Re: [Qemu-devel] [PATCH 8/9] block: vhdx - add log write support

[Jeff Cody <jcody@redhat.com>]

On Tue, Jul 30, 2013 at 11:57:20AM +0800, Fam Zheng wrote:
> On Wed, 07/24 13:54, Jeff Cody wrote:
> > This adds support for writing to the VHDX log.
> > 
> > For spec details, see VHDX Specification Format v1.00:
> > https://www.microsoft.com/en-us/download/details.aspx?id=34750
> > 
> > There are a few limitations to this log support:
> > 1.) There is no caching yet
> > 2.) The log is flushed after each entry
> > 
> > The primary write interface, vhdx_log_write_and_flush(), performs a log
> > write followed by an immediate flush of the log.
> > 
> > As each log entry sector is a minimum of 4KB, partial sector writes are
> > filled in with data from the disk write destination.
> > 
> > If the current file log GUID is 0, a new GUID is generated and updated
> > in the header.
> > 
> > Signed-off-by: Jeff Cody <jcody@redhat.com>
> > ---
> >  block/vhdx-log.c | 273 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
> >  block/vhdx.h     |   3 +
> >  2 files changed, 276 insertions(+)
> > 
> > diff --git a/block/vhdx-log.c b/block/vhdx-log.c
> > index 89b9000..786b393 100644
> > --- a/block/vhdx-log.c
> > +++ b/block/vhdx-log.c
> > @@ -170,6 +170,53 @@ exit:
> >      return ret;
> >  }
> >  
> > +/* Writes num_sectors to the log (all log sectors are 4096 bytes),
> > + * from buffer 'buffer'.  Upon return, *sectors_written will contain
> > + * the number of sectors successfully written.
> > + *
> > + * It is assumed that 'buffer' is at least 4096*num_sectors large.
> > + *
> > + * 0 is returned on success, -errno otherwise */
> > +static int vhdx_log_write_sectors(BlockDriverState *bs, VHDXLogEntries *log,
> > +                                  uint32_t *sectors_written, void *buffer,
> > +                                  uint32_t num_sectors)
> > +{
> > +    int ret = 0;
> > +    uint64_t offset;
> > +    uint32_t write;
> > +    void *buffer_tmp;
> > +    BDRVVHDXState *s = bs->opaque;
> > +
> > +    vhdx_user_visible_write(bs, s);
> > +
> > +    write = log->write;
> > +
> > +    buffer_tmp = buffer;
> > +    while (num_sectors) {
> > +
> > +        offset = log->offset + write;
> > +        write = vhdx_log_inc_idx(write, log->length);
> > +        if (write == log->read) {
> > +            /* full */
> > +            break;
> > +        }
> > +        ret = bdrv_pwrite_sync(bs->file, offset, buffer_tmp,
> > +                               VHDX_LOG_SECTOR_SIZE);
> > +        if (ret < 0) {
> > +            goto exit;
> > +        }
> > +        buffer_tmp += VHDX_LOG_SECTOR_SIZE;
> > +
> > +        log->write = write;
> > +        *sectors_written = *sectors_written + 1;
> > +        num_sectors--;
> > +    }
> > +
> > +exit:
> > +    return ret;
> > +}
> > +
> > +
> >  /* Validates a log entry header */
> >  static bool vhdx_log_hdr_is_valid(VHDXLogEntries *log, VHDXLogEntryHeader *hdr,
> >                                    BDRVVHDXState *s)
> > @@ -732,3 +779,229 @@ exit:
> >      return ret;
> >  }
> >  
> > +
> > +
> > +static void vhdx_log_raw_to_le_sector(VHDXLogDescriptor *desc,
> > +                                      VHDXLogDataSector *sector, void *data,
> > +                                      uint64_t seq)
> > +{
> > +    memcpy(&desc->leading_bytes, data, 8);
> > +    data += 8;
> > +    cpu_to_le64s(&desc->leading_bytes);
> > +    memcpy(sector->data, data, 4084);
> > +    data += 4084;
> > +    memcpy(&desc->trailing_bytes, data, 4);
> > +    cpu_to_le32s(&desc->trailing_bytes);
> > +    data += 4;
> > +
> > +    sector->sequence_high  = (uint32_t) (seq >> 32);
> > +    sector->sequence_low   = (uint32_t) (seq & 0xffffffff);
> > +    sector->data_signature = VHDX_LOG_DATA_SIGNATURE;
> > +
> > +    vhdx_log_desc_le_export(desc);
> > +    vhdx_log_data_le_export(sector);
> > +}
> > +
> > +
> > +static int vhdx_log_write(BlockDriverState *bs, BDRVVHDXState *s,
> > +                          void *data, uint32_t length, uint64_t offset)
> > +{
> > +    int ret = 0;
> > +    void *buffer = NULL;
> > +    void *merged_sector = NULL;
> > +    void *data_tmp, *sector_write;
> > +    unsigned int i;
> > +    int sector_offset;
> > +    uint32_t desc_sectors, sectors, total_length;
> > +    uint32_t sectors_written = 0;
> > +    uint32_t aligned_length;
> > +    uint32_t leading_length = 0;
> > +    uint32_t trailing_length = 0;
> > +    uint32_t partial_sectors = 0;
> > +    uint32_t bytes_written = 0;
> > +    uint64_t file_offset;
> > +    VHDXHeader *header;
> > +    VHDXLogEntryHeader new_hdr;
> > +    VHDXLogDescriptor *new_desc = NULL;
> > +    VHDXLogDataSector *data_sector = NULL;
> > +    MSGUID new_guid = { 0 };
> > +
> > +    header = s->headers[s->curr_header];
> > +
> > +    /* need to have offset read data, and be on 4096 byte boundary */
> > +
> > +    if (length > header->log_length) {
> > +        /* no log present.  we could create a log here instead of failing */
> 
> Does newly created vhdx have allocated log sectors?
> 

I don't know of any way to make Hyper-V create a file without an
allocated log area (I believe with the files I've generated, it
allocates a 1MB log between the header and the BAT region).

The spec says that "LogLength" in the header should be a multiple of
1MB.  And technically, 0 is a multiple of every number, so when
parsing the header I don't fail out on a zero-lengthed log.  In
practice, I don't think Hyper-V creates files with zero-length logs,
but I don't think the spec rules it out.

So we could either allocate a log in the file at this point, or fail.

> > +        ret = -EINVAL;
> > +        goto exit;
> > +    }
> > +
> > +    if (vhdx_log_guid_is_zero(&header->log_guid)) {
> > +        vhdx_guid_generate(&new_guid);
> > +        vhdx_update_headers(bs, s, false, &new_guid);
> > +    } else {
> > +        /* currently, we require that the log be flushed after
> > +         * every write. */
> > +        ret = -ENOTSUP;
> 
> Can we make an assertion here?
>

I don't know if we should assert here - the VM could certainly
continue on if this is not the primary drive.

> > +    }
> > +
> > +    /* 0 is an invalid sequence number, but may also represent the first
> > +     * log write (or a wrapped seq) */
> > +    if (s->log.sequence == 0) {
> > +        s->log.sequence = 1;
> > +    }
> > +
> > +    sector_offset = offset % VHDX_LOG_SECTOR_SIZE;
> > +    file_offset = (offset / VHDX_LOG_SECTOR_SIZE) * VHDX_LOG_SECTOR_SIZE;
> > +
> > +    aligned_length = length;
> > +
> > +    /* add in the unaligned head and tail bytes */
> > +    if (sector_offset) {
> > +        leading_length = (VHDX_LOG_SECTOR_SIZE - sector_offset);
> > +        leading_length = leading_length > length ? length : leading_length;
> > +        aligned_length -= leading_length;
> > +        partial_sectors++;
> > +    }
> > +
> > +    sectors = aligned_length / VHDX_LOG_SECTOR_SIZE;
> > +    trailing_length = aligned_length - (sectors * VHDX_LOG_SECTOR_SIZE);
> > +    if (trailing_length) {
> > +        partial_sectors++;
> > +    }
> > +
> > +    sectors += partial_sectors;
> > +
> > +    /* sectors is now how many sectors the data itself takes, not
> > +     * including the header and descriptor metadata */
> > +
> > +    new_hdr = (VHDXLogEntryHeader) {
> > +                .signature           = VHDX_LOG_SIGNATURE,
> > +                .tail                = s->log.tail,
> > +                .sequence_number     = s->log.sequence,
> > +                .descriptor_count    = sectors,
> > +                .reserved            = 0,
> > +                .flushed_file_offset = bdrv_getlength(bs->file),
> > +                .last_file_offset    = bdrv_getlength(bs->file),
> > +              };
> > +
> > +    memcpy(&new_hdr.log_guid, &header->log_guid, sizeof(MSGUID));
> > +
> > +    desc_sectors = vhdx_compute_desc_sectors(new_hdr.descriptor_count);
> > +
> > +    total_length = (desc_sectors + sectors) * VHDX_LOG_SECTOR_SIZE;
> > +    new_hdr.entry_length = total_length;
> > +
> > +    vhdx_log_entry_hdr_le_export(&new_hdr);
> > +
> > +    buffer = qemu_blockalign(bs, total_length);
> > +    memcpy(buffer, &new_hdr, sizeof(new_hdr));
> > +
> > +    new_desc = (VHDXLogDescriptor *) (buffer + sizeof(new_hdr));
> > +    data_sector = buffer + (desc_sectors * VHDX_LOG_SECTOR_SIZE);
> > +    data_tmp = data;
> > +
> > +    /* All log sectors are 4KB, so for any partial sectors we must
> > +     * merge the data with preexisting data from the final file
> > +     * destination */
> > +    merged_sector = qemu_blockalign(bs, VHDX_LOG_SECTOR_SIZE);
> > +
> > +    for (i = 0; i < sectors; i++) {
> > +        new_desc->signature       = VHDX_LOG_DESC_SIGNATURE;
> > +        new_desc->sequence_number = s->log.sequence;
> > +        new_desc->file_offset     = file_offset;
> > +
> > +        if (i == 0 && leading_length) {
> > +            /* partial sector at the front of the buffer */
> > +            ret = bdrv_pread(bs->file, file_offset, merged_sector,
> > +                             VHDX_LOG_SECTOR_SIZE);
> > +            if (ret < 0) {
> > +                goto exit;
> > +            }
> > +            memcpy(merged_sector + sector_offset, data_tmp, leading_length);
> > +            bytes_written = leading_length;
> > +            sector_write = merged_sector;
> > +        } else if (i == sectors - 1 && trailing_length) {
> > +            /* partial sector at the end of the buffer */
> > +            ret = bdrv_pread(bs->file,
> > +                            file_offset,
> > +                            merged_sector + trailing_length,
> > +                            VHDX_LOG_SECTOR_SIZE - trailing_length);
> > +            if (ret < 0) {
> > +                goto exit;
> > +            }
> > +            memcpy(merged_sector, data_tmp, trailing_length);
> > +            bytes_written = trailing_length;
> > +            sector_write = merged_sector;
> > +        } else {
> > +            bytes_written = VHDX_LOG_SECTOR_SIZE;
> > +            sector_write = data_tmp;
> > +        }
> > +
> > +        /* populate the raw sector data into the proper structures,
> > +         * as well as update the descriptor, and convert to proper
> > +         * endianness */
> > +        vhdx_log_raw_to_le_sector(new_desc, data_sector, sector_write,
> > +                                  s->log.sequence);
> > +
> > +        data_tmp += bytes_written;
> > +        data_sector++;
> > +        new_desc++;
> > +        file_offset += VHDX_LOG_SECTOR_SIZE;
> > +    }
> > +
> > +    /* checksum covers entire entry, from the log header through the
> > +     * last data sector */
> > +    vhdx_update_checksum(buffer, total_length, 4);
> > +    cpu_to_le32s((uint32_t *)(buffer + 4));
> > +
> > +    /* now write to the log */
> > +    vhdx_log_write_sectors(bs, &s->log, &sectors_written, buffer,
> > +                           desc_sectors + sectors);
> > +    if (ret < 0) {
> > +        goto exit;
> > +    }
> > +
> > +    if (sectors_written != desc_sectors + sectors) {
> > +        /* instead of failing, we could flush the log here */
> > +        ret = -EINVAL;
> > +        goto exit;
> > +    }
> > +
> > +    s->log.sequence++;
> > +    /* write new tail */
> > +    s->log.tail = s->log.write;
> > +
> > +exit:
> > +    qemu_vfree(buffer);
> > +    qemu_vfree(merged_sector);
> > +    return ret;
> > +}
> > +
> > +/* Perform a log write, and then immediately flush the entire log */
> > +int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s,
> > +                             void *data, uint32_t length, uint64_t offset)
> > +{
> > +    int ret = 0;
> > +    VHDXLogSequence logs = { .valid = true,
> > +                             .count = 1,
> > +                             .hdr = { 0 } };
> > +
> > +
> > +    ret = vhdx_log_write(bs, s, data, length, offset);
> > +    if (ret < 0) {
> > +        goto exit;
> > +    }
> > +    logs.log = s->log;
> > +
> > +    ret = vhdx_log_flush(bs, s, &logs);
> > +    if (ret < 0) {
> > +        goto exit;
> > +    }
> > +
> > +    s->log = logs.log;
> > +
> > +exit:
> > +    return ret;
> > +}
> > +
> > diff --git a/block/vhdx.h b/block/vhdx.h
> > index 24b126e..b210efc 100644
> > --- a/block/vhdx.h
> > +++ b/block/vhdx.h
> > @@ -393,6 +393,9 @@ bool vhdx_checksum_is_valid(uint8_t *buf, size_t size, int crc_offset);
> >  
> >  int vhdx_parse_log(BlockDriverState *bs, BDRVVHDXState *s);
> >  
> > +int vhdx_log_write_and_flush(BlockDriverState *bs, BDRVVHDXState *s,
> > +                             void *data, uint32_t length, uint64_t offset);
> > +
> >  static inline void leguid_to_cpus(MSGUID *guid)
> >  {
> >      le32_to_cpus(&guid->data1);
> > -- 
> > 1.8.1.4
> > 
> > 
> 
> -- 
> Fam