Re: [Qemu-devel] [PATCH v4 15/19] block: raw-win32 driver reopen support

[Kevin Wolf <kwolf@redhat.com>]

Am 20.09.2012 21:13, schrieb Jeff Cody:
> This is the support for reopen, for win32.  Note that there is a key
> difference in the win32 reopen, namely that it is not guaranteed safe
> like all the other drivers.  Attempts are made to reopen the file, or
> open the file in a new handle prior to closing the old handle.  However,
> this is not always feasible, and there are times when we must close the
> existing file and then open the new file, breaking the transactional nature
> of the reopen.
> 
> Signed-off-by: Jeff Cody <jcody@redhat.com>
> ---
>  block/raw-win32.c | 105 ++++++++++++++++++++++++++++++++++++++++++++++++++++++
>  1 file changed, 105 insertions(+)

I can't really review win32 code, but two comments anyway:

> diff --git a/block/raw-win32.c b/block/raw-win32.c
> index 78c8306..8a698d3 100644
> --- a/block/raw-win32.c
> +++ b/block/raw-win32.c
> @@ -31,6 +31,7 @@
>  #define FTYPE_FILE 0
>  #define FTYPE_CD     1
>  #define FTYPE_HARDDISK 2
> +#define WINDOWS_VISTA 6
>  
>  typedef struct BDRVRawState {
>      HANDLE hfile;
> @@ -38,6 +39,10 @@ typedef struct BDRVRawState {
>      char drive_path[16]; /* format: "d:\" */
>  } BDRVRawState;
>  
> +typedef struct BDRVRawReopenState {
> +    HANDLE hfile;
> +} BDRVRawReopenState;
> +
>  int qemu_ftruncate64(int fd, int64_t length)
>  {
>      LARGE_INTEGER li;
> @@ -117,6 +122,103 @@ static int raw_open(BlockDriverState *bs, const char *filename, int flags)
>      return 0;
>  }
>  
> +static int raw_reopen_prepare(BDRVReopenState *state,
> +                              BlockReopenQueue *queue, Error **errp)
> +{
> +    BDRVRawState *s;
> +    BDRVRawReopenState *raw_s;
> +    int ret = 0;
> +    int access_flags;
> +    DWORD overlapped;
> +    OSVERSIONINFO osvi;
> +
> +    assert(state != NULL);
> +    assert(state->bs != NULL);
> +
> +    s = state->bs->opaque;
> +
> +    state->opaque = g_malloc0(sizeof(BDRVRawReopenState));
> +    raw_s = state->opaque;
> +
> +    raw_parse_flags(state->flags, &access_flags, &overlapped);
> +
> +    ZeroMemory(&osvi, sizeof(OSVERSIONINFO));
> +    osvi.dwOSVersionInfoSize = sizeof(OSVERSIONINFO);
> +
> +    GetVersionEx(&osvi);
> +    raw_s->hfile = INVALID_HANDLE_VALUE;
> +
> +    if (osvi.dwMajorVersion >= WINDOWS_VISTA) {
> +        raw_s->hfile = ReOpenFile(s->hfile, access_flags, FILE_SHARE_READ,
> +                                  overlapped);
> +    }
> +
> +    /* could not reopen the file handle, so fall back to opening
> +     * new file (CreateFile) */
> +    if (raw_s->hfile == INVALID_HANDLE_VALUE) {
> +        raw_s->hfile = CreateFile(state->bs->filename, access_flags,
> +                                  FILE_SHARE_READ, NULL, OPEN_EXISTING,
> +                                  overlapped, NULL);
> +        if (raw_s->hfile == INVALID_HANDLE_VALUE) {
> +            /* this could happen because the access_flags requested are
> +             * incompatible with the existing share mode of s->hfile,
> +             * so our only option now is to close s->hfile, and try again.
> +             * This could end badly */
> +            CloseHandle(s->hfile);

How common is this case?

We do have another option, namely not reopen at all and return an error.
Of course, this only makes sense if it doesn't mean that we almost never
succeed.

> +            s->hfile = INVALID_HANDLE_VALUE;
> +            raw_s->hfile = CreateFile(state->bs->filename, access_flags,
> +                                      FILE_SHARE_READ, NULL, OPEN_EXISTING,
> +                                      overlapped, NULL);
> +            if (raw_s->hfile == INVALID_HANDLE_VALUE) {
> +                /* Unrecoverable error */
> +                error_set(errp, ERROR_CLASS_GENERIC_ERROR,
> +                          "Fatal error reopening %s file; file closed and cannot be opened\n",

This line is longer than 80 characters.

Kevin