCING, /* Currently unsupported */ | |
38 | } VHDXImageType; | |
e8d4e5ff JC |
39 | |
40 | /* Several metadata and region table data entries are identified by | |
41 | * guids in a MS-specific GUID format. */ | |
42 | ||
43 | ||
44 | /* ------- Known Region Table GUIDs ---------------------- */ | |
45 | static const MSGUID bat_guid = { .data1 = 0x2dc27766, | |
46 | .data2 = 0xf623, | |
47 | .data3 = 0x4200, | |
48 | .data4 = { 0x9d, 0x64, 0x11, 0x5e, | |
49 | 0x9b, 0xfd, 0x4a, 0x08} }; | |
50 | ||
51 | static const MSGUID metadata_guid = { .data1 = 0x8b7ca206, | |
52 | .data2 = 0x4790, | |
53 | .data3 = 0x4b9a, | |
54 | .data4 = { 0xb8, 0xfe, 0x57, 0x5f, | |
55 | 0x05, 0x0f, 0x88, 0x6e} }; | |
56 | ||
57 | ||
58 | ||
59 | /* ------- Known Metadata Entry GUIDs ---------------------- */ | |
60 | static const MSGUID file_param_guid = { .data1 = 0xcaa16737, | |
61 | .data2 = 0xfa36, | |
62 | .data3 = 0x4d43, | |
63 | .data4 = { 0xb3, 0xb6, 0x33, 0xf0, | |
64 | 0xaa, 0x44, 0xe7, 0x6b} }; | |
65 | ||
66 | static const MSGUID virtual_size_guid = { .data1 = 0x2FA54224, | |
67 | .data2 = 0xcd1b, | |
68 | .data3 = 0x4876, | |
69 | .data4 = { 0xb2, 0x11, 0x5d, 0xbe, | |
70 | 0xd8, 0x3b, 0xf4, 0xb8} }; | |
71 | ||
72 | static const MSGUID page83_guid = { .data1 = 0xbeca12ab, | |
73 | .data2 = 0xb2e6, | |
74 | .data3 = 0x4523, | |
75 | .data4 = { 0x93, 0xef, 0xc3, 0x09, | |
76 | 0xe0, 0x00, 0xc7, 0x46} }; | |
77 | ||
78 | ||
79 | static const MSGUID phys_sector_guid = { .data1 = 0xcda348c7, | |
80 | .data2 = 0x445d, | |
81 | .data3 = 0x4471, | |
82 | .data4 = { 0x9c, 0xc9, 0xe9, 0x88, | |
83 | 0x52, 0x51, 0xc5, 0x56} }; | |
84 | ||
85 | static const MSGUID parent_locator_guid = { .data1 = 0xa8d35f2d, | |
86 | .data2 = 0xb30b, | |
87 | .data3 = 0x454d, | |
88 | .data4 = { 0xab, 0xf7, 0xd3, | |
89 | 0xd8, 0x48, 0x34, | |
90 | 0xab, 0x0c} }; | |
91 | ||
92 | static const MSGUID logical_sector_guid = { .data1 = 0x8141bf1d, | |
93 | .data2 = 0xa96f, | |
94 | .data3 = 0x4709, | |
95 | .data4 = { 0xba, 0x47, 0xf2, | |
96 | 0x33, 0xa8, 0xfa, | |
97 | 0xab, 0x5f} }; | |
98 | ||
99 | /* Each parent type must have a valid GUID; this is for parent images | |
100 | * of type 'VHDX'. If we were to allow e.g. a QCOW2 parent, we would | |
101 | * need to make up our own QCOW2 GUID type */ | |
102 | static const MSGUID parent_vhdx_guid = { .data1 = 0xb04aefb7, | |
103 | .data2 = 0xd19e, | |
104 | .data3 = 0x4a81, | |
105 | .data4 = { 0xb7, 0x89, 0x25, 0xb8, | |
106 | 0xe9, 0x44, 0x59, 0x13} }; | |
107 | ||
108 | ||
109 | #define META_FILE_PARAMETER_PRESENT 0x01 | |
110 | #define META_VIRTUAL_DISK_SIZE_PRESENT 0x02 | |
111 | #define META_PAGE_83_PRESENT 0x04 | |
112 | #define META_LOGICAL_SECTOR_SIZE_PRESENT 0x08 | |
113 | #define META_PHYS_SECTOR_SIZE_PRESENT 0x10 | |
114 | #define META_PARENT_LOCATOR_PRESENT 0x20 | |
115 | ||
116 | #define META_ALL_PRESENT \ | |
117 | (META_FILE_PARAMETER_PRESENT | META_VIRTUAL_DISK_SIZE_PRESENT | \ | |
118 | META_PAGE_83_PRESENT | META_LOGICAL_SECTOR_SIZE_PRESENT | \ | |
119 | META_PHYS_SECTOR_SIZE_PRESENT) | |
120 | ||
e8d4e5ff | 121 | |
059e2fbb JC |
122 | typedef struct VHDXSectorInfo { |
123 | uint32_t bat_idx; /* BAT entry index */ | |
124 | uint32_t sectors_avail; /* sectors available in payload block */ | |
125 | uint32_t bytes_left; /* bytes left in the block after data to r/w */ | |
126 | uint32_t bytes_avail; /* bytes available in payload block */ | |
127 | uint64_t file_offset; /* absolute offset in bytes, in file */ | |
128 | uint64_t block_offset; /* block offset, in bytes */ | |
129 | } VHDXSectorInfo; | |
130 | ||
4f18b782 JC |
131 | /* Calculates new checksum. |
132 | * | |
133 | * Zero is substituted during crc calculation for the original crc field | |
134 | * crc_offset: byte offset in buf of the buffer crc | |
135 | * buf: buffer pointer | |
136 | * size: size of buffer (must be > crc_offset+4) | |
137 | * | |
138 | * Note: The resulting checksum is in the CPU endianness, not necessarily | |
139 | * in the file format endianness (LE). Any header export to disk should | |
140 | * make sure that vhdx_header_le_export() is used to convert to the | |
141 | * correct endianness | |
142 | */ | |
143 | uint32_t vhdx_update_checksum(uint8_t *buf, size_t size, int crc_offset) | |
144 | { | |
145 | uint32_t crc; | |
146 | ||
147 | assert(buf != NULL); | |
148 | assert(size > (crc_offset + sizeof(crc))); | |
149 | ||
150 | memset(buf + crc_offset, 0, sizeof(crc)); | |
151 | crc = crc32c(0xffffffff, buf, size); | |
152 | memcpy(buf + crc_offset, &crc, sizeof(crc)); | |
153 | ||
154 | return crc; | |
155 | } | |
156 | ||
e8d4e5ff JC |
157 | uint32_t vhdx_checksum_calc(uint32_t crc, uint8_t *buf, size_t size, |
158 | int crc_offset) | |
159 | { | |
160 | uint32_t crc_new; | |
161 | uint32_t crc_orig; | |
162 | assert(buf != NULL); | |
163 | ||
164 | if (crc_offset > 0) { | |
165 | memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig)); | |
166 | memset(buf + crc_offset, 0, sizeof(crc_orig)); | |
167 | } | |
168 | ||
169 | crc_new = crc32c(crc, buf, size); | |
170 | if (crc_offset > 0) { | |
171 | memcpy(buf + crc_offset, &crc_orig, sizeof(crc_orig)); | |
172 | } | |
173 | ||
174 | return crc_new; | |
175 | } | |
176 | ||
177 | /* Validates the checksum of the buffer, with an in-place CRC. | |
178 | * | |
179 | * Zero is substituted during crc calculation for the original crc field, | |
180 | * and the crc field is restored afterwards. But the buffer will be modifed | |
181 | * during the calculation, so this may not be not suitable for multi-threaded | |
182 | * use. | |
183 | * | |
184 | * crc_offset: byte offset in buf of the buffer crc | |
185 | * buf: buffer pointer | |
186 | * size: size of buffer (must be > crc_offset+4) | |
187 | * | |
188 | * returns true if checksum is valid, false otherwise | |
189 | */ | |
190 | bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset) | |
191 | { | |
192 | uint32_t crc_orig; | |
193 | uint32_t crc; | |
194 | ||
195 | assert(buf != NULL); | |
196 | assert(size > (crc_offset + 4)); | |
197 | ||
198 | memcpy(&crc_orig, buf + crc_offset, sizeof(crc_orig)); | |
199 | crc_orig = le32_to_cpu(crc_orig); | |
200 | ||
201 | crc = vhdx_checksum_calc(0xffffffff, buf, size, crc_offset); | |
202 | ||
203 | return crc == crc_orig; | |
204 | } | |
205 | ||
206 | ||
4f18b782 JC |
207 | /* |
208 | * This generates a UUID that is compliant with the MS GUIDs used | |
209 | * in the VHDX spec (and elsewhere). | |
210 | */ | |
211 | void vhdx_guid_generate(MSGUID *guid) | |
212 | { | |
213 | uuid_t uuid; | |
214 | assert(guid != NULL); | |
215 | ||
216 | uuid_generate(uuid); | |
217 | memcpy(guid, uuid, sizeof(MSGUID)); | |
218 | } | |
219 | ||
1a848fd4 JC |
220 | /* Check for region overlaps inside the VHDX image */ |
221 | static int vhdx_region_check(BDRVVHDXState *s, uint64_t start, uint64_t length) | |
222 | { | |
223 | int ret = 0; | |
224 | uint64_t end; | |
225 | VHDXRegionEntry *r; | |
226 | ||
227 | end = start + length; | |
228 | QLIST_FOREACH(r, &s->regions, entries) { | |
229 | if (!((start >= r->end) || (end <= r->start))) { | |
230 | ret = -EINVAL; | |
231 | goto exit; | |
232 | } | |
233 | } | |
234 | ||
235 | exit: | |
236 | return ret; | |
237 | } | |
238 | ||
239 | /* Register a region for future checks */ | |
240 | static void vhdx_region_register(BDRVVHDXState *s, | |
241 | uint64_t start, uint64_t length) | |
242 | { | |
243 | VHDXRegionEntry *r; | |
244 | ||
245 | r = g_malloc0(sizeof(*r)); | |
246 | ||
247 | r->start = start; | |
248 | r->end = start + length; | |
249 | ||
250 | QLIST_INSERT_HEAD(&s->regions, r, entries); | |
251 | } | |
252 | ||
253 | /* Free all registered regions */ | |
254 | static void vhdx_region_unregister_all(BDRVVHDXState *s) | |
255 | { | |
256 | VHDXRegionEntry *r, *r_next; | |
257 | ||
258 | QLIST_FOREACH_SAFE(r, &s->regions, entries, r_next) { | |
259 | QLIST_REMOVE(r, entries); | |
260 | g_free(r); | |
261 | } | |
262 | } | |
263 | ||
1e74a971 JC |
264 | static void vhdx_set_shift_bits(BDRVVHDXState *s) |
265 | { | |
266 | s->logical_sector_size_bits = 31 - clz32(s->logical_sector_size); | |
267 | s->sectors_per_block_bits = 31 - clz32(s->sectors_per_block); | |
268 | s->chunk_ratio_bits = 63 - clz64(s->chunk_ratio); | |
269 | s->block_size_bits = 31 - clz32(s->block_size); | |
270 | } | |
271 | ||
e8d4e5ff JC |
272 | /* |
273 | * Per the MS VHDX Specification, for every VHDX file: | |
274 | * - The header section is fixed size - 1 MB | |
275 | * - The header section is always the first "object" | |
276 | * - The first 64KB of the header is the File Identifier | |
277 | * - The first uint64 (8 bytes) is the VHDX Signature ("vhdxfile") | |
278 | * - The following 512 bytes constitute a UTF-16 string identifiying the | |
279 | * software that created the file, and is optional and diagnostic only. | |
280 | * | |
281 | * Therefore, we probe by looking for the vhdxfile signature "vhdxfile" | |
282 | */ | |
283 | static int vhdx_probe(const uint8_t *buf, int buf_size, const char *filename) | |
284 | { | |
285 | if (buf_size >= 8 && !memcmp(buf, "vhdxfile", 8)) { | |
286 | return 100; | |
287 | } | |
288 | return 0; | |
289 | } | |
290 | ||
1e74a971 JC |
291 | /* |
292 | * Writes the header to the specified offset. | |
293 | * | |
294 | * This will optionally read in buffer data from disk (otherwise zero-fill), | |
295 | * and then update the header checksum. Header is converted to proper | |
296 | * endianness before being written to the specified file offset | |
297 | */ | |
298 | static int vhdx_write_header(BlockDriverState *bs_file, VHDXHeader *hdr, | |
299 | uint64_t offset, bool read) | |
300 | { | |
301 | uint8_t *buffer = NULL; | |
302 | int ret; | |
303 | VHDXHeader header_le; | |
304 | ||
305 | assert(bs_file != NULL); | |
306 | assert(hdr != NULL); | |
307 | ||
308 | /* the header checksum is not over just the packed size of VHDXHeader, | |
309 | * but rather over the entire 'reserved' range for the header, which is | |
310 | * 4KB (VHDX_HEADER_SIZE). */ | |
311 | ||
312 | buffer = qemu_blockalign(bs_file, VHDX_HEADER_SIZE); | |
313 | if (read) { | |
314 | /* if true, we can't assume the extra reserved bytes are 0 */ | |
315 | ret = bdrv_pread(bs_file, offset, buffer, VHDX_HEADER_SIZE); | |
316 | if (ret < 0) { | |
317 | goto exit; | |
318 | } | |
319 | } else { | |
320 | memset(buffer, 0, VHDX_HEADER_SIZE); | |
321 | } | |
322 | ||
323 | /* overwrite the actual VHDXHeader portion */ | |
324 | memcpy(buffer, hdr, sizeof(VHDXHeader)); | |
325 | hdr->checksum = vhdx_update_checksum(buffer, VHDX_HEADER_SIZE, | |
326 | offsetof(VHDXHeader, checksum)); | |
327 | vhdx_header_le_export(hdr, &header_le); | |
328 | ret = bdrv_pwrite_sync(bs_file, offset, &header_le, sizeof(VHDXHeader)); | |
329 | ||
330 | exit: | |
331 | qemu_vfree(buffer); | |
332 | return ret; | |
333 | } | |
334 | ||
4f18b782 JC |
335 | /* Update the VHDX headers |
336 | * | |
337 | * This follows the VHDX spec procedures for header updates. | |
338 | * | |
339 | * - non-current header is updated with largest sequence number | |
340 | */ | |
341 | static int vhdx_update_header(BlockDriverState *bs, BDRVVHDXState *s, | |
c3906c5e | 342 | bool generate_data_write_guid, MSGUID *log_guid) |
4f18b782 JC |
343 | { |
344 | int ret = 0; | |
345 | int hdr_idx = 0; | |
346 | uint64_t header_offset = VHDX_HEADER1_OFFSET; | |
347 | ||
348 | VHDXHeader *active_header; | |
349 | VHDXHeader *inactive_header; | |
4f18b782 JC |
350 | |
351 | /* operate on the non-current header */ | |
352 | if (s->curr_header == 0) { | |
353 | hdr_idx = 1; | |
354 | header_offset = VHDX_HEADER2_OFFSET; | |
355 | } | |
356 | ||
357 | active_header = s->headers[s->curr_header]; | |
358 | inactive_header = s->headers[hdr_idx]; | |
359 | ||
360 | inactive_header->sequence_number = active_header->sequence_number + 1; | |
361 | ||
362 | /* a new file guid must be generated before any file write, including | |
363 | * headers */ | |
364 | inactive_header->file_write_guid = s->session_guid; | |
365 | ||
366 | /* a new data guid only needs to be generated before any guest-visible | |
367 | * writes (i.e. something observable via virtual disk read) */ | |
368 | if (generate_data_write_guid) { | |
369 | vhdx_guid_generate(&inactive_header->data_write_guid); | |
370 | } | |
371 | ||
c3906c5e JC |
372 | /* update the log guid if present */ |
373 | if (log_guid) { | |
374 | inactive_header->log_guid = *log_guid; | |
375 | } | |
376 | ||
f50159fa | 377 | ret = vhdx_write_header(bs->file, inactive_header, header_offset, true); |
4f18b782 JC |
378 | if (ret < 0) { |
379 | goto exit; | |
380 | } | |
381 | s->curr_header = hdr_idx; | |
382 | ||
383 | exit: | |
4f18b782 JC |
384 | return ret; |
385 | } | |
386 | ||
387 | /* | |
388 | * The VHDX spec calls for header updates to be performed twice, so that both | |
389 | * the current and non-current header have valid info | |
390 | */ | |
c3906c5e JC |
391 | int vhdx_update_headers(BlockDriverState *bs, BDRVVHDXState *s, |
392 | bool generate_data_write_guid, MSGUID *log_guid) | |
4f18b782 JC |
393 | { |
394 | int ret; | |
395 | ||
c3906c5e | 396 | ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid); |
4f18b782 JC |
397 | if (ret < 0) { |
398 | return ret; | |
399 | } | |
c3906c5e | 400 | ret = vhdx_update_header(bs, s, generate_data_write_guid, log_guid); |
4f18b782 JC |
401 | return ret; |
402 | } | |
e8d4e5ff JC |
403 | |
404 | /* opens the specified header block from the VHDX file header section */ | |
6890aad4 PB |
405 | static void vhdx_parse_header(BlockDriverState *bs, BDRVVHDXState *s, |
406 | Error **errp) | |
e8d4e5ff | 407 | { |
6890aad4 | 408 | int ret; |
e8d4e5ff JC |
409 | VHDXHeader *header1; |
410 | VHDXHeader *header2; | |
411 | bool h1_valid = false; | |
412 | bool h2_valid = false; | |
413 | uint64_t h1_seq = 0; | |
414 | uint64_t h2_seq = 0; | |
415 | uint8_t *buffer; | |
416 | ||
6e9d290b | 417 | /* header1 & header2 are freed in vhdx_close() */ |
e8d4e5ff JC |
418 | header1 = qemu_blockalign(bs, sizeof(VHDXHeader)); |
419 | header2 = qemu_blockalign(bs, sizeof(VHDXHeader)); | |
420 | ||
421 | buffer = qemu_blockalign(bs, VHDX_HEADER_SIZE); | |
422 | ||
423 | s->headers[0] = header1; | |
424 | s->headers[1] = header2; | |
425 | ||
426 | /* We have to read the whole VHDX_HEADER_SIZE instead of | |
427 | * sizeof(VHDXHeader), because the checksum is over the whole | |
428 | * region */ | |
429 | ret = bdrv_pread(bs->file, VHDX_HEADER1_OFFSET, buffer, VHDX_HEADER_SIZE); | |
430 | if (ret < 0) { | |
431 | goto fail; | |
432 | } | |
433 | /* copy over just the relevant portion that we need */ | |
434 | memcpy(header1, buffer, sizeof(VHDXHeader)); | |
435 | vhdx_header_le_import(header1); | |
436 | ||
437 | if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4) && | |
438 | !memcmp(&header1->signature, "head", 4) && | |
439 | header1->version == 1) { | |
440 | h1_seq = header1->sequence_number; | |
441 | h1_valid = true; | |
442 | } | |
443 | ||
444 | ret = bdrv_pread(bs->file, VHDX_HEADER2_OFFSET, buffer, VHDX_HEADER_SIZE); | |
445 | if (ret < 0) { | |
446 | goto fail; | |
447 | } | |
448 | /* copy over just the relevant portion that we need */ | |
449 | memcpy(header2, buffer, sizeof(VHDXHeader)); | |
450 | vhdx_header_le_import(header2); | |
451 | ||
452 | if (vhdx_checksum_is_valid(buffer, VHDX_HEADER_SIZE, 4) && | |
453 | !memcmp(&header2->signature, "head", 4) && | |
454 | header2->version == 1) { | |
455 | h2_seq = header2->sequence_number; | |
456 | h2_valid = true; | |
457 | } | |
458 | ||
459 | /* If there is only 1 valid header (or no valid headers), we | |
460 | * don't care what the sequence numbers are */ | |
461 | if (h1_valid && !h2_valid) { | |
462 | s->curr_header = 0; | |
463 | } else if (!h1_valid && h2_valid) { | |
464 | s->curr_header = 1; | |
465 | } else if (!h1_valid && !h2_valid) { | |
e8d4e5ff JC |
466 | goto fail; |
467 | } else { | |
468 | /* If both headers are valid, then we choose the active one by the | |
469 | * highest sequence number. If the sequence numbers are equal, that is | |
470 | * invalid */ | |
471 | if (h1_seq > h2_seq) { | |
472 | s->curr_header = 0; | |
473 | } else if (h2_seq > h1_seq) { | |
474 | s->curr_header = 1; | |
475 | } else { | |
e8d4e5ff JC |
476 | goto fail; |
477 | } | |
478 | } | |
479 | ||
1a848fd4 JC |
480 | vhdx_region_register(s, s->headers[s->curr_header]->log_offset, |
481 | s->headers[s->curr_header]->log_length); | |
e8d4e5ff JC |
482 | goto exit; |
483 | ||
484 | fail: | |
6890aad4 | 485 | error_setg_errno(errp, -ret, "No valid VHDX header found"); |
e8d4e5ff JC |
486 | qemu_vfree(header1); |
487 | qemu_vfree(header2); | |
488 | s->headers[0] = NULL; | |
489 | s->headers[1] = NULL; | |
490 | exit: | |
491 | qemu_vfree(buffer); | |
e8d4e5ff JC |
492 | } |
493 | ||
494 | ||
495 | static int vhdx_open_region_tables(BlockDriverState *bs, BDRVVHDXState *s) | |
496 | { | |
497 | int ret = 0; | |
498 | uint8_t *buffer; | |
499 | int offset = 0; | |
500 | VHDXRegionTableEntry rt_entry; | |
501 | uint32_t i; | |
502 | bool bat_rt_found = false; | |
503 | bool metadata_rt_found = false; | |
504 | ||
505 | /* We have to read the whole 64KB block, because the crc32 is over the | |
506 | * whole block */ | |
507 | buffer = qemu_blockalign(bs, VHDX_HEADER_BLOCK_SIZE); | |
508 | ||
509 | ret = bdrv_pread(bs->file, VHDX_REGION_TABLE_OFFSET, buffer, | |
510 | VHDX_HEADER_BLOCK_SIZE); | |
511 | if (ret < 0) { | |
512 | goto fail; | |
513 | } | |
514 | memcpy(&s->rt, buffer, sizeof(s->rt)); | |
c325ee1d | 515 | vhdx_region_header_le_import(&s->rt); |
e8d4e5ff JC |
516 | offset += sizeof(s->rt); |
517 | ||
518 | if (!vhdx_checksum_is_valid(buffer, VHDX_HEADER_BLOCK_SIZE, 4) || | |
519 | memcmp(&s->rt.signature, "regi", 4)) { | |
520 | ret = -EINVAL; | |
521 | goto fail; | |
522 | } | |
523 | ||
524 | /* Per spec, maximum region table entry count is 2047 */ | |
525 | if (s->rt.entry_count > 2047) { | |
526 | ret = -EINVAL; | |
527 | goto fail; | |
528 | } | |
529 | ||
530 | for (i = 0; i < s->rt.entry_count; i++) { | |
531 | memcpy(&rt_entry, buffer + offset, sizeof(rt_entry)); | |
532 | offset += sizeof(rt_entry); | |
533 | ||
c325ee1d | 534 | vhdx_region_entry_le_import(&rt_entry); |
e8d4e5ff | 535 | |
1a848fd4 JC |
536 | /* check for region overlap between these entries, and any |
537 | * other memory regions in the file */ | |
538 | ret = vhdx_region_check(s, rt_entry.file_offset, rt_entry.length); | |
539 | if (ret < 0) { | |
540 | goto fail; | |
541 | } | |
542 | ||
543 | vhdx_region_register(s, rt_entry.file_offset, rt_entry.length); | |
544 | ||
e8d4e5ff JC |
545 | /* see if we recognize the entry */ |
546 | if (guid_eq(rt_entry.guid, bat_guid)) { | |
547 | /* must be unique; if we have already found it this is invalid */ | |
548 | if (bat_rt_found) { | |
549 | ret = -EINVAL; | |
550 | goto fail; | |
551 | } | |
552 | bat_rt_found = true; | |
553 | s->bat_rt = rt_entry; | |
554 | continue; | |
555 | } | |
556 | ||
557 | if (guid_eq(rt_entry.guid, metadata_guid)) { | |
558 | /* must be unique; if we have already found it this is invalid */ | |
559 | if (metadata_rt_found) { | |
560 | ret = -EINVAL; | |
561 | goto fail; | |
562 | } | |
563 | metadata_rt_found = true; | |
564 | s->metadata_rt = rt_entry; | |
565 | continue; | |
566 | } | |
567 | ||
568 | if (rt_entry.data_bits & VHDX_REGION_ENTRY_REQUIRED) { | |
569 | /* cannot read vhdx file - required region table entry that | |
570 | * we do not understand. per spec, we must fail to open */ | |
571 | ret = -ENOTSUP; | |
572 | goto fail; | |
573 | } | |
574 | } | |
1a848fd4 JC |
575 | |
576 | if (!bat_rt_found || !metadata_rt_found) { | |
577 | ret = -EINVAL; | |
578 | goto fail; | |
579 | } | |
580 | ||
e8d4e5ff JC |
581 | ret = 0; |
582 | ||
583 | fail: | |
584 | qemu_vfree(buffer); | |
585 | return ret; | |
586 | } | |
587 | ||
588 | ||
589 | ||
590 | /* Metadata initial parser | |
591 | * | |
592 | * This loads all the metadata entry fields. This may cause additional | |
593 | * fields to be processed (e.g. parent locator, etc..). | |
594 | * | |
595 | * There are 5 Metadata items that are always required: | |
596 | * - File Parameters (block size, has a parent) | |
597 | * - Virtual Disk Size (size, in bytes, of the virtual drive) | |
598 | * - Page 83 Data (scsi page 83 guid) | |
599 | * - Logical Sector Size (logical sector size in bytes, either 512 or | |
600 | * 4096. We only support 512 currently) | |
601 | * - Physical Sector Size (512 or 4096) | |
602 | * | |
603 | * Also, if the File Parameters indicate this is a differencing file, | |
604 | * we must also look for the Parent Locator metadata item. | |
605 | */ | |
606 | static int vhdx_parse_metadata(BlockDriverState *bs, BDRVVHDXState *s) | |
607 | { | |
608 | int ret = 0; | |
609 | uint8_t *buffer; | |
610 | int offset = 0; | |
611 | uint32_t i = 0; | |
612 | VHDXMetadataTableEntry md_entry; | |
613 | ||
614 | buffer = qemu_blockalign(bs, VHDX_METADATA_TABLE_MAX_SIZE); | |
615 | ||
616 | ret = bdrv_pread(bs->file, s->metadata_rt.file_offset, buffer, | |
617 | VHDX_METADATA_TABLE_MAX_SIZE); | |
618 | if (ret < 0) { | |
619 | goto exit; | |
620 | } | |
621 | memcpy(&s->metadata_hdr, buffer, sizeof(s->metadata_hdr)); | |
622 | offset += sizeof(s->metadata_hdr); | |
623 | ||
c325ee1d | 624 | vhdx_metadata_header_le_import(&s->metadata_hdr); |
e8d4e5ff JC |
625 | |
626 | if (memcmp(&s->metadata_hdr.signature, "metadata", 8)) { | |
627 | ret = -EINVAL; | |
628 | goto exit; | |
629 | } | |
630 | ||
631 | s->metadata_entries.present = 0; | |
632 | ||
633 | if ((s->metadata_hdr.entry_count * sizeof(md_entry)) > | |
634 | (VHDX_METADATA_TABLE_MAX_SIZE - offset)) { | |
635 | ret = -EINVAL; | |
636 | goto exit; | |
637 | } | |
638 | ||
639 | for (i = 0; i < s->metadata_hdr.entry_count; i++) { | |
640 | memcpy(&md_entry, buffer + offset, sizeof(md_entry)); | |
641 | offset += sizeof(md_entry); | |
642 | ||
c325ee1d | 643 | vhdx_metadata_entry_le_import(&md_entry); |
e8d4e5ff JC |
644 | |
645 | if (guid_eq(md_entry.item_id, file_param_guid)) { | |
646 | if (s->metadata_entries.present & META_FILE_PARAMETER_PRESENT) { | |
647 | ret = -EINVAL; | |
648 | goto exit; | |
649 | } | |
650 | s->metadata_entries.file_parameters_entry = md_entry; | |
651 | s->metadata_entries.present |= META_FILE_PARAMETER_PRESENT; | |
652 | continue; | |
653 | } | |
654 | ||
655 | if (guid_eq(md_entry.item_id, virtual_size_guid)) { | |
656 | if (s->metadata_entries.present & META_VIRTUAL_DISK_SIZE_PRESENT) { | |
657 | ret = -EINVAL; | |
658 | goto exit; | |
659 | } | |
660 | s->metadata_entries.virtual_disk_size_entry = md_entry; | |
661 | s->metadata_entries.present |= META_VIRTUAL_DISK_SIZE_PRESENT; | |
662 | continue; | |
663 | } | |
664 | ||
665 | if (guid_eq(md_entry.item_id, page83_guid)) { | |
666 | if (s->metadata_entries.present & META_PAGE_83_PRESENT) { | |
667 | ret = -EINVAL; | |
668 | goto exit; | |
669 | } | |
670 | s->metadata_entries.page83_data_entry = md_entry; | |
671 | s->metadata_entries.present |= META_PAGE_83_PRESENT; | |
672 | continue; | |
673 | } | |
674 | ||
675 | if (guid_eq(md_entry.item_id, logical_sector_guid)) { | |
676 | if (s->metadata_entries.present & | |
677 | META_LOGICAL_SECTOR_SIZE_PRESENT) { | |
678 | ret = -EINVAL; | |
679 | goto exit; | |
680 | } | |
681 | s->metadata_entries.logical_sector_size_entry = md_entry; | |
682 | s->metadata_entries.present |= META_LOGICAL_SECTOR_SIZE_PRESENT; | |
683 | continue; | |
684 | } | |
685 | ||
686 | if (guid_eq(md_entry.item_id, phys_sector_guid)) { | |
687 | if (s->metadata_entries.present & META_PHYS_SECTOR_SIZE_PRESENT) { | |
688 | ret = -EINVAL; | |
689 | goto exit; | |
690 | } | |
691 | s->metadata_entries.phys_sector_size_entry = md_entry; | |
692 | s->metadata_entries.present |= META_PHYS_SECTOR_SIZE_PRESENT; | |
693 | continue; | |
694 | } | |
695 | ||
696 | if (guid_eq(md_entry.item_id, parent_locator_guid)) { | |
697 | if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) { | |
698 | ret = -EINVAL; | |
699 | goto exit; | |
700 | } | |
701 | s->metadata_entries.parent_locator_entry = md_entry; | |
702 | s->metadata_entries.present |= META_PARENT_LOCATOR_PRESENT; | |
703 | continue; | |
704 | } | |
705 | ||
706 | if (md_entry.data_bits & VHDX_META_FLAGS_IS_REQUIRED) { | |
707 | /* cannot read vhdx file - required region table entry that | |
708 | * we do not understand. per spec, we must fail to open */ | |
709 | ret = -ENOTSUP; | |
710 | goto exit; | |
711 | } | |
712 | } | |
713 | ||
714 | if (s->metadata_entries.present != META_ALL_PRESENT) { | |
715 | ret = -ENOTSUP; | |
716 | goto exit; | |
717 | } | |
718 | ||
719 | ret = bdrv_pread(bs->file, | |
720 | s->metadata_entries.file_parameters_entry.offset | |
721 | + s->metadata_rt.file_offset, | |
722 | &s->params, | |
723 | sizeof(s->params)); | |
724 | ||
725 | if (ret < 0) { | |
726 | goto exit; | |
727 | } | |
728 | ||
729 | le32_to_cpus(&s->params.block_size); | |
730 | le32_to_cpus(&s->params.data_bits); | |
731 | ||
732 | ||
733 | /* We now have the file parameters, so we can tell if this is a | |
734 | * differencing file (i.e.. has_parent), is dynamic or fixed | |
735 | * sized (leave_blocks_allocated), and the block size */ | |
736 | ||
737 | /* The parent locator required iff the file parameters has_parent set */ | |
738 | if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { | |
739 | if (s->metadata_entries.present & META_PARENT_LOCATOR_PRESENT) { | |
740 | /* TODO: parse parent locator fields */ | |
741 | ret = -ENOTSUP; /* temp, until differencing files are supported */ | |
742 | goto exit; | |
743 | } else { | |
744 | /* if has_parent is set, but there is not parent locator present, | |
745 | * then that is an invalid combination */ | |
746 | ret = -EINVAL; | |
747 | goto exit; | |
748 | } | |
749 | } | |
750 | ||
751 | /* determine virtual disk size, logical sector size, | |
752 | * and phys sector size */ | |
753 | ||
754 | ret = bdrv_pread(bs->file, | |
755 | s->metadata_entries.virtual_disk_size_entry.offset | |
756 | + s->metadata_rt.file_offset, | |
757 | &s->virtual_disk_size, | |
758 | sizeof(uint64_t)); | |
759 | if (ret < 0) { | |
760 | goto exit; | |
761 | } | |
762 | ret = bdrv_pread(bs->file, | |
763 | s->metadata_entries.logical_sector_size_entry.offset | |
764 | + s->metadata_rt.file_offset, | |
765 | &s->logical_sector_size, | |
766 | sizeof(uint32_t)); | |
767 | if (ret < 0) { | |
768 | goto exit; | |
769 | } | |
770 | ret = bdrv_pread(bs->file, | |
771 | s->metadata_entries.phys_sector_size_entry.offset | |
772 | + s->metadata_rt.file_offset, | |
773 | &s->physical_sector_size, | |
774 | sizeof(uint32_t)); | |
775 | if (ret < 0) { | |
776 | goto exit; | |
777 | } | |
778 | ||
779 | le64_to_cpus(&s->virtual_disk_size); | |
780 | le32_to_cpus(&s->logical_sector_size); | |
781 | le32_to_cpus(&s->physical_sector_size); | |
782 | ||
1d7678de JC |
783 | if (s->params.block_size < VHDX_BLOCK_SIZE_MIN || |
784 | s->params.block_size > VHDX_BLOCK_SIZE_MAX) { | |
e8d4e5ff JC |
785 | ret = -EINVAL; |
786 | goto exit; | |
787 | } | |
788 | ||
1d7678de JC |
789 | /* only 2 supported sector sizes */ |
790 | if (s->logical_sector_size != 512 && s->logical_sector_size != 4096) { | |
791 | ret = -EINVAL; | |
792 | goto exit; | |
793 | } | |
794 | ||
795 | /* Both block_size and sector_size are guaranteed powers of 2, below. | |
796 | Due to range checks above, s->sectors_per_block can never be < 256 */ | |
e8d4e5ff JC |
797 | s->sectors_per_block = s->params.block_size / s->logical_sector_size; |
798 | s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) * | |
799 | (uint64_t)s->logical_sector_size / | |
800 | (uint64_t)s->params.block_size; | |
801 | ||
802 | /* These values are ones we will want to use for division / multiplication | |
803 | * later on, and they are all guaranteed (per the spec) to be powers of 2, | |
804 | * so we can take advantage of that for shift operations during | |
805 | * reads/writes */ | |
806 | if (s->logical_sector_size & (s->logical_sector_size - 1)) { | |
807 | ret = -EINVAL; | |
808 | goto exit; | |
809 | } | |
810 | if (s->sectors_per_block & (s->sectors_per_block - 1)) { | |
811 | ret = -EINVAL; | |
812 | goto exit; | |
813 | } | |
814 | if (s->chunk_ratio & (s->chunk_ratio - 1)) { | |
815 | ret = -EINVAL; | |
816 | goto exit; | |
817 | } | |
818 | s->block_size = s->params.block_size; | |
819 | if (s->block_size & (s->block_size - 1)) { | |
820 | ret = -EINVAL; | |
821 | goto exit; | |
822 | } | |
823 | ||
1e74a971 | 824 | vhdx_set_shift_bits(s); |
e8d4e5ff JC |
825 | |
826 | ret = 0; | |
827 | ||
828 | exit: | |
829 | qemu_vfree(buffer); | |
830 | return ret; | |
831 | } | |
832 | ||
1e74a971 JC |
833 | /* |
834 | * Calculate the number of BAT entries, including sector | |
835 | * bitmap entries. | |
836 | */ | |
837 | static void vhdx_calc_bat_entries(BDRVVHDXState *s) | |
838 | { | |
839 | uint32_t data_blocks_cnt, bitmap_blocks_cnt; | |
840 | ||
841 | data_blocks_cnt = s->virtual_disk_size >> s->block_size_bits; | |
842 | if (s->virtual_disk_size - (data_blocks_cnt << s->block_size_bits)) { | |
843 | data_blocks_cnt++; | |
844 | } | |
845 | bitmap_blocks_cnt = data_blocks_cnt >> s->chunk_ratio_bits; | |
846 | if (data_blocks_cnt - (bitmap_blocks_cnt << s->chunk_ratio_bits)) { | |
847 | bitmap_blocks_cnt++; | |
848 | } | |
849 | ||
850 | if (s->parent_entries) { | |
851 | s->bat_entries = bitmap_blocks_cnt * (s->chunk_ratio + 1); | |
852 | } else { | |
853 | s->bat_entries = data_blocks_cnt + | |
854 | ((data_blocks_cnt - 1) >> s->chunk_ratio_bits); | |
855 | } | |
856 | ||
857 | } | |
e8d4e5ff | 858 | |
c46415af JC |
859 | static void vhdx_close(BlockDriverState *bs) |
860 | { | |
861 | BDRVVHDXState *s = bs->opaque; | |
862 | qemu_vfree(s->headers[0]); | |
0a43a1b5 | 863 | s->headers[0] = NULL; |
c46415af | 864 | qemu_vfree(s->headers[1]); |
0a43a1b5 | 865 | s->headers[1] = NULL; |
c46415af | 866 | qemu_vfree(s->bat); |
0a43a1b5 | 867 | s->bat = NULL; |
c46415af | 868 | qemu_vfree(s->parent_entries); |
0a43a1b5 | 869 | s->parent_entries = NULL; |
c46415af JC |
870 | migrate_del_blocker(s->migration_blocker); |
871 | error_free(s->migration_blocker); | |
0a43a1b5 JC |
872 | qemu_vfree(s->log.hdr); |
873 | s->log.hdr = NULL; | |
1a848fd4 | 874 | vhdx_region_unregister_all(s); |
c46415af JC |
875 | } |
876 | ||
015a1036 HR |
877 | static int vhdx_open(BlockDriverState *bs, QDict *options, int flags, |
878 | Error **errp) | |
e8d4e5ff JC |
879 | { |
880 | BDRVVHDXState *s = bs->opaque; | |
881 | int ret = 0; | |
882 | uint32_t i; | |
883 | uint64_t signature; | |
6890aad4 | 884 | Error *local_err = NULL; |
e8d4e5ff JC |
885 | |
886 | s->bat = NULL; | |
c3906c5e | 887 | s->first_visible_write = true; |
e8d4e5ff JC |
888 | |
889 | qemu_co_mutex_init(&s->lock); | |
1a848fd4 | 890 | QLIST_INIT(&s->regions); |
e8d4e5ff JC |
891 | |
892 | /* validate the file signature */ | |
893 | ret = bdrv_pread(bs->file, 0, &signature, sizeof(uint64_t)); | |
894 | if (ret < 0) { | |
895 | goto fail; | |
896 | } | |
897 | if (memcmp(&signature, "vhdxfile", 8)) { | |
898 | ret = -EINVAL; | |
899 | goto fail; | |
900 | } | |
901 | ||
4f18b782 JC |
902 | /* This is used for any header updates, for the file_write_guid. |
903 | * The spec dictates that a new value should be used for the first | |
904 | * header update */ | |
905 | vhdx_guid_generate(&s->session_guid); | |
906 | ||
6890aad4 PB |
907 | vhdx_parse_header(bs, s, &local_err); |
908 | if (local_err != NULL) { | |
909 | error_propagate(errp, local_err); | |
910 | ret = -EINVAL; | |
e8d4e5ff JC |
911 | goto fail; |
912 | } | |
913 | ||
7e30e6a6 | 914 | ret = vhdx_parse_log(bs, s, &s->log_replayed_on_open, errp); |
0a43a1b5 | 915 | if (ret < 0) { |
e8d4e5ff JC |
916 | goto fail; |
917 | } | |
918 | ||
919 | ret = vhdx_open_region_tables(bs, s); | |
0a43a1b5 | 920 | if (ret < 0) { |
e8d4e5ff JC |
921 | goto fail; |
922 | } | |
923 | ||
924 | ret = vhdx_parse_metadata(bs, s); | |
0a43a1b5 | 925 | if (ret < 0) { |
e8d4e5ff JC |
926 | goto fail; |
927 | } | |
0a43a1b5 | 928 | |
e8d4e5ff JC |
929 | s->block_size = s->params.block_size; |
930 | ||
931 | /* the VHDX spec dictates that virtual_disk_size is always a multiple of | |
932 | * logical_sector_size */ | |
933 | bs->total_sectors = s->virtual_disk_size >> s->logical_sector_size_bits; | |
934 | ||
1e74a971 | 935 | vhdx_calc_bat_entries(s); |
e8d4e5ff JC |
936 | |
937 | s->bat_offset = s->bat_rt.file_offset; | |
938 | ||
939 | if (s->bat_entries > s->bat_rt.length / sizeof(VHDXBatEntry)) { | |
940 | /* BAT allocation is not large enough for all entries */ | |
941 | ret = -EINVAL; | |
942 | goto fail; | |
943 | } | |
944 | ||
6e9d290b | 945 | /* s->bat is freed in vhdx_close() */ |
e8d4e5ff JC |
946 | s->bat = qemu_blockalign(bs, s->bat_rt.length); |
947 | ||
948 | ret = bdrv_pread(bs->file, s->bat_offset, s->bat, s->bat_rt.length); | |
949 | if (ret < 0) { | |
950 | goto fail; | |
951 | } | |
952 | ||
1a848fd4 JC |
953 | uint64_t payblocks = s->chunk_ratio; |
954 | /* endian convert, and verify populated BAT field file offsets against | |
955 | * region table and log entries */ | |
e8d4e5ff JC |
956 | for (i = 0; i < s->bat_entries; i++) { |
957 | le64_to_cpus(&s->bat[i]); | |
1a848fd4 JC |
958 | if (payblocks--) { |
959 | /* payload bat entries */ | |
960 | if ((s->bat[i] & VHDX_BAT_STATE_BIT_MASK) == | |
d92aa883 | 961 | PAYLOAD_BLOCK_FULLY_PRESENT) { |
1a848fd4 JC |
962 | ret = vhdx_region_check(s, s->bat[i] & VHDX_BAT_FILE_OFF_MASK, |
963 | s->block_size); | |
964 | if (ret < 0) { | |
965 | goto fail; | |
966 | } | |
967 | } | |
968 | } else { | |
969 | payblocks = s->chunk_ratio; | |
970 | /* Once differencing files are supported, verify sector bitmap | |
971 | * blocks here */ | |
972 | } | |
e8d4e5ff JC |
973 | } |
974 | ||
975 | if (flags & BDRV_O_RDWR) { | |
c3906c5e | 976 | ret = vhdx_update_headers(bs, s, false, NULL); |
4f18b782 JC |
977 | if (ret < 0) { |
978 | goto fail; | |
979 | } | |
e8d4e5ff JC |
980 | } |
981 | ||
d92aa883 | 982 | /* TODO: differencing files */ |
e8d4e5ff | 983 | |
5641bf40 JC |
984 | /* Disable migration when VHDX images are used */ |
985 | error_set(&s->migration_blocker, | |
986 | QERR_BLOCK_FORMAT_FEATURE_NOT_SUPPORTED, | |
987 | "vhdx", bs->device_name, "live migration"); | |
988 | migrate_add_blocker(s->migration_blocker); | |
989 | ||
e8d4e5ff JC |
990 | return 0; |
991 | fail: | |
0a43a1b5 | 992 | vhdx_close(bs); |
e8d4e5ff JC |
993 | return ret; |
994 | } | |
995 | ||
996 | static int vhdx_reopen_prepare(BDRVReopenState *state, | |
997 | BlockReopenQueue *queue, Error **errp) | |
998 | { | |
999 | return 0; | |
1000 | } | |
1001 | ||
1002 | ||
059e2fbb JC |
1003 | /* |
1004 | * Perform sector to block offset translations, to get various | |
1005 | * sector and file offsets into the image. See VHDXSectorInfo | |
1006 | */ | |
1007 | static void vhdx_block_translate(BDRVVHDXState *s, int64_t sector_num, | |
1008 | int nb_sectors, VHDXSectorInfo *sinfo) | |
1009 | { | |
1010 | uint32_t block_offset; | |
1011 | ||
1012 | sinfo->bat_idx = sector_num >> s->sectors_per_block_bits; | |
1013 | /* effectively a modulo - this gives us the offset into the block | |
1014 | * (in sector sizes) for our sector number */ | |
1015 | block_offset = sector_num - (sinfo->bat_idx << s->sectors_per_block_bits); | |
1016 | /* the chunk ratio gives us the interleaving of the sector | |
1017 | * bitmaps, so we need to advance our page block index by the | |
1018 | * sector bitmaps entry number */ | |
1019 | sinfo->bat_idx += sinfo->bat_idx >> s->chunk_ratio_bits; | |
1020 | ||
1021 | /* the number of sectors we can read/write in this cycle */ | |
1022 | sinfo->sectors_avail = s->sectors_per_block - block_offset; | |
1023 | ||
1024 | sinfo->bytes_left = sinfo->sectors_avail << s->logical_sector_size_bits; | |
1025 | ||
1026 | if (sinfo->sectors_avail > nb_sectors) { | |
1027 | sinfo->sectors_avail = nb_sectors; | |
1028 | } | |
1029 | ||
1030 | sinfo->bytes_avail = sinfo->sectors_avail << s->logical_sector_size_bits; | |
1031 | ||
0b7da092 | 1032 | sinfo->file_offset = s->bat[sinfo->bat_idx] & VHDX_BAT_FILE_OFF_MASK; |
059e2fbb JC |
1033 | |
1034 | sinfo->block_offset = block_offset << s->logical_sector_size_bits; | |
1035 | ||
1036 | /* The file offset must be past the header section, so must be > 0 */ | |
1037 | if (sinfo->file_offset == 0) { | |
1038 | return; | |
1039 | } | |
1040 | ||
1041 | /* block offset is the offset in vhdx logical sectors, in | |
1042 | * the payload data block. Convert that to a byte offset | |
1043 | * in the block, and add in the payload data block offset | |
1044 | * in the file, in bytes, to get the final read address */ | |
1045 | ||
059e2fbb JC |
1046 | sinfo->file_offset += sinfo->block_offset; |
1047 | } | |
1048 | ||
1049 | ||
97b00e28 PB |
1050 | static int vhdx_get_info(BlockDriverState *bs, BlockDriverInfo *bdi) |
1051 | { | |
1052 | BDRVVHDXState *s = bs->opaque; | |
1053 | ||
1054 | bdi->cluster_size = s->block_size; | |
1055 | ||
95de6d70 PB |
1056 | bdi->unallocated_blocks_are_zero = |
1057 | (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) == 0; | |
1058 | ||
97b00e28 PB |
1059 | return 0; |
1060 | } | |
1061 | ||
059e2fbb | 1062 | |
e8d4e5ff JC |
1063 | static coroutine_fn int vhdx_co_readv(BlockDriverState *bs, int64_t sector_num, |
1064 | int nb_sectors, QEMUIOVector *qiov) | |
1065 | { | |
059e2fbb JC |
1066 | BDRVVHDXState *s = bs->opaque; |
1067 | int ret = 0; | |
1068 | VHDXSectorInfo sinfo; | |
1069 | uint64_t bytes_done = 0; | |
1070 | QEMUIOVector hd_qiov; | |
1071 | ||
1072 | qemu_iovec_init(&hd_qiov, qiov->niov); | |
1073 | ||
1074 | qemu_co_mutex_lock(&s->lock); | |
1075 | ||
1076 | while (nb_sectors > 0) { | |
1077 | /* We are a differencing file, so we need to inspect the sector bitmap | |
1078 | * to see if we have the data or not */ | |
1079 | if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { | |
1080 | /* not supported yet */ | |
1081 | ret = -ENOTSUP; | |
1082 | goto exit; | |
1083 | } else { | |
1084 | vhdx_block_translate(s, sector_num, nb_sectors, &sinfo); | |
1085 | ||
1086 | qemu_iovec_reset(&hd_qiov); | |
1087 | qemu_iovec_concat(&hd_qiov, qiov, bytes_done, sinfo.bytes_avail); | |
1088 | ||
1089 | /* check the payload block state */ | |
1090 | switch (s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK) { | |
1091 | case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */ | |
1092 | case PAYLOAD_BLOCK_UNDEFINED: /* fall through */ | |
1093 | case PAYLOAD_BLOCK_UNMAPPED: /* fall through */ | |
1094 | case PAYLOAD_BLOCK_ZERO: | |
1095 | /* return zero */ | |
1096 | qemu_iovec_memset(&hd_qiov, 0, 0, sinfo.bytes_avail); | |
1097 | break; | |
d92aa883 | 1098 | case PAYLOAD_BLOCK_FULLY_PRESENT: |
059e2fbb JC |
1099 | qemu_co_mutex_unlock(&s->lock); |
1100 | ret = bdrv_co_readv(bs->file, | |
1101 | sinfo.file_offset >> BDRV_SECTOR_BITS, | |
1102 | sinfo.sectors_avail, &hd_qiov); | |
1103 | qemu_co_mutex_lock(&s->lock); | |
1104 | if (ret < 0) { | |
1105 | goto exit; | |
1106 | } | |
1107 | break; | |
1108 | case PAYLOAD_BLOCK_PARTIALLY_PRESENT: | |
1109 | /* we don't yet support difference files, fall through | |
1110 | * to error */ | |
1111 | default: | |
1112 | ret = -EIO; | |
1113 | goto exit; | |
1114 | break; | |
1115 | } | |
1116 | nb_sectors -= sinfo.sectors_avail; | |
1117 | sector_num += sinfo.sectors_avail; | |
1118 | bytes_done += sinfo.bytes_avail; | |
1119 | } | |
1120 | } | |
1121 | ret = 0; | |
1122 | exit: | |
1123 | qemu_co_mutex_unlock(&s->lock); | |
1124 | qemu_iovec_destroy(&hd_qiov); | |
1125 | return ret; | |
e8d4e5ff JC |
1126 | } |
1127 | ||
d92aa883 JC |
1128 | /* |
1129 | * Allocate a new payload block at the end of the file. | |
1130 | * | |
1131 | * Allocation will happen at 1MB alignment inside the file | |
1132 | * | |
1133 | * Returns the file offset start of the new payload block | |
1134 | */ | |
1135 | static int vhdx_allocate_block(BlockDriverState *bs, BDRVVHDXState *s, | |
1136 | uint64_t *new_offset) | |
1137 | { | |
1138 | *new_offset = bdrv_getlength(bs->file); | |
1139 | ||
1140 | /* per the spec, the address for a block is in units of 1MB */ | |
1141 | *new_offset = ROUND_UP(*new_offset, 1024 * 1024); | |
1142 | ||
1143 | return bdrv_truncate(bs->file, *new_offset + s->block_size); | |
1144 | } | |
1145 | ||
1146 | /* | |
1147 | * Update the BAT table entry with the new file offset, and the new entry | |
1148 | * state */ | |
1149 | static void vhdx_update_bat_table_entry(BlockDriverState *bs, BDRVVHDXState *s, | |
1150 | VHDXSectorInfo *sinfo, | |
1151 | uint64_t *bat_entry_le, | |
1152 | uint64_t *bat_offset, int state) | |
1153 | { | |
1154 | /* The BAT entry is a uint64, with 44 bits for the file offset in units of | |
1155 | * 1MB, and 3 bits for the block state. */ | |
0b7da092 | 1156 | s->bat[sinfo->bat_idx] = sinfo->file_offset; |
e8d4e5ff | 1157 | |
d92aa883 JC |
1158 | s->bat[sinfo->bat_idx] |= state & VHDX_BAT_STATE_BIT_MASK; |
1159 | ||
1160 | *bat_entry_le = cpu_to_le64(s->bat[sinfo->bat_idx]); | |
1161 | *bat_offset = s->bat_offset + sinfo->bat_idx * sizeof(VHDXBatEntry); | |
1162 | ||
1163 | } | |
e8d4e5ff | 1164 | |
c3906c5e JC |
1165 | /* Per the spec, on the first write of guest-visible data to the file the |
1166 | * data write guid must be updated in the header */ | |
1167 | int vhdx_user_visible_write(BlockDriverState *bs, BDRVVHDXState *s) | |
1168 | { | |
1169 | int ret = 0; | |
1170 | if (s->first_visible_write) { | |
1171 | s->first_visible_write = false; | |
1172 | ret = vhdx_update_headers(bs, s, true, NULL); | |
1173 | } | |
1174 | return ret; | |
1175 | } | |
1176 | ||
e8d4e5ff JC |
1177 | static coroutine_fn int vhdx_co_writev(BlockDriverState *bs, int64_t sector_num, |
1178 | int nb_sectors, QEMUIOVector *qiov) | |
1179 | { | |
d92aa883 JC |
1180 | int ret = -ENOTSUP; |
1181 | BDRVVHDXState *s = bs->opaque; | |
1182 | VHDXSectorInfo sinfo; | |
1183 | uint64_t bytes_done = 0; | |
1184 | uint64_t bat_entry = 0; | |
1185 | uint64_t bat_entry_offset = 0; | |
1186 | QEMUIOVector hd_qiov; | |
1187 | struct iovec iov1 = { 0 }; | |
1188 | struct iovec iov2 = { 0 }; | |
1189 | int sectors_to_write; | |
1190 | int bat_state; | |
1191 | uint64_t bat_prior_offset = 0; | |
1192 | bool bat_update = false; | |
1193 | ||
1194 | qemu_iovec_init(&hd_qiov, qiov->niov); | |
1195 | ||
1196 | qemu_co_mutex_lock(&s->lock); | |
1197 | ||
1198 | ret = vhdx_user_visible_write(bs, s); | |
1199 | if (ret < 0) { | |
1200 | goto exit; | |
1201 | } | |
1202 | ||
1203 | while (nb_sectors > 0) { | |
1204 | bool use_zero_buffers = false; | |
1205 | bat_update = false; | |
1206 | if (s->params.data_bits & VHDX_PARAMS_HAS_PARENT) { | |
1207 | /* not supported yet */ | |
1208 | ret = -ENOTSUP; | |
1209 | goto exit; | |
1210 | } else { | |
1211 | vhdx_block_translate(s, sector_num, nb_sectors, &sinfo); | |
1212 | sectors_to_write = sinfo.sectors_avail; | |
1213 | ||
1214 | qemu_iovec_reset(&hd_qiov); | |
1215 | /* check the payload block state */ | |
1216 | bat_state = s->bat[sinfo.bat_idx] & VHDX_BAT_STATE_BIT_MASK; | |
1217 | switch (bat_state) { | |
1218 | case PAYLOAD_BLOCK_ZERO: | |
1219 | /* in this case, we need to preserve zero writes for | |
1220 | * data that is not part of this write, so we must pad | |
1221 | * the rest of the buffer to zeroes */ | |
1222 | ||
1223 | /* if we are on a posix system with ftruncate() that extends | |
1224 | * a file, then it is zero-filled for us. On Win32, the raw | |
1225 | * layer uses SetFilePointer and SetFileEnd, which does not | |
1226 | * zero fill AFAIK */ | |
1227 | ||
1228 | /* Queue another write of zero buffers if the underlying file | |
1229 | * does not zero-fill on file extension */ | |
1230 | ||
1231 | if (bdrv_has_zero_init(bs->file) == 0) { | |
1232 | use_zero_buffers = true; | |
1233 | ||
1234 | /* zero fill the front, if any */ | |
1235 | if (sinfo.block_offset) { | |
1236 | iov1.iov_len = sinfo.block_offset; | |
1237 | iov1.iov_base = qemu_blockalign(bs, iov1.iov_len); | |
1238 | memset(iov1.iov_base, 0, iov1.iov_len); | |
1239 | qemu_iovec_concat_iov(&hd_qiov, &iov1, 1, 0, | |
1240 | sinfo.block_offset); | |
1241 | sectors_to_write += iov1.iov_len >> BDRV_SECTOR_BITS; | |
1242 | } | |
1243 | ||
1244 | /* our actual data */ | |
1245 | qemu_iovec_concat(&hd_qiov, qiov, bytes_done, | |
1246 | sinfo.bytes_avail); | |
1247 | ||
1248 | /* zero fill the back, if any */ | |
1249 | if ((sinfo.bytes_avail - sinfo.block_offset) < | |
1250 | s->block_size) { | |
1251 | iov2.iov_len = s->block_size - | |
1252 | (sinfo.bytes_avail + sinfo.block_offset); | |
1253 | iov2.iov_base = qemu_blockalign(bs, iov2.iov_len); | |
1254 | memset(iov2.iov_base, 0, iov2.iov_len); | |
1255 | qemu_iovec_concat_iov(&hd_qiov, &iov2, 1, 0, | |
1256 | sinfo.block_offset); | |
1257 | sectors_to_write += iov2.iov_len >> BDRV_SECTOR_BITS; | |
1258 | } | |
1259 | } | |
1260 | ||
1261 | /* fall through */ | |
1262 | case PAYLOAD_BLOCK_NOT_PRESENT: /* fall through */ | |
1263 | case PAYLOAD_BLOCK_UNMAPPED: /* fall through */ | |
1264 | case PAYLOAD_BLOCK_UNDEFINED: /* fall through */ | |
1265 | bat_prior_offset = sinfo.file_offset; | |
1266 | ret = vhdx_allocate_block(bs, s, &sinfo.file_offset); | |
1267 | if (ret < 0) { | |
1268 | goto exit; | |
1269 | } | |
1270 | /* once we support differencing files, this may also be | |
1271 | * partially present */ | |
1272 | /* update block state to the newly specified state */ | |
1273 | vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry, | |
1274 | &bat_entry_offset, | |
1275 | PAYLOAD_BLOCK_FULLY_PRESENT); | |
1276 | bat_update = true; | |
1277 | /* since we just allocated a block, file_offset is the | |
1278 | * beginning of the payload block. It needs to be the | |
1279 | * write address, which includes the offset into the block */ | |
1280 | if (!use_zero_buffers) { | |
1281 | sinfo.file_offset += sinfo.block_offset; | |
1282 | } | |
1283 | /* fall through */ | |
1284 | case PAYLOAD_BLOCK_FULLY_PRESENT: | |
1285 | /* if the file offset address is in the header zone, | |
1286 | * there is a problem */ | |
1287 | if (sinfo.file_offset < (1024 * 1024)) { | |
1288 | ret = -EFAULT; | |
1289 | goto error_bat_restore; | |
1290 | } | |
1291 | ||
1292 | if (!use_zero_buffers) { | |
1293 | qemu_iovec_concat(&hd_qiov, qiov, bytes_done, | |
1294 | sinfo.bytes_avail); | |
1295 | } | |
1296 | /* block exists, so we can just overwrite it */ | |
1297 | qemu_co_mutex_unlock(&s->lock); | |
1298 | ret = bdrv_co_writev(bs->file, | |
1299 | sinfo.file_offset >> BDRV_SECTOR_BITS, | |
1300 | sectors_to_write, &hd_qiov); | |
1301 | qemu_co_mutex_lock(&s->lock); | |
1302 | if (ret < 0) { | |
1303 | goto error_bat_restore; | |
1304 | } | |
1305 | break; | |
1306 | case PAYLOAD_BLOCK_PARTIALLY_PRESENT: | |
1307 | /* we don't yet support difference files, fall through | |
1308 | * to error */ | |
1309 | default: | |
1310 | ret = -EIO; | |
1311 | goto exit; | |
1312 | break; | |
1313 | } | |
1314 | ||
1315 | if (bat_update) { | |
1316 | /* this will update the BAT entry into the log journal, and | |
1317 | * then flush the log journal out to disk */ | |
1318 | ret = vhdx_log_write_and_flush(bs, s, &bat_entry, | |
1319 | sizeof(VHDXBatEntry), | |
1320 | bat_entry_offset); | |
1321 | if (ret < 0) { | |
1322 | goto exit; | |
1323 | } | |
1324 | } | |
1325 | ||
1326 | nb_sectors -= sinfo.sectors_avail; | |
1327 | sector_num += sinfo.sectors_avail; | |
1328 | bytes_done += sinfo.bytes_avail; | |
1329 | ||
1330 | } | |
1331 | } | |
1332 | ||
1333 | goto exit; | |
1334 | ||
1335 | error_bat_restore: | |
1336 | if (bat_update) { | |
1337 | /* keep metadata in sync, and restore the bat entry state | |
1338 | * if error. */ | |
1339 | sinfo.file_offset = bat_prior_offset; | |
1340 | vhdx_update_bat_table_entry(bs, s, &sinfo, &bat_entry, | |
1341 | &bat_entry_offset, bat_state); | |
1342 | } | |
1343 | exit: | |
1344 | qemu_vfree(iov1.iov_base); | |
1345 | qemu_vfree(iov2.iov_base); | |
1346 | qemu_co_mutex_unlock(&s->lock); | |
1347 | qemu_iovec_destroy(&hd_qiov); | |
1348 | return ret; | |
e8d4e5ff JC |
1349 | } |
1350 | ||
1351 | ||
3412f7b1 JC |
1352 | |
1353 | /* | |
1354 | * Create VHDX Headers | |
1355 | * | |
1356 | * There are 2 headers, and the highest sequence number will represent | |
1357 | * the active header | |
1358 | */ | |
1359 | static int vhdx_create_new_headers(BlockDriverState *bs, uint64_t image_size, | |
1360 | uint32_t log_size) | |
1361 | { | |
1362 | int ret = 0; | |
1363 | VHDXHeader *hdr = NULL; | |
1364 | ||
1365 | hdr = g_malloc0(sizeof(VHDXHeader)); | |
1366 | ||
1367 | hdr->signature = VHDX_HEADER_SIGNATURE; | |
1368 | hdr->sequence_number = g_random_int(); | |
1369 | hdr->log_version = 0; | |
1370 | hdr->version = 1; | |
1371 | hdr->log_length = log_size; | |
1372 | hdr->log_offset = VHDX_HEADER_SECTION_END; | |
1373 | vhdx_guid_generate(&hdr->file_write_guid); | |
1374 | vhdx_guid_generate(&hdr->data_write_guid); | |
1375 | ||
1376 | ret = vhdx_write_header(bs, hdr, VHDX_HEADER1_OFFSET, false); | |
1377 | if (ret < 0) { | |
1378 | goto exit; | |
1379 | } | |
1380 | hdr->sequence_number++; | |
1381 | ret = vhdx_write_header(bs, hdr, VHDX_HEADER2_OFFSET, false); | |
1382 | if (ret < 0) { | |
1383 | goto exit; | |
1384 | } | |
1385 | ||
1386 | exit: | |
1387 | g_free(hdr); | |
1388 | return ret; | |
1389 | } | |
1390 | ||
1391 | ||
1392 | /* | |
1393 | * Create the Metadata entries. | |
1394 | * | |
1395 | * For more details on the entries, see section 3.5 (pg 29) in the | |
1396 | * VHDX 1.00 specification. | |
1397 | * | |
1398 | * We support 5 metadata entries (all required by spec): | |
1399 | * File Parameters, | |
1400 | * Virtual Disk Size, | |
1401 | * Page 83 Data, | |
1402 | * Logical Sector Size, | |
1403 | * Physical Sector Size | |
1404 | * | |
1405 | * The first 64KB of the Metadata section is reserved for the metadata | |
1406 | * header and entries; beyond that, the metadata items themselves reside. | |
1407 | */ | |
1408 | static int vhdx_create_new_metadata(BlockDriverState *bs, | |
1409 | uint64_t image_size, | |
1410 | uint32_t block_size, | |
1411 | uint32_t sector_size, | |
1412 | uint64_t metadata_offset, | |
1413 | VHDXImageType type) | |
1414 | { | |
1415 | int ret = 0; | |
1416 | uint32_t offset = 0; | |
1417 | void *buffer = NULL; | |
1418 | void *entry_buffer; | |
1419 | VHDXMetadataTableHeader *md_table;; | |
1420 | VHDXMetadataTableEntry *md_table_entry; | |
1421 | ||
1422 | /* Metadata entries */ | |
1423 | VHDXFileParameters *mt_file_params; | |
1424 | VHDXVirtualDiskSize *mt_virtual_size; | |
1425 | VHDXPage83Data *mt_page83; | |
1426 | VHDXVirtualDiskLogicalSectorSize *mt_log_sector_size; | |
1427 | VHDXVirtualDiskPhysicalSectorSize *mt_phys_sector_size; | |
1428 | ||
1429 | entry_buffer = g_malloc0(sizeof(VHDXFileParameters) + | |
1430 | sizeof(VHDXVirtualDiskSize) + | |
1431 | sizeof(VHDXPage83Data) + | |
1432 | sizeof(VHDXVirtualDiskLogicalSectorSize) + | |
1433 | sizeof(VHDXVirtualDiskPhysicalSectorSize)); | |
1434 | ||
1435 | mt_file_params = entry_buffer; | |
1436 | offset += sizeof(VHDXFileParameters); | |
1437 | mt_virtual_size = entry_buffer + offset; | |
1438 | offset += sizeof(VHDXVirtualDiskSize); | |
1439 | mt_page83 = entry_buffer + offset; | |
1440 | offset += sizeof(VHDXPage83Data); | |
1441 | mt_log_sector_size = entry_buffer + offset; | |
1442 | offset += sizeof(VHDXVirtualDiskLogicalSectorSize); | |
1443 | mt_phys_sector_size = entry_buffer + offset; | |
1444 | ||
1445 | mt_file_params->block_size = cpu_to_le32(block_size); | |
1446 | if (type == VHDX_TYPE_FIXED) { | |
1447 | mt_file_params->data_bits |= VHDX_PARAMS_LEAVE_BLOCKS_ALLOCED; | |
1448 | cpu_to_le32s(&mt_file_params->data_bits); | |
1449 | } | |
1450 | ||
1451 | vhdx_guid_generate(&mt_page83->page_83_data); | |
1452 | cpu_to_leguids(&mt_page83->page_83_data); | |
1453 | mt_virtual_size->virtual_disk_size = cpu_to_le64(image_size); | |
1454 | mt_log_sector_size->logical_sector_size = cpu_to_le32(sector_size); | |
1455 | mt_phys_sector_size->physical_sector_size = cpu_to_le32(sector_size); | |
1456 | ||
1457 | buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE); | |
1458 | md_table = buffer; | |
1459 | ||
1460 | md_table->signature = VHDX_METADATA_SIGNATURE; | |
1461 | md_table->entry_count = 5; | |
1462 | vhdx_metadata_header_le_export(md_table); | |
1463 | ||
1464 | ||
1465 | /* This will reference beyond the reserved table portion */ | |
1466 | offset = 64 * KiB; | |
1467 | ||
1468 | md_table_entry = buffer + sizeof(VHDXMetadataTableHeader); | |
1469 | ||
1470 | md_table_entry[0].item_id = file_param_guid; | |
1471 | md_table_entry[0].offset = offset; | |
1472 | md_table_entry[0].length = sizeof(VHDXFileParameters); | |
1473 | md_table_entry[0].data_bits |= VHDX_META_FLAGS_IS_REQUIRED; | |
1474 | offset += md_table_entry[0].length; | |
1475 | vhdx_metadata_entry_le_export(&md_table_entry[0]); | |
1476 | ||
1477 | md_table_entry[1].item_id = virtual_size_guid; | |
1478 | md_table_entry[1].offset = offset; | |
1479 | md_table_entry[1].length = sizeof(VHDXVirtualDiskSize); | |
1480 | md_table_entry[1].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | | |
1481 | VHDX_META_FLAGS_IS_VIRTUAL_DISK; | |
1482 | offset += md_table_entry[1].length; | |
1483 | vhdx_metadata_entry_le_export(&md_table_entry[1]); | |
1484 | ||
1485 | md_table_entry[2].item_id = page83_guid; | |
1486 | md_table_entry[2].offset = offset; | |
1487 | md_table_entry[2].length = sizeof(VHDXPage83Data); | |
1488 | md_table_entry[2].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | | |
1489 | VHDX_META_FLAGS_IS_VIRTUAL_DISK; | |
1490 | offset += md_table_entry[2].length; | |
1491 | vhdx_metadata_entry_le_export(&md_table_entry[2]); | |
1492 | ||
1493 | md_table_entry[3].item_id = logical_sector_guid; | |
1494 | md_table_entry[3].offset = offset; | |
1495 | md_table_entry[3].length = sizeof(VHDXVirtualDiskLogicalSectorSize); | |
1496 | md_table_entry[3].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | | |
1497 | VHDX_META_FLAGS_IS_VIRTUAL_DISK; | |
1498 | offset += md_table_entry[3].length; | |
1499 | vhdx_metadata_entry_le_export(&md_table_entry[3]); | |
1500 | ||
1501 | md_table_entry[4].item_id = phys_sector_guid; | |
1502 | md_table_entry[4].offset = offset; | |
1503 | md_table_entry[4].length = sizeof(VHDXVirtualDiskPhysicalSectorSize); | |
1504 | md_table_entry[4].data_bits |= VHDX_META_FLAGS_IS_REQUIRED | | |
1505 | VHDX_META_FLAGS_IS_VIRTUAL_DISK; | |
1506 | vhdx_metadata_entry_le_export(&md_table_entry[4]); | |
1507 | ||
1508 | ret = bdrv_pwrite(bs, metadata_offset, buffer, VHDX_HEADER_BLOCK_SIZE); | |
1509 | if (ret < 0) { | |
1510 | goto exit; | |
1511 | } | |
1512 | ||
1513 | ret = bdrv_pwrite(bs, metadata_offset + (64 * KiB), entry_buffer, | |
1514 | VHDX_HEADER_BLOCK_SIZE); | |
1515 | if (ret < 0) { | |
1516 | goto exit; | |
1517 | } | |
1518 | ||
1519 | ||
1520 | exit: | |
1521 | g_free(buffer); | |
1522 | g_free(entry_buffer); | |
1523 | return ret; | |
1524 | } | |
1525 | ||
1526 | /* This create the actual BAT itself. We currently only support | |
1527 | * 'Dynamic' and 'Fixed' image types. | |
1528 | * | |
1529 | * Dynamic images: default state of the BAT is all zeroes. | |
1530 | * | |
1531 | * Fixed images: default state of the BAT is fully populated, with | |
1532 | * file offsets and state PAYLOAD_BLOCK_FULLY_PRESENT. | |
1533 | */ | |
1534 | static int vhdx_create_bat(BlockDriverState *bs, BDRVVHDXState *s, | |
1535 | uint64_t image_size, VHDXImageType type, | |
1536 | bool use_zero_blocks, VHDXRegionTableEntry *rt_bat) | |
1537 | { | |
1538 | int ret = 0; | |
1539 | uint64_t data_file_offset; | |
1540 | uint64_t total_sectors = 0; | |
1541 | uint64_t sector_num = 0; | |
1542 | uint64_t unused; | |
1543 | int block_state; | |
1544 | VHDXSectorInfo sinfo; | |
1545 | ||
1546 | assert(s->bat == NULL); | |
1547 | ||
1548 | /* this gives a data start after BAT/bitmap entries, and well | |
1549 | * past any metadata entries (with a 4 MB buffer for future | |
1550 | * expansion */ | |
1551 | data_file_offset = rt_bat->file_offset + rt_bat->length + 5 * MiB; | |
1552 | total_sectors = image_size >> s->logical_sector_size_bits; | |
1553 | ||
1554 | if (type == VHDX_TYPE_DYNAMIC) { | |
1555 | /* All zeroes, so we can just extend the file - the end of the BAT | |
1556 | * is the furthest thing we have written yet */ | |
1557 | ret = bdrv_truncate(bs, data_file_offset); | |
1558 | if (ret < 0) { | |
1559 | goto exit; | |
1560 | } | |
1561 | } else if (type == VHDX_TYPE_FIXED) { | |
1562 | ret = bdrv_truncate(bs, data_file_offset + image_size); | |
1563 | if (ret < 0) { | |
1564 | goto exit; | |
1565 | } | |
1566 | } else { | |
1567 | ret = -ENOTSUP; | |
1568 | goto exit; | |
1569 | } | |
1570 | ||
1571 | if (type == VHDX_TYPE_FIXED || | |
1572 | use_zero_blocks || | |
1573 | bdrv_has_zero_init(bs) == 0) { | |
1574 | /* for a fixed file, the default BAT entry is not zero */ | |
1575 | s->bat = g_malloc0(rt_bat->length); | |
1576 | block_state = type == VHDX_TYPE_FIXED ? PAYLOAD_BLOCK_FULLY_PRESENT : | |
1577 | PAYLOAD_BLOCK_NOT_PRESENT; | |
1578 | block_state = use_zero_blocks ? PAYLOAD_BLOCK_ZERO : block_state; | |
1579 | /* fill the BAT by emulating sector writes of sectors_per_block size */ | |
1580 | while (sector_num < total_sectors) { | |
1581 | vhdx_block_translate(s, sector_num, s->sectors_per_block, &sinfo); | |
1582 | sinfo.file_offset = data_file_offset + | |
1583 | (sector_num << s->logical_sector_size_bits); | |
1584 | sinfo.file_offset = ROUND_UP(sinfo.file_offset, MiB); | |
1585 | vhdx_update_bat_table_entry(bs, s, &sinfo, &unused, &unused, | |
1586 | block_state); | |
1587 | cpu_to_le64s(&s->bat[sinfo.bat_idx]); | |
1588 | sector_num += s->sectors_per_block; | |
1589 | } | |
1590 | ret = bdrv_pwrite(bs, rt_bat->file_offset, s->bat, rt_bat->length); | |
1591 | if (ret < 0) { | |
1592 | goto exit; | |
1593 | } | |
1594 | } | |
1595 | ||
1596 | ||
1597 | ||
1598 | exit: | |
1599 | g_free(s->bat); | |
1600 | return ret; | |
1601 | } | |
1602 | ||
1603 | /* Creates the region table header, and region table entries. | |
1604 | * There are 2 supported region table entries: BAT, and Metadata/ | |
1605 | * | |
1606 | * As the calculations for the BAT region table are also needed | |
1607 | * to create the BAT itself, we will also cause the BAT to be | |
1608 | * created. | |
1609 | */ | |
1610 | static int vhdx_create_new_region_table(BlockDriverState *bs, | |
1611 | uint64_t image_size, | |
1612 | uint32_t block_size, | |
1613 | uint32_t sector_size, | |
1614 | uint32_t log_size, | |
1615 | bool use_zero_blocks, | |
1616 | VHDXImageType type, | |
1617 | uint64_t *metadata_offset) | |
1618 | { | |
1619 | int ret = 0; | |
1620 | uint32_t offset = 0; | |
1621 | void *buffer = NULL; | |
1622 | BDRVVHDXState *s = NULL; | |
1623 | VHDXRegionTableHeader *region_table; | |
1624 | VHDXRegionTableEntry *rt_bat; | |
1625 | VHDXRegionTableEntry *rt_metadata; | |
1626 | ||
1627 | assert(metadata_offset != NULL); | |
1628 | ||
1629 | /* Populate enough of the BDRVVHDXState to be able to use the | |
1630 | * pre-existing BAT calculation, translation, and update functions */ | |
1631 | s = g_malloc0(sizeof(BDRVVHDXState)); | |
1632 | ||
1633 | s->chunk_ratio = (VHDX_MAX_SECTORS_PER_BLOCK) * | |
1634 | (uint64_t) sector_size / (uint64_t) block_size; | |
1635 | ||
1636 | s->sectors_per_block = block_size / sector_size; | |
1637 | s->virtual_disk_size = image_size; | |
1638 | s->block_size = block_size; | |
1639 | s->logical_sector_size = sector_size; | |
1640 | ||
1641 | vhdx_set_shift_bits(s); | |
1642 | ||
1643 | vhdx_calc_bat_entries(s); | |
1644 | ||
1645 | /* At this point the VHDX state is populated enough for creation */ | |
1646 | ||
1647 | /* a single buffer is used so we can calculate the checksum over the | |
1648 | * entire 64KB block */ | |
1649 | buffer = g_malloc0(VHDX_HEADER_BLOCK_SIZE); | |
1650 | region_table = buffer; | |
1651 | offset += sizeof(VHDXRegionTableHeader); | |
1652 | rt_bat = buffer + offset; | |
1653 | offset += sizeof(VHDXRegionTableEntry); | |
1654 | rt_metadata = buffer + offset; | |
1655 | ||
1656 | region_table->signature = VHDX_REGION_SIGNATURE; | |
1657 | region_table->entry_count = 2; /* BAT and Metadata */ | |
1658 | ||
1659 | rt_bat->guid = bat_guid; | |
1660 | rt_bat->length = ROUND_UP(s->bat_entries * sizeof(VHDXBatEntry), MiB); | |
1661 | rt_bat->file_offset = ROUND_UP(VHDX_HEADER_SECTION_END + log_size, MiB); | |
1662 | s->bat_offset = rt_bat->file_offset; | |
1663 | ||
1664 | rt_metadata->guid = metadata_guid; | |
1665 | rt_metadata->file_offset = ROUND_UP(rt_bat->file_offset + rt_bat->length, | |
1666 | MiB); | |
1667 | rt_metadata->length = 1 * MiB; /* min size, and more than enough */ | |
1668 | *metadata_offset = rt_metadata->file_offset; | |
1669 | ||
1670 | vhdx_update_checksum(buffer, VHDX_HEADER_BLOCK_SIZE, | |
1671 | offsetof(VHDXRegionTableHeader, checksum)); | |
1672 | ||
1673 | ||
1674 | /* The region table gives us the data we need to create the BAT, | |
1675 | * so do that now */ | |
1676 | ret = vhdx_create_bat(bs, s, image_size, type, use_zero_blocks, rt_bat); | |
1677 | ||
1678 | /* Now write out the region headers to disk */ | |
1679 | vhdx_region_header_le_export(region_table); | |
1680 | vhdx_region_entry_le_export(rt_bat); | |
1681 | vhdx_region_entry_le_export(rt_metadata); | |
1682 | ||
1683 | ret = bdrv_pwrite(bs, VHDX_REGION_TABLE_OFFSET, buffer, | |
1684 | VHDX_HEADER_BLOCK_SIZE); | |
1685 | if (ret < 0) { | |
1686 | goto exit; | |
1687 | } | |
1688 | ||
1689 | ret = bdrv_pwrite(bs, VHDX_REGION_TABLE2_OFFSET, buffer, | |
1690 | VHDX_HEADER_BLOCK_SIZE); | |
1691 | if (ret < 0) { | |
1692 | goto exit; | |
1693 | } | |
1694 | ||
1695 | ||
1696 | exit: | |
1697 | g_free(s); | |
1698 | g_free(buffer); | |
1699 | return ret; | |
1700 | } | |
1701 | ||
1702 | /* We need to create the following elements: | |
1703 | * | |
1704 | * .-----------------------------------------------------------------. | |
1705 | * | (A) | (B) | (C) | (D) | (E) | | |
1706 | * | File ID | Header1 | Header 2 | Region Tbl 1 | Region Tbl 2 | | |
1707 | * | | | | | | | |
1708 | * .-----------------------------------------------------------------. | |
1709 | * 0 64KB 128KB 192KB 256KB 320KB | |
1710 | * | |
1711 | * | |
1712 | * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------. | |
1713 | * | (F) | (G) | (H) | | | |
1714 | * | Journal Log | BAT / Bitmap | Metadata | .... data ...... | | |
1715 | * | | | | | | |
1716 | * .---- ~ ----------- ~ ------------ ~ ---------------- ~ -----------. | |
1717 | * 1MB | |
1718 | */ | |
1719 | static int vhdx_create(const char *filename, QEMUOptionParameter *options, | |
1720 | Error **errp) | |
1721 | { | |
1722 | int ret = 0; | |
1723 | uint64_t image_size = (uint64_t) 2 * GiB; | |
1724 | uint32_t log_size = 1 * MiB; | |
1725 | uint32_t block_size = 0; | |
1726 | uint64_t signature; | |
1727 | uint64_t metadata_offset; | |
1728 | bool use_zero_blocks = false; | |
1729 | ||
1730 | gunichar2 *creator = NULL; | |
1731 | glong creator_items; | |
1732 | BlockDriverState *bs; | |
1733 | const char *type = NULL; | |
1734 | VHDXImageType image_type; | |
1735 | Error *local_err = NULL; | |
1736 | ||
1737 | while (options && options->name) { | |
1738 | if (!strcmp(options->name, BLOCK_OPT_SIZE)) { | |
1739 | image_size = options->value.n; | |
1740 | } else if (!strcmp(options->name, VHDX_BLOCK_OPT_LOG_SIZE)) { | |
1741 | log_size = options->value.n; | |
1742 | } else if (!strcmp(options->name, VHDX_BLOCK_OPT_BLOCK_SIZE)) { | |
1743 | block_size = options->value.n; | |
1744 | } else if (!strcmp(options->name, BLOCK_OPT_SUBFMT)) { | |
1745 | type = options->value.s; | |
1746 | } else if (!strcmp(options->name, VHDX_BLOCK_OPT_ZERO)) { | |
1747 | use_zero_blocks = options->value.n != 0; | |
1748 | } | |
1749 | options++; | |
1750 | } | |
1751 | ||
1752 | if (image_size > VHDX_MAX_IMAGE_SIZE) { | |
1753 | error_setg_errno(errp, EINVAL, "Image size too large; max of 64TB"); | |
1754 | ret = -EINVAL; | |
1755 | goto exit; | |
1756 | } | |
1757 | ||
1758 | if (type == NULL) { | |
1759 | type = "dynamic"; | |
1760 | } | |
1761 | ||
1762 | if (!strcmp(type, "dynamic")) { | |
1763 | image_type = VHDX_TYPE_DYNAMIC; | |
1764 | } else if (!strcmp(type, "fixed")) { | |
1765 | image_type = VHDX_TYPE_FIXED; | |
1766 | } else if (!strcmp(type, "differencing")) { | |
1767 | error_setg_errno(errp, ENOTSUP, | |
1768 | "Differencing files not yet supported"); | |
1769 | ret = -ENOTSUP; | |
1770 | goto exit; | |
1771 | } else { | |
1772 | ret = -EINVAL; | |
1773 | goto exit; | |
1774 | } | |
1775 | ||
1776 | /* These are pretty arbitrary, and mainly designed to keep the BAT | |
1777 | * size reasonable to load into RAM */ | |
1778 | if (block_size == 0) { | |
1779 | if (image_size > 32 * TiB) { | |
1780 | block_size = 64 * MiB; | |
1781 | } else if (image_size > (uint64_t) 100 * GiB) { | |
1782 | block_size = 32 * MiB; | |
1783 | } else if (image_size > 1 * GiB) { | |
1784 | block_size = 16 * MiB; | |
1785 | } else { | |
1786 | block_size = 8 * MiB; | |
1787 | } | |
1788 | } | |
1789 | ||
1790 | ||
1791 | /* make the log size close to what was specified, but must be | |
1792 | * min 1MB, and multiple of 1MB */ | |
1793 | log_size = ROUND_UP(log_size, MiB); | |
1794 | ||
1795 | block_size = ROUND_UP(block_size, MiB); | |
1796 | block_size = block_size > VHDX_BLOCK_SIZE_MAX ? VHDX_BLOCK_SIZE_MAX : | |
1797 | block_size; | |
1798 | ||
1799 | ret = bdrv_create_file(filename, options, &local_err); | |
1800 | if (ret < 0) { | |
1801 | error_propagate(errp, local_err); | |
1802 | goto exit; | |
1803 | } | |
1804 | ||
2e40134b HR |
1805 | bs = NULL; |
1806 | ret = bdrv_open(&bs, filename, NULL, NULL, BDRV_O_RDWR | BDRV_O_PROTOCOL, | |
1807 | NULL, &local_err); | |
3412f7b1 JC |
1808 | if (ret < 0) { |
1809 | error_propagate(errp, local_err); | |
1810 | goto exit; | |
1811 | } | |
1812 | ||
1813 | /* Create (A) */ | |
1814 | ||
1815 | /* The creator field is optional, but may be useful for | |
1816 | * debugging / diagnostics */ | |
1817 | creator = g_utf8_to_utf16("QEMU v" QEMU_VERSION, -1, NULL, | |
1818 | &creator_items, NULL); | |
1819 | signature = cpu_to_le64(VHDX_FILE_SIGNATURE); | |
f50159fa | 1820 | ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET, &signature, sizeof(signature)); |
3412f7b1 JC |
1821 | if (ret < 0) { |
1822 | goto delete_and_exit; | |
1823 | } | |
1824 | if (creator) { | |
f50159fa MA |
1825 | ret = bdrv_pwrite(bs, VHDX_FILE_ID_OFFSET + sizeof(signature), |
1826 | creator, creator_items * sizeof(gunichar2)); | |
3412f7b1 JC |
1827 | if (ret < 0) { |
1828 | goto delete_and_exit; | |
1829 | } | |
1830 | } | |
1831 | ||
1832 | ||
1833 | /* Creates (B),(C) */ | |
1834 | ret = vhdx_create_new_headers(bs, image_size, log_size); | |
1835 | if (ret < 0) { | |
1836 | goto delete_and_exit; | |
1837 | } | |
1838 | ||
1839 | /* Creates (D),(E),(G) explicitly. (F) created as by-product */ | |
1840 | ret = vhdx_create_new_region_table(bs, image_size, block_size, 512, | |
1841 | log_size, use_zero_blocks, image_type, | |
1842 | &metadata_offset); | |
1843 | if (ret < 0) { | |
1844 | goto delete_and_exit; | |
1845 | } | |
1846 | ||
1847 | /* Creates (H) */ | |
1848 | ret = vhdx_create_new_metadata(bs, image_size, block_size, 512, | |
1849 | metadata_offset, image_type); | |
1850 | if (ret < 0) { | |
1851 | goto delete_and_exit; | |
1852 | } | |
1853 | ||
1854 | ||
1855 | ||
1856 | delete_and_exit: | |
1857 | bdrv_unref(bs); | |
1858 | exit: | |
1859 | g_free(creator); | |
1860 | return ret; | |
1861 | } | |
1862 | ||
7e30e6a6 JC |
1863 | /* If opened r/w, the VHDX driver will automatically replay the log, |
1864 | * if one is present, inside the vhdx_open() call. | |
1865 | * | |
1866 | * If qemu-img check -r all is called, the image is automatically opened | |
1867 | * r/w and any log has already been replayed, so there is nothing (currently) | |
1868 | * for us to do here | |
1869 | */ | |
1870 | static int vhdx_check(BlockDriverState *bs, BdrvCheckResult *result, | |
1871 | BdrvCheckMode fix) | |
1872 | { | |
1873 | BDRVVHDXState *s = bs->opaque; | |
1874 | ||
1875 | if (s->log_replayed_on_open) { | |
1876 | result->corruptions_fixed++; | |
1877 | } | |
1878 | return 0; | |
1879 | } | |
1880 | ||
3412f7b1 JC |
1881 | static QEMUOptionParameter vhdx_create_options[] = { |
1882 | { | |
1883 | .name = BLOCK_OPT_SIZE, | |
1884 | .type = OPT_SIZE, | |
1885 | .help = "Virtual disk size; max of 64TB." | |
1886 | }, | |
1887 | { | |
1888 | .name = VHDX_BLOCK_OPT_LOG_SIZE, | |
1889 | .type = OPT_SIZE, | |
1890 | .value.n = 1 * MiB, | |
1891 | .help = "Log size; min 1MB." | |
1892 | }, | |
1893 | { | |
1894 | .name = VHDX_BLOCK_OPT_BLOCK_SIZE, | |
1895 | .type = OPT_SIZE, | |
1896 | .value.n = 0, | |
1897 | .help = "Block Size; min 1MB, max 256MB. " \ | |
1898 | "0 means auto-calculate based on image size." | |
1899 | }, | |
1900 | { | |
1901 | .name = BLOCK_OPT_SUBFMT, | |
1902 | .type = OPT_STRING, | |
1903 | .help = "VHDX format type, can be either 'dynamic' or 'fixed'. "\ | |
1904 | "Default is 'dynamic'." | |
1905 | }, | |
1906 | { | |
1907 | .name = VHDX_BLOCK_OPT_ZERO, | |
1908 | .type = OPT_FLAG, | |
1909 | .help = "Force use of payload blocks of type 'ZERO'. Non-standard." | |
1910 | }, | |
1911 | { NULL } | |
1912 | }; | |
1913 | ||
e8d4e5ff JC |
1914 | static BlockDriver bdrv_vhdx = { |
1915 | .format_name = "vhdx", | |
1916 | .instance_size = sizeof(BDRVVHDXState), | |
1917 | .bdrv_probe = vhdx_probe, | |
1918 | .bdrv_open = vhdx_open, | |
1919 | .bdrv_close = vhdx_close, | |
1920 | .bdrv_reopen_prepare = vhdx_reopen_prepare, | |
1921 | .bdrv_co_readv = vhdx_co_readv, | |
1922 | .bdrv_co_writev = vhdx_co_writev, | |
3412f7b1 | 1923 | .bdrv_create = vhdx_create, |
97b00e28 | 1924 | .bdrv_get_info = vhdx_get_info, |
7e30e6a6 | 1925 | .bdrv_check = vhdx_check, |
3412f7b1 JC |
1926 | |
1927 | .create_options = vhdx_create_options, | |
e8d4e5ff JC |
1928 | }; |
1929 | ||
1930 | static void bdrv_vhdx_init(void) | |
1931 | { | |
1932 | bdrv_register(&bdrv_vhdx); | |
1933 | } | |
1934 | ||
1935 | block_init(bdrv_vhdx_init); |