Am 11.08.2011 18:28, schrieb Corey Bryant:
On 07/26/2011 08:51 AM, Corey Bryant wrote:
+static int raw_open_fd(BlockDriverState *bs, const char *filename, int flags) +{ + BDRVRawState *s = bs->opaque; + const char *fd_str; + int fd; + + /* extract the file descriptor - fail if it's not fd: */ + if (!strstart(filename, "fd:",&fd_str)) { + return -EINVAL; + } + + if (!qemu_isdigit(fd_str[0])) { + /* get fd from monitor */ + fd = qemu_get_fd(fd_str); + if (fd == -1) { + return -EBADF; + } + } else { + char *endptr = NULL; + + fd = strtol(fd_str,&endptr, 10); + if (*endptr || (fd == 0&& fd_str == endptr)) { + return -EBADF; + } + } + + s->fd = fd; + s->type = FTYPE_FILE; + + return raw_open_common(bs, filename, flags, 0); +} + +static BlockDriver bdrv_file_fd = { + .format_name = "file", + .protocol_name = "fd", + .instance_size = sizeof(BDRVRawState), + .bdrv_probe = NULL, /* no probe for protocols */ + .bdrv_file_open = raw_open_fd, + .bdrv_read = raw_read, + .bdrv_write = raw_write, + .bdrv_close = raw_close, + .bdrv_flush = raw_flush, + .bdrv_discard = raw_discard, + + .bdrv_aio_readv = raw_aio_readv, + .bdrv_aio_writev = raw_aio_writev, + .bdrv_aio_flush = raw_aio_flush, + + .bdrv_truncate = raw_truncate, + .bdrv_getlength = raw_getlength, + + .create_options = raw_create_options, +}; +
I'm a bit unsure of how to support CD-ROM with the fd: protocol.
I don't think use of bdrv_file_fd is an option. For example, while raw_open_fd may work, there is no eject support.
Another approach is to have 2 BlockDriver structs that support the fd protocol. For example, creating a new BlockDriver struct that mirrors bdrv_host_cdrom. So you would have bdrv_host_cdrom_fd with a corresponding cdrom_open_fd function. But that doesn't appear possible as it appears that the protocol must be unique among all BlockDriver structs.
Do we need to introduce a similar protocol (maybe "cdfd") to support passing of CD-ROM file descriptors?
I would try to avoid that. I think you can determine the kind of device during fd_open and then implement an fd_eject etc. that checks this type and calls the appropriate ioctl. Maybe the same should be done for file/host_device/host_cdrom instead of having separate protocols. Anyway, I don't think this is important to get a first version merged. It will still take a while anyway until we have full support that libvirt would want to use. Kevin