block-qcow.c

xen虚拟机源代码安装包

	/*We store a local record of the request*/
	while (nb_sectors > 0) {
		cluster_offset = 
			get_cluster_offset(s, sector << 9, 0, 0, 0, 0);
		index_in_cluster = sector & (s->cluster_sectors - 1);
		n = s->cluster_sectors - index_in_cluster;
		if (n > nb_sectors)
			n = nb_sectors;

		if (s->aio.iocb_free_count == 0 || !tap_aio_lock(&s->aio, sector)) 
			return cb(dd, -EBUSY, sector, nb_sectors, id, private);
		
		if(!cluster_offset) {
			tap_aio_unlock(&s->aio, sector);
			ret = cb(dd, BLK_NOT_ALLOCATED, 
				 sector, n, id, private);
			if (ret == -EBUSY) {
				/* mark remainder of request
				 * as busy and try again later */
				return cb(dd, -EBUSY, sector + n,
					  nb_sectors - n, id, private);
			} else
				rsp += ret;
		} else if (cluster_offset & QCOW_OFLAG_COMPRESSED) {
			tap_aio_unlock(&s->aio, sector);
			if (decompress_cluster(s, cluster_offset) < 0) {
				rsp += cb(dd, -EIO, sector, 
					  nb_sectors, id, private);
				goto done;
			}
			memcpy(buf, s->cluster_cache + index_in_cluster * 512, 
			       512 * n);
			rsp += cb(dd, 0, sector, n, id, private);
		} else {
			tap_aio_read(&s->aio, s->fd, n * 512, 
				   (cluster_offset + index_in_cluster * 512),
				   buf, cb, id, sector, private);
		}
		nb_sectors -= n;
		sector += n;
		buf += n * 512;
	}
done:
	return rsp;
}

static int tdqcow_queue_write(struct disk_driver *dd, uint64_t sector,
		       int nb_sectors, char *buf, td_callback_t cb,
		       int id, void *private)
{
	struct tdqcow_state *s = (struct tdqcow_state *)dd->private;
	int ret = 0, index_in_cluster, n, i;
	uint64_t cluster_offset, sec, nr_secs;

	sec     = sector;
	nr_secs = nb_sectors;

	/*Check we can get a lock*/
	for (i = 0; i < nb_sectors; i++)
		if (!tap_aio_can_lock(&s->aio, sector + i))  
			return cb(dd, -EBUSY, sector, nb_sectors, id, private);
		   
	/*We store a local record of the request*/
	while (nb_sectors > 0) {
		index_in_cluster = sector & (s->cluster_sectors - 1);
		n = s->cluster_sectors - index_in_cluster;
		if (n > nb_sectors)
			n = nb_sectors;

		if (s->aio.iocb_free_count == 0 || !tap_aio_lock(&s->aio, sector))
			return cb(dd, -EBUSY, sector, nb_sectors, id, private);

		cluster_offset = get_cluster_offset(s, sector << 9, 1, 0,
						    index_in_cluster, 
						    index_in_cluster+n);
		if (!cluster_offset) {
			DPRINTF("Ooops, no write cluster offset!\n");
			tap_aio_unlock(&s->aio, sector);
			return cb(dd, -EIO, sector, nb_sectors, id, private);
		}

		if (s->crypt_method) {
			encrypt_sectors(s, sector, s->cluster_data, 
					(unsigned char *)buf, n, 1,
					&s->aes_encrypt_key);
			tap_aio_write(&s->aio, s->fd, n * 512, 
				    (cluster_offset + index_in_cluster*512),
				    (char *)s->cluster_data, cb, id, sector, 
				    private);
		} else {
			tap_aio_write(&s->aio, s->fd, n * 512, 
				    (cluster_offset + index_in_cluster*512),
				    buf, cb, id, sector, private);
		}
		
		nb_sectors -= n;
		sector += n;
		buf += n * 512;
	}
	s->cluster_cache_offset = -1; /* disable compressed cache */

	return 0;
}
 		
static int tdqcow_submit(struct disk_driver *dd)
{
        struct tdqcow_state *prv = (struct tdqcow_state *)dd->private;

	return tap_aio_submit(&prv->aio);
}

static int tdqcow_close(struct disk_driver *dd)
{
	struct tdqcow_state *s = (struct tdqcow_state *)dd->private;
	uint32_t cksum, out;
	int fd, offset;

	/*Update the hdr cksum*/
	if(s->min_cluster_alloc == s->l2_size) {
		cksum = gen_cksum((char *)s->l1_table, s->l1_size * sizeof(uint64_t));
		printf("Writing cksum: %d",cksum);
		fd = open(s->name, O_WRONLY | O_LARGEFILE); /*Open without O_DIRECT*/
		offset = sizeof(QCowHeader) + sizeof(uint32_t);
		lseek(fd, offset, SEEK_SET);
		out = cpu_to_be32(cksum);
		if (write(fd, &out, sizeof(uint32_t))) ;
		close(fd);
	}

	io_destroy(s->aio.aio_ctx.aio_ctx);
	free(s->name);
	free(s->l1_table);
	free(s->l2_cache);
	free(s->cluster_cache);
	free(s->cluster_data);
	close(s->fd);	
	return 0;
}

static int tdqcow_do_callbacks(struct disk_driver *dd, int sid)
{
        int ret, i, nr_events, rsp = 0,*ptr;
        struct io_event *ep;
        struct tdqcow_state *prv = (struct tdqcow_state *)dd->private;

        if (sid > MAX_IOFD) return 1;

        nr_events = tap_aio_get_events(&prv->aio.aio_ctx);
repeat:
        for (ep = prv->aio.aio_events, i = nr_events; i-- > 0; ep++) {
                struct iocb        *io  = ep->obj;
                struct pending_aio *pio;

                pio = &prv->aio.pending_aio[(long)io->data];

		tap_aio_unlock(&prv->aio, pio->sector);

		if (prv->crypt_method)
			encrypt_sectors(prv, pio->sector, 
					(unsigned char *)pio->buf, 
					(unsigned char *)pio->buf, 
					pio->nb_sectors, 0, 
					&prv->aes_decrypt_key);

		rsp += pio->cb(dd, ep->res == io->u.c.nbytes ? 0 : 1, 
			       pio->sector, pio->nb_sectors,
			       pio->id, pio->private);

                prv->aio.iocb_free[prv->aio.iocb_free_count++] = io;
        }

        if (nr_events) {
                nr_events = tap_aio_more_events(&prv->aio.aio_ctx);
                goto repeat;
        }

        tap_aio_continue(&prv->aio.aio_ctx);

        return rsp;
}

int qcow_create(const char *filename, uint64_t total_size,
		const char *backing_file, int sparse)
{
	int fd, header_size, backing_filename_len, l1_size, i;
	int shift, length, adjust, flags = 0, ret = 0;
	QCowHeader header;
	QCowHeader_ext exthdr;
	char backing_filename[PATH_MAX], *ptr;
	uint64_t tmp, size, total_length;
	struct stat st;

	DPRINTF("Qcow_create: size %llu\n",(long long unsigned)total_size);

	fd = open(filename, 
		  O_WRONLY | O_CREAT | O_TRUNC | O_BINARY | O_LARGEFILE,
		  0644);
	if (fd < 0)
		return -1;

	memset(&header, 0, sizeof(header));
	header.magic = cpu_to_be32(QCOW_MAGIC);
	header.version = cpu_to_be32(QCOW_VERSION);

	/*Create extended header fields*/
	exthdr.xmagic = cpu_to_be32(XEN_MAGIC);

	header_size = sizeof(header) + sizeof(QCowHeader_ext);
	backing_filename_len = 0;
	size = (total_size >> SECTOR_SHIFT);
	if (backing_file) {
		if (strcmp(backing_file, "fat:")) {
			const char *p;
			/* XXX: this is a hack: we do not attempt to 
			 *check for URL like syntax */
			p = strchr(backing_file, ':');
			if (p && (p - backing_file) >= 2) {
				/* URL like but exclude "c:" like filenames */
				strncpy(backing_filename, backing_file,
					sizeof(backing_filename));
			} else {
				if (realpath(backing_file, backing_filename) == NULL ||
				    stat(backing_filename, &st) != 0) {
					return -1;
				}
			}
			header.backing_file_offset = cpu_to_be64(header_size);
			backing_filename_len = strlen(backing_filename);
			header.backing_file_size = cpu_to_be32(
				backing_filename_len);
			header_size += backing_filename_len;
			
			/*Set to the backing file size*/
			if(get_filesize(backing_filename, &size, &st)) {
				return -1;
			}
			DPRINTF("Backing file size detected: %lld sectors" 
				"(total %lld [%lld MB])\n", 
				(long long)size, 
				(long long)(size << SECTOR_SHIFT), 
				(long long)(size >> 11));
		} else {
			backing_file = NULL;
			DPRINTF("Setting file size: %lld (total %lld)\n", 
				(long long) total_size, 
				(long long) (total_size << SECTOR_SHIFT));
		}
		header.mtime = cpu_to_be32(st.st_mtime);
		header.cluster_bits = 9; /* 512 byte cluster to avoid copying
					    unmodifyed sectors */
		header.l2_bits = 12; /* 32 KB L2 tables */
		exthdr.min_cluster_alloc = cpu_to_be32(1);
	} else {
		DPRINTF("Setting file size: %lld sectors" 
			"(total %lld [%lld MB])\n", 
			(long long) size, 
			(long long) (size << SECTOR_SHIFT), 
			(long long) (size >> 11));
		header.cluster_bits = 12; /* 4 KB clusters */
		header.l2_bits = 9; /* 4 KB L2 tables */
		exthdr.min_cluster_alloc = cpu_to_be32(1 << 9);
	}
	/*Set the header size value*/
	header.size = cpu_to_be64(size * 512);
	
	header_size = (header_size + 7) & ~7;
	if (header_size % 4096 > 0) {
		header_size = ((header_size >> 12) + 1) << 12;
	}

	shift = header.cluster_bits + header.l2_bits;
	l1_size = ((size * 512) + (1LL << shift) - 1) >> shift;

	header.l1_table_offset = cpu_to_be64(header_size);
	DPRINTF("L1 Table offset: %d, size %d\n",
		header_size,
		(int)(l1_size * sizeof(uint64_t)));
	header.crypt_method = cpu_to_be32(QCOW_CRYPT_NONE);

	ptr = calloc(1, l1_size * sizeof(uint64_t));
	exthdr.cksum = cpu_to_be32(gen_cksum(ptr, l1_size * sizeof(uint64_t)));
	printf("Created cksum: %d\n",exthdr.cksum);
	free(ptr);

	/*adjust file length to system page size boundary*/
	length = ROUNDUP(header_size + (l1_size * sizeof(uint64_t)),
		getpagesize());
	if (qtruncate(fd, length, 0)!=0) {
		DPRINTF("ERROR truncating file\n");
		return -1;
	}

	if (sparse == 0) {
		/*Filesize is length+l1_size*(1 << s->l2_bits)+(size*512)*/
		total_length = length + (l1_size * (1 << 9)) + (size * 512);
		if (qtruncate(fd, total_length, 0)!=0) {
                        DPRINTF("ERROR truncating file\n");
                        return -1;
		}
		printf("File truncated to length %"PRIu64"\n",total_length);
	} else
		flags = SPARSE_FILE;

	exthdr.flags = cpu_to_be32(flags);
	
	/* write all the data */
	lseek(fd, 0, SEEK_SET);
	ret += write(fd, &header, sizeof(header));
	ret += write(fd, &exthdr, sizeof(exthdr));
	if (backing_file)
		ret += write(fd, backing_filename, backing_filename_len);

	lseek(fd, header_size, SEEK_SET);
	tmp = 0;
	for (i = 0;i < l1_size; i++) {
		ret += write(fd, &tmp, sizeof(tmp));
	}

	close(fd);

	return 0;
}

static int qcow_make_empty(struct tdqcow_state *s)
{
	uint32_t l1_length = s->l1_size * sizeof(uint64_t);

	memset(s->l1_table, 0, l1_length);
	lseek(s->fd, s->l1_table_offset, SEEK_SET);
	if (write(s->fd, s->l1_table, l1_length) < 0)
		return -1;
	if (qtruncate(s->fd, s->l1_table_offset + l1_length, s->sparse)!=0) {
		DPRINTF("ERROR truncating file\n");
		return -1;
	}

	memset(s->l2_cache, 0, s->l2_size * L2_CACHE_SIZE * sizeof(uint64_t));
	memset(s->l2_cache_offsets, 0, L2_CACHE_SIZE * sizeof(uint64_t));
	memset(s->l2_cache_counts, 0, L2_CACHE_SIZE * sizeof(uint32_t));

	return 0;
}

static int qcow_get_cluster_size(struct tdqcow_state *s)
{
	return s->cluster_size;
}

/* XXX: put compressed sectors first, then all the cluster aligned
   tables to avoid losing bytes in alignment */
static int qcow_compress_cluster(struct tdqcow_state *s, int64_t sector_num, 
                          const uint8_t *buf)
{
	z_stream strm;
	int ret, out_len;
	uint8_t *out_buf;
	uint64_t cluster_offset;

	out_buf = malloc(s->cluster_size + (s->cluster_size / 1000) + 128);
	if (!out_buf)
		return -1;

	/* best compression, small window, no zlib header */
	memset(&strm, 0, sizeof(strm));
	ret = deflateInit2(&strm, Z_DEFAULT_COMPRESSION,
			   Z_DEFLATED, -12, 
			   9, Z_DEFAULT_STRATEGY);
	if (ret != 0) {
		free(out_buf);
		return -1;
	}

	strm.avail_in = s->cluster_size;
	strm.next_in = (uint8_t *)buf;
	strm.avail_out = s->cluster_size;
	strm.next_out = out_buf;

	ret = deflate(&strm, Z_FINISH);
	if (ret != Z_STREAM_END && ret != Z_OK) {
		free(out_buf);
		deflateEnd(&strm);
		return -1;
	}
	out_len = strm.next_out - out_buf;

	deflateEnd(&strm);

	if (ret != Z_STREAM_END || out_len >= s->cluster_size) {
		/* could not compress: write normal cluster */
		//tdqcow_queue_write(bs, sector_num, buf, s->cluster_sectors);
	} else {
		cluster_offset = get_cluster_offset(s, sector_num << 9, 2, 
                                            out_len, 0, 0);
		cluster_offset &= s->cluster_offset_mask;
		lseek(s->fd, cluster_offset, SEEK_SET);
		if (write(s->fd, out_buf, out_len) != out_len) {
			free(out_buf);
			return -1;
		}
	}
	
	free(out_buf);
	return 0;
}

static int tdqcow_get_parent_id(struct disk_driver *dd, struct disk_id *id)
{
	off_t off;
	char *buf, *filename;
	int len, secs, err = -EINVAL;
	struct tdqcow_state *child  = (struct tdqcow_state *)dd->private;

	if (!child->backing_file_offset)
		return TD_NO_PARENT;

	/* read the backing file name */
	len  = child->backing_file_size;
	off  = child->backing_file_offset - (child->backing_file_offset % 512);
	secs = (len + (child->backing_file_offset - off) + 511) >> 9;

	if (posix_memalign((void **)&buf, 512, secs << 9)) 
		return -1;

	if (lseek(child->fd, off, SEEK_SET) == (off_t)-1)
		goto out;

	if (read(child->fd, buf, secs << 9) != secs << 9)
		goto out;
	filename       = buf + (child->backing_file_offset - off);
	filename[len]  = '\0';

	id->name       = strdup(filename);
	id->drivertype = DISK_TYPE_QCOW;
	err            = 0;
 out:
	free(buf);
	return err;
}

static int tdqcow_validate_parent(struct disk_driver *child,
			   struct disk_driver *parent, td_flag_t flags)
{
	struct stat stats;
	uint64_t psize, csize;
	struct tdqcow_state *c = (struct tdqcow_state *)child->private;
	struct tdqcow_state *p = (struct tdqcow_state *)parent->private;
	
	if (stat(p->name, &stats))
		return -EINVAL;
	if (get_filesize(p->name, &psize, &stats))
		return -EINVAL;

	if (stat(c->name, &stats))
		return -EINVAL;
	if (get_filesize(c->name, &csize, &stats))
		return -EINVAL;

	if (csize != psize)
		return -EINVAL;

	return 0;
}

struct tap_disk tapdisk_qcow = {
	.disk_type           = "tapdisk_qcow",
	.private_data_size   = sizeof(struct tdqcow_state),
	.td_open             = tdqcow_open,
	.td_queue_read       = tdqcow_queue_read,
	.td_queue_write      = tdqcow_queue_write,
	.td_submit           = tdqcow_submit,
	.td_close            = tdqcow_close,
	.td_do_callbacks     = tdqcow_do_callbacks,
	.td_get_parent_id    = tdqcow_get_parent_id,
	.td_validate_parent  = tdqcow_validate_parent
};