[Qemu-devel] [PATCH] Add support for DictZip enabled gzip files

[Alexander Graf]

DictZip is an extension to the gzip format that allows random seeks in gzip
compressed files by cutting the file into pieces and storing the piece offsets
in the "extra" header of the gzip format.

Thanks to that extension, we can use gzip compressed files as block backend,
though only in read mode.

This makes a lot of sense when stacked with tar files that can then be shipped
to VM users. If a VM image is inside a tar file that is inside a DictZip
enabled gzip file, the user can run the tar.gz file as is without having to
extract the image first.

Tar patch will follow.

Signed-off-by: Alexander Graf <address@hidden>
---
 Makefile        |    2 +-
 block/dictzip.c |  492 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 493 insertions(+), 1 deletions(-)
 create mode 100644 block/dictzip.c

diff --git a/Makefile b/Makefile
index d3f999e..288190d 100644
--- a/Makefile
+++ b/Makefile
@@ -73,7 +73,7 @@ block-obj-$(CONFIG_AIO) += posix-aio-compat.o
 
 block-nested-y += cow.o qcow.o vmdk.o cloop.o dmg.o bochs.o vpc.o vvfat.o
 block-nested-y += qcow2.o qcow2-refcount.o qcow2-cluster.o qcow2-snapshot.o
-block-nested-y += parallels.o nbd.o
+block-nested-y += parallels.o nbd.o dictzip.o
 block-nested-$(CONFIG_WIN32) += raw-win32.o
 block-nested-$(CONFIG_POSIX) += raw-posix.o
 block-nested-$(CONFIG_CURL) += curl.o
diff --git a/block/dictzip.c b/block/dictzip.c
new file mode 100644
index 0000000..27c0a29
--- /dev/null
+++ b/block/dictzip.c
@@ -0,0 +1,492 @@
+/*
+ * DictZip Block driver for dictzip enabled gzip files
+ *
+ * Use the "dictzip" tool from the "dictd" package to create gzip files that
+ * contain the extra DictZip headers.
+ *
+ * dictzip(1) is a compression program which creates compressed files in the
+ * gzip format (see RFC 1952). However, unlike gzip(1), dictzip(1) compresses
+ * the file in pieces and stores an index to the pieces in the gzip header.
+ * This allows random access to the file at the granularity of the compressed
+ * pieces (currently about 64kB) while maintaining good compression ratios
+ * (within 5% of the expected ratio for dictionary data).
+ * dictd(8) uses files stored in this format.
+ *
+ * For details on DictZip see http://dict.org/.
+ *
+ * Copyright (c) 2009 Alexander Graf <address@hidden>
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to 
deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 
FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ */
+
+#include "qemu-common.h"
+#include "block_int.h"
+#include <zlib.h>
+
+// #define DEBUG
+
+#ifdef DEBUG
+#define dprintf(fmt, ...) do { printf("dzip: " fmt, ## __VA_ARGS__); } while 
(0)
+#else
+#define dprintf(fmt, ...) do { } while (0)
+#endif
+
+#define SECTOR_SIZE 512
+#define Z_STREAM_COUNT 4
+#define CACHE_COUNT 20
+
+/* magic values */
+
+#define GZ_MAGIC1     0x1f
+#define GZ_MAGIC2     0x8b
+#define DZ_MAGIC1      'R'
+#define DZ_MAGIC2      'A'
+
+#define GZ_FEXTRA     0x04      /* Optional field (random access index)    */
+#define GZ_FNAME      0x08      /* Original name                           */
+#define GZ_COMMENT    0x10      /* Zero-terminated, human-readable comment */
+#define GZ_FHCRC      0x02      /* Header CRC16                            */
+
+/* offsets */
+
+#define GZ_ID            0      /* GZ_MAGIC (16bit)                        */
+#define GZ_FLG           3      /* FLaGs (see above)                       */
+#define GZ_XLEN         10      /* eXtra LENgth (16bit)                    */
+#define GZ_SI           12      /* Subfield ID (16bit)                     */
+#define GZ_VERSION      16      /* Version for subfield format             */
+#define GZ_CHUNKLEN     18      /* Chunk length (16bit)                    */
+#define GZ_CHUNKCNT     20      /* Number of chunks (16bit)                */
+#define GZ_RNDDATA      22      /* Random access data (16bit)              */
+
+
+struct BDRVDictZipState;
+
+typedef struct DictZipAIOCB {
+    BlockDriverAIOCB common;
+    struct BDRVDictZipState *s;
+    QEMUIOVector *qiov;          /* QIOV of the original request */
+    QEMUIOVector *qiov_gz;       /* QIOV of the gz subrequest */
+    QEMUBH *bh;                  /* BH for cache */
+    z_stream *zStream;           /* stream to use for decoding */
+    int zStream_id;              /* stream id of the above pointer */
+    size_t start;                /* offset into the uncompressed file */
+    size_t len;                  /* uncompressed bytes to read */
+    uint8_t *gzipped;            /* the gzipped data */
+    uint8_t *buf;                /* cached result */
+    size_t gz_len;               /* amount of gzip data */
+    size_t gz_start;             /* uncompressed starting point of gzip data */
+    int offset;                  /* offset for "start" into the uncompressed 
chunk */
+    int chunks_len;              /* amount of uncompressed data in all gzip 
data */
+} DictZipAIOCB;
+
+typedef struct dict_cache {
+    size_t start;
+    size_t len;
+    uint8_t *buf;
+} DictCache;
+
+typedef struct BDRVDictZipState {
+    BlockDriverState *hd;
+    z_stream zStream[Z_STREAM_COUNT];
+    DictCache cache[CACHE_COUNT];
+    int cache_index;
+    uint8_t  stream_in_use;
+    uint16_t chunk_len;
+    uint16_t chunk_cnt;
+    uint16_t *chunks;
+    uint32_t *offsets;
+    uint32_t file_len;
+} BDRVDictZipState;
+
+static int dictzip_probe(const uint8_t *buf, int buf_size, const char 
*filename)
+{
+    if (buf_size < 2)
+        return 0;
+
+    /* We match on every gzip file */
+    if ((buf[0] == GZ_MAGIC1) && (buf[1] == GZ_MAGIC2))
+        return 100;
+
+    return 0;
+}
+
+static int start_zStream(z_stream *zStream)
+{
+    zStream->zalloc    = NULL;
+    zStream->zfree     = NULL;
+    zStream->opaque    = NULL;
+    zStream->next_in   = 0;
+    zStream->avail_in  = 0;
+    zStream->next_out  = NULL;
+    zStream->avail_out = 0;
+
+    return inflateInit2( zStream, -15 );
+}
+
+static int dictzip_open(BlockDriverState *bs, const char *filename, int flags)
+{
+    BDRVDictZipState *s = bs->opaque;
+    const char *err = "Unknown (read error?)";
+    uint8_t magic[2];
+    char buf[100];
+    uint8_t header_flags;
+    uint16_t header_ver;
+    uint16_t tmp_short;
+    uint32_t offset;
+    int chunks_len;
+    int headerLength = GZ_XLEN - 1;
+    int ret;
+    int i;
+
+    ret = bdrv_file_open(&s->hd, filename, flags);
+    if (ret < 0)
+        return ret;
+
+    /* initialize zlib streams */
+    for (i = 0; i < Z_STREAM_COUNT; i++) {
+        if (start_zStream( &s->zStream[i] ) != Z_OK) {
+            err = s->zStream[i].msg;
+            goto fail;
+        }
+    }
+
+    /* gzip header */
+    if (bdrv_pread(s->hd, GZ_ID, &magic, sizeof(magic)) != sizeof(magic))
+        goto fail;
+
+    if (!((magic[0] == GZ_MAGIC1) && (magic[1] == GZ_MAGIC2))) {
+        err = "No gzip file";
+        goto fail;
+    }
+
+    /* dzip header */
+    if (bdrv_pread(s->hd, GZ_FLG, &header_flags, 1) != 1)
+        goto fail;
+
+    if (!(header_flags & GZ_FEXTRA)) {
+        err = "Not a dictzip file (wrong flags)";
+        goto fail;
+    }
+
+    /* extra length */
+    if (bdrv_pread(s->hd, GZ_XLEN, &tmp_short, 2) != 2)
+        goto fail;
+
+    headerLength += le16_to_cpu(tmp_short) + 2;
+
+    /* DictZip magic */
+    if (bdrv_pread(s->hd, GZ_SI, &magic, 2) != 2)
+        goto fail;
+
+    if (magic[0] != DZ_MAGIC1 || magic[1] != DZ_MAGIC2) {
+        err = "Not a dictzip file (missing extra magic)";
+        goto fail;
+    }
+
+    /* DictZip version */
+    if (bdrv_pread(s->hd, GZ_VERSION, &header_ver, 2) != 2)
+        goto fail;
+
+    header_ver = le16_to_cpu(header_ver);
+
+    if (header_ver != 1) {
+        err = "Invalid DictZip version";
+        goto fail;
+    }
+
+    /* chunk len */
+    if (bdrv_pread(s->hd, GZ_CHUNKLEN, &s->chunk_len, 2) != 2)
+        goto fail;
+
+    s->chunk_len = le16_to_cpu(s->chunk_len);
+
+    /* chunk count */
+    if (bdrv_pread(s->hd, GZ_CHUNKCNT, &s->chunk_cnt, 2) != 2)
+        goto fail;
+
+    s->chunk_cnt = le16_to_cpu(s->chunk_cnt);
+
+    /* random access data */
+    chunks_len = sizeof(*s->chunks) * s->chunk_cnt;
+    s->chunks = qemu_malloc(chunks_len);
+
+    if (bdrv_pread(s->hd, GZ_RNDDATA, s->chunks, chunks_len) != chunks_len)
+        goto fail;
+
+    /* orig filename */
+    if (header_flags & GZ_FNAME) {
+        if (bdrv_pread(s->hd, headerLength + 1, buf, sizeof(buf)) != 
sizeof(buf))
+            goto fail;
+
+        buf[sizeof(buf) - 1] = '\0';
+        headerLength += strlen(buf) + 1;
+
+        if (strlen(buf) == sizeof(buf))
+            goto fail;
+
+        dprintf("filename: %s\n", buf);
+    }
+
+    /* comment field */
+    if (header_flags & GZ_COMMENT) {
+        if (bdrv_pread(s->hd, headerLength, buf, sizeof(buf)) != sizeof(buf))
+            goto fail;
+
+        buf[sizeof(buf) - 1] = '\0';
+        headerLength += strlen(buf) + 1;
+
+        if (strlen(buf) == sizeof(buf))
+            goto fail;
+
+        dprintf("comment: %s\n", buf);
+    }
+
+    if (header_flags & GZ_FHCRC)
+        headerLength += 2;
+
+    /* uncompressed file length*/
+    if (bdrv_pread(s->hd, bdrv_getlength(s->hd) - 4, &s->file_len, 4) != 4)
+        goto fail;
+
+    s->file_len = le32_to_cpu(s->file_len);
+
+    /* compute offsets */
+    s->offsets = qemu_malloc(sizeof( *s->offsets ) * s->chunk_cnt);
+
+    for (offset = headerLength + 1, i = 0; i < s->chunk_cnt; i++) {
+        s->offsets[i] = offset;
+        offset += s->chunks[i];
+    }
+
+    return 0;
+
+fail:
+    fprintf(stderr, "DictZip: Error opening file: %s\n", err);
+    bdrv_delete(s->hd);
+    if (s->chunks)
+        qemu_free(s->chunks);
+    return -EINVAL;
+}
+
+static AIOPool dictzip_aio_pool = {
+    .aiocb_size         = sizeof(DictZipAIOCB),
+};
+
+/* This callback gets invoked when we have the result in cache already */
+static void dictzip_cache_cb(void *opaque)
+{
+    DictZipAIOCB *acb = (DictZipAIOCB *)opaque;
+
+    qemu_iovec_from_buffer(acb->qiov, acb->buf, acb->len);
+    acb->common.cb(acb->common.opaque, 0);
+    qemu_bh_delete(acb->bh);
+    qemu_aio_release(acb);
+}
+
+/* This callback gets invoked by the underlying block reader when we have
+ * all compressed data. We uncompress in here. */
+static void dictzip_read_cb(void *opaque, int ret)
+{
+    DictZipAIOCB *acb = (DictZipAIOCB *)opaque;
+    struct BDRVDictZipState *s = acb->s;
+    uint8_t *buf;
+    DictCache *cache;
+
+    buf = qemu_malloc(acb->chunks_len);
+
+    /* uncompress the chunk */
+    acb->zStream->next_in   = acb->gzipped;
+    acb->zStream->avail_in  = acb->gz_len;
+    acb->zStream->next_out  = buf;
+    acb->zStream->avail_out = acb->chunks_len;
+    if (inflate( acb->zStream,  Z_PARTIAL_FLUSH ) != Z_OK)
+        fprintf(stderr, "Error inflating: %s\n", acb->zStream->msg);
+    s->stream_in_use &= ~(1 << acb->zStream_id);
+
+    /* nofity the caller */
+    qemu_iovec_from_buffer(acb->qiov, buf + acb->offset, acb->len);
+    acb->common.cb(acb->common.opaque, 0);
+
+    /* fill the cache */
+    cache = &s->cache[s->cache_index];
+    s->cache_index++;
+    if (s->cache_index == CACHE_COUNT)
+        s->cache_index = 0;
+    
+    cache->len = 0;
+    if (cache[s->cache_index].buf)
+        qemu_free(cache->buf);
+    cache->start = acb->gz_start;
+    cache->buf = buf;
+    cache->len = acb->chunks_len;
+
+    /* free occupied ressources */
+    qemu_free(acb->qiov_gz);
+    qemu_aio_release(acb);
+}
+
+/* This is where we get a request from a caller to read something */
+static BlockDriverAIOCB *dictzip_aio_readv(BlockDriverState *bs,
+        int64_t sector_num, QEMUIOVector *qiov, int nb_sectors,
+        BlockDriverCompletionFunc *cb, void *opaque)
+{
+    BDRVDictZipState *s = bs->opaque;
+    DictZipAIOCB *acb;
+    QEMUIOVector *qiov_gz;
+    struct iovec *iov;
+    uint8_t *buf;
+    size_t  start = sector_num * SECTOR_SIZE;
+    size_t  len = nb_sectors * SECTOR_SIZE;
+    size_t  end = start + len;
+    size_t  gz_start;
+    size_t  gz_len;
+    int64_t gz_sector_num;
+    int     gz_nb_sectors;
+    int     first_chunk, last_chunk;
+    int     first_offset, last_offset;
+    int     i;
+
+    acb = qemu_aio_get(&dictzip_aio_pool, bs, cb, opaque);
+    if (!acb)
+        return NULL;
+
+    /* Search Cache */
+    for (i = 0; i < CACHE_COUNT; i++) {
+        if (!s->cache[i].len)
+            continue;
+
+        if ((start >= s->cache[i].start) &&
+            (end <= (s->cache[i].start + s->cache[i].len))) {
+            acb->buf = s->cache[i].buf + (start - s->cache[i].start);
+            acb->len = len;
+            acb->qiov = qiov;
+            acb->bh = qemu_bh_new(dictzip_cache_cb, acb);
+            qemu_bh_schedule(acb->bh);
+
+            return &acb->common;
+        }
+    }
+
+    /* No cache, so let's decode */
+    do {
+        for (i = 0; i < Z_STREAM_COUNT; i++) {
+            if (!(s->stream_in_use & (1 << i))) {
+                s->stream_in_use |= (1 << i);
+                acb->zStream_id = i;
+                acb->zStream = &s->zStream[i];
+                break;
+            }
+        }
+    } while(!acb->zStream);
+
+    /* We need to read these chunks */
+    first_chunk  = start / s->chunk_len;
+    first_offset = start - first_chunk * s->chunk_len;
+    last_chunk   = end / s->chunk_len;
+    last_offset  = end - last_chunk * s->chunk_len;
+
+    gz_start = s->offsets[first_chunk];
+    gz_len = 0;
+    for (i = first_chunk; i <= last_chunk; i++) {
+        gz_len += s->chunks[i];
+    }
+
+    gz_sector_num = gz_start / SECTOR_SIZE;
+    gz_nb_sectors = (gz_len / SECTOR_SIZE);
+
+    /* account for tail and heads */
+    while ((gz_start + gz_len) > ((gz_sector_num + gz_nb_sectors) * 
SECTOR_SIZE))
+        gz_nb_sectors++;
+
+    /* Allocate qiov, iov and buf in one chunk so we only need to free qiov */
+    qiov_gz = qemu_mallocz(sizeof(QEMUIOVector) + sizeof(struct iovec) +
+                           (gz_nb_sectors * SECTOR_SIZE));
+    iov = (struct iovec *)(((char *)qiov_gz) + sizeof(QEMUIOVector));
+    buf = ((uint8_t *)iov) + sizeof(struct iovec *);
+
+    /* Kick off the read by the backing file, so we can start decompressing */
+    iov->iov_base = (void *)buf;
+    iov->iov_len = gz_nb_sectors * 512;
+    qemu_iovec_init_external(qiov_gz, iov, 1);
+
+    acb->s = s;
+    acb->qiov = qiov;
+    acb->qiov_gz = qiov_gz;
+    acb->start = start;
+    acb->len = len;
+    acb->gzipped = buf + (gz_start % SECTOR_SIZE);
+    acb->gz_len = gz_len;
+    acb->gz_start = first_chunk * s->chunk_len;
+    acb->offset = first_offset;
+    acb->chunks_len = (last_chunk - first_chunk + 1) * s->chunk_len;
+
+    return bdrv_aio_readv(s->hd, gz_sector_num, qiov_gz, gz_nb_sectors,
+                          dictzip_read_cb, acb);
+}
+
+static void dictzip_close(BlockDriverState *bs)
+{
+    BDRVDictZipState *s = bs->opaque;
+    int i;
+
+    for (i = 0; i < CACHE_COUNT; i++) {
+        if (!s->cache[i].len)
+            continue;
+
+        qemu_free(s->cache[i].buf);
+    }
+
+    for (i = 0; i < Z_STREAM_COUNT; i++) {
+        inflateEnd(&s->zStream[i]);
+    }
+
+    if (s->chunks)
+        qemu_free(s->chunks);
+
+    if (s->offsets)
+        qemu_free(s->offsets);
+
+    dprintf("Close\n");
+}
+
+static int64_t dictzip_getlength(BlockDriverState *bs)
+{
+    BDRVDictZipState *s = bs->opaque;
+    dprintf("getlength -> %d\n", s->file_len);
+    return s->file_len;
+}
+
+static BlockDriver bdrv_dictzip = {
+    .format_name     = "dzip",
+
+    .instance_size   = sizeof(BDRVDictZipState),
+    .bdrv_open       = dictzip_open,
+    .bdrv_close      = dictzip_close,
+    .bdrv_getlength  = dictzip_getlength,
+    .bdrv_probe      = dictzip_probe,
+
+    .bdrv_aio_readv  = dictzip_aio_readv,
+};
+
+static void dictzip_block_init(void)
+{
+    bdrv_register(&bdrv_dictzip);
+}
+
+block_init(dictzip_block_init);
-- 
1.6.0.2