rcu: actually register threads that have RCU read-side critical sections
[qemu/ar7.git] / block / vhdx-log.c
blob47fec63c616cc8431c12cb7313bb6efc50c3a8be
1 /*
2 * Block driver for Hyper-V VHDX Images
4 * Copyright (c) 2013 Red Hat, Inc.,
6 * Authors:
7 * Jeff Cody <jcody@redhat.com>
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
13 * This file covers the functionality of the metadata log writing, parsing, and
14 * replay.
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.
20 #include "qemu-common.h"
21 #include "block/block_int.h"
22 #include "qemu/error-report.h"
23 #include "qemu/module.h"
24 #include "block/vhdx.h"
27 typedef struct VHDXLogSequence {
28 bool valid;
29 uint32_t count;
30 VHDXLogEntries log;
31 VHDXLogEntryHeader hdr;
32 } VHDXLogSequence;
34 typedef struct VHDXLogDescEntries {
35 VHDXLogEntryHeader hdr;
36 VHDXLogDescriptor desc[];
37 } VHDXLogDescEntries;
39 static const MSGUID zero_guid = { 0 };
41 /* The log located on the disk is circular buffer containing
42 * sectors of 4096 bytes each.
44 * It is assumed for the read/write functions below that the
45 * circular buffer scheme uses a 'one sector open' to indicate
46 * the buffer is full. Given the validation methods used for each
47 * sector, this method should be compatible with other methods that
48 * do not waste a sector.
52 /* Allow peeking at the hdr entry at the beginning of the current
53 * read index, without advancing the read index */
54 static int vhdx_log_peek_hdr(BlockDriverState *bs, VHDXLogEntries *log,
55 VHDXLogEntryHeader *hdr)
57 int ret = 0;
58 uint64_t offset;
59 uint32_t read;
61 assert(hdr != NULL);
63 /* peek is only supported on sector boundaries */
64 if (log->read % VHDX_LOG_SECTOR_SIZE) {
65 ret = -EFAULT;
66 goto exit;
69 read = log->read;
70 /* we are guaranteed that a) log sectors are 4096 bytes,
71 * and b) the log length is a multiple of 1MB. So, there
72 * is always a round number of sectors in the buffer */
73 if ((read + sizeof(VHDXLogEntryHeader)) > log->length) {
74 read = 0;
77 if (read == log->write) {
78 ret = -EINVAL;
79 goto exit;
82 offset = log->offset + read;
84 ret = bdrv_pread(bs->file, offset, hdr, sizeof(VHDXLogEntryHeader));
85 if (ret < 0) {
86 goto exit;
88 vhdx_log_entry_hdr_le_import(hdr);
90 exit:
91 return ret;
94 /* Index increment for log, based on sector boundaries */
95 static int vhdx_log_inc_idx(uint32_t idx, uint64_t length)
97 idx += VHDX_LOG_SECTOR_SIZE;
98 /* we are guaranteed that a) log sectors are 4096 bytes,
99 * and b) the log length is a multiple of 1MB. So, there
100 * is always a round number of sectors in the buffer */
101 return idx >= length ? 0 : idx;
105 /* Reset the log to empty */
106 static void vhdx_log_reset(BlockDriverState *bs, BDRVVHDXState *s)
108 MSGUID guid = { 0 };
109 s->log.read = s->log.write = 0;
110 /* a log guid of 0 indicates an empty log to any parser of v0
111 * VHDX logs */
112 vhdx_update_headers(bs, s, false, &guid);
115 /* Reads num_sectors from the log (all log sectors are 4096 bytes),
116 * into buffer 'buffer'. Upon return, *sectors_read will contain
117 * the number of sectors successfully read.
119 * It is assumed that 'buffer' is already allocated, and of sufficient
120 * size (i.e. >= 4096*num_sectors).
122 * If 'peek' is true, then the tail (read) pointer for the circular buffer is
123 * not modified.
125 * 0 is returned on success, -errno otherwise. */
126 static int vhdx_log_read_sectors(BlockDriverState *bs, VHDXLogEntries *log,
127 uint32_t *sectors_read, void *buffer,
128 uint32_t num_sectors, bool peek)
130 int ret = 0;
131 uint64_t offset;
132 uint32_t read;
134 read = log->read;
136 *sectors_read = 0;
137 while (num_sectors) {
138 if (read == log->write) {
139 /* empty */
140 break;
142 offset = log->offset + read;
144 ret = bdrv_pread(bs->file, offset, buffer, VHDX_LOG_SECTOR_SIZE);
145 if (ret < 0) {
146 goto exit;
148 read = vhdx_log_inc_idx(read, log->length);
150 *sectors_read = *sectors_read + 1;
151 num_sectors--;
154 exit:
155 if (!peek) {
156 log->read = read;
158 return ret;
161 /* Writes num_sectors to the log (all log sectors are 4096 bytes),
162 * from buffer 'buffer'. Upon return, *sectors_written will contain
163 * the number of sectors successfully written.
165 * It is assumed that 'buffer' is at least 4096*num_sectors large.
167 * 0 is returned on success, -errno otherwise */
168 static int vhdx_log_write_sectors(BlockDriverState *bs, VHDXLogEntries *log,
169 uint32_t *sectors_written, void *buffer,
170 uint32_t num_sectors)
172 int ret = 0;
173 uint64_t offset;
174 uint32_t write;
175 void *buffer_tmp;
176 BDRVVHDXState *s = bs->opaque;
178 ret = vhdx_user_visible_write(bs, s);
179 if (ret < 0) {
180 goto exit;
183 write = log->write;
185 buffer_tmp = buffer;
186 while (num_sectors) {
188 offset = log->offset + write;
189 write = vhdx_log_inc_idx(write, log->length);
190 if (write == log->read) {
191 /* full */
192 break;
194 ret = bdrv_pwrite(bs->file, offset, buffer_tmp, VHDX_LOG_SECTOR_SIZE);
195 if (ret < 0) {
196 goto exit;
198 buffer_tmp += VHDX_LOG_SECTOR_SIZE;
200 log->write = write;
201 *sectors_written = *sectors_written + 1;
202 num_sectors--;
205 exit:
206 return ret;
210 /* Validates a log entry header */
211 static bool vhdx_log_hdr_is_valid(VHDXLogEntries *log, VHDXLogEntryHeader *hdr,
212 BDRVVHDXState *s)
214 int valid = false;
216 if (hdr->signature != VHDX_LOG_SIGNATURE) {
217 goto exit;
220 /* if the individual entry length is larger than the whole log
221 * buffer, that is obviously invalid */
222 if (log->length < hdr->entry_length) {
223 goto exit;
226 /* length of entire entry must be in units of 4KB (log sector size) */
227 if (hdr->entry_length % (VHDX_LOG_SECTOR_SIZE)) {
228 goto exit;
231 /* per spec, sequence # must be > 0 */
232 if (hdr->sequence_number == 0) {
233 goto exit;
236 /* log entries are only valid if they match the file-wide log guid
237 * found in the active header */
238 if (!guid_eq(hdr->log_guid, s->headers[s->curr_header]->log_guid)) {
239 goto exit;
242 if (hdr->descriptor_count * sizeof(VHDXLogDescriptor) > hdr->entry_length) {
243 goto exit;
246 valid = true;
248 exit:
249 return valid;
253 * Given a log header, this will validate that the descriptors and the
254 * corresponding data sectors (if applicable)
256 * Validation consists of:
257 * 1. Making sure the sequence numbers matches the entry header
258 * 2. Verifying a valid signature ('zero' or 'desc' for descriptors)
259 * 3. File offset field is a multiple of 4KB
260 * 4. If a data descriptor, the corresponding data sector
261 * has its signature ('data') and matching sequence number
263 * @desc: the data buffer containing the descriptor
264 * @hdr: the log entry header
266 * Returns true if valid
268 static bool vhdx_log_desc_is_valid(VHDXLogDescriptor *desc,
269 VHDXLogEntryHeader *hdr)
271 bool ret = false;
273 if (desc->sequence_number != hdr->sequence_number) {
274 goto exit;
276 if (desc->file_offset % VHDX_LOG_SECTOR_SIZE) {
277 goto exit;
280 if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) {
281 if (desc->zero_length % VHDX_LOG_SECTOR_SIZE == 0) {
282 /* valid */
283 ret = true;
285 } else if (desc->signature == VHDX_LOG_DESC_SIGNATURE) {
286 /* valid */
287 ret = true;
290 exit:
291 return ret;
295 /* Prior to sector data for a log entry, there is the header
296 * and the descriptors referenced in the header:
298 * [] = 4KB sector
300 * [ hdr, desc ][ desc ][ ... ][ data ][ ... ]
302 * The first sector in a log entry has a 64 byte header, and
303 * up to 126 32-byte descriptors. If more descriptors than
304 * 126 are required, then subsequent sectors can have up to 128
305 * descriptors. Each sector is 4KB. Data follows the descriptor
306 * sectors.
308 * This will return the number of sectors needed to encompass
309 * the passed number of descriptors in desc_cnt.
311 * This will never return 0, even if desc_cnt is 0.
313 static int vhdx_compute_desc_sectors(uint32_t desc_cnt)
315 uint32_t desc_sectors;
317 desc_cnt += 2; /* account for header in first sector */
318 desc_sectors = desc_cnt / 128;
319 if (desc_cnt % 128) {
320 desc_sectors++;
323 return desc_sectors;
327 /* Reads the log header, and subsequent descriptors (if any). This
328 * will allocate all the space for buffer, which must be NULL when
329 * passed into this function. Each descriptor will also be validated,
330 * and error returned if any are invalid. */
331 static int vhdx_log_read_desc(BlockDriverState *bs, BDRVVHDXState *s,
332 VHDXLogEntries *log, VHDXLogDescEntries **buffer,
333 bool convert_endian)
335 int ret = 0;
336 uint32_t desc_sectors;
337 uint32_t sectors_read;
338 VHDXLogEntryHeader hdr;
339 VHDXLogDescEntries *desc_entries = NULL;
340 VHDXLogDescriptor desc;
341 int i;
343 assert(*buffer == NULL);
345 ret = vhdx_log_peek_hdr(bs, log, &hdr);
346 if (ret < 0) {
347 goto exit;
350 if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) {
351 ret = -EINVAL;
352 goto exit;
355 desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count);
356 desc_entries = qemu_try_blockalign(bs->file,
357 desc_sectors * VHDX_LOG_SECTOR_SIZE);
358 if (desc_entries == NULL) {
359 ret = -ENOMEM;
360 goto exit;
363 ret = vhdx_log_read_sectors(bs, log, &sectors_read, desc_entries,
364 desc_sectors, false);
365 if (ret < 0) {
366 goto free_and_exit;
368 if (sectors_read != desc_sectors) {
369 ret = -EINVAL;
370 goto free_and_exit;
373 /* put in proper endianness, and validate each desc */
374 for (i = 0; i < hdr.descriptor_count; i++) {
375 desc = desc_entries->desc[i];
376 vhdx_log_desc_le_import(&desc);
377 if (convert_endian) {
378 desc_entries->desc[i] = desc;
380 if (vhdx_log_desc_is_valid(&desc, &hdr) == false) {
381 ret = -EINVAL;
382 goto free_and_exit;
385 if (convert_endian) {
386 desc_entries->hdr = hdr;
389 *buffer = desc_entries;
390 goto exit;
392 free_and_exit:
393 qemu_vfree(desc_entries);
394 exit:
395 return ret;
399 /* Flushes the descriptor described by desc to the VHDX image file.
400 * If the descriptor is a data descriptor, than 'data' must be non-NULL,
401 * and >= 4096 bytes (VHDX_LOG_SECTOR_SIZE), containing the data to be
402 * written.
404 * Verification is performed to make sure the sequence numbers of a data
405 * descriptor match the sequence number in the desc.
407 * For a zero descriptor, it may describe multiple sectors to fill with zeroes.
408 * In this case, it should be noted that zeroes are written to disk, and the
409 * image file is not extended as a sparse file. */
410 static int vhdx_log_flush_desc(BlockDriverState *bs, VHDXLogDescriptor *desc,
411 VHDXLogDataSector *data)
413 int ret = 0;
414 uint64_t seq, file_offset;
415 uint32_t offset = 0;
416 void *buffer = NULL;
417 uint64_t count = 1;
418 int i;
420 buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
422 if (desc->signature == VHDX_LOG_DESC_SIGNATURE) {
423 /* data sector */
424 if (data == NULL) {
425 ret = -EFAULT;
426 goto exit;
429 /* The sequence number of the data sector must match that
430 * in the descriptor */
431 seq = data->sequence_high;
432 seq <<= 32;
433 seq |= data->sequence_low & 0xffffffff;
435 if (seq != desc->sequence_number) {
436 ret = -EINVAL;
437 goto exit;
440 /* Each data sector is in total 4096 bytes, however the first
441 * 8 bytes, and last 4 bytes, are located in the descriptor */
442 memcpy(buffer, &desc->leading_bytes, 8);
443 offset += 8;
445 memcpy(buffer+offset, data->data, 4084);
446 offset += 4084;
448 memcpy(buffer+offset, &desc->trailing_bytes, 4);
450 } else if (desc->signature == VHDX_LOG_ZERO_SIGNATURE) {
451 /* write 'count' sectors of sector */
452 memset(buffer, 0, VHDX_LOG_SECTOR_SIZE);
453 count = desc->zero_length / VHDX_LOG_SECTOR_SIZE;
454 } else {
455 error_report("Invalid VHDX log descriptor entry signature 0x%" PRIx32,
456 desc->signature);
457 ret = -EINVAL;
458 goto exit;
461 file_offset = desc->file_offset;
463 /* count is only > 1 if we are writing zeroes */
464 for (i = 0; i < count; i++) {
465 ret = bdrv_pwrite_sync(bs->file, file_offset, buffer,
466 VHDX_LOG_SECTOR_SIZE);
467 if (ret < 0) {
468 goto exit;
470 file_offset += VHDX_LOG_SECTOR_SIZE;
473 exit:
474 qemu_vfree(buffer);
475 return ret;
478 /* Flush the entire log (as described by 'logs') to the VHDX image
479 * file, and then set the log to 'empty' status once complete.
481 * The log entries should be validate prior to flushing */
482 static int vhdx_log_flush(BlockDriverState *bs, BDRVVHDXState *s,
483 VHDXLogSequence *logs)
485 int ret = 0;
486 int i;
487 uint32_t cnt, sectors_read;
488 uint64_t new_file_size;
489 void *data = NULL;
490 VHDXLogDescEntries *desc_entries = NULL;
491 VHDXLogEntryHeader hdr_tmp = { 0 };
493 cnt = logs->count;
495 data = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
497 ret = vhdx_user_visible_write(bs, s);
498 if (ret < 0) {
499 goto exit;
502 /* each iteration represents one log sequence, which may span multiple
503 * sectors */
504 while (cnt--) {
505 ret = vhdx_log_peek_hdr(bs, &logs->log, &hdr_tmp);
506 if (ret < 0) {
507 goto exit;
509 /* if the log shows a FlushedFileOffset larger than our current file
510 * size, then that means the file has been truncated / corrupted, and
511 * we must refused to open it / use it */
512 if (hdr_tmp.flushed_file_offset > bdrv_getlength(bs->file)) {
513 ret = -EINVAL;
514 goto exit;
517 ret = vhdx_log_read_desc(bs, s, &logs->log, &desc_entries, true);
518 if (ret < 0) {
519 goto exit;
522 for (i = 0; i < desc_entries->hdr.descriptor_count; i++) {
523 if (desc_entries->desc[i].signature == VHDX_LOG_DESC_SIGNATURE) {
524 /* data sector, so read a sector to flush */
525 ret = vhdx_log_read_sectors(bs, &logs->log, &sectors_read,
526 data, 1, false);
527 if (ret < 0) {
528 goto exit;
530 if (sectors_read != 1) {
531 ret = -EINVAL;
532 goto exit;
534 vhdx_log_data_le_import(data);
537 ret = vhdx_log_flush_desc(bs, &desc_entries->desc[i], data);
538 if (ret < 0) {
539 goto exit;
542 if (bdrv_getlength(bs->file) < desc_entries->hdr.last_file_offset) {
543 new_file_size = desc_entries->hdr.last_file_offset;
544 if (new_file_size % (1024*1024)) {
545 /* round up to nearest 1MB boundary */
546 new_file_size = ((new_file_size >> 20) + 1) << 20;
547 bdrv_truncate(bs->file, new_file_size);
550 qemu_vfree(desc_entries);
551 desc_entries = NULL;
554 bdrv_flush(bs);
555 /* once the log is fully flushed, indicate that we have an empty log
556 * now. This also sets the log guid to 0, to indicate an empty log */
557 vhdx_log_reset(bs, s);
559 exit:
560 qemu_vfree(data);
561 qemu_vfree(desc_entries);
562 return ret;
565 static int vhdx_validate_log_entry(BlockDriverState *bs, BDRVVHDXState *s,
566 VHDXLogEntries *log, uint64_t seq,
567 bool *valid, VHDXLogEntryHeader *entry)
569 int ret = 0;
570 VHDXLogEntryHeader hdr;
571 void *buffer = NULL;
572 uint32_t i, desc_sectors, total_sectors, crc;
573 uint32_t sectors_read = 0;
574 VHDXLogDescEntries *desc_buffer = NULL;
576 *valid = false;
578 ret = vhdx_log_peek_hdr(bs, log, &hdr);
579 if (ret < 0) {
580 goto inc_and_exit;
583 if (vhdx_log_hdr_is_valid(log, &hdr, s) == false) {
584 goto inc_and_exit;
587 if (seq > 0) {
588 if (hdr.sequence_number != seq + 1) {
589 goto inc_and_exit;
593 desc_sectors = vhdx_compute_desc_sectors(hdr.descriptor_count);
595 /* Read all log sectors, and calculate log checksum */
597 total_sectors = hdr.entry_length / VHDX_LOG_SECTOR_SIZE;
600 /* read_desc() will increment the read idx */
601 ret = vhdx_log_read_desc(bs, s, log, &desc_buffer, false);
602 if (ret < 0) {
603 goto free_and_exit;
606 crc = vhdx_checksum_calc(0xffffffff, (void *)desc_buffer,
607 desc_sectors * VHDX_LOG_SECTOR_SIZE, 4);
608 crc ^= 0xffffffff;
610 buffer = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
611 if (total_sectors > desc_sectors) {
612 for (i = 0; i < total_sectors - desc_sectors; i++) {
613 sectors_read = 0;
614 ret = vhdx_log_read_sectors(bs, log, &sectors_read, buffer,
615 1, false);
616 if (ret < 0 || sectors_read != 1) {
617 goto free_and_exit;
619 crc = vhdx_checksum_calc(crc, buffer, VHDX_LOG_SECTOR_SIZE, -1);
620 crc ^= 0xffffffff;
623 crc ^= 0xffffffff;
624 if (crc != hdr.checksum) {
625 goto free_and_exit;
628 *valid = true;
629 *entry = hdr;
630 goto free_and_exit;
632 inc_and_exit:
633 log->read = vhdx_log_inc_idx(log->read, log->length);
635 free_and_exit:
636 qemu_vfree(buffer);
637 qemu_vfree(desc_buffer);
638 return ret;
641 /* Search through the log circular buffer, and find the valid, active
642 * log sequence, if any exists
643 * */
644 static int vhdx_log_search(BlockDriverState *bs, BDRVVHDXState *s,
645 VHDXLogSequence *logs)
647 int ret = 0;
648 uint32_t tail;
649 bool seq_valid = false;
650 VHDXLogSequence candidate = { 0 };
651 VHDXLogEntryHeader hdr = { 0 };
652 VHDXLogEntries curr_log;
654 memcpy(&curr_log, &s->log, sizeof(VHDXLogEntries));
655 curr_log.write = curr_log.length; /* assume log is full */
656 curr_log.read = 0;
659 /* now we will go through the whole log sector by sector, until
660 * we find a valid, active log sequence, or reach the end of the
661 * log buffer */
662 for (;;) {
663 uint64_t curr_seq = 0;
664 VHDXLogSequence current = { 0 };
666 tail = curr_log.read;
668 ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq,
669 &seq_valid, &hdr);
670 if (ret < 0) {
671 goto exit;
674 if (seq_valid) {
675 current.valid = true;
676 current.log = curr_log;
677 current.log.read = tail;
678 current.log.write = curr_log.read;
679 current.count = 1;
680 current.hdr = hdr;
683 for (;;) {
684 ret = vhdx_validate_log_entry(bs, s, &curr_log, curr_seq,
685 &seq_valid, &hdr);
686 if (ret < 0) {
687 goto exit;
689 if (seq_valid == false) {
690 break;
692 current.log.write = curr_log.read;
693 current.count++;
695 curr_seq = hdr.sequence_number;
699 if (current.valid) {
700 if (candidate.valid == false ||
701 current.hdr.sequence_number > candidate.hdr.sequence_number) {
702 candidate = current;
706 if (curr_log.read < tail) {
707 break;
711 *logs = candidate;
713 if (candidate.valid) {
714 /* this is the next sequence number, for writes */
715 s->log.sequence = candidate.hdr.sequence_number + 1;
719 exit:
720 return ret;
723 /* Parse the replay log. Per the VHDX spec, if the log is present
724 * it must be replayed prior to opening the file, even read-only.
726 * If read-only, we must replay the log in RAM (or refuse to open
727 * a dirty VHDX file read-only) */
728 int vhdx_parse_log(BlockDriverState *bs, BDRVVHDXState *s, bool *flushed,
729 Error **errp)
731 int ret = 0;
732 VHDXHeader *hdr;
733 VHDXLogSequence logs = { 0 };
735 hdr = s->headers[s->curr_header];
737 *flushed = false;
739 /* s->log.hdr is freed in vhdx_close() */
740 if (s->log.hdr == NULL) {
741 s->log.hdr = qemu_blockalign(bs, sizeof(VHDXLogEntryHeader));
744 s->log.offset = hdr->log_offset;
745 s->log.length = hdr->log_length;
747 if (s->log.offset < VHDX_LOG_MIN_SIZE ||
748 s->log.offset % VHDX_LOG_MIN_SIZE) {
749 ret = -EINVAL;
750 goto exit;
753 /* per spec, only log version of 0 is supported */
754 if (hdr->log_version != 0) {
755 ret = -EINVAL;
756 goto exit;
759 /* If either the log guid, or log length is zero,
760 * then a replay log is not present */
761 if (guid_eq(hdr->log_guid, zero_guid)) {
762 goto exit;
765 if (hdr->log_length == 0) {
766 goto exit;
769 if (hdr->log_length % VHDX_LOG_MIN_SIZE) {
770 ret = -EINVAL;
771 goto exit;
775 /* The log is present, we need to find if and where there is an active
776 * sequence of valid entries present in the log. */
778 ret = vhdx_log_search(bs, s, &logs);
779 if (ret < 0) {
780 goto exit;
783 if (logs.valid) {
784 if (bs->read_only) {
785 ret = -EPERM;
786 error_setg_errno(errp, EPERM,
787 "VHDX image file '%s' opened read-only, but "
788 "contains a log that needs to be replayed. To "
789 "replay the log, execute:\n qemu-img check -r "
790 "all '%s'",
791 bs->filename, bs->filename);
792 goto exit;
794 /* now flush the log */
795 ret = vhdx_log_flush(bs, s, &logs);
796 if (ret < 0) {
797 goto exit;
799 *flushed = true;
803 exit:
804 return ret;
809 static void vhdx_log_raw_to_le_sector(VHDXLogDescriptor *desc,
810 VHDXLogDataSector *sector, void *data,
811 uint64_t seq)
813 /* 8 + 4084 + 4 = 4096, 1 log sector */
814 memcpy(&desc->leading_bytes, data, 8);
815 data += 8;
816 cpu_to_le64s(&desc->leading_bytes);
817 memcpy(sector->data, data, 4084);
818 data += 4084;
819 memcpy(&desc->trailing_bytes, data, 4);
820 cpu_to_le32s(&desc->trailing_bytes);
821 data += 4;
823 sector->sequence_high = (uint32_t) (seq >> 32);
824 sector->sequence_low = (uint32_t) (seq & 0xffffffff);
825 sector->data_signature = VHDX_LOG_DATA_SIGNATURE;
827 vhdx_log_desc_le_export(desc);
828 vhdx_log_data_le_export(sector);
832 static int vhdx_log_write(BlockDriverState *bs, BDRVVHDXState *s,
833 void *data, uint32_t length, uint64_t offset)
835 int ret = 0;
836 void *buffer = NULL;
837 void *merged_sector = NULL;
838 void *data_tmp, *sector_write;
839 unsigned int i;
840 int sector_offset;
841 uint32_t desc_sectors, sectors, total_length;
842 uint32_t sectors_written = 0;
843 uint32_t aligned_length;
844 uint32_t leading_length = 0;
845 uint32_t trailing_length = 0;
846 uint32_t partial_sectors = 0;
847 uint32_t bytes_written = 0;
848 uint64_t file_offset;
849 VHDXHeader *header;
850 VHDXLogEntryHeader new_hdr;
851 VHDXLogDescriptor *new_desc = NULL;
852 VHDXLogDataSector *data_sector = NULL;
853 MSGUID new_guid = { 0 };
855 header = s->headers[s->curr_header];
857 /* need to have offset read data, and be on 4096 byte boundary */
859 if (length > header->log_length) {
860 /* no log present. we could create a log here instead of failing */
861 ret = -EINVAL;
862 goto exit;
865 if (guid_eq(header->log_guid, zero_guid)) {
866 vhdx_guid_generate(&new_guid);
867 vhdx_update_headers(bs, s, false, &new_guid);
868 } else {
869 /* currently, we require that the log be flushed after
870 * every write. */
871 ret = -ENOTSUP;
872 goto exit;
875 /* 0 is an invalid sequence number, but may also represent the first
876 * log write (or a wrapped seq) */
877 if (s->log.sequence == 0) {
878 s->log.sequence = 1;
881 sector_offset = offset % VHDX_LOG_SECTOR_SIZE;
882 file_offset = (offset / VHDX_LOG_SECTOR_SIZE) * VHDX_LOG_SECTOR_SIZE;
884 aligned_length = length;
886 /* add in the unaligned head and tail bytes */
887 if (sector_offset) {
888 leading_length = (VHDX_LOG_SECTOR_SIZE - sector_offset);
889 leading_length = leading_length > length ? length : leading_length;
890 aligned_length -= leading_length;
891 partial_sectors++;
894 sectors = aligned_length / VHDX_LOG_SECTOR_SIZE;
895 trailing_length = aligned_length - (sectors * VHDX_LOG_SECTOR_SIZE);
896 if (trailing_length) {
897 partial_sectors++;
900 sectors += partial_sectors;
902 /* sectors is now how many sectors the data itself takes, not
903 * including the header and descriptor metadata */
905 new_hdr = (VHDXLogEntryHeader) {
906 .signature = VHDX_LOG_SIGNATURE,
907 .tail = s->log.tail,
908 .sequence_number = s->log.sequence,
909 .descriptor_count = sectors,
910 .reserved = 0,
911 .flushed_file_offset = bdrv_getlength(bs->file),
912 .last_file_offset = bdrv_getlength(bs->file),
915 new_hdr.log_guid = header->log_guid;
917 desc_sectors = vhdx_compute_desc_sectors(new_hdr.descriptor_count);
919 total_length = (desc_sectors + sectors) * VHDX_LOG_SECTOR_SIZE;
920 new_hdr.entry_length = total_length;
922 vhdx_log_entry_hdr_le_export(&new_hdr);
924 buffer = qemu_blockalign(bs, total_length);
925 memcpy(buffer, &new_hdr, sizeof(new_hdr));
927 new_desc = buffer + sizeof(new_hdr);
928 data_sector = buffer + (desc_sectors * VHDX_LOG_SECTOR_SIZE);
929 data_tmp = data;
931 /* All log sectors are 4KB, so for any partial sectors we must
932 * merge the data with preexisting data from the final file
933 * destination */
934 merged_sector = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
936 for (i = 0; i < sectors; i++) {
937 new_desc->signature = VHDX_LOG_DESC_SIGNATURE;
938 new_desc->sequence_number = s->log.sequence;
939 new_desc->file_offset = file_offset;
941 if (i == 0 && leading_length) {
942 /* partial sector at the front of the buffer */
943 ret = bdrv_pread(bs->file, file_offset, merged_sector,
944 VHDX_LOG_SECTOR_SIZE);
945 if (ret < 0) {
946 goto exit;
948 memcpy(merged_sector + sector_offset, data_tmp, leading_length);
949 bytes_written = leading_length;
950 sector_write = merged_sector;
951 } else if (i == sectors - 1 && trailing_length) {
952 /* partial sector at the end of the buffer */
953 ret = bdrv_pread(bs->file,
954 file_offset,
955 merged_sector + trailing_length,
956 VHDX_LOG_SECTOR_SIZE - trailing_length);
957 if (ret < 0) {
958 goto exit;
960 memcpy(merged_sector, data_tmp, trailing_length);
961 bytes_written = trailing_length;
962 sector_write = merged_sector;
963 } else {
964 bytes_written = VHDX_LOG_SECTOR_SIZE;
965 sector_write = data_tmp;
968 /* populate the raw sector data into the proper structures,
969 * as well as update the descriptor, and convert to proper
970 * endianness */
971 vhdx_log_raw_to_le_sector(new_desc, data_sector, sector_write,
972 s->log.sequence);
974 data_tmp += bytes_written;
975 data_sector++;
976 new_desc++;
977 file_offset += VHDX_LOG_SECTOR_SIZE;
980 /* checksum covers entire entry, from the log header through the
981 * last data sector */
982 vhdx_update_checksum(buffer, total_length,
983 offsetof(VHDXLogEntryHeader, checksum));
985 /* now write to the log */
986 ret = vhdx_log_write_sectors(bs, &s->log, &sectors_written, buffer,
987 desc_sectors + sectors);
988 if (ret < 0) {
989 goto exit;
992 if (sectors_written != desc_sectors + sectors) {
993 /* instead of failing, we could flush the log here */
994 ret = -EINVAL;
995 goto exit;
998 s->log.sequence++;
999 /* write new tail */
1000 s->log.tail = s->log.write;
1002 exit:
1003 qemu_vfree(buffer);
1004 qemu_vfree(merged_sector);
1005 return ret;
1008 /* Perform a log write, and then immediately flush the entire log */
1009 int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s,
1010 void *data, uint32_t length, uint64_t offset)
1012 int ret = 0;
1013 VHDXLogSequence logs = { .valid = true,
1014 .count = 1,
1015 .hdr = { 0 } };
1018 /* Make sure data written (new and/or changed blocks) is stable
1019 * on disk, before creating log entry */
1020 bdrv_flush(bs);
1021 ret = vhdx_log_write(bs, s, data, length, offset);
1022 if (ret < 0) {
1023 goto exit;
1025 logs.log = s->log;
1027 /* Make sure log is stable on disk */
1028 bdrv_flush(bs);
1029 ret = vhdx_log_flush(bs, s, &logs);
1030 if (ret < 0) {
1031 goto exit;
1034 s->log = logs.log;
1036 exit:
1037 return ret;