summaryrefslogtreecommitdiffstats
path: root/xlators/features/compress/src/cdc-helper.c
diff options
context:
space:
mode:
authorPrashanth Pai <ppai@redhat.com>2012-03-26 14:33:41 +0530
committerVijay Bellur <vbellur@redhat.com>2013-11-11 19:35:01 -0800
commitb18257183d137c7ea8a76b4d57812fe26d9ea850 (patch)
tree2f50aa44455255afb3f43b18b9a0066331e25cc1 /xlators/features/compress/src/cdc-helper.c
parent27935ee84c484fff3d8d29218c197812701957ed (diff)
features/compress: Compression/DeCompression translator
* When a writev call occurs, the client compresses the data before sending it to server. On the server, compressed data is decompressed. Similarly, when a readv call occurs, the server compresses the data before sending it to client. On the client, the compressed data is decompressed. Thus the amount of data sent over the wire is minimized. * Compression/Decompression is done using Zlib library. * During normal operation, this is the format of data sent over wire : <compressed-data> + trailer(8) The trailer contains the CRC32 checksum and length of original uncompressed data. This is used for validation. HOW TO USE ---------- Turning on compression xlator: gluster volume set <vol_name> compress on Configurable options: gluster volume set <vol_name> compress.compression-level 8 gluster volume set <vol_name> compress.min-size 50 Change-Id: Ib7a66b6f1f70fe002b7c513588cdf75c69370805 BUG: 923540 Original-author : Venky Shankar <vshankar@redhat.com> Signed-off-by: Venky Shankar <vshankar@redhat.com> Signed-off-by: Prashanth Pai <nullpai@gmail.com> Signed-off-by: Prashanth Pai <ppai@redhat.com> Reviewed-on: http://review.gluster.org/3251 Tested-by: Gluster Build System <jenkins@build.gluster.com> Reviewed-by: Vijay Bellur <vbellur@redhat.com>
Diffstat (limited to 'xlators/features/compress/src/cdc-helper.c')
-rw-r--r--xlators/features/compress/src/cdc-helper.c547
1 files changed, 547 insertions, 0 deletions
diff --git a/xlators/features/compress/src/cdc-helper.c b/xlators/features/compress/src/cdc-helper.c
new file mode 100644
index 00000000000..54432ff455c
--- /dev/null
+++ b/xlators/features/compress/src/cdc-helper.c
@@ -0,0 +1,547 @@
+/*
+ Copyright (c) 2013 Red Hat, Inc. <http://www.redhat.com>
+ This file is part of GlusterFS.
+
+ This file is licensed to you under your choice of the GNU Lesser
+ General Public License, version 3 or any later version (LGPLv3 or
+ later), or the GNU General Public License, version 2 (GPLv2), in all
+ cases as published by the Free Software Foundation.
+*/
+
+#ifndef _CONFIG_H
+#define _CONFIG_H
+#include "config.h"
+#endif
+
+#include "glusterfs.h"
+#include "logging.h"
+
+#include "cdc.h"
+#include "cdc-mem-types.h"
+
+#ifdef HAVE_LIB_Z
+#include "zlib.h"
+#endif
+
+#ifdef HAVE_LIB_Z
+/* gzip header looks something like this
+ * (RFC 1950)
+ *
+ * +---+---+---+---+---+---+---+---+---+---+
+ * |ID1|ID2|CM |FLG| MTIME |XFL|OS |
+ * +---+---+---+---+---+---+---+---+---+---+
+ *
+ * Data is usually sent without this header i.e
+ * Data sent = <compressed-data> + trailer(8)
+ * The trailer contains the checksum.
+ *
+ * gzip_header is added only during debugging.
+ * Refer to the function cdc_dump_iovec_to_disk
+ */
+static const char gzip_header[10] =
+ {
+ '\037', '\213', Z_DEFLATED, 0,
+ 0, 0, 0, 0,
+ 0, GF_CDC_OS_ID
+ };
+
+static int32_t
+cdc_next_iovec (xlator_t *this, cdc_info_t *ci)
+{
+ int ret = -1;
+
+ ci->ncount++;
+ /* check for iovec overflow -- should not happen */
+ if (ci->ncount == MAX_IOVEC) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Zlib output buffer overflow"
+ " ->ncount (%d) | ->MAX_IOVEC (%d)",
+ ci->ncount, MAX_IOVEC);
+ goto out;
+ }
+
+ ret = 0;
+
+ out:
+ return ret;
+}
+
+static void
+cdc_put_long (unsigned char *string, unsigned long x)
+{
+ string[0] = (unsigned char) (x & 0xff);
+ string[1] = (unsigned char) ((x & 0xff00) >> 8);
+ string[2] = (unsigned char) ((x & 0xff0000) >> 16);
+ string[3] = (unsigned char) ((x & 0xff000000) >> 24);
+}
+
+static unsigned long
+cdc_get_long (unsigned char *buf)
+{
+ return ((unsigned long) buf[0])
+ | (((unsigned long) buf[1]) << 8)
+ | (((unsigned long) buf[2]) << 16)
+ | (((unsigned long) buf[3]) << 24);
+}
+
+static int32_t
+cdc_init_gzip_trailer (xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci)
+{
+ int ret = -1;
+ char *buf = NULL;
+
+ ret = cdc_next_iovec (this, ci);
+ if (ret)
+ goto out;
+
+ buf = CURR_VEC(ci).iov_base =
+ (char *) GF_CALLOC (1, GF_CDC_VALIDATION_SIZE,
+ gf_cdc_mt_gzip_trailer_t);
+
+ if (!CURR_VEC(ci).iov_base)
+ goto out;
+
+ CURR_VEC(ci).iov_len = GF_CDC_VALIDATION_SIZE;
+
+ cdc_put_long ((unsigned char *)&buf[0], ci->crc);
+ cdc_put_long ((unsigned char *)&buf[4], ci->stream.total_in);
+
+ ret = 0;
+
+ out:
+ return ret;
+}
+
+static int32_t
+cdc_alloc_iobuf_and_init_vec (xlator_t *this,
+ cdc_priv_t *priv, cdc_info_t *ci,
+ int size)
+{
+ int ret = -1;
+ int alloc_len = 0;
+ struct iobuf *iobuf = NULL;
+
+ ret = cdc_next_iovec (this, ci);
+ if (ret)
+ goto out;
+
+ alloc_len = size ? size : ci->buffer_size;
+
+ iobuf = iobuf_get2 (this->ctx->iobuf_pool, alloc_len);
+ if (!iobuf)
+ goto out;
+
+ ret = iobref_add (ci->iobref, iobuf);
+ if (ret)
+ goto out;
+
+ /* Initialize this iovec */
+ CURR_VEC(ci).iov_base = iobuf->ptr;
+ CURR_VEC(ci).iov_len = alloc_len;
+
+ ret = 0;
+
+ out:
+ return ret;
+}
+
+static void
+cdc_init_zlib_output_stream (cdc_priv_t *priv, cdc_info_t *ci, int size)
+{
+ ci->stream.next_out = (unsigned char *) CURR_VEC(ci).iov_base;
+ ci->stream.avail_out = size ? size : ci->buffer_size;
+}
+
+/* This routine is for testing and debugging only.
+ * Data written = header(10) + <compressed-data> + trailer(8)
+ * So each gzip dump file is at least 18 bytes in size.
+ */
+void
+cdc_dump_iovec_to_disk (xlator_t *this, cdc_info_t *ci, const char *file)
+{
+ int i = 0;
+ int fd = 0;
+ size_t writen = 0;
+ size_t total_writen = 0;
+
+ fd = open (file, O_WRONLY|O_CREAT|O_TRUNC, 0777 );
+ if (fd < 0) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Cannot open file: %s", file);
+ return;
+ }
+
+ writen = write (fd, (char *) gzip_header, 10);
+ total_writen += writen;
+ for (i = 0; i < ci->ncount; i++) {
+ writen = write (fd, (char *) ci->vec[i].iov_base, ci->vec[i].iov_len);
+ total_writen += writen;
+ }
+
+ gf_log (this->name, GF_LOG_DEBUG,
+ "dump'd %zu bytes to %s", total_writen, GF_CDC_DEBUG_DUMP_FILE );
+
+ close (fd);
+}
+
+static int32_t
+cdc_flush_libz_buffer (cdc_priv_t *priv, xlator_t *this, cdc_info_t *ci,
+ int (*libz_func)(z_streamp, int),
+ int flush)
+{
+ int32_t ret = Z_OK;
+ int done = 0;
+ unsigned int deflate_len = 0;
+
+ for (;;) {
+ deflate_len = ci->buffer_size - ci->stream.avail_out;
+
+ if (deflate_len != 0) {
+ CURR_VEC(ci).iov_len = deflate_len;
+
+ ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+ if (ret) {
+ ret = Z_MEM_ERROR;
+ break;
+ }
+
+ /* Re-position Zlib output buffer */
+ cdc_init_zlib_output_stream (priv, ci, 0);
+ }
+
+ if (done) {
+ ci->ncount--;
+ break;
+ }
+
+ ret = libz_func (&ci->stream, flush);
+
+ if (ret == Z_BUF_ERROR) {
+ ret = Z_OK;
+ ci->ncount--;
+ break;
+ }
+
+ done = (ci->stream.avail_out != 0 || ret == Z_STREAM_END);
+
+ if (ret != Z_OK && ret != Z_STREAM_END)
+ break;
+ }
+
+ return ret;
+}
+
+static int32_t
+do_cdc_compress (struct iovec *vec, xlator_t *this, cdc_priv_t *priv,
+ cdc_info_t *ci)
+{
+ int ret = -1;
+
+ /* Initialize defalte */
+ ret = deflateInit2 (&ci->stream, priv->cdc_level, Z_DEFLATED,
+ priv->window_size, priv->mem_level,
+ Z_DEFAULT_STRATEGY);
+
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "unable to init Zlib (retval: %d)", ret);
+ goto out;
+ }
+
+ ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+ if (ret)
+ goto out;
+
+ /* setup output buffer */
+ cdc_init_zlib_output_stream (priv, ci, 0);
+
+ /* setup input buffer */
+ ci->stream.next_in = (unsigned char *) vec->iov_base;
+ ci->stream.avail_in = vec->iov_len;
+
+ ci->crc = crc32 (ci->crc, (const Bytef *) vec->iov_base, vec->iov_len);
+
+ gf_log (this->name, GF_LOG_DEBUG, "crc=%lu len=%d buffer_size=%d",
+ ci->crc, ci->stream.avail_in, ci->buffer_size);
+
+ /* compress !! */
+ while (ci->stream.avail_in != 0) {
+ if (ci->stream.avail_out == 0) {
+
+ CURR_VEC(ci).iov_len = ci->buffer_size;
+
+ ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+ if (ret)
+ break;
+
+ /* Re-position Zlib output buffer */
+ cdc_init_zlib_output_stream (priv, ci, 0);
+ }
+
+ ret = deflate (&ci->stream, Z_NO_FLUSH);
+ if (ret != Z_OK)
+ break;
+ }
+
+ out:
+ return ret;
+}
+
+int32_t
+cdc_compress (xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci,
+ dict_t **xdata)
+{
+ int ret = -1;
+ int i = 0;
+
+ ci->iobref = iobref_new ();
+ if (!ci->iobref)
+ goto out;
+
+ if (!*xdata) {
+ *xdata = dict_new ();
+ if (!*xdata) {
+ gf_log (this->name, GF_LOG_ERROR, "Cannot allocate xdata"
+ " dict");
+ goto out;
+ }
+ }
+
+ /* data */
+ for (i = 0; i < ci->count; i++) {
+ ret = do_cdc_compress (&ci->vector[i], this, priv, ci);
+ if (ret != Z_OK)
+ goto deflate_cleanup_out;
+ }
+
+ /* flush zlib buffer */
+ ret = cdc_flush_libz_buffer (priv, this, ci, deflate, Z_FINISH);
+ if (!(ret == Z_OK || ret == Z_STREAM_END)) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Compression Error: ret (%d)", ret);
+ ret = -1;
+ goto deflate_cleanup_out;
+ }
+
+ /* trailer */
+ ret = cdc_init_gzip_trailer (this, priv, ci);
+ if (ret)
+ goto deflate_cleanup_out;
+
+ gf_log (this->name, GF_LOG_DEBUG,
+ "Compressed %ld to %ld bytes",
+ ci->stream.total_in, ci->stream.total_out);
+
+ ci->nbytes = ci->stream.total_out + GF_CDC_VALIDATION_SIZE;
+
+ /* set deflated canary value for identification */
+ ret = dict_set_int32 (*xdata, GF_CDC_DEFLATE_CANARY_VAL, 1);
+ if (ret) {
+ /* Send uncompressed data if we can't _tell_ the client
+ * that deflated data is on it's way. So, we just log
+ * the faliure and continue as usual.
+ */
+ gf_log (this->name, GF_LOG_ERROR,
+ "Data deflated, but could not set canary"
+ " value in dict for identification");
+ }
+
+ /* This is to be used in testing */
+ if ( priv->debug ) {
+ cdc_dump_iovec_to_disk (this, ci, GF_CDC_DEBUG_DUMP_FILE );
+ }
+
+ deflate_cleanup_out:
+ (void) deflateEnd(&ci->stream);
+
+ out:
+ return ret;
+}
+
+
+/* deflate content is checked by the presence of a canary
+ * value in the dict as the key
+ */
+static int32_t
+cdc_check_content_for_deflate (dict_t *xdata)
+{
+ return dict_get (xdata, GF_CDC_DEFLATE_CANARY_VAL) ? -1 : 0;
+}
+
+static unsigned long
+cdc_extract_crc (char *trailer)
+{
+ return cdc_get_long ((unsigned char *) &trailer[0]);
+}
+
+static unsigned long
+cdc_extract_size (char *trailer)
+{
+ return cdc_get_long ((unsigned char *) &trailer[4]);
+}
+
+static int32_t
+cdc_validate_inflate (cdc_info_t *ci, unsigned long crc,
+ unsigned long len)
+{
+ return !((crc == ci->crc)
+ /* inflated length is hidden inside
+ * Zlib stream struct */
+ && (len == ci->stream.total_out));
+}
+
+static int32_t
+do_cdc_decompress (xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci)
+{
+ int ret = -1;
+ int i = 0;
+ int len = 0;
+ char *inflte = NULL;
+ char *trailer = NULL;
+ struct iovec vec = {0,};
+ unsigned long computed_crc = 0;
+ unsigned long computed_len = 0;
+
+ ret = inflateInit2 (&ci->stream, priv->window_size);
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Zlib: Unable to initialize inflate");
+ goto out;
+ }
+
+ vec = THIS_VEC(ci, 0);
+
+ trailer = (char *) (((char *) vec.iov_base) + vec.iov_len
+ - GF_CDC_VALIDATION_SIZE);
+
+ /* CRC of uncompressed data */
+ computed_crc = cdc_extract_crc (trailer);
+
+ /* size of uncomrpessed data */
+ computed_len = cdc_extract_size (trailer);
+
+ gf_log (this->name, GF_LOG_DEBUG, "crc=%lu len=%lu buffer_size=%d",
+ computed_crc, computed_len, ci->buffer_size);
+
+ inflte = vec.iov_base ;
+ len = vec.iov_len - GF_CDC_VALIDATION_SIZE;
+
+ /* allocate buffer of the original length of the data */
+ ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+ if (ret)
+ goto out;
+
+ /* setup output buffer */
+ cdc_init_zlib_output_stream (priv, ci, 0);
+
+ /* setup input buffer */
+ ci->stream.next_in = (unsigned char *) inflte;
+ ci->stream.avail_in = len;
+
+ while (ci->stream.avail_in != 0) {
+ if (ci->stream.avail_out == 0) {
+ CURR_VEC(ci).iov_len = ci->buffer_size;
+
+ ret = cdc_alloc_iobuf_and_init_vec (this, priv, ci, 0);
+ if (ret)
+ break;
+
+ /* Re-position Zlib output buffer */
+ cdc_init_zlib_output_stream (priv, ci, 0);
+ }
+
+ ret = inflate (&ci->stream, Z_NO_FLUSH);
+ if (ret == Z_STREAM_ERROR)
+ break;
+ }
+
+ /* flush zlib buffer */
+ ret = cdc_flush_libz_buffer (priv, this, ci, inflate, Z_SYNC_FLUSH);
+ if (!(ret == Z_OK || ret == Z_STREAM_END)) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Decompression Error: ret (%d)", ret);
+ ret = -1;
+ goto out;
+ }
+
+ /* compute CRC of the uncompresses data to check for
+ * correctness */
+
+ for (i = 0; i < ci->ncount; i++) {
+ ci->crc = crc32 (ci->crc,
+ (const Bytef *) ci->vec[i].iov_base,
+ ci->vec[i].iov_len);
+ }
+
+ /* validate inflated data */
+ ret = cdc_validate_inflate (ci, computed_crc, computed_len);
+ if (ret) {
+ gf_log (this->name, GF_LOG_ERROR,
+ "Checksum or length mismatched in inflated data");
+ }
+
+ out:
+ return ret;
+}
+
+int32_t
+cdc_decompress (xlator_t *this, cdc_priv_t *priv, cdc_info_t *ci,
+ dict_t *xdata)
+{
+ int32_t ret = -1;
+
+ /* check for deflate content */
+ if (!cdc_check_content_for_deflate (xdata)) {
+ gf_log (this->name, GF_LOG_DEBUG,
+ "Content not deflated, passing through ...");
+ goto passthrough_out;
+ }
+
+ ci->iobref = iobref_new ();
+ if (!ci->iobref)
+ goto passthrough_out;
+
+ /* do we need to do this? can we assume that one iovec
+ * will hold per request data everytime?
+ *
+ * server/client protocol seems to deal with a single
+ * iovec even if op_ret > 1M. So, it looks ok to
+ * assume that a single iovec will contain all the
+ * data (This saves us a lot from finding the trailer
+ * and the data since it could have been split-up onto
+ * two adjacent iovec's.
+ *
+ * But, in case this translator is loaded above quick-read
+ * for some reason, then it's entirely possible that we get
+ * multiple iovec's...
+ *
+ * This case (handled below) is not tested. (by loading the
+ * xlator below quick-read)
+ */
+
+ /* @@ I_HOPE_THIS_IS_NEVER_HIT */
+ if (ci->count > 1) {
+ gf_log (this->name, GF_LOG_WARNING, "unable to handle"
+ " multiple iovecs (%d in number)", ci->count);
+ goto inflate_cleanup_out;
+ /* TODO: coallate all iovecs in one */
+ }
+
+ ret = do_cdc_decompress (this, priv, ci);
+ if (ret)
+ goto inflate_cleanup_out;
+
+ ci->nbytes = ci->stream.total_out;
+
+ gf_log (this->name, GF_LOG_DEBUG,
+ "Inflated %ld to %ld bytes",
+ ci->stream.total_in, ci->stream.total_out);
+
+ inflate_cleanup_out:
+ (void) inflateEnd (&ci->stream);
+
+ passthrough_out:
+ return ret;
+}
+
+#endif