]>
Commit | Line | Data |
---|---|---|
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 | * | |
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 | */ | |
20 | ||
21 | #include "qemu/osdep.h" | |
22 | #include "qapi/error.h" | |
23 | #include "block/block-io.h" | |
24 | #include "block/block_int.h" | |
25 | #include "qemu/error-report.h" | |
26 | #include "qemu/bswap.h" | |
27 | #include "qemu/memalign.h" | |
28 | #include "vhdx.h" | |
29 | ||
30 | ||
31 | typedef struct VHDXLogSequence { | |
32 | bool valid; | |
33 | uint32_t count; | |
34 | VHDXLogEntries log; | |
35 | VHDXLogEntryHeader hdr; | |
36 | } VHDXLogSequence; | |
37 | ||
38 | typedef struct VHDXLogDescEntries { | |
39 | VHDXLogEntryHeader hdr; | |
40 | VHDXLogDescriptor desc[]; | |
41 | } VHDXLogDescEntries; | |
42 | ||
43 | static const MSGUID zero_guid = { 0 }; | |
44 | ||
45 | /* The log located on the disk is circular buffer containing | |
46 | * sectors of 4096 bytes each. | |
47 | * | |
48 | * It is assumed for the read/write functions below that the | |
49 | * circular buffer scheme uses a 'one sector open' to indicate | |
50 | * the buffer is full. Given the validation methods used for each | |
51 | * sector, this method should be compatible with other methods that | |
52 | * do not waste a sector. | |
53 | */ | |
54 | ||
55 | ||
56 | /* Allow peeking at the hdr entry at the beginning of the current | |
57 | * read index, without advancing the read index */ | |
58 | static int vhdx_log_peek_hdr(BlockDriverState *bs, VHDXLogEntries *log, | |
59 | VHDXLogEntryHeader *hdr) | |
60 | { | |
61 | int ret = 0; | |
62 | uint64_t offset; | |
63 | uint32_t read; | |
64 | ||
65 | assert(hdr != NULL); | |
66 | ||
67 | /* peek is only supported on sector boundaries */ | |
68 | if (log->read % VHDX_LOG_SECTOR_SIZE) { | |
69 | ret = -EFAULT; | |
70 | goto exit; | |
71 | } | |
72 | ||
73 | read = log->read; | |
74 | /* we are guaranteed that a) log sectors are 4096 bytes, | |
75 | * and b) the log length is a multiple of 1MB. So, there | |
76 | * is always a round number of sectors in the buffer */ | |
77 | if ((read + sizeof(VHDXLogEntryHeader)) > log->length) { | |
78 | read = 0; | |
79 | } | |
80 | ||
81 | if (read == log->write) { | |
82 | ret = -EINVAL; | |
83 | goto exit; | |
84 | } | |
85 | ||
86 | offset = log->offset + read; | |
87 | ||
88 | ret = bdrv_pread(bs->file, offset, sizeof(VHDXLogEntryHeader), hdr, 0); | |
89 | if (ret < 0) { | |
90 | goto exit; | |
91 | } | |
92 | vhdx_log_entry_hdr_le_import(hdr); | |
93 | ||
94 | exit: | |
95 | return ret; | |
96 | } | |
97 | ||
98 | /* Index increment for log, based on sector boundaries */ | |
99 | static int vhdx_log_inc_idx(uint32_t idx, uint64_t length) | |
100 | { | |
101 | idx += VHDX_LOG_SECTOR_SIZE; | |
102 | /* we are guaranteed that a) log sectors are 4096 bytes, | |
103 | * and b) the log length is a multiple of 1MB. So, there | |
104 | * is always a round number of sectors in the buffer */ | |
105 | return idx >= length ? 0 : idx; | |
106 | } | |
107 | ||
108 | ||
109 | /* Reset the log to empty */ | |
110 | static void vhdx_log_reset(BlockDriverState *bs, BDRVVHDXState *s) | |
111 | { | |
112 | MSGUID guid = { 0 }; | |
113 | s->log.read = s->log.write = 0; | |
114 | /* a log guid of 0 indicates an empty log to any parser of v0 | |
115 | * VHDX logs */ | |
116 | vhdx_update_headers(bs, s, false, &guid); | |
117 | } | |
118 | ||
119 | /* Reads num_sectors from the log (all log sectors are 4096 bytes), | |
120 | * into buffer 'buffer'. Upon return, *sectors_read will contain | |
121 | * the number of sectors successfully read. | |
122 | * | |
123 | * It is assumed that 'buffer' is already allocated, and of sufficient | |
124 | * size (i.e. >= 4096*num_sectors). | |
125 | * | |
126 | * If 'peek' is true, then the tail (read) pointer for the circular buffer is | |
127 | * not modified. | |
128 | * | |
129 | * 0 is returned on success, -errno otherwise. */ | |
130 | static int vhdx_log_read_sectors(BlockDriverState *bs, VHDXLogEntries *log, | |
131 | uint32_t *sectors_read, void *buffer, | |
132 | uint32_t num_sectors, bool peek) | |
133 | { | |
134 | int ret = 0; | |
135 | uint64_t offset; | |
136 | uint32_t read; | |
137 | ||
138 | read = log->read; | |
139 | ||
140 | *sectors_read = 0; | |
141 | while (num_sectors) { | |
142 | if (read == log->write) { | |
143 | /* empty */ | |
144 | break; | |
145 | } | |
146 | offset = log->offset + read; | |
147 | ||
148 | ret = bdrv_pread(bs->file, offset, VHDX_LOG_SECTOR_SIZE, buffer, 0); | |
149 | if (ret < 0) { | |
150 | goto exit; | |
151 | } | |
152 | read = vhdx_log_inc_idx(read, log->length); | |
153 | ||
154 | *sectors_read = *sectors_read + 1; | |
155 | num_sectors--; | |
156 | } | |
157 | ||
158 | exit: | |
159 | if (!peek) { | |
160 | log->read = read; | |
161 | } | |
162 | return ret; | |
163 | } | |
164 | ||
165 | /* Writes num_sectors to the log (all log sectors are 4096 bytes), | |
166 | * from buffer 'buffer'. Upon return, *sectors_written will contain | |
167 | * the number of sectors successfully written. | |
168 | * | |
169 | * It is assumed that 'buffer' is at least 4096*num_sectors large. | |
170 | * | |
171 | * 0 is returned on success, -errno otherwise */ | |
172 | static int vhdx_log_write_sectors(BlockDriverState *bs, VHDXLogEntries *log, | |
173 | uint32_t *sectors_written, void *buffer, | |
174 | uint32_t num_sectors) | |
175 | { | |
176 | int ret = 0; | |
177 | uint64_t offset; | |
178 | uint32_t write; | |
179 | void *buffer_tmp; | |
180 | BDRVVHDXState *s = bs->opaque; | |
181 | ||
182 | ret = vhdx_user_visible_write(bs, s); | |
183 | if (ret < 0) { | |
184 | goto exit; | |
185 | } | |
186 | ||
187 | write = log->write; | |
188 | ||
189 | buffer_tmp = buffer; | |
190 | while (num_sectors) { | |
191 | ||
192 | offset = log->offset + write; | |
193 | write = vhdx_log_inc_idx(write, log->length); | |
194 | if (write == log->read) { | |
195 | /* full */ | |
196 | break; | |
197 | } | |
198 | ret = bdrv_pwrite(bs->file, offset, VHDX_LOG_SECTOR_SIZE, buffer_tmp, | |
199 | 0); | |
200 | if (ret < 0) { | |
201 | goto exit; | |
202 | } | |
203 | buffer_tmp += VHDX_LOG_SECTOR_SIZE; | |
204 | ||
205 | log->write = write; | |
206 | *sectors_written = *sectors_written + 1; | |
207 | num_sectors--; | |
208 | } | |
209 | ||
210 | exit: | |
211 | return ret; | |
212 | } | |
213 | ||
214 | ||
215 | /* Validates a log entry header */ | |
216 | static bool vhdx_log_hdr_is_valid(VHDXLogEntries *log, VHDXLogEntryHeader *hdr, | |
217 | BDRVVHDXState *s) | |
218 | { | |
219 | int valid = false; | |
220 | ||
221 | if (hdr->signature != VHDX_LOG_SIGNATURE) { | |
222 | goto exit; | |
223 | } | |
224 | ||
225 | /* if the individual entry length is larger than the whole log | |
226 | * buffer, that is obviously invalid */ | |
227 | if (log->length < hdr->entry_length) { | |
228 | goto exit; | |
229 | } | |
230 | ||
231 | /* length of entire entry must be in units of 4KB (log sector size) */ | |
232 | if (hdr->entry_length % (VHDX_LOG_SECTOR_SIZE)) { | |
233 | goto exit; | |
234 | } | |
235 | ||
236 | /* per spec, sequence # must be > 0 */ | |
237 | if (hdr->sequence_number == 0) { | |
238 | goto exit; | |
239 | } | |
240 | ||
241 | /* log entries are only valid if they match the file-wide log guid | |
242 | * found in the active header */ | |
243 | if (!guid_eq(hdr->log_guid, s->headers[s->curr_header]->log_guid)) { | |
244 | goto exit; | |
245 | } | |
246 | ||
247 | if (hdr->descriptor_count * sizeof(VHDXLogDescriptor) > hdr->entry_length) { | |
248 | goto exit; | |
249 | } | |
250 | ||
251 | valid = true; | |
252 | ||
253 | exit: | |
254 | return valid; | |
255 | } | |
256 | ||
257 | /* | |
258 | * Given a log header, this will validate that the descriptors and the | |
259 | * corresponding data sectors (if applicable) | |
260 | * | |
261 | * Validation consists of: | |
262 | * 1. Making sure the sequence numbers matches the entry header | |
263 | * 2. Verifying a valid signature ('zero' or 'desc' for descriptors) | |
264 | * 3. File offset field is a multiple of 4KB | |
265 | * 4. If a data descriptor, the corresponding data sector | |
266 | * has its signature ('data') and matching sequence number | |
267 | * | |
268 | * @desc: the data buffer containing the descriptor | |
269 | * @hdr: the log entry header | |
270 | * | |
271 | * Returns true if valid | |
272 | */ | |
273 | static bool vhdx_log_desc_is_valid(VHDXLogDescriptor *desc, | |
274 | VHDXLogEntryHeader *hdr) | |
275 | { | |
276 | bool ret = false; | |
277 | ||
278 | if (desc->sequence_number != hdr->sequence_number) { | |
279 | goto exit; | |
280 | } | |
281 | if (desc->file_offset % VHDX_LOG_SECTOR_SIZE) { | |
282 | goto exit; | |
283 | } | |
284 | ||
285 | if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) { | |
286 | if (desc->zero_length % VHDX_LOG_SECTOR_SIZE == 0) { | |
287 | /* valid */ | |
288 | ret = true; | |
289 | } | |
290 | } else if (desc->signature == VHDX_LOG_DESC_SIGNATURE) { | |
291 | /* valid */ | |
292 | ret = true; | |
293 | } | |
294 | ||
295 | exit: | |
296 | return ret; | |
297 | } | |
298 | ||
299 | ||
300 | /* Prior to sector data for a log entry, there is the header | |
301 | * and the descriptors referenced in the header: | |
302 | * | |
303 | * [] = 4KB sector | |
304 | * | |
305 | * [ hdr, desc ][ desc ][ ... ][ data ][ ... ] | |
306 | * | |
307 | * The first sector in a log entry has a 64 byte header, and | |
308 | * up to 126 32-byte descriptors. If more descriptors than | |
309 | * 126 are required, then subsequent sectors can have up to 128 | |
310 | * descriptors. Each sector is 4KB. Data follows the descriptor | |
311 | * sectors. | |
312 | * | |
313 | * This will return the number of sectors needed to encompass | |
314 | * the passed number of descriptors in desc_cnt. | |
315 | * | |
316 | * This will never return 0, even if desc_cnt is 0. | |
317 | */ | |
318 | static int vhdx_compute_desc_sectors(uint32_t desc_cnt) | |
319 | { | |
320 | uint32_t desc_sectors; | |
321 | ||
322 | desc_cnt += 2; /* account for header in first sector */ | |
323 | desc_sectors = desc_cnt / 128; | |
324 | if (desc_cnt % 128) { | |
325 | desc_sectors++; | |
326 | } | |
327 | ||
328 | return desc_sectors; | |
329 | } | |
330 | ||
331 | ||
332 | /* Reads the log header, and subsequent descriptors (if any). This | |
333 | * will allocate all the space for buffer, which must be NULL when | |
334 | * passed into this function. Each descriptor will also be validated, | |
335 | * and error returned if any are invalid. */ | |
336 | static int vhdx_log_read_desc(BlockDriverState *bs, BDRVVHDXState *s, | |
337 | VHDXLogEntries *log, VHDXLogDescEntries **buffer, | |
338 | bool convert_endian) | |
339 | { | |
340 | int ret = 0; | |
341 | uint32_t desc_sectors; | |
342 | uint32_t sectors_read; | |
343 | VHDXLogEntryHeader hdr; | |
344 | VHDXLogDescEntries *desc_entries = NULL; | |
345 | VHDXLogDescriptor desc; | |
346 | int i; | |
347 | ||
348 | assert(*buffer == NULL); | |
349 | ||
350 | ret = vhdx_log_peek_hdr(bs, log, &hdr); | |
351 | if (ret < 0) { | |
352 | goto exit; | |
353 | } | |
354 | ||
355 | if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) { | |
356 | ret = -EINVAL; | |
357 | goto exit; | |
358 | } | |
359 | ||
360 | desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count); | |
361 | desc_entries = qemu_try_blockalign(bs->file->bs, | |
362 | desc_sectors * VHDX_LOG_SECTOR_SIZE); | |
363 | if (desc_entries == NULL) { | |
364 | ret = -ENOMEM; | |
365 | goto exit; | |
366 | } | |
367 | ||
368 | ret = vhdx_log_read_sectors(bs, log, §ors_read, desc_entries, | |
369 | desc_sectors, false); | |
370 | if (ret < 0) { | |
371 | goto free_and_exit; | |
372 | } | |
373 | if (sectors_read != desc_sectors) { | |
374 | ret = -EINVAL; | |
375 | goto free_and_exit; | |
376 | } | |
377 | ||
378 | /* put in proper endianness, and validate each desc */ | |
379 | for (i = 0; i < hdr.descriptor_count; i++) { | |
380 | desc = desc_entries->desc[i]; | |
381 | vhdx_log_desc_le_import(&desc); | |
382 | if (convert_endian) { | |
383 | desc_entries->desc[i] = desc; | |
384 | } | |
385 | if (vhdx_log_desc_is_valid(&desc, &hdr) == false) { | |
386 | ret = -EINVAL; | |
387 | goto free_and_exit; | |
388 | } | |
389 | } | |
390 | if (convert_endian) { | |
391 | desc_entries->hdr = hdr; | |
392 | } | |
393 | ||
394 | *buffer = desc_entries; | |
395 | goto exit; | |
396 | ||
397 | free_and_exit: | |
398 | qemu_vfree(desc_entries); | |
399 | exit: | |
400 | return ret; | |
401 | } | |
402 | ||
403 | ||
404 | /* Flushes the descriptor described by desc to the VHDX image file. | |
405 | * If the descriptor is a data descriptor, than 'data' must be non-NULL, | |
406 | * and >= 4096 bytes (VHDX_LOG_SECTOR_SIZE), containing the data to be | |
407 | * written. | |
408 | * | |
409 | * Verification is performed to make sure the sequence numbers of a data | |
410 | * descriptor match the sequence number in the desc. | |
411 | * | |
412 | * For a zero descriptor, it may describe multiple sectors to fill with zeroes. | |
413 | * In this case, it should be noted that zeroes are written to disk, and the | |
414 | * image file is not extended as a sparse file. */ | |
415 | static int vhdx_log_flush_desc(BlockDriverState *bs, VHDXLogDescriptor *desc, | |
416 | VHDXLogDataSector *data) | |
417 | { | |
418 | int ret = 0; | |
419 | uint64_t seq, file_offset; | |
420 | uint32_t offset = 0; | |
421 | void *buffer = NULL; | |
422 | uint64_t count = 1; | |
423 | int i; | |
424 | ||
425 | buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE); | |
426 | ||
427 | if (desc->signature == VHDX_LOG_DESC_SIGNATURE) { | |
428 | /* data sector */ | |
429 | if (data == NULL) { | |
430 | ret = -EFAULT; | |
431 | goto exit; | |
432 | } | |
433 | ||
434 | /* The sequence number of the data sector must match that | |
435 | * in the descriptor */ | |
436 | seq = data->sequence_high; | |
437 | seq <<= 32; | |
438 | seq |= data->sequence_low & 0xffffffff; | |
439 | ||
440 | if (seq != desc->sequence_number) { | |
441 | ret = -EINVAL; | |
442 | goto exit; | |
443 | } | |
444 | ||
445 | /* Each data sector is in total 4096 bytes, however the first | |
446 | * 8 bytes, and last 4 bytes, are located in the descriptor */ | |
447 | memcpy(buffer, &desc->leading_bytes, 8); | |
448 | offset += 8; | |
449 | ||
450 | memcpy(buffer+offset, data->data, 4084); | |
451 | offset += 4084; | |
452 | ||
453 | memcpy(buffer+offset, &desc->trailing_bytes, 4); | |
454 | ||
455 | } else if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) { | |
456 | /* write 'count' sectors of sector */ | |
457 | memset(buffer, 0, VHDX_LOG_SECTOR_SIZE); | |
458 | count = desc->zero_length / VHDX_LOG_SECTOR_SIZE; | |
459 | } else { | |
460 | error_report("Invalid VHDX log descriptor entry signature 0x%" PRIx32, | |
461 | desc->signature); | |
462 | ret = -EINVAL; | |
463 | goto exit; | |
464 | } | |
465 | ||
466 | file_offset = desc->file_offset; | |
467 | ||
468 | /* count is only > 1 if we are writing zeroes */ | |
469 | for (i = 0; i < count; i++) { | |
470 | ret = bdrv_pwrite_sync(bs->file, file_offset, VHDX_LOG_SECTOR_SIZE, | |
471 | buffer, 0); | |
472 | if (ret < 0) { | |
473 | goto exit; | |
474 | } | |
475 | file_offset += VHDX_LOG_SECTOR_SIZE; | |
476 | } | |
477 | ||
478 | exit: | |
479 | qemu_vfree(buffer); | |
480 | return ret; | |
481 | } | |
482 | ||
483 | /* Flush the entire log (as described by 'logs') to the VHDX image | |
484 | * file, and then set the log to 'empty' status once complete. | |
485 | * | |
486 | * The log entries should be validate prior to flushing */ | |
487 | static int vhdx_log_flush(BlockDriverState *bs, BDRVVHDXState *s, | |
488 | VHDXLogSequence *logs) | |
489 | { | |
490 | int ret = 0; | |
491 | int i; | |
492 | uint32_t cnt, sectors_read; | |
493 | uint64_t new_file_size; | |
494 | void *data = NULL; | |
495 | int64_t file_length; | |
496 | VHDXLogDescEntries *desc_entries = NULL; | |
497 | VHDXLogEntryHeader hdr_tmp = { 0 }; | |
498 | ||
499 | cnt = logs->count; | |
500 | ||
501 | data = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE); | |
502 | ||
503 | ret = vhdx_user_visible_write(bs, s); | |
504 | if (ret < 0) { | |
505 | goto exit; | |
506 | } | |
507 | ||
508 | /* each iteration represents one log sequence, which may span multiple | |
509 | * sectors */ | |
510 | while (cnt--) { | |
511 | ret = vhdx_log_peek_hdr(bs, &logs->log, &hdr_tmp); | |
512 | if (ret < 0) { | |
513 | goto exit; | |
514 | } | |
515 | file_length = bdrv_getlength(bs->file->bs); | |
516 | if (file_length < 0) { | |
517 | ret = file_length; | |
518 | goto exit; | |
519 | } | |
520 | /* if the log shows a FlushedFileOffset larger than our current file | |
521 | * size, then that means the file has been truncated / corrupted, and | |
522 | * we must refused to open it / use it */ | |
523 | if (hdr_tmp.flushed_file_offset > file_length) { | |
524 | ret = -EINVAL; | |
525 | goto exit; | |
526 | } | |
527 | ||
528 | ret = vhdx_log_read_desc(bs, s, &logs->log, &desc_entries, true); | |
529 | if (ret < 0) { | |
530 | goto exit; | |
531 | } | |
532 | ||
533 | for (i = 0; i < desc_entries->hdr.descriptor_count; i++) { | |
534 | if (desc_entries->desc[i].signature == VHDX_LOG_DESC_SIGNATURE) { | |
535 | /* data sector, so read a sector to flush */ | |
536 | ret = vhdx_log_read_sectors(bs, &logs->log, §ors_read, | |
537 | data, 1, false); | |
538 | if (ret < 0) { | |
539 | goto exit; | |
540 | } | |
541 | if (sectors_read != 1) { | |
542 | ret = -EINVAL; | |
543 | goto exit; | |
544 | } | |
545 | vhdx_log_data_le_import(data); | |
546 | } | |
547 | ||
548 | ret = vhdx_log_flush_desc(bs, &desc_entries->desc[i], data); | |
549 | if (ret < 0) { | |
550 | goto exit; | |
551 | } | |
552 | } | |
553 | if (file_length < desc_entries->hdr.last_file_offset) { | |
554 | new_file_size = desc_entries->hdr.last_file_offset; | |
555 | if (new_file_size % (1 * MiB)) { | |
556 | /* round up to nearest 1MB boundary */ | |
557 | new_file_size = QEMU_ALIGN_UP(new_file_size, MiB); | |
558 | if (new_file_size > INT64_MAX) { | |
559 | ret = -EINVAL; | |
560 | goto exit; | |
561 | } | |
562 | ret = bdrv_truncate(bs->file, new_file_size, false, | |
563 | PREALLOC_MODE_OFF, 0, NULL); | |
564 | if (ret < 0) { | |
565 | goto exit; | |
566 | } | |
567 | } | |
568 | } | |
569 | qemu_vfree(desc_entries); | |
570 | desc_entries = NULL; | |
571 | } | |
572 | ||
573 | ret = bdrv_flush(bs); | |
574 | if (ret < 0) { | |
575 | goto exit; | |
576 | } | |
577 | /* once the log is fully flushed, indicate that we have an empty log | |
578 | * now. This also sets the log guid to 0, to indicate an empty log */ | |
579 | vhdx_log_reset(bs, s); | |
580 | ||
581 | exit: | |
582 | qemu_vfree(data); | |
583 | qemu_vfree(desc_entries); | |
584 | return ret; | |
585 | } | |
586 | ||
587 | static int vhdx_validate_log_entry(BlockDriverState *bs, BDRVVHDXState *s, | |
588 | VHDXLogEntries *log, uint64_t seq, | |
589 | bool *valid, VHDXLogEntryHeader *entry) | |
590 | { | |
591 | int ret = 0; | |
592 | VHDXLogEntryHeader hdr; | |
593 | void *buffer = NULL; | |
594 | uint32_t i, desc_sectors, total_sectors, crc; | |
595 | uint32_t sectors_read = 0; | |
596 | VHDXLogDescEntries *desc_buffer = NULL; | |
597 | ||
598 | *valid = false; | |
599 | ||
600 | ret = vhdx_log_peek_hdr(bs, log, &hdr); | |
601 | if (ret < 0) { | |
602 | goto inc_and_exit; | |
603 | } | |
604 | ||
605 | if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) { | |
606 | goto inc_and_exit; | |
607 | } | |
608 | ||
609 | if (seq > 0) { | |
610 | if (hdr.sequence_number != seq + 1) { | |
611 | goto inc_and_exit; | |
612 | } | |
613 | } | |
614 | ||
615 | desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count); | |
616 | ||
617 | /* Read all log sectors, and calculate log checksum */ | |
618 | ||
619 | total_sectors = hdr.entry_length / VHDX_LOG_SECTOR_SIZE; | |
620 | ||
621 | ||
622 | /* read_desc() will increment the read idx */ | |
623 | ret = vhdx_log_read_desc(bs, s, log, &desc_buffer, false); | |
624 | if (ret < 0) { | |
625 | goto free_and_exit; | |
626 | } | |
627 | ||
628 | crc = vhdx_checksum_calc(0xffffffff, (void *)desc_buffer, | |
629 | desc_sectors * VHDX_LOG_SECTOR_SIZE, 4); | |
630 | crc ^= 0xffffffff; | |
631 | ||
632 | buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE); | |
633 | if (total_sectors > desc_sectors) { | |
634 | for (i = 0; i < total_sectors - desc_sectors; i++) { | |
635 | sectors_read = 0; | |
636 | ret = vhdx_log_read_sectors(bs, log, §ors_read, buffer, | |
637 | 1, false); | |
638 | if (ret < 0 || sectors_read != 1) { | |
639 | goto free_and_exit; | |
640 | } | |
641 | crc = vhdx_checksum_calc(crc, buffer, VHDX_LOG_SECTOR_SIZE, -1); | |
642 | crc ^= 0xffffffff; | |
643 | } | |
644 | } | |
645 | crc ^= 0xffffffff; | |
646 | if (crc != hdr.checksum) { | |
647 | goto free_and_exit; | |
648 | } | |
649 | ||
650 | *valid = true; | |
651 | *entry = hdr; | |
652 | goto free_and_exit; | |
653 | ||
654 | inc_and_exit: | |
655 | log->read = vhdx_log_inc_idx(log->read, log->length); | |
656 | ||
657 | free_and_exit: | |
658 | qemu_vfree(buffer); | |
659 | qemu_vfree(desc_buffer); | |
660 | return ret; | |
661 | } | |
662 | ||
663 | /* Search through the log circular buffer, and find the valid, active | |
664 | * log sequence, if any exists | |
665 | * */ | |
666 | static int vhdx_log_search(BlockDriverState *bs, BDRVVHDXState *s, | |
667 | VHDXLogSequence *logs) | |
668 | { | |
669 | int ret = 0; | |
670 | uint32_t tail; | |
671 | bool seq_valid = false; | |
672 | VHDXLogSequence candidate = { 0 }; | |
673 | VHDXLogEntryHeader hdr = { 0 }; | |
674 | VHDXLogEntries curr_log; | |
675 | ||
676 | memcpy(&curr_log, &s->log, sizeof(VHDXLogEntries)); | |
677 | curr_log.write = curr_log.length; /* assume log is full */ | |
678 | curr_log.read = 0; | |
679 | ||
680 | ||
681 | /* now we will go through the whole log sector by sector, until | |
682 | * we find a valid, active log sequence, or reach the end of the | |
683 | * log buffer */ | |
684 | for (;;) { | |
685 | uint64_t curr_seq = 0; | |
686 | VHDXLogSequence current = { 0 }; | |
687 | ||
688 | tail = curr_log.read; | |
689 | ||
690 | ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq, | |
691 | &seq_valid, &hdr); | |
692 | if (ret < 0) { | |
693 | goto exit; | |
694 | } | |
695 | ||
696 | if (seq_valid) { | |
697 | current.valid = true; | |
698 | current.log = curr_log; | |
699 | current.log.read = tail; | |
700 | current.log.write = curr_log.read; | |
701 | current.count = 1; | |
702 | current.hdr = hdr; | |
703 | ||
704 | ||
705 | for (;;) { | |
706 | ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq, | |
707 | &seq_valid, &hdr); | |
708 | if (ret < 0) { | |
709 | goto exit; | |
710 | } | |
711 | if (seq_valid == false) { | |
712 | break; | |
713 | } | |
714 | current.log.write = curr_log.read; | |
715 | current.count++; | |
716 | ||
717 | curr_seq = hdr.sequence_number; | |
718 | } | |
719 | } | |
720 | ||
721 | if (current.valid) { | |
722 | if (candidate.valid == false || | |
723 | current.hdr.sequence_number > candidate.hdr.sequence_number) { | |
724 | candidate = current; | |
725 | } | |
726 | } | |
727 | ||
728 | if (curr_log.read < tail) { | |
729 | break; | |
730 | } | |
731 | } | |
732 | ||
733 | *logs = candidate; | |
734 | ||
735 | if (candidate.valid) { | |
736 | /* this is the next sequence number, for writes */ | |
737 | s->log.sequence = candidate.hdr.sequence_number + 1; | |
738 | } | |
739 | ||
740 | ||
741 | exit: | |
742 | return ret; | |
743 | } | |
744 | ||
745 | /* Parse the replay log. Per the VHDX spec, if the log is present | |
746 | * it must be replayed prior to opening the file, even read-only. | |
747 | * | |
748 | * If read-only, we must replay the log in RAM (or refuse to open | |
749 | * a dirty VHDX file read-only) */ | |
750 | int vhdx_parse_log(BlockDriverState *bs, BDRVVHDXState *s, bool *flushed, | |
751 | Error **errp) | |
752 | { | |
753 | int ret = 0; | |
754 | VHDXHeader *hdr; | |
755 | VHDXLogSequence logs = { 0 }; | |
756 | ||
757 | hdr = s->headers[s->curr_header]; | |
758 | ||
759 | *flushed = false; | |
760 | ||
761 | /* s->log.hdr is freed in vhdx_close() */ | |
762 | if (s->log.hdr == NULL) { | |
763 | s->log.hdr = qemu_blockalign(bs, sizeof(VHDXLogEntryHeader)); | |
764 | } | |
765 | ||
766 | s->log.offset = hdr->log_offset; | |
767 | s->log.length = hdr->log_length; | |
768 | ||
769 | if (s->log.offset < VHDX_LOG_MIN_SIZE || | |
770 | s->log.offset % VHDX_LOG_MIN_SIZE) { | |
771 | ret = -EINVAL; | |
772 | goto exit; | |
773 | } | |
774 | ||
775 | /* per spec, only log version of 0 is supported */ | |
776 | if (hdr->log_version != 0) { | |
777 | ret = -EINVAL; | |
778 | goto exit; | |
779 | } | |
780 | ||
781 | /* If either the log guid, or log length is zero, | |
782 | * then a replay log is not present */ | |
783 | if (guid_eq(hdr->log_guid, zero_guid)) { | |
784 | goto exit; | |
785 | } | |
786 | ||
787 | if (hdr->log_length == 0) { | |
788 | goto exit; | |
789 | } | |
790 | ||
791 | if (hdr->log_length % VHDX_LOG_MIN_SIZE) { | |
792 | ret = -EINVAL; | |
793 | goto exit; | |
794 | } | |
795 | ||
796 | ||
797 | /* The log is present, we need to find if and where there is an active | |
798 | * sequence of valid entries present in the log. */ | |
799 | ||
800 | ret = vhdx_log_search(bs, s, &logs); | |
801 | if (ret < 0) { | |
802 | goto exit; | |
803 | } | |
804 | ||
805 | if (logs.valid) { | |
806 | if (bdrv_is_read_only(bs)) { | |
807 | bdrv_refresh_filename(bs); | |
808 | ret = -EPERM; | |
809 | error_setg(errp, | |
810 | "VHDX image file '%s' opened read-only, but " | |
811 | "contains a log that needs to be replayed", | |
812 | bs->filename); | |
813 | error_append_hint(errp, "To replay the log, run:\n" | |
814 | "qemu-img check -r all '%s'\n", | |
815 | bs->filename); | |
816 | goto exit; | |
817 | } | |
818 | /* now flush the log */ | |
819 | ret = vhdx_log_flush(bs, s, &logs); | |
820 | if (ret < 0) { | |
821 | goto exit; | |
822 | } | |
823 | *flushed = true; | |
824 | } | |
825 | ||
826 | ||
827 | exit: | |
828 | return ret; | |
829 | } | |
830 | ||
831 | ||
832 | ||
833 | static void vhdx_log_raw_to_le_sector(VHDXLogDescriptor *desc, | |
834 | VHDXLogDataSector *sector, void *data, | |
835 | uint64_t seq) | |
836 | { | |
837 | /* 8 + 4084 + 4 = 4096, 1 log sector */ | |
838 | memcpy(&desc->leading_bytes, data, 8); | |
839 | data += 8; | |
840 | desc->leading_bytes = cpu_to_le64(desc->leading_bytes); | |
841 | memcpy(sector->data, data, 4084); | |
842 | data += 4084; | |
843 | memcpy(&desc->trailing_bytes, data, 4); | |
844 | desc->trailing_bytes = cpu_to_le32(desc->trailing_bytes); | |
845 | data += 4; | |
846 | ||
847 | sector->sequence_high = (uint32_t) (seq >> 32); | |
848 | sector->sequence_low = (uint32_t) (seq & 0xffffffff); | |
849 | sector->data_signature = VHDX_LOG_DATA_SIGNATURE; | |
850 | ||
851 | vhdx_log_desc_le_export(desc); | |
852 | vhdx_log_data_le_export(sector); | |
853 | } | |
854 | ||
855 | ||
856 | static int vhdx_log_write(BlockDriverState *bs, BDRVVHDXState *s, | |
857 | void *data, uint32_t length, uint64_t offset) | |
858 | { | |
859 | int ret = 0; | |
860 | void *buffer = NULL; | |
861 | void *merged_sector = NULL; | |
862 | void *data_tmp, *sector_write; | |
863 | unsigned int i; | |
864 | int sector_offset; | |
865 | uint32_t desc_sectors, sectors, total_length; | |
866 | uint32_t sectors_written = 0; | |
867 | uint32_t aligned_length; | |
868 | uint32_t leading_length = 0; | |
869 | uint32_t trailing_length = 0; | |
870 | uint32_t partial_sectors = 0; | |
871 | uint32_t bytes_written = 0; | |
872 | uint64_t file_offset; | |
873 | int64_t file_length; | |
874 | VHDXHeader *header; | |
875 | VHDXLogEntryHeader new_hdr; | |
876 | VHDXLogDescriptor *new_desc = NULL; | |
877 | VHDXLogDataSector *data_sector = NULL; | |
878 | MSGUID new_guid = { 0 }; | |
879 | ||
880 | header = s->headers[s->curr_header]; | |
881 | ||
882 | /* need to have offset read data, and be on 4096 byte boundary */ | |
883 | ||
884 | if (length > header->log_length) { | |
885 | /* no log present. we could create a log here instead of failing */ | |
886 | ret = -EINVAL; | |
887 | goto exit; | |
888 | } | |
889 | ||
890 | if (guid_eq(header->log_guid, zero_guid)) { | |
891 | vhdx_guid_generate(&new_guid); | |
892 | vhdx_update_headers(bs, s, false, &new_guid); | |
893 | } else { | |
894 | /* currently, we require that the log be flushed after | |
895 | * every write. */ | |
896 | ret = -ENOTSUP; | |
897 | goto exit; | |
898 | } | |
899 | ||
900 | /* 0 is an invalid sequence number, but may also represent the first | |
901 | * log write (or a wrapped seq) */ | |
902 | if (s->log.sequence == 0) { | |
903 | s->log.sequence = 1; | |
904 | } | |
905 | ||
906 | sector_offset = offset % VHDX_LOG_SECTOR_SIZE; | |
907 | file_offset = QEMU_ALIGN_DOWN(offset, VHDX_LOG_SECTOR_SIZE); | |
908 | ||
909 | aligned_length = length; | |
910 | ||
911 | /* add in the unaligned head and tail bytes */ | |
912 | if (sector_offset) { | |
913 | leading_length = (VHDX_LOG_SECTOR_SIZE - sector_offset); | |
914 | leading_length = leading_length > length ? length : leading_length; | |
915 | aligned_length -= leading_length; | |
916 | partial_sectors++; | |
917 | } | |
918 | ||
919 | sectors = aligned_length / VHDX_LOG_SECTOR_SIZE; | |
920 | trailing_length = aligned_length - (sectors * VHDX_LOG_SECTOR_SIZE); | |
921 | if (trailing_length) { | |
922 | partial_sectors++; | |
923 | } | |
924 | ||
925 | sectors += partial_sectors; | |
926 | ||
927 | file_length = bdrv_getlength(bs->file->bs); | |
928 | if (file_length < 0) { | |
929 | ret = file_length; | |
930 | goto exit; | |
931 | } | |
932 | ||
933 | /* sectors is now how many sectors the data itself takes, not | |
934 | * including the header and descriptor metadata */ | |
935 | ||
936 | new_hdr = (VHDXLogEntryHeader) { | |
937 | .signature = VHDX_LOG_SIGNATURE, | |
938 | .tail = s->log.tail, | |
939 | .sequence_number = s->log.sequence, | |
940 | .descriptor_count = sectors, | |
941 | .reserved = 0, | |
942 | .flushed_file_offset = file_length, | |
943 | .last_file_offset = file_length, | |
944 | .log_guid = header->log_guid, | |
945 | }; | |
946 | ||
947 | ||
948 | desc_sectors = vhdx_compute_desc_sectors(new_hdr.descriptor_count); | |
949 | ||
950 | total_length = (desc_sectors + sectors) * VHDX_LOG_SECTOR_SIZE; | |
951 | new_hdr.entry_length = total_length; | |
952 | ||
953 | vhdx_log_entry_hdr_le_export(&new_hdr); | |
954 | ||
955 | buffer = qemu_blockalign(bs, total_length); | |
956 | memcpy(buffer, &new_hdr, sizeof(new_hdr)); | |
957 | ||
958 | new_desc = buffer + sizeof(new_hdr); | |
959 | data_sector = buffer + (desc_sectors * VHDX_LOG_SECTOR_SIZE); | |
960 | data_tmp = data; | |
961 | ||
962 | /* All log sectors are 4KB, so for any partial sectors we must | |
963 | * merge the data with preexisting data from the final file | |
964 | * destination */ | |
965 | merged_sector = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE); | |
966 | ||
967 | for (i = 0; i < sectors; i++) { | |
968 | new_desc->signature = VHDX_LOG_DESC_SIGNATURE; | |
969 | new_desc->sequence_number = s->log.sequence; | |
970 | new_desc->file_offset = file_offset; | |
971 | ||
972 | if (i == 0 && leading_length) { | |
973 | /* partial sector at the front of the buffer */ | |
974 | ret = bdrv_pread(bs->file, file_offset, VHDX_LOG_SECTOR_SIZE, | |
975 | merged_sector, 0); | |
976 | if (ret < 0) { | |
977 | goto exit; | |
978 | } | |
979 | memcpy(merged_sector + sector_offset, data_tmp, leading_length); | |
980 | bytes_written = leading_length; | |
981 | sector_write = merged_sector; | |
982 | } else if (i == sectors - 1 && trailing_length) { | |
983 | /* partial sector at the end of the buffer */ | |
984 | ret = bdrv_pread(bs->file, file_offset + trailing_length, | |
985 | VHDX_LOG_SECTOR_SIZE - trailing_length, | |
986 | merged_sector + trailing_length, 0); | |
987 | if (ret < 0) { | |
988 | goto exit; | |
989 | } | |
990 | memcpy(merged_sector, data_tmp, trailing_length); | |
991 | bytes_written = trailing_length; | |
992 | sector_write = merged_sector; | |
993 | } else { | |
994 | bytes_written = VHDX_LOG_SECTOR_SIZE; | |
995 | sector_write = data_tmp; | |
996 | } | |
997 | ||
998 | /* populate the raw sector data into the proper structures, | |
999 | * as well as update the descriptor, and convert to proper | |
1000 | * endianness */ | |
1001 | vhdx_log_raw_to_le_sector(new_desc, data_sector, sector_write, | |
1002 | s->log.sequence); | |
1003 | ||
1004 | data_tmp += bytes_written; | |
1005 | data_sector++; | |
1006 | new_desc++; | |
1007 | file_offset += VHDX_LOG_SECTOR_SIZE; | |
1008 | } | |
1009 | ||
1010 | /* checksum covers entire entry, from the log header through the | |
1011 | * last data sector */ | |
1012 | vhdx_update_checksum(buffer, total_length, | |
1013 | offsetof(VHDXLogEntryHeader, checksum)); | |
1014 | ||
1015 | /* now write to the log */ | |
1016 | ret = vhdx_log_write_sectors(bs, &s->log, §ors_written, buffer, | |
1017 | desc_sectors + sectors); | |
1018 | if (ret < 0) { | |
1019 | goto exit; | |
1020 | } | |
1021 | ||
1022 | if (sectors_written != desc_sectors + sectors) { | |
1023 | /* instead of failing, we could flush the log here */ | |
1024 | ret = -EINVAL; | |
1025 | goto exit; | |
1026 | } | |
1027 | ||
1028 | s->log.sequence++; | |
1029 | /* write new tail */ | |
1030 | s->log.tail = s->log.write; | |
1031 | ||
1032 | exit: | |
1033 | qemu_vfree(buffer); | |
1034 | qemu_vfree(merged_sector); | |
1035 | return ret; | |
1036 | } | |
1037 | ||
1038 | /* Perform a log write, and then immediately flush the entire log */ | |
1039 | int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s, | |
1040 | void *data, uint32_t length, uint64_t offset) | |
1041 | { | |
1042 | int ret = 0; | |
1043 | VHDXLogSequence logs = { .valid = true, | |
1044 | .count = 1, | |
1045 | .hdr = { 0 } }; | |
1046 | ||
1047 | ||
1048 | /* Make sure data written (new and/or changed blocks) is stable | |
1049 | * on disk, before creating log entry */ | |
1050 | ret = bdrv_flush(bs); | |
1051 | if (ret < 0) { | |
1052 | goto exit; | |
1053 | } | |
1054 | ||
1055 | ret = vhdx_log_write(bs, s, data, length, offset); | |
1056 | if (ret < 0) { | |
1057 | goto exit; | |
1058 | } | |
1059 | logs.log = s->log; | |
1060 | ||
1061 | /* Make sure log is stable on disk */ | |
1062 | ret = bdrv_flush(bs); | |
1063 | if (ret < 0) { | |
1064 | goto exit; | |
1065 | } | |
1066 | ||
1067 | ret = vhdx_log_flush(bs, s, &logs); | |
1068 | if (ret < 0) { | |
1069 | goto exit; | |
1070 | } | |
1071 | ||
1072 | s->log = logs.log; | |
1073 | ||
1074 | exit: | |
1075 | return ret; | |
1076 | } | |
1077 |