]> git.proxmox.com Git - mirror_qemu.git/blame - block/vhdx-log.c
Merge tag 'win32-pull-request' of https://gitlab.com/marcandre.lureau/qemu into staging
[mirror_qemu.git] / block / vhdx-log.c
CommitLineData
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 <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 */
0b8fa32f 20
80c71a24 21#include "qemu/osdep.h"
da34e65c 22#include "qapi/error.h"
0a43a1b5 23#include "block/block_int.h"
d49b6836 24#include "qemu/error-report.h"
58369e22 25#include "qemu/bswap.h"
5df022cf 26#include "qemu/memalign.h"
0d8c41da 27#include "vhdx.h"
0a43a1b5
JC
28
29
30typedef struct VHDXLogSequence {
31 bool valid;
32 uint32_t count;
33 VHDXLogEntries log;
34 VHDXLogEntryHeader hdr;
35} VHDXLogSequence;
36
37typedef struct VHDXLogDescEntries {
38 VHDXLogEntryHeader hdr;
39 VHDXLogDescriptor desc[];
40} VHDXLogDescEntries;
41
42static const MSGUID zero_guid = { 0 };
43
44/* The log located on the disk is circular buffer containing
45 * sectors of 4096 bytes each.
46 *
47 * It is assumed for the read/write functions below that the
48 * circular buffer scheme uses a 'one sector open' to indicate
49 * the buffer is full. Given the validation methods used for each
50 * sector, this method should be compatible with other methods that
51 * do not waste a sector.
52 */
53
54
55/* Allow peeking at the hdr entry at the beginning of the current
56 * read index, without advancing the read index */
57static int vhdx_log_peek_hdr(BlockDriverState *bs, VHDXLogEntries *log,
58 VHDXLogEntryHeader *hdr)
59{
60 int ret = 0;
61 uint64_t offset;
62 uint32_t read;
63
64 assert(hdr != NULL);
65
66 /* peek is only supported on sector boundaries */
67 if (log->read % VHDX_LOG_SECTOR_SIZE) {
68 ret = -EFAULT;
69 goto exit;
70 }
71
72 read = log->read;
73 /* we are guaranteed that a) log sectors are 4096 bytes,
74 * and b) the log length is a multiple of 1MB. So, there
75 * is always a round number of sectors in the buffer */
76 if ((read + sizeof(VHDXLogEntryHeader)) > log->length) {
77 read = 0;
78 }
79
80 if (read == log->write) {
81 ret = -EINVAL;
82 goto exit;
83 }
84
85 offset = log->offset + read;
86
32cc71de 87 ret = bdrv_pread(bs->file, offset, sizeof(VHDXLogEntryHeader), hdr, 0);
0a43a1b5
JC
88 if (ret < 0) {
89 goto exit;
90 }
4f75b52a 91 vhdx_log_entry_hdr_le_import(hdr);
0a43a1b5
JC
92
93exit:
94 return ret;
95}
96
97/* Index increment for log, based on sector boundaries */
98static int vhdx_log_inc_idx(uint32_t idx, uint64_t length)
99{
100 idx += VHDX_LOG_SECTOR_SIZE;
101 /* we are guaranteed that a) log sectors are 4096 bytes,
102 * and b) the log length is a multiple of 1MB. So, there
103 * is always a round number of sectors in the buffer */
104 return idx >= length ? 0 : idx;
105}
106
107
108/* Reset the log to empty */
109static void vhdx_log_reset(BlockDriverState *bs, BDRVVHDXState *s)
110{
111 MSGUID guid = { 0 };
112 s->log.read = s->log.write = 0;
113 /* a log guid of 0 indicates an empty log to any parser of v0
114 * VHDX logs */
115 vhdx_update_headers(bs, s, false, &guid);
116}
117
118/* Reads num_sectors from the log (all log sectors are 4096 bytes),
119 * into buffer 'buffer'. Upon return, *sectors_read will contain
120 * the number of sectors successfully read.
121 *
122 * It is assumed that 'buffer' is already allocated, and of sufficient
123 * size (i.e. >= 4096*num_sectors).
124 *
125 * If 'peek' is true, then the tail (read) pointer for the circular buffer is
126 * not modified.
127 *
128 * 0 is returned on success, -errno otherwise. */
129static int vhdx_log_read_sectors(BlockDriverState *bs, VHDXLogEntries *log,
130 uint32_t *sectors_read, void *buffer,
131 uint32_t num_sectors, bool peek)
132{
133 int ret = 0;
134 uint64_t offset;
135 uint32_t read;
136
137 read = log->read;
138
139 *sectors_read = 0;
140 while (num_sectors) {
141 if (read == log->write) {
142 /* empty */
143 break;
144 }
145 offset = log->offset + read;
146
32cc71de 147 ret = bdrv_pread(bs->file, offset, VHDX_LOG_SECTOR_SIZE, buffer, 0);
0a43a1b5
JC
148 if (ret < 0) {
149 goto exit;
150 }
151 read = vhdx_log_inc_idx(read, log->length);
152
153 *sectors_read = *sectors_read + 1;
154 num_sectors--;
155 }
156
157exit:
158 if (!peek) {
159 log->read = read;
160 }
161 return ret;
162}
163
8adc5233
JC
164/* Writes num_sectors to the log (all log sectors are 4096 bytes),
165 * from buffer 'buffer'. Upon return, *sectors_written will contain
166 * the number of sectors successfully written.
167 *
168 * It is assumed that 'buffer' is at least 4096*num_sectors large.
169 *
170 * 0 is returned on success, -errno otherwise */
171static int vhdx_log_write_sectors(BlockDriverState *bs, VHDXLogEntries *log,
172 uint32_t *sectors_written, void *buffer,
173 uint32_t num_sectors)
174{
175 int ret = 0;
176 uint64_t offset;
177 uint32_t write;
178 void *buffer_tmp;
179 BDRVVHDXState *s = bs->opaque;
180
181 ret = vhdx_user_visible_write(bs, s);
182 if (ret < 0) {
183 goto exit;
184 }
185
186 write = log->write;
187
188 buffer_tmp = buffer;
189 while (num_sectors) {
190
191 offset = log->offset + write;
192 write = vhdx_log_inc_idx(write, log->length);
193 if (write == log->read) {
194 /* full */
195 break;
196 }
32cc71de 197 ret = bdrv_pwrite(bs->file, offset, VHDX_LOG_SECTOR_SIZE, buffer_tmp,
53fb7844 198 0);
8adc5233
JC
199 if (ret < 0) {
200 goto exit;
201 }
202 buffer_tmp += VHDX_LOG_SECTOR_SIZE;
203
204 log->write = write;
205 *sectors_written = *sectors_written + 1;
206 num_sectors--;
207 }
208
209exit:
210 return ret;
211}
212
213
0a43a1b5
JC
214/* Validates a log entry header */
215static bool vhdx_log_hdr_is_valid(VHDXLogEntries *log, VHDXLogEntryHeader *hdr,
216 BDRVVHDXState *s)
217{
218 int valid = false;
219
4f75b52a 220 if (hdr->signature != VHDX_LOG_SIGNATURE) {
0a43a1b5
JC
221 goto exit;
222 }
223
224 /* if the individual entry length is larger than the whole log
225 * buffer, that is obviously invalid */
226 if (log->length < hdr->entry_length) {
227 goto exit;
228 }
229
230 /* length of entire entry must be in units of 4KB (log sector size) */
231 if (hdr->entry_length % (VHDX_LOG_SECTOR_SIZE)) {
232 goto exit;
233 }
234
235 /* per spec, sequence # must be > 0 */
236 if (hdr->sequence_number == 0) {
237 goto exit;
238 }
239
240 /* log entries are only valid if they match the file-wide log guid
241 * found in the active header */
242 if (!guid_eq(hdr->log_guid, s->headers[s->curr_header]->log_guid)) {
243 goto exit;
244 }
245
246 if (hdr->descriptor_count * sizeof(VHDXLogDescriptor) > hdr->entry_length) {
247 goto exit;
248 }
249
250 valid = true;
251
252exit:
253 return valid;
254}
255
256/*
257 * Given a log header, this will validate that the descriptors and the
258 * corresponding data sectors (if applicable)
259 *
260 * Validation consists of:
261 * 1. Making sure the sequence numbers matches the entry header
262 * 2. Verifying a valid signature ('zero' or 'desc' for descriptors)
263 * 3. File offset field is a multiple of 4KB
264 * 4. If a data descriptor, the corresponding data sector
265 * has its signature ('data') and matching sequence number
266 *
267 * @desc: the data buffer containing the descriptor
268 * @hdr: the log entry header
269 *
270 * Returns true if valid
271 */
272static bool vhdx_log_desc_is_valid(VHDXLogDescriptor *desc,
273 VHDXLogEntryHeader *hdr)
274{
275 bool ret = false;
276
277 if (desc->sequence_number != hdr->sequence_number) {
278 goto exit;
279 }
280 if (desc->file_offset % VHDX_LOG_SECTOR_SIZE) {
281 goto exit;
282 }
283
4f75b52a 284 if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) {
0a43a1b5
JC
285 if (desc->zero_length % VHDX_LOG_SECTOR_SIZE == 0) {
286 /* valid */
287 ret = true;
288 }
4f75b52a 289 } else if (desc->signature == VHDX_LOG_DESC_SIGNATURE) {
0a43a1b5
JC
290 /* valid */
291 ret = true;
292 }
293
294exit:
295 return ret;
296}
297
298
299/* Prior to sector data for a log entry, there is the header
300 * and the descriptors referenced in the header:
301 *
302 * [] = 4KB sector
303 *
304 * [ hdr, desc ][ desc ][ ... ][ data ][ ... ]
305 *
306 * The first sector in a log entry has a 64 byte header, and
307 * up to 126 32-byte descriptors. If more descriptors than
308 * 126 are required, then subsequent sectors can have up to 128
309 * descriptors. Each sector is 4KB. Data follows the descriptor
310 * sectors.
311 *
312 * This will return the number of sectors needed to encompass
313 * the passed number of descriptors in desc_cnt.
314 *
315 * This will never return 0, even if desc_cnt is 0.
316 */
317static int vhdx_compute_desc_sectors(uint32_t desc_cnt)
318{
319 uint32_t desc_sectors;
320
321 desc_cnt += 2; /* account for header in first sector */
322 desc_sectors = desc_cnt / 128;
323 if (desc_cnt % 128) {
324 desc_sectors++;
325 }
326
327 return desc_sectors;
328}
329
330
331/* Reads the log header, and subsequent descriptors (if any). This
332 * will allocate all the space for buffer, which must be NULL when
333 * passed into this function. Each descriptor will also be validated,
334 * and error returned if any are invalid. */
335static int vhdx_log_read_desc(BlockDriverState *bs, BDRVVHDXState *s,
4f75b52a
JC
336 VHDXLogEntries *log, VHDXLogDescEntries **buffer,
337 bool convert_endian)
0a43a1b5
JC
338{
339 int ret = 0;
340 uint32_t desc_sectors;
341 uint32_t sectors_read;
342 VHDXLogEntryHeader hdr;
343 VHDXLogDescEntries *desc_entries = NULL;
4f75b52a 344 VHDXLogDescriptor desc;
0a43a1b5
JC
345 int i;
346
347 assert(*buffer == NULL);
348
349 ret = vhdx_log_peek_hdr(bs, log, &hdr);
350 if (ret < 0) {
351 goto exit;
352 }
4f75b52a 353
0a43a1b5
JC
354 if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) {
355 ret = -EINVAL;
356 goto exit;
357 }
358
359 desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count);
9a4f4c31 360 desc_entries = qemu_try_blockalign(bs->file->bs,
a67e128a
KW
361 desc_sectors * VHDX_LOG_SECTOR_SIZE);
362 if (desc_entries == NULL) {
363 ret = -ENOMEM;
364 goto exit;
365 }
0a43a1b5
JC
366
367 ret = vhdx_log_read_sectors(bs, log, &sectors_read, desc_entries,
368 desc_sectors, false);
369 if (ret < 0) {
370 goto free_and_exit;
371 }
372 if (sectors_read != desc_sectors) {
373 ret = -EINVAL;
374 goto free_and_exit;
375 }
376
377 /* put in proper endianness, and validate each desc */
378 for (i = 0; i < hdr.descriptor_count; i++) {
4f75b52a
JC
379 desc = desc_entries->desc[i];
380 vhdx_log_desc_le_import(&desc);
381 if (convert_endian) {
382 desc_entries->desc[i] = desc;
383 }
384 if (vhdx_log_desc_is_valid(&desc, &hdr) == false) {
0a43a1b5
JC
385 ret = -EINVAL;
386 goto free_and_exit;
387 }
388 }
4f75b52a
JC
389 if (convert_endian) {
390 desc_entries->hdr = hdr;
391 }
0a43a1b5
JC
392
393 *buffer = desc_entries;
394 goto exit;
395
396free_and_exit:
397 qemu_vfree(desc_entries);
398exit:
399 return ret;
400}
401
402
403/* Flushes the descriptor described by desc to the VHDX image file.
404 * If the descriptor is a data descriptor, than 'data' must be non-NULL,
405 * and >= 4096 bytes (VHDX_LOG_SECTOR_SIZE), containing the data to be
406 * written.
407 *
408 * Verification is performed to make sure the sequence numbers of a data
409 * descriptor match the sequence number in the desc.
410 *
411 * For a zero descriptor, it may describe multiple sectors to fill with zeroes.
412 * In this case, it should be noted that zeroes are written to disk, and the
413 * image file is not extended as a sparse file. */
414static int vhdx_log_flush_desc(BlockDriverState *bs, VHDXLogDescriptor *desc,
415 VHDXLogDataSector *data)
416{
417 int ret = 0;
418 uint64_t seq, file_offset;
419 uint32_t offset = 0;
420 void *buffer = NULL;
421 uint64_t count = 1;
422 int i;
423
424 buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
425
4f75b52a 426 if (desc->signature == VHDX_LOG_DESC_SIGNATURE) {
0a43a1b5
JC
427 /* data sector */
428 if (data == NULL) {
429 ret = -EFAULT;
430 goto exit;
431 }
432
433 /* The sequence number of the data sector must match that
434 * in the descriptor */
435 seq = data->sequence_high;
436 seq <<= 32;
437 seq |= data->sequence_low & 0xffffffff;
438
439 if (seq != desc->sequence_number) {
440 ret = -EINVAL;
441 goto exit;
442 }
443
444 /* Each data sector is in total 4096 bytes, however the first
445 * 8 bytes, and last 4 bytes, are located in the descriptor */
446 memcpy(buffer, &desc->leading_bytes, 8);
447 offset += 8;
448
449 memcpy(buffer+offset, data->data, 4084);
450 offset += 4084;
451
452 memcpy(buffer+offset, &desc->trailing_bytes, 4);
453
4f75b52a 454 } else if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) {
0a43a1b5
JC
455 /* write 'count' sectors of sector */
456 memset(buffer, 0, VHDX_LOG_SECTOR_SIZE);
457 count = desc->zero_length / VHDX_LOG_SECTOR_SIZE;
349592e0
JC
458 } else {
459 error_report("Invalid VHDX log descriptor entry signature 0x%" PRIx32,
460 desc->signature);
461 ret = -EINVAL;
462 goto exit;
0a43a1b5
JC
463 }
464
465 file_offset = desc->file_offset;
466
467 /* count is only > 1 if we are writing zeroes */
468 for (i = 0; i < count; i++) {
32cc71de
AF
469 ret = bdrv_pwrite_sync(bs->file, file_offset, VHDX_LOG_SECTOR_SIZE,
470 buffer, 0);
0a43a1b5
JC
471 if (ret < 0) {
472 goto exit;
473 }
474 file_offset += VHDX_LOG_SECTOR_SIZE;
475 }
476
477exit:
478 qemu_vfree(buffer);
479 return ret;
480}
481
482/* Flush the entire log (as described by 'logs') to the VHDX image
483 * file, and then set the log to 'empty' status once complete.
484 *
485 * The log entries should be validate prior to flushing */
486static int vhdx_log_flush(BlockDriverState *bs, BDRVVHDXState *s,
487 VHDXLogSequence *logs)
488{
489 int ret = 0;
490 int i;
491 uint32_t cnt, sectors_read;
492 uint64_t new_file_size;
493 void *data = NULL;
3f910692 494 int64_t file_length;
0a43a1b5
JC
495 VHDXLogDescEntries *desc_entries = NULL;
496 VHDXLogEntryHeader hdr_tmp = { 0 };
497
498 cnt = logs->count;
499
500 data = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
501
502 ret = vhdx_user_visible_write(bs, s);
503 if (ret < 0) {
504 goto exit;
505 }
506
507 /* each iteration represents one log sequence, which may span multiple
508 * sectors */
509 while (cnt--) {
510 ret = vhdx_log_peek_hdr(bs, &logs->log, &hdr_tmp);
511 if (ret < 0) {
512 goto exit;
513 }
3f910692
JC
514 file_length = bdrv_getlength(bs->file->bs);
515 if (file_length < 0) {
516 ret = file_length;
517 goto exit;
518 }
0a43a1b5
JC
519 /* if the log shows a FlushedFileOffset larger than our current file
520 * size, then that means the file has been truncated / corrupted, and
521 * we must refused to open it / use it */
3f910692 522 if (hdr_tmp.flushed_file_offset > file_length) {
0a43a1b5
JC
523 ret = -EINVAL;
524 goto exit;
525 }
526
4f75b52a 527 ret = vhdx_log_read_desc(bs, s, &logs->log, &desc_entries, true);
0a43a1b5
JC
528 if (ret < 0) {
529 goto exit;
530 }
531
532 for (i = 0; i < desc_entries->hdr.descriptor_count; i++) {
4f75b52a 533 if (desc_entries->desc[i].signature == VHDX_LOG_DESC_SIGNATURE) {
0a43a1b5
JC
534 /* data sector, so read a sector to flush */
535 ret = vhdx_log_read_sectors(bs, &logs->log, &sectors_read,
536 data, 1, false);
537 if (ret < 0) {
538 goto exit;
539 }
540 if (sectors_read != 1) {
541 ret = -EINVAL;
542 goto exit;
543 }
4f75b52a 544 vhdx_log_data_le_import(data);
0a43a1b5
JC
545 }
546
547 ret = vhdx_log_flush_desc(bs, &desc_entries->desc[i], data);
548 if (ret < 0) {
549 goto exit;
550 }
551 }
3f910692 552 if (file_length < desc_entries->hdr.last_file_offset) {
0a43a1b5 553 new_file_size = desc_entries->hdr.last_file_offset;
0cb98af2 554 if (new_file_size % (1 * MiB)) {
0a43a1b5 555 /* round up to nearest 1MB boundary */
27539ac5
JC
556 new_file_size = QEMU_ALIGN_UP(new_file_size, MiB);
557 if (new_file_size > INT64_MAX) {
558 ret = -EINVAL;
559 goto exit;
560 }
c80d8b06 561 ret = bdrv_truncate(bs->file, new_file_size, false,
7b8e4857 562 PREALLOC_MODE_OFF, 0, NULL);
95d72983
JC
563 if (ret < 0) {
564 goto exit;
565 }
0a43a1b5
JC
566 }
567 }
568 qemu_vfree(desc_entries);
569 desc_entries = NULL;
570 }
571
c6572fa0
JC
572 ret = bdrv_flush(bs);
573 if (ret < 0) {
574 goto exit;
575 }
0a43a1b5
JC
576 /* once the log is fully flushed, indicate that we have an empty log
577 * now. This also sets the log guid to 0, to indicate an empty log */
578 vhdx_log_reset(bs, s);
579
580exit:
581 qemu_vfree(data);
582 qemu_vfree(desc_entries);
583 return ret;
584}
585
586static int vhdx_validate_log_entry(BlockDriverState *bs, BDRVVHDXState *s,
587 VHDXLogEntries *log, uint64_t seq,
588 bool *valid, VHDXLogEntryHeader *entry)
589{
590 int ret = 0;
591 VHDXLogEntryHeader hdr;
592 void *buffer = NULL;
593 uint32_t i, desc_sectors, total_sectors, crc;
594 uint32_t sectors_read = 0;
595 VHDXLogDescEntries *desc_buffer = NULL;
596
597 *valid = false;
598
599 ret = vhdx_log_peek_hdr(bs, log, &hdr);
600 if (ret < 0) {
601 goto inc_and_exit;
602 }
603
0a43a1b5
JC
604 if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) {
605 goto inc_and_exit;
606 }
607
608 if (seq > 0) {
609 if (hdr.sequence_number != seq + 1) {
610 goto inc_and_exit;
611 }
612 }
613
614 desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count);
615
4f75b52a 616 /* Read all log sectors, and calculate log checksum */
0a43a1b5
JC
617
618 total_sectors = hdr.entry_length / VHDX_LOG_SECTOR_SIZE;
619
620
dc6fb73d 621 /* read_desc() will increment the read idx */
4f75b52a 622 ret = vhdx_log_read_desc(bs, s, log, &desc_buffer, false);
0a43a1b5
JC
623 if (ret < 0) {
624 goto free_and_exit;
625 }
626
627 crc = vhdx_checksum_calc(0xffffffff, (void *)desc_buffer,
628 desc_sectors * VHDX_LOG_SECTOR_SIZE, 4);
629 crc ^= 0xffffffff;
630
631 buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
632 if (total_sectors > desc_sectors) {
633 for (i = 0; i < total_sectors - desc_sectors; i++) {
634 sectors_read = 0;
635 ret = vhdx_log_read_sectors(bs, log, &sectors_read, buffer,
636 1, false);
637 if (ret < 0 || sectors_read != 1) {
638 goto free_and_exit;
639 }
640 crc = vhdx_checksum_calc(crc, buffer, VHDX_LOG_SECTOR_SIZE, -1);
641 crc ^= 0xffffffff;
642 }
643 }
644 crc ^= 0xffffffff;
4f75b52a 645 if (crc != hdr.checksum) {
0a43a1b5
JC
646 goto free_and_exit;
647 }
648
649 *valid = true;
650 *entry = hdr;
651 goto free_and_exit;
652
653inc_and_exit:
654 log->read = vhdx_log_inc_idx(log->read, log->length);
655
656free_and_exit:
657 qemu_vfree(buffer);
658 qemu_vfree(desc_buffer);
659 return ret;
660}
661
662/* Search through the log circular buffer, and find the valid, active
663 * log sequence, if any exists
664 * */
665static int vhdx_log_search(BlockDriverState *bs, BDRVVHDXState *s,
666 VHDXLogSequence *logs)
667{
668 int ret = 0;
669 uint32_t tail;
670 bool seq_valid = false;
671 VHDXLogSequence candidate = { 0 };
672 VHDXLogEntryHeader hdr = { 0 };
673 VHDXLogEntries curr_log;
674
675 memcpy(&curr_log, &s->log, sizeof(VHDXLogEntries));
676 curr_log.write = curr_log.length; /* assume log is full */
677 curr_log.read = 0;
678
679
680 /* now we will go through the whole log sector by sector, until
681 * we find a valid, active log sequence, or reach the end of the
682 * log buffer */
683 for (;;) {
684 uint64_t curr_seq = 0;
685 VHDXLogSequence current = { 0 };
686
687 tail = curr_log.read;
688
689 ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq,
690 &seq_valid, &hdr);
691 if (ret < 0) {
692 goto exit;
693 }
694
695 if (seq_valid) {
696 current.valid = true;
697 current.log = curr_log;
698 current.log.read = tail;
699 current.log.write = curr_log.read;
700 current.count = 1;
701 current.hdr = hdr;
702
703
704 for (;;) {
705 ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq,
706 &seq_valid, &hdr);
707 if (ret < 0) {
708 goto exit;
709 }
710 if (seq_valid == false) {
711 break;
712 }
713 current.log.write = curr_log.read;
714 current.count++;
715
716 curr_seq = hdr.sequence_number;
717 }
718 }
719
720 if (current.valid) {
721 if (candidate.valid == false ||
722 current.hdr.sequence_number > candidate.hdr.sequence_number) {
723 candidate = current;
724 }
725 }
726
727 if (curr_log.read < tail) {
728 break;
729 }
730 }
731
732 *logs = candidate;
733
734 if (candidate.valid) {
735 /* this is the next sequence number, for writes */
736 s->log.sequence = candidate.hdr.sequence_number + 1;
737 }
738
739
740exit:
741 return ret;
742}
743
744/* Parse the replay log. Per the VHDX spec, if the log is present
745 * it must be replayed prior to opening the file, even read-only.
746 *
747 * If read-only, we must replay the log in RAM (or refuse to open
748 * a dirty VHDX file read-only) */
7e30e6a6
JC
749int vhdx_parse_log(BlockDriverState *bs, BDRVVHDXState *s, bool *flushed,
750 Error **errp)
0a43a1b5
JC
751{
752 int ret = 0;
753 VHDXHeader *hdr;
754 VHDXLogSequence logs = { 0 };
755
756 hdr = s->headers[s->curr_header];
757
758 *flushed = false;
759
760 /* s->log.hdr is freed in vhdx_close() */
761 if (s->log.hdr == NULL) {
762 s->log.hdr = qemu_blockalign(bs, sizeof(VHDXLogEntryHeader));
763 }
764
765 s->log.offset = hdr->log_offset;
766 s->log.length = hdr->log_length;
767
768 if (s->log.offset < VHDX_LOG_MIN_SIZE ||
769 s->log.offset % VHDX_LOG_MIN_SIZE) {
770 ret = -EINVAL;
771 goto exit;
772 }
773
774 /* per spec, only log version of 0 is supported */
775 if (hdr->log_version != 0) {
776 ret = -EINVAL;
777 goto exit;
778 }
779
780 /* If either the log guid, or log length is zero,
781 * then a replay log is not present */
782 if (guid_eq(hdr->log_guid, zero_guid)) {
783 goto exit;
784 }
785
786 if (hdr->log_length == 0) {
787 goto exit;
788 }
789
790 if (hdr->log_length % VHDX_LOG_MIN_SIZE) {
791 ret = -EINVAL;
792 goto exit;
793 }
794
795
796 /* The log is present, we need to find if and where there is an active
797 * sequence of valid entries present in the log. */
798
799 ret = vhdx_log_search(bs, s, &logs);
800 if (ret < 0) {
801 goto exit;
802 }
803
804 if (logs.valid) {
307261b2 805 if (bdrv_is_read_only(bs)) {
f30c66ba 806 bdrv_refresh_filename(bs);
7e30e6a6 807 ret = -EPERM;
bf89e874
MA
808 error_setg(errp,
809 "VHDX image file '%s' opened read-only, but "
810 "contains a log that needs to be replayed",
811 bs->filename);
812 error_append_hint(errp, "To replay the log, run:\n"
813 "qemu-img check -r all '%s'\n",
814 bs->filename);
7e30e6a6
JC
815 goto exit;
816 }
0a43a1b5
JC
817 /* now flush the log */
818 ret = vhdx_log_flush(bs, s, &logs);
819 if (ret < 0) {
820 goto exit;
821 }
822 *flushed = true;
823 }
824
825
826exit:
827 return ret;
828}
829
830
8adc5233
JC
831
832static void vhdx_log_raw_to_le_sector(VHDXLogDescriptor *desc,
833 VHDXLogDataSector *sector, void *data,
834 uint64_t seq)
835{
836 /* 8 + 4084 + 4 = 4096, 1 log sector */
837 memcpy(&desc->leading_bytes, data, 8);
838 data += 8;
1229e46d 839 desc->leading_bytes = cpu_to_le64(desc->leading_bytes);
8adc5233
JC
840 memcpy(sector->data, data, 4084);
841 data += 4084;
842 memcpy(&desc->trailing_bytes, data, 4);
1229e46d 843 desc->trailing_bytes = cpu_to_le32(desc->trailing_bytes);
8adc5233
JC
844 data += 4;
845
846 sector->sequence_high = (uint32_t) (seq >> 32);
847 sector->sequence_low = (uint32_t) (seq & 0xffffffff);
848 sector->data_signature = VHDX_LOG_DATA_SIGNATURE;
849
850 vhdx_log_desc_le_export(desc);
851 vhdx_log_data_le_export(sector);
852}
853
854
855static int vhdx_log_write(BlockDriverState *bs, BDRVVHDXState *s,
856 void *data, uint32_t length, uint64_t offset)
857{
858 int ret = 0;
859 void *buffer = NULL;
860 void *merged_sector = NULL;
861 void *data_tmp, *sector_write;
862 unsigned int i;
863 int sector_offset;
864 uint32_t desc_sectors, sectors, total_length;
865 uint32_t sectors_written = 0;
866 uint32_t aligned_length;
867 uint32_t leading_length = 0;
868 uint32_t trailing_length = 0;
869 uint32_t partial_sectors = 0;
870 uint32_t bytes_written = 0;
871 uint64_t file_offset;
3f910692 872 int64_t file_length;
8adc5233
JC
873 VHDXHeader *header;
874 VHDXLogEntryHeader new_hdr;
875 VHDXLogDescriptor *new_desc = NULL;
876 VHDXLogDataSector *data_sector = NULL;
877 MSGUID new_guid = { 0 };
878
879 header = s->headers[s->curr_header];
880
881 /* need to have offset read data, and be on 4096 byte boundary */
882
883 if (length > header->log_length) {
884 /* no log present. we could create a log here instead of failing */
885 ret = -EINVAL;
886 goto exit;
887 }
888
889 if (guid_eq(header->log_guid, zero_guid)) {
890 vhdx_guid_generate(&new_guid);
891 vhdx_update_headers(bs, s, false, &new_guid);
892 } else {
893 /* currently, we require that the log be flushed after
894 * every write. */
895 ret = -ENOTSUP;
896 goto exit;
897 }
898
899 /* 0 is an invalid sequence number, but may also represent the first
900 * log write (or a wrapped seq) */
901 if (s->log.sequence == 0) {
902 s->log.sequence = 1;
903 }
904
905 sector_offset = offset % VHDX_LOG_SECTOR_SIZE;
cf7a09c1 906 file_offset = QEMU_ALIGN_DOWN(offset, VHDX_LOG_SECTOR_SIZE);
8adc5233
JC
907
908 aligned_length = length;
909
910 /* add in the unaligned head and tail bytes */
911 if (sector_offset) {
912 leading_length = (VHDX_LOG_SECTOR_SIZE - sector_offset);
913 leading_length = leading_length > length ? length : leading_length;
914 aligned_length -= leading_length;
915 partial_sectors++;
916 }
917
918 sectors = aligned_length / VHDX_LOG_SECTOR_SIZE;
919 trailing_length = aligned_length - (sectors * VHDX_LOG_SECTOR_SIZE);
920 if (trailing_length) {
921 partial_sectors++;
922 }
923
924 sectors += partial_sectors;
925
3f910692
JC
926 file_length = bdrv_getlength(bs->file->bs);
927 if (file_length < 0) {
928 ret = file_length;
929 goto exit;
930 }
931
8adc5233
JC
932 /* sectors is now how many sectors the data itself takes, not
933 * including the header and descriptor metadata */
934
935 new_hdr = (VHDXLogEntryHeader) {
936 .signature = VHDX_LOG_SIGNATURE,
937 .tail = s->log.tail,
938 .sequence_number = s->log.sequence,
939 .descriptor_count = sectors,
940 .reserved = 0,
3f910692
JC
941 .flushed_file_offset = file_length,
942 .last_file_offset = file_length,
943 .log_guid = header->log_guid,
8adc5233
JC
944 };
945
8adc5233
JC
946
947 desc_sectors = vhdx_compute_desc_sectors(new_hdr.descriptor_count);
948
949 total_length = (desc_sectors + sectors) * VHDX_LOG_SECTOR_SIZE;
950 new_hdr.entry_length = total_length;
951
952 vhdx_log_entry_hdr_le_export(&new_hdr);
953
954 buffer = qemu_blockalign(bs, total_length);
955 memcpy(buffer, &new_hdr, sizeof(new_hdr));
956
d4df3dbc 957 new_desc = buffer + sizeof(new_hdr);
8adc5233
JC
958 data_sector = buffer + (desc_sectors * VHDX_LOG_SECTOR_SIZE);
959 data_tmp = data;
960
961 /* All log sectors are 4KB, so for any partial sectors we must
962 * merge the data with preexisting data from the final file
963 * destination */
964 merged_sector = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
965
966 for (i = 0; i < sectors; i++) {
967 new_desc->signature = VHDX_LOG_DESC_SIGNATURE;
968 new_desc->sequence_number = s->log.sequence;
969 new_desc->file_offset = file_offset;
970
971 if (i == 0 && leading_length) {
972 /* partial sector at the front of the buffer */
32cc71de
AF
973 ret = bdrv_pread(bs->file, file_offset, VHDX_LOG_SECTOR_SIZE,
974 merged_sector, 0);
8adc5233
JC
975 if (ret < 0) {
976 goto exit;
977 }
978 memcpy(merged_sector + sector_offset, data_tmp, leading_length);
979 bytes_written = leading_length;
980 sector_write = merged_sector;
981 } else if (i == sectors - 1 && trailing_length) {
982 /* partial sector at the end of the buffer */
53fb7844 983 ret = bdrv_pread(bs->file, file_offset,
32cc71de
AF
984 VHDX_LOG_SECTOR_SIZE - trailing_length,
985 merged_sector + trailing_length, 0);
8adc5233
JC
986 if (ret < 0) {
987 goto exit;
988 }
989 memcpy(merged_sector, data_tmp, trailing_length);
990 bytes_written = trailing_length;
991 sector_write = merged_sector;
992 } else {
993 bytes_written = VHDX_LOG_SECTOR_SIZE;
994 sector_write = data_tmp;
995 }
996
997 /* populate the raw sector data into the proper structures,
998 * as well as update the descriptor, and convert to proper
999 * endianness */
1000 vhdx_log_raw_to_le_sector(new_desc, data_sector, sector_write,
1001 s->log.sequence);
1002
1003 data_tmp += bytes_written;
1004 data_sector++;
1005 new_desc++;
1006 file_offset += VHDX_LOG_SECTOR_SIZE;
1007 }
1008
1009 /* checksum covers entire entry, from the log header through the
1010 * last data sector */
1011 vhdx_update_checksum(buffer, total_length,
1012 offsetof(VHDXLogEntryHeader, checksum));
8adc5233
JC
1013
1014 /* now write to the log */
f50159fa
MA
1015 ret = vhdx_log_write_sectors(bs, &s->log, &sectors_written, buffer,
1016 desc_sectors + sectors);
8adc5233
JC
1017 if (ret < 0) {
1018 goto exit;
1019 }
1020
1021 if (sectors_written != desc_sectors + sectors) {
1022 /* instead of failing, we could flush the log here */
1023 ret = -EINVAL;
1024 goto exit;
1025 }
1026
1027 s->log.sequence++;
1028 /* write new tail */
1029 s->log.tail = s->log.write;
1030
1031exit:
1032 qemu_vfree(buffer);
1033 qemu_vfree(merged_sector);
1034 return ret;
1035}
1036
1037/* Perform a log write, and then immediately flush the entire log */
1038int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s,
1039 void *data, uint32_t length, uint64_t offset)
1040{
1041 int ret = 0;
1042 VHDXLogSequence logs = { .valid = true,
1043 .count = 1,
1044 .hdr = { 0 } };
1045
1046
1047 /* Make sure data written (new and/or changed blocks) is stable
1048 * on disk, before creating log entry */
c6572fa0
JC
1049 ret = bdrv_flush(bs);
1050 if (ret < 0) {
1051 goto exit;
1052 }
1053
8adc5233
JC
1054 ret = vhdx_log_write(bs, s, data, length, offset);
1055 if (ret < 0) {
1056 goto exit;
1057 }
1058 logs.log = s->log;
1059
1060 /* Make sure log is stable on disk */
c6572fa0
JC
1061 ret = bdrv_flush(bs);
1062 if (ret < 0) {
1063 goto exit;
1064 }
1065
8adc5233
JC
1066 ret = vhdx_log_flush(bs, s, &logs);
1067 if (ret < 0) {
1068 goto exit;
1069 }
1070
1071 s->log = logs.log;
1072
1073exit:
1074 return ret;
1075}
1076