]>
Commit | Line | Data |
---|---|---|
0a43a1b5 JC |
1 | /* |
2 | * Block driver for Hyper-V VHDX Images | |
3 | * | |
4 | * Copyright (c) 2013 Red Hat, Inc., | |
5 | * | |
6 | * Authors: | |
7 | * Jeff Cody <[email protected]> | |
8 | * | |
9 | * This is based on the "VHDX Format Specification v1.00", published 8/25/2012 | |
10 | * by Microsoft: | |
11 | * https://www.microsoft.com/en-us/download/details.aspx?id=34750 | |
12 | * | |
13 | * This file covers the functionality of the metadata log writing, parsing, and | |
14 | * replay. | |
15 | * | |
16 | * This work is licensed under the terms of the GNU LGPL, version 2 or later. | |
17 | * See the COPYING.LIB file in the top-level directory. | |
18 | * | |
19 | */ | |
80c71a24 | 20 | #include "qemu/osdep.h" |
0a43a1b5 JC |
21 | #include "qemu-common.h" |
22 | #include "block/block_int.h" | |
d49b6836 | 23 | #include "qemu/error-report.h" |
0a43a1b5 JC |
24 | #include "qemu/module.h" |
25 | #include "block/vhdx.h" | |
26 | ||
27 | ||
28 | typedef struct VHDXLogSequence { | |
29 | bool valid; | |
30 | uint32_t count; | |
31 | VHDXLogEntries log; | |
32 | VHDXLogEntryHeader hdr; | |
33 | } VHDXLogSequence; | |
34 | ||
35 | typedef struct VHDXLogDescEntries { | |
36 | VHDXLogEntryHeader hdr; | |
37 | VHDXLogDescriptor desc[]; | |
38 | } VHDXLogDescEntries; | |
39 | ||
40 | static const MSGUID zero_guid = { 0 }; | |
41 | ||
42 | /* The log located on the disk is circular buffer containing | |
43 | * sectors of 4096 bytes each. | |
44 | * | |
45 | * It is assumed for the read/write functions below that the | |
46 | * circular buffer scheme uses a 'one sector open' to indicate | |
47 | * the buffer is full. Given the validation methods used for each | |
48 | * sector, this method should be compatible with other methods that | |
49 | * do not waste a sector. | |
50 | */ | |
51 | ||
52 | ||
53 | /* Allow peeking at the hdr entry at the beginning of the current | |
54 | * read index, without advancing the read index */ | |
55 | static int vhdx_log_peek_hdr(BlockDriverState *bs, VHDXLogEntries *log, | |
56 | VHDXLogEntryHeader *hdr) | |
57 | { | |
58 | int ret = 0; | |
59 | uint64_t offset; | |
60 | uint32_t read; | |
61 | ||
62 | assert(hdr != NULL); | |
63 | ||
64 | /* peek is only supported on sector boundaries */ | |
65 | if (log->read % VHDX_LOG_SECTOR_SIZE) { | |
66 | ret = -EFAULT; | |
67 | goto exit; | |
68 | } | |
69 | ||
70 | read = log->read; | |
71 | /* we are guaranteed that a) log sectors are 4096 bytes, | |
72 | * and b) the log length is a multiple of 1MB. So, there | |
73 | * is always a round number of sectors in the buffer */ | |
74 | if ((read + sizeof(VHDXLogEntryHeader)) > log->length) { | |
75 | read = 0; | |
76 | } | |
77 | ||
78 | if (read == log->write) { | |
79 | ret = -EINVAL; | |
80 | goto exit; | |
81 | } | |
82 | ||
83 | offset = log->offset + read; | |
84 | ||
9a4f4c31 | 85 | ret = bdrv_pread(bs->file->bs, offset, hdr, sizeof(VHDXLogEntryHeader)); |
0a43a1b5 JC |
86 | if (ret < 0) { |
87 | goto exit; | |
88 | } | |
4f75b52a | 89 | vhdx_log_entry_hdr_le_import(hdr); |
0a43a1b5 JC |
90 | |
91 | exit: | |
92 | return ret; | |
93 | } | |
94 | ||
95 | /* Index increment for log, based on sector boundaries */ | |
96 | static int vhdx_log_inc_idx(uint32_t idx, uint64_t length) | |
97 | { | |
98 | idx += VHDX_LOG_SECTOR_SIZE; | |
99 | /* we are guaranteed that a) log sectors are 4096 bytes, | |
100 | * and b) the log length is a multiple of 1MB. So, there | |
101 | * is always a round number of sectors in the buffer */ | |
102 | return idx >= length ? 0 : idx; | |
103 | } | |
104 | ||
105 | ||
106 | /* Reset the log to empty */ | |
107 | static void vhdx_log_reset(BlockDriverState *bs, BDRVVHDXState *s) | |
108 | { | |
109 | MSGUID guid = { 0 }; | |
110 | s->log.read = s->log.write = 0; | |
111 | /* a log guid of 0 indicates an empty log to any parser of v0 | |
112 | * VHDX logs */ | |
113 | vhdx_update_headers(bs, s, false, &guid); | |
114 | } | |
115 | ||
116 | /* Reads num_sectors from the log (all log sectors are 4096 bytes), | |
117 | * into buffer 'buffer'. Upon return, *sectors_read will contain | |
118 | * the number of sectors successfully read. | |
119 | * | |
120 | * It is assumed that 'buffer' is already allocated, and of sufficient | |
121 | * size (i.e. >= 4096*num_sectors). | |
122 | * | |
123 | * If 'peek' is true, then the tail (read) pointer for the circular buffer is | |
124 | * not modified. | |
125 | * | |
126 | * 0 is returned on success, -errno otherwise. */ | |
127 | static int vhdx_log_read_sectors(BlockDriverState *bs, VHDXLogEntries *log, | |
128 | uint32_t *sectors_read, void *buffer, | |
129 | uint32_t num_sectors, bool peek) | |
130 | { | |
131 | int ret = 0; | |
132 | uint64_t offset; | |
133 | uint32_t read; | |
134 | ||
135 | read = log->read; | |
136 | ||
137 | *sectors_read = 0; | |
138 | while (num_sectors) { | |
139 | if (read == log->write) { | |
140 | /* empty */ | |
141 | break; | |
142 | } | |
143 | offset = log->offset + read; | |
144 | ||
9a4f4c31 | 145 | ret = bdrv_pread(bs->file->bs, offset, buffer, VHDX_LOG_SECTOR_SIZE); |
0a43a1b5 JC |
146 | if (ret < 0) { |
147 | goto exit; | |
148 | } | |
149 | read = vhdx_log_inc_idx(read, log->length); | |
150 | ||
151 | *sectors_read = *sectors_read + 1; | |
152 | num_sectors--; | |
153 | } | |
154 | ||
155 | exit: | |
156 | if (!peek) { | |
157 | log->read = read; | |
158 | } | |
159 | return ret; | |
160 | } | |
161 | ||
8adc5233 JC |
162 | /* Writes num_sectors to the log (all log sectors are 4096 bytes), |
163 | * from buffer 'buffer'. Upon return, *sectors_written will contain | |
164 | * the number of sectors successfully written. | |
165 | * | |
166 | * It is assumed that 'buffer' is at least 4096*num_sectors large. | |
167 | * | |
168 | * 0 is returned on success, -errno otherwise */ | |
169 | static int vhdx_log_write_sectors(BlockDriverState *bs, VHDXLogEntries *log, | |
170 | uint32_t *sectors_written, void *buffer, | |
171 | uint32_t num_sectors) | |
172 | { | |
173 | int ret = 0; | |
174 | uint64_t offset; | |
175 | uint32_t write; | |
176 | void *buffer_tmp; | |
177 | BDRVVHDXState *s = bs->opaque; | |
178 | ||
179 | ret = vhdx_user_visible_write(bs, s); | |
180 | if (ret < 0) { | |
181 | goto exit; | |
182 | } | |
183 | ||
184 | write = log->write; | |
185 | ||
186 | buffer_tmp = buffer; | |
187 | while (num_sectors) { | |
188 | ||
189 | offset = log->offset + write; | |
190 | write = vhdx_log_inc_idx(write, log->length); | |
191 | if (write == log->read) { | |
192 | /* full */ | |
193 | break; | |
194 | } | |
9a4f4c31 KW |
195 | ret = bdrv_pwrite(bs->file->bs, offset, buffer_tmp, |
196 | VHDX_LOG_SECTOR_SIZE); | |
8adc5233 JC |
197 | if (ret < 0) { |
198 | goto exit; | |
199 | } | |
200 | buffer_tmp += VHDX_LOG_SECTOR_SIZE; | |
201 | ||
202 | log->write = write; | |
203 | *sectors_written = *sectors_written + 1; | |
204 | num_sectors--; | |
205 | } | |
206 | ||
207 | exit: | |
208 | return ret; | |
209 | } | |
210 | ||
211 | ||
0a43a1b5 JC |
212 | /* Validates a log entry header */ |
213 | static bool vhdx_log_hdr_is_valid(VHDXLogEntries *log, VHDXLogEntryHeader *hdr, | |
214 | BDRVVHDXState *s) | |
215 | { | |
216 | int valid = false; | |
217 | ||
4f75b52a | 218 | if (hdr->signature != VHDX_LOG_SIGNATURE) { |
0a43a1b5 JC |
219 | goto exit; |
220 | } | |
221 | ||
222 | /* if the individual entry length is larger than the whole log | |
223 | * buffer, that is obviously invalid */ | |
224 | if (log->length < hdr->entry_length) { | |
225 | goto exit; | |
226 | } | |
227 | ||
228 | /* length of entire entry must be in units of 4KB (log sector size) */ | |
229 | if (hdr->entry_length % (VHDX_LOG_SECTOR_SIZE)) { | |
230 | goto exit; | |
231 | } | |
232 | ||
233 | /* per spec, sequence # must be > 0 */ | |
234 | if (hdr->sequence_number == 0) { | |
235 | goto exit; | |
236 | } | |
237 | ||
238 | /* log entries are only valid if they match the file-wide log guid | |
239 | * found in the active header */ | |
240 | if (!guid_eq(hdr->log_guid, s->headers[s->curr_header]->log_guid)) { | |
241 | goto exit; | |
242 | } | |
243 | ||
244 | if (hdr->descriptor_count * sizeof(VHDXLogDescriptor) > hdr->entry_length) { | |
245 | goto exit; | |
246 | } | |
247 | ||
248 | valid = true; | |
249 | ||
250 | exit: | |
251 | return valid; | |
252 | } | |
253 | ||
254 | /* | |
255 | * Given a log header, this will validate that the descriptors and the | |
256 | * corresponding data sectors (if applicable) | |
257 | * | |
258 | * Validation consists of: | |
259 | * 1. Making sure the sequence numbers matches the entry header | |
260 | * 2. Verifying a valid signature ('zero' or 'desc' for descriptors) | |
261 | * 3. File offset field is a multiple of 4KB | |
262 | * 4. If a data descriptor, the corresponding data sector | |
263 | * has its signature ('data') and matching sequence number | |
264 | * | |
265 | * @desc: the data buffer containing the descriptor | |
266 | * @hdr: the log entry header | |
267 | * | |
268 | * Returns true if valid | |
269 | */ | |
270 | static bool vhdx_log_desc_is_valid(VHDXLogDescriptor *desc, | |
271 | VHDXLogEntryHeader *hdr) | |
272 | { | |
273 | bool ret = false; | |
274 | ||
275 | if (desc->sequence_number != hdr->sequence_number) { | |
276 | goto exit; | |
277 | } | |
278 | if (desc->file_offset % VHDX_LOG_SECTOR_SIZE) { | |
279 | goto exit; | |
280 | } | |
281 | ||
4f75b52a | 282 | if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) { |
0a43a1b5 JC |
283 | if (desc->zero_length % VHDX_LOG_SECTOR_SIZE == 0) { |
284 | /* valid */ | |
285 | ret = true; | |
286 | } | |
4f75b52a | 287 | } else if (desc->signature == VHDX_LOG_DESC_SIGNATURE) { |
0a43a1b5 JC |
288 | /* valid */ |
289 | ret = true; | |
290 | } | |
291 | ||
292 | exit: | |
293 | return ret; | |
294 | } | |
295 | ||
296 | ||
297 | /* Prior to sector data for a log entry, there is the header | |
298 | * and the descriptors referenced in the header: | |
299 | * | |
300 | * [] = 4KB sector | |
301 | * | |
302 | * [ hdr, desc ][ desc ][ ... ][ data ][ ... ] | |
303 | * | |
304 | * The first sector in a log entry has a 64 byte header, and | |
305 | * up to 126 32-byte descriptors. If more descriptors than | |
306 | * 126 are required, then subsequent sectors can have up to 128 | |
307 | * descriptors. Each sector is 4KB. Data follows the descriptor | |
308 | * sectors. | |
309 | * | |
310 | * This will return the number of sectors needed to encompass | |
311 | * the passed number of descriptors in desc_cnt. | |
312 | * | |
313 | * This will never return 0, even if desc_cnt is 0. | |
314 | */ | |
315 | static int vhdx_compute_desc_sectors(uint32_t desc_cnt) | |
316 | { | |
317 | uint32_t desc_sectors; | |
318 | ||
319 | desc_cnt += 2; /* account for header in first sector */ | |
320 | desc_sectors = desc_cnt / 128; | |
321 | if (desc_cnt % 128) { | |
322 | desc_sectors++; | |
323 | } | |
324 | ||
325 | return desc_sectors; | |
326 | } | |
327 | ||
328 | ||
329 | /* Reads the log header, and subsequent descriptors (if any). This | |
330 | * will allocate all the space for buffer, which must be NULL when | |
331 | * passed into this function. Each descriptor will also be validated, | |
332 | * and error returned if any are invalid. */ | |
333 | static int vhdx_log_read_desc(BlockDriverState *bs, BDRVVHDXState *s, | |
4f75b52a JC |
334 | VHDXLogEntries *log, VHDXLogDescEntries **buffer, |
335 | bool convert_endian) | |
0a43a1b5 JC |
336 | { |
337 | int ret = 0; | |
338 | uint32_t desc_sectors; | |
339 | uint32_t sectors_read; | |
340 | VHDXLogEntryHeader hdr; | |
341 | VHDXLogDescEntries *desc_entries = NULL; | |
4f75b52a | 342 | VHDXLogDescriptor desc; |
0a43a1b5 JC |
343 | int i; |
344 | ||
345 | assert(*buffer == NULL); | |
346 | ||
347 | ret = vhdx_log_peek_hdr(bs, log, &hdr); | |
348 | if (ret < 0) { | |
349 | goto exit; | |
350 | } | |
4f75b52a | 351 | |
0a43a1b5 JC |
352 | if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) { |
353 | ret = -EINVAL; | |
354 | goto exit; | |
355 | } | |
356 | ||
357 | desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count); | |
9a4f4c31 | 358 | desc_entries = qemu_try_blockalign(bs->file->bs, |
a67e128a KW |
359 | desc_sectors * VHDX_LOG_SECTOR_SIZE); |
360 | if (desc_entries == NULL) { | |
361 | ret = -ENOMEM; | |
362 | goto exit; | |
363 | } | |
0a43a1b5 JC |
364 | |
365 | ret = vhdx_log_read_sectors(bs, log, §ors_read, desc_entries, | |
366 | desc_sectors, false); | |
367 | if (ret < 0) { | |
368 | goto free_and_exit; | |
369 | } | |
370 | if (sectors_read != desc_sectors) { | |
371 | ret = -EINVAL; | |
372 | goto free_and_exit; | |
373 | } | |
374 | ||
375 | /* put in proper endianness, and validate each desc */ | |
376 | for (i = 0; i < hdr.descriptor_count; i++) { | |
4f75b52a JC |
377 | desc = desc_entries->desc[i]; |
378 | vhdx_log_desc_le_import(&desc); | |
379 | if (convert_endian) { | |
380 | desc_entries->desc[i] = desc; | |
381 | } | |
382 | if (vhdx_log_desc_is_valid(&desc, &hdr) == false) { | |
0a43a1b5 JC |
383 | ret = -EINVAL; |
384 | goto free_and_exit; | |
385 | } | |
386 | } | |
4f75b52a JC |
387 | if (convert_endian) { |
388 | desc_entries->hdr = hdr; | |
389 | } | |
0a43a1b5 JC |
390 | |
391 | *buffer = desc_entries; | |
392 | goto exit; | |
393 | ||
394 | free_and_exit: | |
395 | qemu_vfree(desc_entries); | |
396 | exit: | |
397 | return ret; | |
398 | } | |
399 | ||
400 | ||
401 | /* Flushes the descriptor described by desc to the VHDX image file. | |
402 | * If the descriptor is a data descriptor, than 'data' must be non-NULL, | |
403 | * and >= 4096 bytes (VHDX_LOG_SECTOR_SIZE), containing the data to be | |
404 | * written. | |
405 | * | |
406 | * Verification is performed to make sure the sequence numbers of a data | |
407 | * descriptor match the sequence number in the desc. | |
408 | * | |
409 | * For a zero descriptor, it may describe multiple sectors to fill with zeroes. | |
410 | * In this case, it should be noted that zeroes are written to disk, and the | |
411 | * image file is not extended as a sparse file. */ | |
412 | static int vhdx_log_flush_desc(BlockDriverState *bs, VHDXLogDescriptor *desc, | |
413 | VHDXLogDataSector *data) | |
414 | { | |
415 | int ret = 0; | |
416 | uint64_t seq, file_offset; | |
417 | uint32_t offset = 0; | |
418 | void *buffer = NULL; | |
419 | uint64_t count = 1; | |
420 | int i; | |
421 | ||
422 | buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE); | |
423 | ||
4f75b52a | 424 | if (desc->signature == VHDX_LOG_DESC_SIGNATURE) { |
0a43a1b5 JC |
425 | /* data sector */ |
426 | if (data == NULL) { | |
427 | ret = -EFAULT; | |
428 | goto exit; | |
429 | } | |
430 | ||
431 | /* The sequence number of the data sector must match that | |
432 | * in the descriptor */ | |
433 | seq = data->sequence_high; | |
434 | seq <<= 32; | |
435 | seq |= data->sequence_low & 0xffffffff; | |
436 | ||
437 | if (seq != desc->sequence_number) { | |
438 | ret = -EINVAL; | |
439 | goto exit; | |
440 | } | |
441 | ||
442 | /* Each data sector is in total 4096 bytes, however the first | |
443 | * 8 bytes, and last 4 bytes, are located in the descriptor */ | |
444 | memcpy(buffer, &desc->leading_bytes, 8); | |
445 | offset += 8; | |
446 | ||
447 | memcpy(buffer+offset, data->data, 4084); | |
448 | offset += 4084; | |
449 | ||
450 | memcpy(buffer+offset, &desc->trailing_bytes, 4); | |
451 | ||
4f75b52a | 452 | } else if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) { |
0a43a1b5 JC |
453 | /* write 'count' sectors of sector */ |
454 | memset(buffer, 0, VHDX_LOG_SECTOR_SIZE); | |
455 | count = desc->zero_length / VHDX_LOG_SECTOR_SIZE; | |
349592e0 JC |
456 | } else { |
457 | error_report("Invalid VHDX log descriptor entry signature 0x%" PRIx32, | |
458 | desc->signature); | |
459 | ret = -EINVAL; | |
460 | goto exit; | |
0a43a1b5 JC |
461 | } |
462 | ||
463 | file_offset = desc->file_offset; | |
464 | ||
465 | /* count is only > 1 if we are writing zeroes */ | |
466 | for (i = 0; i < count; i++) { | |
9a4f4c31 | 467 | ret = bdrv_pwrite_sync(bs->file->bs, file_offset, buffer, |
0a43a1b5 JC |
468 | VHDX_LOG_SECTOR_SIZE); |
469 | if (ret < 0) { | |
470 | goto exit; | |
471 | } | |
472 | file_offset += VHDX_LOG_SECTOR_SIZE; | |
473 | } | |
474 | ||
475 | exit: | |
476 | qemu_vfree(buffer); | |
477 | return ret; | |
478 | } | |
479 | ||
480 | /* Flush the entire log (as described by 'logs') to the VHDX image | |
481 | * file, and then set the log to 'empty' status once complete. | |
482 | * | |
483 | * The log entries should be validate prior to flushing */ | |
484 | static int vhdx_log_flush(BlockDriverState *bs, BDRVVHDXState *s, | |
485 | VHDXLogSequence *logs) | |
486 | { | |
487 | int ret = 0; | |
488 | int i; | |
489 | uint32_t cnt, sectors_read; | |
490 | uint64_t new_file_size; | |
491 | void *data = NULL; | |
492 | VHDXLogDescEntries *desc_entries = NULL; | |
493 | VHDXLogEntryHeader hdr_tmp = { 0 }; | |
494 | ||
495 | cnt = logs->count; | |
496 | ||
497 | data = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE); | |
498 | ||
499 | ret = vhdx_user_visible_write(bs, s); | |
500 | if (ret < 0) { | |
501 | goto exit; | |
502 | } | |
503 | ||
504 | /* each iteration represents one log sequence, which may span multiple | |
505 | * sectors */ | |
506 | while (cnt--) { | |
507 | ret = vhdx_log_peek_hdr(bs, &logs->log, &hdr_tmp); | |
508 | if (ret < 0) { | |
509 | goto exit; | |
510 | } | |
511 | /* if the log shows a FlushedFileOffset larger than our current file | |
512 | * size, then that means the file has been truncated / corrupted, and | |
513 | * we must refused to open it / use it */ | |
9a4f4c31 | 514 | if (hdr_tmp.flushed_file_offset > bdrv_getlength(bs->file->bs)) { |
0a43a1b5 JC |
515 | ret = -EINVAL; |
516 | goto exit; | |
517 | } | |
518 | ||
4f75b52a | 519 | ret = vhdx_log_read_desc(bs, s, &logs->log, &desc_entries, true); |
0a43a1b5 JC |
520 | if (ret < 0) { |
521 | goto exit; | |
522 | } | |
523 | ||
524 | for (i = 0; i < desc_entries->hdr.descriptor_count; i++) { | |
4f75b52a | 525 | if (desc_entries->desc[i].signature == VHDX_LOG_DESC_SIGNATURE) { |
0a43a1b5 JC |
526 | /* data sector, so read a sector to flush */ |
527 | ret = vhdx_log_read_sectors(bs, &logs->log, §ors_read, | |
528 | data, 1, false); | |
529 | if (ret < 0) { | |
530 | goto exit; | |
531 | } | |
532 | if (sectors_read != 1) { | |
533 | ret = -EINVAL; | |
534 | goto exit; | |
535 | } | |
4f75b52a | 536 | vhdx_log_data_le_import(data); |
0a43a1b5 JC |
537 | } |
538 | ||
539 | ret = vhdx_log_flush_desc(bs, &desc_entries->desc[i], data); | |
540 | if (ret < 0) { | |
541 | goto exit; | |
542 | } | |
543 | } | |
9a4f4c31 | 544 | if (bdrv_getlength(bs->file->bs) < desc_entries->hdr.last_file_offset) { |
0a43a1b5 JC |
545 | new_file_size = desc_entries->hdr.last_file_offset; |
546 | if (new_file_size % (1024*1024)) { | |
547 | /* round up to nearest 1MB boundary */ | |
548 | new_file_size = ((new_file_size >> 20) + 1) << 20; | |
9a4f4c31 | 549 | bdrv_truncate(bs->file->bs, new_file_size); |
0a43a1b5 JC |
550 | } |
551 | } | |
552 | qemu_vfree(desc_entries); | |
553 | desc_entries = NULL; | |
554 | } | |
555 | ||
556 | bdrv_flush(bs); | |
557 | /* once the log is fully flushed, indicate that we have an empty log | |
558 | * now. This also sets the log guid to 0, to indicate an empty log */ | |
559 | vhdx_log_reset(bs, s); | |
560 | ||
561 | exit: | |
562 | qemu_vfree(data); | |
563 | qemu_vfree(desc_entries); | |
564 | return ret; | |
565 | } | |
566 | ||
567 | static int vhdx_validate_log_entry(BlockDriverState *bs, BDRVVHDXState *s, | |
568 | VHDXLogEntries *log, uint64_t seq, | |
569 | bool *valid, VHDXLogEntryHeader *entry) | |
570 | { | |
571 | int ret = 0; | |
572 | VHDXLogEntryHeader hdr; | |
573 | void *buffer = NULL; | |
574 | uint32_t i, desc_sectors, total_sectors, crc; | |
575 | uint32_t sectors_read = 0; | |
576 | VHDXLogDescEntries *desc_buffer = NULL; | |
577 | ||
578 | *valid = false; | |
579 | ||
580 | ret = vhdx_log_peek_hdr(bs, log, &hdr); | |
581 | if (ret < 0) { | |
582 | goto inc_and_exit; | |
583 | } | |
584 | ||
0a43a1b5 JC |
585 | if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) { |
586 | goto inc_and_exit; | |
587 | } | |
588 | ||
589 | if (seq > 0) { | |
590 | if (hdr.sequence_number != seq + 1) { | |
591 | goto inc_and_exit; | |
592 | } | |
593 | } | |
594 | ||
595 | desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count); | |
596 | ||
4f75b52a | 597 | /* Read all log sectors, and calculate log checksum */ |
0a43a1b5 JC |
598 | |
599 | total_sectors = hdr.entry_length / VHDX_LOG_SECTOR_SIZE; | |
600 | ||
601 | ||
dc6fb73d | 602 | /* read_desc() will increment the read idx */ |
4f75b52a | 603 | ret = vhdx_log_read_desc(bs, s, log, &desc_buffer, false); |
0a43a1b5 JC |
604 | if (ret < 0) { |
605 | goto free_and_exit; | |
606 | } | |
607 | ||
608 | crc = vhdx_checksum_calc(0xffffffff, (void *)desc_buffer, | |
609 | desc_sectors * VHDX_LOG_SECTOR_SIZE, 4); | |
610 | crc ^= 0xffffffff; | |
611 | ||
612 | buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE); | |
613 | if (total_sectors > desc_sectors) { | |
614 | for (i = 0; i < total_sectors - desc_sectors; i++) { | |
615 | sectors_read = 0; | |
616 | ret = vhdx_log_read_sectors(bs, log, §ors_read, buffer, | |
617 | 1, false); | |
618 | if (ret < 0 || sectors_read != 1) { | |
619 | goto free_and_exit; | |
620 | } | |
621 | crc = vhdx_checksum_calc(crc, buffer, VHDX_LOG_SECTOR_SIZE, -1); | |
622 | crc ^= 0xffffffff; | |
623 | } | |
624 | } | |
625 | crc ^= 0xffffffff; | |
4f75b52a | 626 | if (crc != hdr.checksum) { |
0a43a1b5 JC |
627 | goto free_and_exit; |
628 | } | |
629 | ||
630 | *valid = true; | |
631 | *entry = hdr; | |
632 | goto free_and_exit; | |
633 | ||
634 | inc_and_exit: | |
635 | log->read = vhdx_log_inc_idx(log->read, log->length); | |
636 | ||
637 | free_and_exit: | |
638 | qemu_vfree(buffer); | |
639 | qemu_vfree(desc_buffer); | |
640 | return ret; | |
641 | } | |
642 | ||
643 | /* Search through the log circular buffer, and find the valid, active | |
644 | * log sequence, if any exists | |
645 | * */ | |
646 | static int vhdx_log_search(BlockDriverState *bs, BDRVVHDXState *s, | |
647 | VHDXLogSequence *logs) | |
648 | { | |
649 | int ret = 0; | |
650 | uint32_t tail; | |
651 | bool seq_valid = false; | |
652 | VHDXLogSequence candidate = { 0 }; | |
653 | VHDXLogEntryHeader hdr = { 0 }; | |
654 | VHDXLogEntries curr_log; | |
655 | ||
656 | memcpy(&curr_log, &s->log, sizeof(VHDXLogEntries)); | |
657 | curr_log.write = curr_log.length; /* assume log is full */ | |
658 | curr_log.read = 0; | |
659 | ||
660 | ||
661 | /* now we will go through the whole log sector by sector, until | |
662 | * we find a valid, active log sequence, or reach the end of the | |
663 | * log buffer */ | |
664 | for (;;) { | |
665 | uint64_t curr_seq = 0; | |
666 | VHDXLogSequence current = { 0 }; | |
667 | ||
668 | tail = curr_log.read; | |
669 | ||
670 | ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq, | |
671 | &seq_valid, &hdr); | |
672 | if (ret < 0) { | |
673 | goto exit; | |
674 | } | |
675 | ||
676 | if (seq_valid) { | |
677 | current.valid = true; | |
678 | current.log = curr_log; | |
679 | current.log.read = tail; | |
680 | current.log.write = curr_log.read; | |
681 | current.count = 1; | |
682 | current.hdr = hdr; | |
683 | ||
684 | ||
685 | for (;;) { | |
686 | ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq, | |
687 | &seq_valid, &hdr); | |
688 | if (ret < 0) { | |
689 | goto exit; | |
690 | } | |
691 | if (seq_valid == false) { | |
692 | break; | |
693 | } | |
694 | current.log.write = curr_log.read; | |
695 | current.count++; | |
696 | ||
697 | curr_seq = hdr.sequence_number; | |
698 | } | |
699 | } | |
700 | ||
701 | if (current.valid) { | |
702 | if (candidate.valid == false || | |
703 | current.hdr.sequence_number > candidate.hdr.sequence_number) { | |
704 | candidate = current; | |
705 | } | |
706 | } | |
707 | ||
708 | if (curr_log.read < tail) { | |
709 | break; | |
710 | } | |
711 | } | |
712 | ||
713 | *logs = candidate; | |
714 | ||
715 | if (candidate.valid) { | |
716 | /* this is the next sequence number, for writes */ | |
717 | s->log.sequence = candidate.hdr.sequence_number + 1; | |
718 | } | |
719 | ||
720 | ||
721 | exit: | |
722 | return ret; | |
723 | } | |
724 | ||
725 | /* Parse the replay log. Per the VHDX spec, if the log is present | |
726 | * it must be replayed prior to opening the file, even read-only. | |
727 | * | |
728 | * If read-only, we must replay the log in RAM (or refuse to open | |
729 | * a dirty VHDX file read-only) */ | |
7e30e6a6 JC |
730 | int vhdx_parse_log(BlockDriverState *bs, BDRVVHDXState *s, bool *flushed, |
731 | Error **errp) | |
0a43a1b5 JC |
732 | { |
733 | int ret = 0; | |
734 | VHDXHeader *hdr; | |
735 | VHDXLogSequence logs = { 0 }; | |
736 | ||
737 | hdr = s->headers[s->curr_header]; | |
738 | ||
739 | *flushed = false; | |
740 | ||
741 | /* s->log.hdr is freed in vhdx_close() */ | |
742 | if (s->log.hdr == NULL) { | |
743 | s->log.hdr = qemu_blockalign(bs, sizeof(VHDXLogEntryHeader)); | |
744 | } | |
745 | ||
746 | s->log.offset = hdr->log_offset; | |
747 | s->log.length = hdr->log_length; | |
748 | ||
749 | if (s->log.offset < VHDX_LOG_MIN_SIZE || | |
750 | s->log.offset % VHDX_LOG_MIN_SIZE) { | |
751 | ret = -EINVAL; | |
752 | goto exit; | |
753 | } | |
754 | ||
755 | /* per spec, only log version of 0 is supported */ | |
756 | if (hdr->log_version != 0) { | |
757 | ret = -EINVAL; | |
758 | goto exit; | |
759 | } | |
760 | ||
761 | /* If either the log guid, or log length is zero, | |
762 | * then a replay log is not present */ | |
763 | if (guid_eq(hdr->log_guid, zero_guid)) { | |
764 | goto exit; | |
765 | } | |
766 | ||
767 | if (hdr->log_length == 0) { | |
768 | goto exit; | |
769 | } | |
770 | ||
771 | if (hdr->log_length % VHDX_LOG_MIN_SIZE) { | |
772 | ret = -EINVAL; | |
773 | goto exit; | |
774 | } | |
775 | ||
776 | ||
777 | /* The log is present, we need to find if and where there is an active | |
778 | * sequence of valid entries present in the log. */ | |
779 | ||
780 | ret = vhdx_log_search(bs, s, &logs); | |
781 | if (ret < 0) { | |
782 | goto exit; | |
783 | } | |
784 | ||
785 | if (logs.valid) { | |
7e30e6a6 JC |
786 | if (bs->read_only) { |
787 | ret = -EPERM; | |
bf89e874 MA |
788 | error_setg(errp, |
789 | "VHDX image file '%s' opened read-only, but " | |
790 | "contains a log that needs to be replayed", | |
791 | bs->filename); | |
792 | error_append_hint(errp, "To replay the log, run:\n" | |
793 | "qemu-img check -r all '%s'\n", | |
794 | bs->filename); | |
7e30e6a6 JC |
795 | goto exit; |
796 | } | |
0a43a1b5 JC |
797 | /* now flush the log */ |
798 | ret = vhdx_log_flush(bs, s, &logs); | |
799 | if (ret < 0) { | |
800 | goto exit; | |
801 | } | |
802 | *flushed = true; | |
803 | } | |
804 | ||
805 | ||
806 | exit: | |
807 | return ret; | |
808 | } | |
809 | ||
810 | ||
8adc5233 JC |
811 | |
812 | static void vhdx_log_raw_to_le_sector(VHDXLogDescriptor *desc, | |
813 | VHDXLogDataSector *sector, void *data, | |
814 | uint64_t seq) | |
815 | { | |
816 | /* 8 + 4084 + 4 = 4096, 1 log sector */ | |
817 | memcpy(&desc->leading_bytes, data, 8); | |
818 | data += 8; | |
819 | cpu_to_le64s(&desc->leading_bytes); | |
820 | memcpy(sector->data, data, 4084); | |
821 | data += 4084; | |
822 | memcpy(&desc->trailing_bytes, data, 4); | |
823 | cpu_to_le32s(&desc->trailing_bytes); | |
824 | data += 4; | |
825 | ||
826 | sector->sequence_high = (uint32_t) (seq >> 32); | |
827 | sector->sequence_low = (uint32_t) (seq & 0xffffffff); | |
828 | sector->data_signature = VHDX_LOG_DATA_SIGNATURE; | |
829 | ||
830 | vhdx_log_desc_le_export(desc); | |
831 | vhdx_log_data_le_export(sector); | |
832 | } | |
833 | ||
834 | ||
835 | static int vhdx_log_write(BlockDriverState *bs, BDRVVHDXState *s, | |
836 | void *data, uint32_t length, uint64_t offset) | |
837 | { | |
838 | int ret = 0; | |
839 | void *buffer = NULL; | |
840 | void *merged_sector = NULL; | |
841 | void *data_tmp, *sector_write; | |
842 | unsigned int i; | |
843 | int sector_offset; | |
844 | uint32_t desc_sectors, sectors, total_length; | |
845 | uint32_t sectors_written = 0; | |
846 | uint32_t aligned_length; | |
847 | uint32_t leading_length = 0; | |
848 | uint32_t trailing_length = 0; | |
849 | uint32_t partial_sectors = 0; | |
850 | uint32_t bytes_written = 0; | |
851 | uint64_t file_offset; | |
852 | VHDXHeader *header; | |
853 | VHDXLogEntryHeader new_hdr; | |
854 | VHDXLogDescriptor *new_desc = NULL; | |
855 | VHDXLogDataSector *data_sector = NULL; | |
856 | MSGUID new_guid = { 0 }; | |
857 | ||
858 | header = s->headers[s->curr_header]; | |
859 | ||
860 | /* need to have offset read data, and be on 4096 byte boundary */ | |
861 | ||
862 | if (length > header->log_length) { | |
863 | /* no log present. we could create a log here instead of failing */ | |
864 | ret = -EINVAL; | |
865 | goto exit; | |
866 | } | |
867 | ||
868 | if (guid_eq(header->log_guid, zero_guid)) { | |
869 | vhdx_guid_generate(&new_guid); | |
870 | vhdx_update_headers(bs, s, false, &new_guid); | |
871 | } else { | |
872 | /* currently, we require that the log be flushed after | |
873 | * every write. */ | |
874 | ret = -ENOTSUP; | |
875 | goto exit; | |
876 | } | |
877 | ||
878 | /* 0 is an invalid sequence number, but may also represent the first | |
879 | * log write (or a wrapped seq) */ | |
880 | if (s->log.sequence == 0) { | |
881 | s->log.sequence = 1; | |
882 | } | |
883 | ||
884 | sector_offset = offset % VHDX_LOG_SECTOR_SIZE; | |
885 | file_offset = (offset / VHDX_LOG_SECTOR_SIZE) * VHDX_LOG_SECTOR_SIZE; | |
886 | ||
887 | aligned_length = length; | |
888 | ||
889 | /* add in the unaligned head and tail bytes */ | |
890 | if (sector_offset) { | |
891 | leading_length = (VHDX_LOG_SECTOR_SIZE - sector_offset); | |
892 | leading_length = leading_length > length ? length : leading_length; | |
893 | aligned_length -= leading_length; | |
894 | partial_sectors++; | |
895 | } | |
896 | ||
897 | sectors = aligned_length / VHDX_LOG_SECTOR_SIZE; | |
898 | trailing_length = aligned_length - (sectors * VHDX_LOG_SECTOR_SIZE); | |
899 | if (trailing_length) { | |
900 | partial_sectors++; | |
901 | } | |
902 | ||
903 | sectors += partial_sectors; | |
904 | ||
905 | /* sectors is now how many sectors the data itself takes, not | |
906 | * including the header and descriptor metadata */ | |
907 | ||
908 | new_hdr = (VHDXLogEntryHeader) { | |
909 | .signature = VHDX_LOG_SIGNATURE, | |
910 | .tail = s->log.tail, | |
911 | .sequence_number = s->log.sequence, | |
912 | .descriptor_count = sectors, | |
913 | .reserved = 0, | |
9a4f4c31 KW |
914 | .flushed_file_offset = bdrv_getlength(bs->file->bs), |
915 | .last_file_offset = bdrv_getlength(bs->file->bs), | |
8adc5233 JC |
916 | }; |
917 | ||
918 | new_hdr.log_guid = header->log_guid; | |
919 | ||
920 | desc_sectors = vhdx_compute_desc_sectors(new_hdr.descriptor_count); | |
921 | ||
922 | total_length = (desc_sectors + sectors) * VHDX_LOG_SECTOR_SIZE; | |
923 | new_hdr.entry_length = total_length; | |
924 | ||
925 | vhdx_log_entry_hdr_le_export(&new_hdr); | |
926 | ||
927 | buffer = qemu_blockalign(bs, total_length); | |
928 | memcpy(buffer, &new_hdr, sizeof(new_hdr)); | |
929 | ||
d4df3dbc | 930 | new_desc = buffer + sizeof(new_hdr); |
8adc5233 JC |
931 | data_sector = buffer + (desc_sectors * VHDX_LOG_SECTOR_SIZE); |
932 | data_tmp = data; | |
933 | ||
934 | /* All log sectors are 4KB, so for any partial sectors we must | |
935 | * merge the data with preexisting data from the final file | |
936 | * destination */ | |
937 | merged_sector = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE); | |
938 | ||
939 | for (i = 0; i < sectors; i++) { | |
940 | new_desc->signature = VHDX_LOG_DESC_SIGNATURE; | |
941 | new_desc->sequence_number = s->log.sequence; | |
942 | new_desc->file_offset = file_offset; | |
943 | ||
944 | if (i == 0 && leading_length) { | |
945 | /* partial sector at the front of the buffer */ | |
9a4f4c31 | 946 | ret = bdrv_pread(bs->file->bs, file_offset, merged_sector, |
8adc5233 JC |
947 | VHDX_LOG_SECTOR_SIZE); |
948 | if (ret < 0) { | |
949 | goto exit; | |
950 | } | |
951 | memcpy(merged_sector + sector_offset, data_tmp, leading_length); | |
952 | bytes_written = leading_length; | |
953 | sector_write = merged_sector; | |
954 | } else if (i == sectors - 1 && trailing_length) { | |
955 | /* partial sector at the end of the buffer */ | |
9a4f4c31 | 956 | ret = bdrv_pread(bs->file->bs, |
8adc5233 JC |
957 | file_offset, |
958 | merged_sector + trailing_length, | |
959 | VHDX_LOG_SECTOR_SIZE - trailing_length); | |
960 | if (ret < 0) { | |
961 | goto exit; | |
962 | } | |
963 | memcpy(merged_sector, data_tmp, trailing_length); | |
964 | bytes_written = trailing_length; | |
965 | sector_write = merged_sector; | |
966 | } else { | |
967 | bytes_written = VHDX_LOG_SECTOR_SIZE; | |
968 | sector_write = data_tmp; | |
969 | } | |
970 | ||
971 | /* populate the raw sector data into the proper structures, | |
972 | * as well as update the descriptor, and convert to proper | |
973 | * endianness */ | |
974 | vhdx_log_raw_to_le_sector(new_desc, data_sector, sector_write, | |
975 | s->log.sequence); | |
976 | ||
977 | data_tmp += bytes_written; | |
978 | data_sector++; | |
979 | new_desc++; | |
980 | file_offset += VHDX_LOG_SECTOR_SIZE; | |
981 | } | |
982 | ||
983 | /* checksum covers entire entry, from the log header through the | |
984 | * last data sector */ | |
985 | vhdx_update_checksum(buffer, total_length, | |
986 | offsetof(VHDXLogEntryHeader, checksum)); | |
8adc5233 JC |
987 | |
988 | /* now write to the log */ | |
f50159fa MA |
989 | ret = vhdx_log_write_sectors(bs, &s->log, §ors_written, buffer, |
990 | desc_sectors + sectors); | |
8adc5233 JC |
991 | if (ret < 0) { |
992 | goto exit; | |
993 | } | |
994 | ||
995 | if (sectors_written != desc_sectors + sectors) { | |
996 | /* instead of failing, we could flush the log here */ | |
997 | ret = -EINVAL; | |
998 | goto exit; | |
999 | } | |
1000 | ||
1001 | s->log.sequence++; | |
1002 | /* write new tail */ | |
1003 | s->log.tail = s->log.write; | |
1004 | ||
1005 | exit: | |
1006 | qemu_vfree(buffer); | |
1007 | qemu_vfree(merged_sector); | |
1008 | return ret; | |
1009 | } | |
1010 | ||
1011 | /* Perform a log write, and then immediately flush the entire log */ | |
1012 | int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s, | |
1013 | void *data, uint32_t length, uint64_t offset) | |
1014 | { | |
1015 | int ret = 0; | |
1016 | VHDXLogSequence logs = { .valid = true, | |
1017 | .count = 1, | |
1018 | .hdr = { 0 } }; | |
1019 | ||
1020 | ||
1021 | /* Make sure data written (new and/or changed blocks) is stable | |
1022 | * on disk, before creating log entry */ | |
1023 | bdrv_flush(bs); | |
1024 | ret = vhdx_log_write(bs, s, data, length, offset); | |
1025 | if (ret < 0) { | |
1026 | goto exit; | |
1027 | } | |
1028 | logs.log = s->log; | |
1029 | ||
1030 | /* Make sure log is stable on disk */ | |
1031 | bdrv_flush(bs); | |
1032 | ret = vhdx_log_flush(bs, s, &logs); | |
1033 | if (ret < 0) { | |
1034 | goto exit; | |
1035 | } | |
1036 | ||
1037 | s->log = logs.log; | |
1038 | ||
1039 | exit: | |
1040 | return ret; | |
1041 | } | |
1042 |