Allow to compress on-the-wire data using two algorithms:

- HOLE - it simply turns all-zero blocks into few bytes header;
	it is extremely fast, so it is turned on by default;
	it is mostly intended to speed up initial synchronization
	where we expect many zeros;
- LZF - very fast algorithm by Marc Alexander Lehmann, which shows
	very decent compression ratio and has BSD license.

MFC after:	2 weeks

14 files changed, 1037 insertions, 4 deletions

diff --git a/sbin/hastctl/Makefile b/sbin/hastctl/Makefile
index 3137a75..c266666 100644
--- a/sbin/hastctl/Makefile
+++ b/sbin/hastctl/Makefile
@@ -8,7 +8,8 @@ PROG=	hastctl
 SRCS=	activemap.c
 SRCS+=	crc32.c
 SRCS+=	ebuf.c
-SRCS+=	hast_checksum.c hast_proto.c hastctl.c
+SRCS+=	hast_checksum.c hast_compression.c hast_proto.c hastctl.c
+SRCS+=	lzf.c
 SRCS+=	metadata.c
 SRCS+=	nv.c
 SRCS+=	parse.y pjdlog.c
diff --git a/sbin/hastd/Makefile b/sbin/hastd/Makefile
index 6d8dd71..8172c26 100644
--- a/sbin/hastd/Makefile
+++ b/sbin/hastd/Makefile
@@ -6,7 +6,8 @@ PROG=	hastd
 SRCS=	activemap.c
 SRCS+=	control.c crc32.c
 SRCS+=	ebuf.c event.c
-SRCS+=	hast_checksum.c hast_proto.c hastd.c hooks.c
+SRCS+=	hast_checksum.c hast_compression.c hast_proto.c hastd.c hooks.c
+SRCS+=	lzf.c
 SRCS+=	metadata.c
 SRCS+=	nv.c
 SRCS+=	secondary.c
diff --git a/sbin/hastd/control.c b/sbin/hastd/control.c
index bdbb34a..8fcd5ef 100644
--- a/sbin/hastd/control.c
+++ b/sbin/hastd/control.c
@@ -44,6 +44,7 @@ __FBSDID("$FreeBSD$");
 #include "hast.h"
 #include "hastd.h"
 #include "hast_checksum.h"
+#include "hast_compression.h"
 #include "hast_proto.h"
 #include "hooks.h"
 #include "nv.h"
@@ -249,6 +250,8 @@ control_status(struct hastd_config *cfg, struct nv *nvout,
 	}
 	nv_add_string(nvout, checksum_name(res->hr_checksum),
 	    "checksum%u", no);
+	nv_add_string(nvout, compression_name(res->hr_compression),
+	    "compression%u", no);
 	nv_add_string(nvout, role2str(res->hr_role), "role%u", no);
 
 	switch (res->hr_role) {
diff --git a/sbin/hastd/hast.conf.5 b/sbin/hastd/hast.conf.5
index 3fd2fc6..f4d0754 100644
--- a/sbin/hastd/hast.conf.5
+++ b/sbin/hastd/hast.conf.5
@@ -60,6 +60,7 @@ control <addr>
 listen <addr>
 replication <mode>
 checksum <algorithm>
+compression <algorithm>
 timeout <seconds>
 exec <path>
 
@@ -79,6 +80,7 @@ resource <name> {
 	# Resource section
 	replication <mode>
 	checksum <algorithm>
+	compression <algorithm>
 	name <name>
 	local <path>
 	timeout <seconds>
@@ -215,6 +217,24 @@ CRC32 checksum will be calculated.
 .It Ic sha256
 SHA256 checksum will be calculated.
 .El
+.It Ic compression Aq algorithm
+.Pp
+Compression algorithm should be one of the following:
+.Bl -tag -width ".Ic none"
+.It Ic none
+Data send over the network will not be compressed.
+.It Ic hole
+Only blocks that contain all zeros will be compressed.
+This is very useful for initial synchronization where potentially many blocks
+are still all zeros.
+There should be no measurable performance overhead when this algorithm is being
+used.
+This is the default setting.
+.It Ic lzf
+The LZF algorithm by Marc Alexander Lehmann will be used to compress the data
+send over the network.
+LZF is very fast, general purpose compression algorithm.
+.El
 .It Ic timeout Aq seconds
 .Pp
 Connection timeout in seconds.
diff --git a/sbin/hastd/hast.h b/sbin/hastd/hast.h
index fe9cf0d..8c64395 100644
--- a/sbin/hastd/hast.h
+++ b/sbin/hastd/hast.h
@@ -117,6 +117,10 @@ struct hastd_config {
 #define	HAST_REPLICATION_MEMSYNC	1
 #define	HAST_REPLICATION_ASYNC		2
 
+#define	HAST_COMPRESSION_NONE	0
+#define	HAST_COMPRESSION_HOLE	1
+#define	HAST_COMPRESSION_LZF	2
+
 #define	HAST_CHECKSUM_NONE	0
 #define	HAST_CHECKSUM_CRC32	1
 #define	HAST_CHECKSUM_SHA256	2
@@ -137,6 +141,8 @@ struct hast_resource {
 	int	hr_keepdirty;
 	/* Path to a program to execute on various events. */
 	char	hr_exec[PATH_MAX];
+	/* Compression algorithm. */
+	int	hr_compression;
 	/* Checksum algorithm. */
 	int	hr_checksum;
 
diff --git a/sbin/hastd/hast_compression.c b/sbin/hastd/hast_compression.c
new file mode 100644
index 0000000..4f4a93a
--- /dev/null
+++ b/sbin/hastd/hast_compression.c
@@ -0,0 +1,283 @@
+/*-
+ * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ */
+
+#include <sys/cdefs.h>
+__FBSDID("$FreeBSD$");
+
+#include <sys/endian.h>
+
+#include <errno.h>
+#include <string.h>
+#include <strings.h>
+
+#include <hast.h>
+#include <lzf.h>
+#include <nv.h>
+#include <pjdlog.h>
+
+#include "hast_compression.h"
+
+static bool
+allzeros(const void *data, size_t size)
+{
+	const uint64_t *p = data;
+	unsigned int i;
+	uint64_t v;
+
+	PJDLOG_ASSERT((size % sizeof(*p)) == 0);
+
+	/*
+	 * This is the fastest method I found for checking if the given
+	 * buffer contain all zeros.
+	 * Because inside the loop we don't check at every step, we would
+	 * get an answer only after walking through entire buffer.
+	 * To return early if the buffer doesn't contain all zeros, we probe
+	 * 8 bytes at the begining, in the middle and at the end of the buffer
+	 * first.
+	 */
+
+	size >>= 3;	/* divide by 8 */
+	if ((p[0] | p[size >> 1] | p[size - 1]) != 0)
+		return (false);
+	v = 0;
+	for (i = 0; i < size; i++)
+		v |= *p++;
+	return (v == 0);
+}
+
+static void *
+hast_hole_compress(const unsigned char *data, size_t *sizep)
+{
+	uint32_t size;
+	void *newbuf;
+
+	if (!allzeros(data, *sizep))
+		return (NULL);
+
+	newbuf = malloc(sizeof(size));
+	if (newbuf == NULL) {
+		pjdlog_warning("Unable to compress (no memory: %zu).",
+		    (size_t)*sizep);
+		return (NULL);
+	}
+	size = htole32((uint32_t)*sizep);
+	bcopy(&size, newbuf, sizeof(size));
+	*sizep = sizeof(size);
+
+	return (newbuf);
+}
+
+static void *
+hast_hole_decompress(const unsigned char *data, size_t *sizep)
+{
+	uint32_t size;
+	void *newbuf;
+
+	if (*sizep != sizeof(size)) {
+		pjdlog_error("Unable to decompress (invalid size: %zu).",
+		    *sizep);
+		return (NULL);
+	}
+
+	bcopy(data, &size, sizeof(size));
+	size = le32toh(size);
+
+	newbuf = malloc(size);
+	if (newbuf == NULL) {
+		pjdlog_error("Unable to decompress (no memory: %zu).",
+		    (size_t)size);
+		return (NULL);
+	}
+	bzero(newbuf, size);
+	*sizep = size;
+
+	return (newbuf);
+}
+
+/* Minimum block size to try to compress. */
+#define	HAST_LZF_COMPRESS_MIN	1024
+
+static void *
+hast_lzf_compress(const unsigned char *data, size_t *sizep)
+{
+	unsigned char *newbuf;
+	uint32_t origsize;
+	size_t newsize;
+
+	origsize = *sizep;
+
+	if (origsize <= HAST_LZF_COMPRESS_MIN)
+		return (NULL);
+
+	newsize = sizeof(origsize) + origsize - HAST_LZF_COMPRESS_MIN;
+	newbuf = malloc(newsize);
+	if (newbuf == NULL) {
+		pjdlog_warning("Unable to compress (no memory: %zu).",
+		    newsize);
+		return (NULL);
+	}
+	newsize = lzf_compress(data, *sizep, newbuf + sizeof(origsize),
+	    newsize - sizeof(origsize));
+	if (newsize == 0) {
+		free(newbuf);
+		return (NULL);
+	}
+	origsize = htole32(origsize);
+	bcopy(&origsize, newbuf, sizeof(origsize));
+
+	*sizep = sizeof(origsize) + newsize;
+	return (newbuf);
+}
+
+static void *
+hast_lzf_decompress(const unsigned char *data, size_t *sizep)
+{
+	unsigned char *newbuf;
+	uint32_t origsize;
+	size_t newsize;
+
+	PJDLOG_ASSERT(*sizep > sizeof(origsize));
+
+	bcopy(data, &origsize, sizeof(origsize));
+	origsize = le32toh(origsize);
+	PJDLOG_ASSERT(origsize > HAST_LZF_COMPRESS_MIN);
+
+	newbuf = malloc(origsize);
+	if (newbuf == NULL) {
+		pjdlog_error("Unable to decompress (no memory: %zu).",
+		    (size_t)origsize);
+		return (NULL);
+	}
+	newsize = lzf_decompress(data + sizeof(origsize),
+	    *sizep - sizeof(origsize), newbuf, origsize);
+	if (newsize == 0) {
+		free(newbuf);
+		pjdlog_error("Unable to decompress.");
+		return (NULL);
+	}
+	PJDLOG_ASSERT(newsize == origsize);
+
+	*sizep = newsize;
+	return (newbuf);
+}
+
+const char *
+compression_name(int num)
+{
+
+	switch (num) {
+	case HAST_COMPRESSION_NONE:
+		return ("none");
+	case HAST_COMPRESSION_HOLE:
+		return ("hole");
+	case HAST_COMPRESSION_LZF:
+		return ("lzf");
+	}
+	return ("unknown");
+}
+
+int
+compression_send(const struct hast_resource *res, struct nv *nv, void **datap,
+    size_t *sizep, bool *freedatap)
+{
+	unsigned char *newbuf;
+	int compression;
+	size_t size;
+
+	size = *sizep;
+	compression = res->hr_compression;
+
+	switch (compression) {
+	case HAST_COMPRESSION_NONE:
+		return (0);
+	case HAST_COMPRESSION_HOLE:
+		newbuf = hast_hole_compress(*datap, &size);
+		break;
+	case HAST_COMPRESSION_LZF:
+		/* Try 'hole' compression first. */
+		newbuf = hast_hole_compress(*datap, &size);
+		if (newbuf != NULL)
+			compression = HAST_COMPRESSION_HOLE;
+		else
+			newbuf = hast_lzf_compress(*datap, &size);
+		break;
+	default:
+		PJDLOG_ABORT("Invalid compression: %d.", res->hr_compression);
+	}
+
+	if (newbuf == NULL) {
+		/* Unable to compress the data. */
+		return (0);
+	}
+	nv_add_string(nv, compression_name(compression), "compression");
+	if (nv_error(nv) != 0) {
+		free(newbuf);
+		errno = nv_error(nv);
+		return (-1);
+	}
+	if (*freedatap)
+		free(*datap);
+	*freedatap = true;
+	*datap = newbuf;
+	*sizep = size;
+
+	return (0);
+}
+
+int
+compression_recv(const struct hast_resource *res __unused, struct nv *nv,
+    void **datap, size_t *sizep, bool *freedatap)
+{
+	unsigned char *newbuf;
+	const char *algo;
+	size_t size;
+
+	algo = nv_get_string(nv, "compression");
+	if (algo == NULL)
+		return (0);	/* No compression. */
+
+	newbuf = NULL;
+	size = *sizep;
+
+	if (strcmp(algo, "hole") == 0)
+		newbuf = hast_hole_decompress(*datap, &size);
+	else if (strcmp(algo, "lzf") == 0)
+		newbuf = hast_lzf_decompress(*datap, &size);
+	else {
+		pjdlog_error("Unknown compression algorithm '%s'.", algo);
+		return (-1);	/* Unknown compression algorithm. */
+	}
+
+	if (newbuf == NULL)
+		return (-1);
+	if (*freedatap)
+		free(*datap);
+	*freedatap = true;
+	*datap = newbuf;
+	*sizep = size;
+
+	return (0);
+}
diff --git a/sbin/hastd/hast_compression.h b/sbin/hastd/hast_compression.h
new file mode 100644
index 0000000..eabdfb2
--- /dev/null
+++ b/sbin/hastd/hast_compression.h
@@ -0,0 +1,44 @@
+/*-
+ * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHORS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHORS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * $FreeBSD$
+ */
+
+#ifndef	_HAST_COMPRESSION_H_
+#define	_HAST_COMPRESSION_H_
+
+#include <stdlib.h>	/* size_t */
+
+#include <hast.h>
+#include <nv.h>
+
+const char *compression_name(int num);
+
+int compression_send(const struct hast_resource *res, struct nv *nv,
+    void **datap, size_t *sizep, bool *freedatap);
+int compression_recv(const struct hast_resource *res, struct nv *nv,
+    void **datap, size_t *sizep, bool *freedatap);
+
+#endif	/* !_HAST_COMPRESSION_H_ */
diff --git a/sbin/hastd/hast_proto.c b/sbin/hastd/hast_proto.c
index 3f6c336..a7e1449 100644
--- a/sbin/hastd/hast_proto.c
+++ b/sbin/hastd/hast_proto.c
@@ -46,6 +46,7 @@ __FBSDID("$FreeBSD$");
 #ifdef HAVE_CRYPTO
 #include "hast_checksum.h"
 #endif
+#include "hast_compression.h"
 #include "hast_proto.h"
 
 struct hast_main_header {
@@ -67,6 +68,7 @@ struct hast_pipe_stage {
 };
 
 static struct hast_pipe_stage pipeline[] = {
+	{ "compression", compression_send, compression_recv },
 	{ "checksum", checksum_send, checksum_recv }
 };
 
diff --git a/sbin/hastd/hastd.c b/sbin/hastd/hastd.c
index c6db672..dd53eff 100644
--- a/sbin/hastd/hastd.c
+++ b/sbin/hastd/hastd.c
@@ -363,6 +363,8 @@ resource_needs_restart(const struct hast_resource *res0,
 			return (true);
 		if (res0->hr_checksum != res1->hr_checksum)
 			return (true);
+		if (res0->hr_compression != res1->hr_compression)
+			return (true);
 		if (res0->hr_timeout != res1->hr_timeout)
 			return (true);
 		if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
@@ -389,6 +391,8 @@ resource_needs_reload(const struct hast_resource *res0,
 		return (true);
 	if (res0->hr_checksum != res1->hr_checksum)
 		return (true);
+	if (res0->hr_compression != res1->hr_compression)
+		return (true);
 	if (res0->hr_timeout != res1->hr_timeout)
 		return (true);
 	if (strcmp(res0->hr_exec, res1->hr_exec) != 0)
@@ -409,6 +413,7 @@ resource_reload(const struct hast_resource *res)
 	nv_add_string(nvout, res->hr_remoteaddr, "remoteaddr");
 	nv_add_int32(nvout, (int32_t)res->hr_replication, "replication");
 	nv_add_int32(nvout, (int32_t)res->hr_checksum, "checksum");
+	nv_add_int32(nvout, (int32_t)res->hr_compression, "compression");
 	nv_add_int32(nvout, (int32_t)res->hr_timeout, "timeout");
 	nv_add_string(nvout, res->hr_exec, "exec");
 	if (nv_error(nvout) != 0) {
@@ -568,6 +573,7 @@ hastd_reload(void)
 			    sizeof(cres->hr_remoteaddr));
 			cres->hr_replication = nres->hr_replication;
 			cres->hr_checksum = nres->hr_checksum;
+			cres->hr_compression = nres->hr_compression;
 			cres->hr_timeout = nres->hr_timeout;
 			strlcpy(cres->hr_exec, nres->hr_exec,
 			    sizeof(cres->hr_exec));
diff --git a/sbin/hastd/lzf.c b/sbin/hastd/lzf.c
new file mode 100644
index 0000000..cca6a17
--- /dev/null
+++ b/sbin/hastd/lzf.c
@@ -0,0 +1,406 @@
+/*
+ * Copyright (c) 2000-2008 Marc Alexander Lehmann <schmorp@schmorp.de>
+ * 
+ * Redistribution and use in source and binary forms, with or without modifica-
+ * tion, are permitted provided that the following conditions are met:
+ * 
+ *   1.  Redistributions of source code must retain the above copyright notice,
+ *       this list of conditions and the following disclaimer.
+ * 
+ *   2.  Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
+ * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO
+ * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
+ * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
+ * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
+ * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Alternatively, the contents of this file may be used under the terms of
+ * the GNU General Public License ("GPL") version 2 or any later version,
+ * in which case the provisions of the GPL are applicable instead of
+ * the above. If you wish to allow the use of your version of this file
+ * only under the terms of the GPL and not to allow others to use your
+ * version of this file under the BSD license, indicate your decision
+ * by deleting the provisions above and replace them with the notice
+ * and other provisions required by the GPL. If you do not delete the
+ * provisions above, a recipient may use your version of this file under
+ * either the BSD or the GPL.
+ */
+
+#include "lzf.h"
+
+#define HSIZE (1 << (HLOG))
+
+/*
+ * don't play with this unless you benchmark!
+ * decompression is not dependent on the hash function
+ * the hashing function might seem strange, just believe me
+ * it works ;)
+ */
+#ifndef FRST
+# define FRST(p) (((p[0]) << 8) | p[1])
+# define NEXT(v,p) (((v) << 8) | p[2])
+# if ULTRA_FAST
+#  define IDX(h) ((( h             >> (3*8 - HLOG)) - h  ) & (HSIZE - 1))
+# elif VERY_FAST
+#  define IDX(h) ((( h             >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
+# else
+#  define IDX(h) ((((h ^ (h << 5)) >> (3*8 - HLOG)) - h*5) & (HSIZE - 1))
+# endif
+#endif
+/*
+ * IDX works because it is very similar to a multiplicative hash, e.g.
+ * ((h * 57321 >> (3*8 - HLOG)) & (HSIZE - 1))
+ * the latter is also quite fast on newer CPUs, and compresses similarly.
+ *
+ * the next one is also quite good, albeit slow ;)
+ * (int)(cos(h & 0xffffff) * 1e6)
+ */
+
+#if 0
+/* original lzv-like hash function, much worse and thus slower */
+# define FRST(p) (p[0] << 5) ^ p[1]
+# define NEXT(v,p) ((v) << 5) ^ p[2]
+# define IDX(h) ((h) & (HSIZE - 1))
+#endif
+
+#define        MAX_LIT        (1 <<  5)
+#define        MAX_OFF        (1 << 13)
+#define        MAX_REF        ((1 << 8) + (1 << 3))
+
+#if __GNUC__ >= 3
+# define expect(expr,value)         __builtin_expect ((expr),(value))
+# define inline                     inline
+#else
+# define expect(expr,value)         (expr)
+# define inline                     static
+#endif
+
+#define expect_false(expr) expect ((expr) != 0, 0)
+#define expect_true(expr)  expect ((expr) != 0, 1)
+
+/*
+ * compressed format
+ *
+ * 000LLLLL <L+1>    ; literal
+ * LLLooooo oooooooo ; backref L
+ * 111ooooo LLLLLLLL oooooooo ; backref L+7
+ *
+ */
+
+unsigned int
+lzf_compress (const void *const in_data, unsigned int in_len,
+	      void *out_data, unsigned int out_len
+#if LZF_STATE_ARG
+              , LZF_STATE htab
+#endif
+              )
+{
+#if !LZF_STATE_ARG
+  LZF_STATE htab;
+#endif
+  const u8 **hslot;
+  const u8 *ip = (const u8 *)in_data;
+        u8 *op = (u8 *)out_data;
+  const u8 *in_end  = ip + in_len;
+        u8 *out_end = op + out_len;
+  const u8 *ref;
+
+  /* off requires a type wide enough to hold a general pointer difference.
+   * ISO C doesn't have that (size_t might not be enough and ptrdiff_t only
+   * works for differences within a single object). We also assume that no
+   * no bit pattern traps. Since the only platform that is both non-POSIX
+   * and fails to support both assumptions is windows 64 bit, we make a
+   * special workaround for it.
+   */
+#if defined (WIN32) && defined (_M_X64)
+  unsigned _int64 off; /* workaround for missing POSIX compliance */
+#else
+  unsigned long off;
+#endif
+  unsigned int hval;
+  int lit;
+
+  if (!in_len || !out_len)
+    return 0;
+
+#if INIT_HTAB
+  memset (htab, 0, sizeof (htab));
+# if 0
+  for (hslot = htab; hslot < htab + HSIZE; hslot++)
+    *hslot++ = ip;
+# endif
+#endif
+
+  lit = 0; op++; /* start run */
+
+  hval = FRST (ip);
+  while (ip < in_end - 2)
+    {
+      hval = NEXT (hval, ip);
+      hslot = htab + IDX (hval);
+      ref = *hslot; *hslot = ip;
+
+      if (1
+#if INIT_HTAB
+          && ref < ip /* the next test will actually take care of this, but this is faster */
+#endif
+          && (off = ip - ref - 1) < MAX_OFF
+          && ip + 4 < in_end
+          && ref > (const u8 *)in_data
+#if STRICT_ALIGN
+          && ref[0] == ip[0]
+          && ref[1] == ip[1]
+          && ref[2] == ip[2]
+#else
+          && *(const u16 *)ref == *(const u16 *)ip
+          && ref[2] == ip[2]
+#endif
+        )
+        {
+          /* match found at *ref++ */
+          unsigned int len = 2;
+          unsigned int maxlen = in_end - ip - len;
+          maxlen = maxlen > MAX_REF ? MAX_REF : maxlen;
+
+          if (expect_false (op + 3 + 1 >= out_end)) /* first a faster conservative test */
+            if (op - !lit + 3 + 1 >= out_end) /* second the exact but rare test */
+              return 0;
+
+          op [- lit - 1] = lit - 1; /* stop run */
+          op -= !lit; /* undo run if length is zero */
+
+          for (;;)
+            {
+              if (expect_true (maxlen > 16))
+                {
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                  len++; if (ref [len] != ip [len]) break;
+                }
+
+              do
+                len++;
+              while (len < maxlen && ref[len] == ip[len]);
+
+              break;
+            }
+
+          len -= 2; /* len is now #octets - 1 */
+          ip++;
+
+          if (len < 7)
+            {
+              *op++ = (off >> 8) + (len << 5);
+            }
+          else
+            {
+              *op++ = (off >> 8) + (  7 << 5);
+              *op++ = len - 7;
+            }
+
+          *op++ = off;
+          lit = 0; op++; /* start run */
+
+          ip += len + 1;
+
+          if (expect_false (ip >= in_end - 2))
+            break;
+
+#if ULTRA_FAST || VERY_FAST
+          --ip;
+# if VERY_FAST && !ULTRA_FAST
+          --ip;
+# endif
+          hval = FRST (ip);
+
+          hval = NEXT (hval, ip);
+          htab[IDX (hval)] = ip;
+          ip++;
+
+# if VERY_FAST && !ULTRA_FAST
+          hval = NEXT (hval, ip);
+          htab[IDX (hval)] = ip;
+          ip++;
+# endif
+#else
+          ip -= len + 1;
+
+          do
+            {
+              hval = NEXT (hval, ip);
+              htab[IDX (hval)] = ip;
+              ip++;
+            }
+          while (len--);
+#endif
+        }
+      else
+        {
+          /* one more literal byte we must copy */
+          if (expect_false (op >= out_end))
+            return 0;
+
+          lit++; *op++ = *ip++;
+
+          if (expect_false (lit == MAX_LIT))
+            {
+              op [- lit - 1] = lit - 1; /* stop run */
+              lit = 0; op++; /* start run */
+            }
+        }
+    }
+
+  if (op + 3 > out_end) /* at most 3 bytes can be missing here */
+    return 0;
+
+  while (ip < in_end)
+    {
+      lit++; *op++ = *ip++;
+
+      if (expect_false (lit == MAX_LIT))
+        {
+          op [- lit - 1] = lit - 1; /* stop run */
+          lit = 0; op++; /* start run */
+        }
+    }
+
+  op [- lit - 1] = lit - 1; /* end run */
+  op -= !lit; /* undo run if length is zero */
+
+  return op - (u8 *)out_data;
+}
+
+#if AVOID_ERRNO
+# define SET_ERRNO(n)
+#else
+# include <errno.h>
+# define SET_ERRNO(n) errno = (n)
+#endif
+
+#if (__i386 || __amd64) && __GNUC__ >= 3
+# define lzf_movsb(dst, src, len)                \
+   asm ("rep movsb"                              \
+        : "=D" (dst), "=S" (src), "=c" (len)     \
+        :  "0" (dst),  "1" (src),  "2" (len));
+#endif
+
+unsigned int 
+lzf_decompress (const void *const in_data,  unsigned int in_len,
+                void             *out_data, unsigned int out_len)
+{
+  u8 const *ip = (const u8 *)in_data;
+  u8       *op = (u8 *)out_data;
+  u8 const *const in_end  = ip + in_len;
+  u8       *const out_end = op + out_len;
+
+  do
+    {
+      unsigned int ctrl = *ip++;
+
+      if (ctrl < (1 << 5)) /* literal run */
+        {
+          ctrl++;
+
+          if (op + ctrl > out_end)
+            {
+              SET_ERRNO (E2BIG);
+              return 0;
+            }
+
+#if CHECK_INPUT
+          if (ip + ctrl > in_end)
+            {
+              SET_ERRNO (EINVAL);
+              return 0;
+            }
+#endif
+
+#ifdef lzf_movsb
+          lzf_movsb (op, ip, ctrl);
+#else
+          do
+            *op++ = *ip++;
+          while (--ctrl);
+#endif
+        }
+      else /* back reference */
+        {
+          unsigned int len = ctrl >> 5;
+
+          u8 *ref = op - ((ctrl & 0x1f) << 8) - 1;
+
+#if CHECK_INPUT
+          if (ip >= in_end)
+            {
+              SET_ERRNO (EINVAL);
+              return 0;
+            }
+#endif
+          if (len == 7)
+            {
+              len += *ip++;
+#if CHECK_INPUT
+              if (ip >= in_end)
+                {
+                  SET_ERRNO (EINVAL);
+                  return 0;
+                }
+#endif
+            }
+
+          ref -= *ip++;
+
+          if (op + len + 2 > out_end)
+            {
+              SET_ERRNO (E2BIG);
+              return 0;
+            }
+
+          if (ref < (u8 *)out_data)
+            {
+              SET_ERRNO (EINVAL);
+              return 0;
+            }
+
+#ifdef lzf_movsb
+          len += 2;
+          lzf_movsb (op, ref, len);
+#else
+          *op++ = *ref++;
+          *op++ = *ref++;
+
+          do
+            *op++ = *ref++;
+          while (--len);
+#endif
+        }
+    }
+  while (ip < in_end);
+
+  return op - (u8 *)out_data;
+}
+
diff --git a/sbin/hastd/lzf.h b/sbin/hastd/lzf.h
new file mode 100644
index 0000000..92fdd0f
--- /dev/null
+++ b/sbin/hastd/lzf.h
@@ -0,0 +1,211 @@
+/*
+ * Copyright (c) 2000-2008 Marc Alexander Lehmann <schmorp@schmorp.de>
+ * 
+ * Redistribution and use in source and binary forms, with or without modifica-
+ * tion, are permitted provided that the following conditions are met:
+ * 
+ *   1.  Redistributions of source code must retain the above copyright notice,
+ *       this list of conditions and the following disclaimer.
+ * 
+ *   2.  Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ * 
+ * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
+ * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
+ * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO
+ * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
+ * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
+ * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
+ * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
+ * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
+ * OF THE POSSIBILITY OF SUCH DAMAGE.
+ *
+ * Alternatively, the contents of this file may be used under the terms of
+ * the GNU General Public License ("GPL") version 2 or any later version,
+ * in which case the provisions of the GPL are applicable instead of
+ * the above. If you wish to allow the use of your version of this file
+ * only under the terms of the GPL and not to allow others to use your
+ * version of this file under the BSD license, indicate your decision
+ * by deleting the provisions above and replace them with the notice
+ * and other provisions required by the GPL. If you do not delete the
+ * provisions above, a recipient may use your version of this file under
+ * either the BSD or the GPL.
+ */
+
+#ifndef LZF_H
+#define LZF_H
+
+/***********************************************************************
+**
+**	lzf -- an extremely fast/free compression/decompression-method
+**	http://liblzf.plan9.de/
+**
+**	This algorithm is believed to be patent-free.
+**
+***********************************************************************/
+
+#define LZF_VERSION 0x0105 /* 1.5, API version */
+
+/*
+ * Compress in_len bytes stored at the memory block starting at
+ * in_data and write the result to out_data, up to a maximum length
+ * of out_len bytes.
+ *
+ * If the output buffer is not large enough or any error occurs return 0,
+ * otherwise return the number of bytes used, which might be considerably
+ * more than in_len (but less than 104% of the original size), so it
+ * makes sense to always use out_len == in_len - 1), to ensure _some_
+ * compression, and store the data uncompressed otherwise (with a flag, of
+ * course.
+ *
+ * lzf_compress might use different algorithms on different systems and
+ * even different runs, thus might result in different compressed strings
+ * depending on the phase of the moon or similar factors. However, all
+ * these strings are architecture-independent and will result in the
+ * original data when decompressed using lzf_decompress.
+ *
+ * The buffers must not be overlapping.
+ *
+ * If the option LZF_STATE_ARG is enabled, an extra argument must be
+ * supplied which is not reflected in this header file. Refer to lzfP.h
+ * and lzf_c.c.
+ *
+ */
+unsigned int 
+lzf_compress (const void *const in_data,  unsigned int in_len,
+              void             *out_data, unsigned int out_len);
+
+/*
+ * Decompress data compressed with some version of the lzf_compress
+ * function and stored at location in_data and length in_len. The result
+ * will be stored at out_data up to a maximum of out_len characters.
+ *
+ * If the output buffer is not large enough to hold the decompressed
+ * data, a 0 is returned and errno is set to E2BIG. Otherwise the number
+ * of decompressed bytes (i.e. the original length of the data) is
+ * returned.
+ *
+ * If an error in the compressed data is detected, a zero is returned and
+ * errno is set to EINVAL.
+ *
+ * This function is very fast, about as fast as a copying loop.
+ */
+unsigned int 
+lzf_decompress (const void *const in_data,  unsigned int in_len,
+                void             *out_data, unsigned int out_len);
+
+/*
+ * Size of hashtable is (1 << HLOG) * sizeof (char *)
+ * decompression is independent of the hash table size
+ * the difference between 15 and 14 is very small
+ * for small blocks (and 14 is usually a bit faster).
+ * For a low-memory/faster configuration, use HLOG == 13;
+ * For best compression, use 15 or 16 (or more, up to 23).
+ */
+#ifndef HLOG
+# define HLOG 16
+#endif
+
+/*
+ * Sacrifice very little compression quality in favour of compression speed.
+ * This gives almost the same compression as the default code, and is
+ * (very roughly) 15% faster. This is the preferred mode of operation.
+ */
+#ifndef VERY_FAST
+# define VERY_FAST 1
+#endif
+
+/*
+ * Sacrifice some more compression quality in favour of compression speed.
+ * (roughly 1-2% worse compression for large blocks and
+ * 9-10% for small, redundant, blocks and >>20% better speed in both cases)
+ * In short: when in need for speed, enable this for binary data,
+ * possibly disable this for text data.
+ */
+#ifndef ULTRA_FAST
+# define ULTRA_FAST 0
+#endif
+
+/*
+ * Unconditionally aligning does not cost very much, so do it if unsure
+ */
+#ifndef STRICT_ALIGN
+# define STRICT_ALIGN !(defined(__i386) || defined (__amd64))
+#endif
+
+/*
+ * You may choose to pre-set the hash table (might be faster on some
+ * modern cpus and large (>>64k) blocks, and also makes compression
+ * deterministic/repeatable when the configuration otherwise is the same).
+ */
+#ifndef INIT_HTAB
+# define INIT_HTAB 1
+#endif
+
+/*
+ * Avoid assigning values to errno variable? for some embedding purposes
+ * (linux kernel for example), this is neccessary. NOTE: this breaks
+ * the documentation in lzf.h.
+ */
+#ifndef AVOID_ERRNO
+# define AVOID_ERRNO 0
+#endif
+
+/*
+ * Wether to pass the LZF_STATE variable as argument, or allocate it
+ * on the stack. For small-stack environments, define this to 1.
+ * NOTE: this breaks the prototype in lzf.h.
+ */
+#ifndef LZF_STATE_ARG
+# define LZF_STATE_ARG 0
+#endif
+
+/*
+ * Wether to add extra checks for input validity in lzf_decompress
+ * and return EINVAL if the input stream has been corrupted. This
+ * only shields against overflowing the input buffer and will not
+ * detect most corrupted streams.
+ * This check is not normally noticable on modern hardware
+ * (<1% slowdown), but might slow down older cpus considerably.
+ */
+#ifndef CHECK_INPUT
+# define CHECK_INPUT 1
+#endif
+
+/*****************************************************************************/
+/* nothing should be changed below */
+
+typedef unsigned char u8;
+
+typedef const u8 *LZF_STATE[1 << (HLOG)];
+
+#if !STRICT_ALIGN
+/* for unaligned accesses we need a 16 bit datatype. */
+# include <limits.h>
+# if USHRT_MAX == 65535
+    typedef unsigned short u16;
+# elif UINT_MAX == 65535
+    typedef unsigned int u16;
+# else
+#  undef STRICT_ALIGN
+#  define STRICT_ALIGN 1
+# endif
+#endif
+
+#if ULTRA_FAST
+# if defined(VERY_FAST)
+#  undef VERY_FAST
+# endif
+#endif
+
+#if INIT_HTAB
+# ifdef __cplusplus
+#  include <cstring>
+# else
+#  include <string.h>
+# endif
+#endif
+
+#endif
diff --git a/sbin/hastd/parse.y b/sbin/hastd/parse.y
index c5a39b7..be2cc1e 100644
--- a/sbin/hastd/parse.y
+++ b/sbin/hastd/parse.y
@@ -62,6 +62,7 @@ static char depth0_control[HAST_ADDRSIZE];
 static char depth0_listen[HAST_ADDRSIZE];
 static int depth0_replication;
 static int depth0_checksum;
+static int depth0_compression;
 static int depth0_timeout;
 static char depth0_exec[PATH_MAX];
 
@@ -170,6 +171,7 @@ yy_config_parse(const char *config, bool exitonerror)
 	depth0_timeout = HAST_TIMEOUT;
 	depth0_replication = HAST_REPLICATION_MEMSYNC;
 	depth0_checksum = HAST_CHECKSUM_NONE;
+	depth0_compression = HAST_COMPRESSION_HOLE;
 	strlcpy(depth0_control, HAST_CONTROL, sizeof(depth0_control));
 	strlcpy(depth0_listen, HASTD_LISTEN, sizeof(depth0_listen));
 	depth0_exec[0] = '\0';
@@ -233,6 +235,13 @@ yy_config_parse(const char *config, bool exitonerror)
 			 */
 			curres->hr_checksum = depth0_checksum;
 		}
+		if (curres->hr_compression == -1) {
+			/*
+			 * Compression is not set at resource-level.
+			 * Use global or default setting.
+			 */
+			curres->hr_compression = depth0_compression;
+		}
 		if (curres->hr_timeout == -1) {
 			/*
 			 * Timeout is not set at resource-level.
@@ -266,13 +275,14 @@ yy_config_free(struct hastd_config *config)
 }
 %}
 
-%token CONTROL LISTEN PORT REPLICATION CHECKSUM
+%token CONTROL LISTEN PORT REPLICATION CHECKSUM COMPRESSION
 %token TIMEOUT EXEC EXTENTSIZE RESOURCE NAME LOCAL REMOTE ON
-%token FULLSYNC MEMSYNC ASYNC NONE CRC32 SHA256
+%token FULLSYNC MEMSYNC ASYNC NONE CRC32 SHA256 HOLE LZF
 %token NUM STR OB CB
 
 %type <num> replication_type
 %type <num> checksum_type
+%type <num> compression_type
 
 %union
 {
@@ -299,6 +309,8 @@ statement:
 	|
 	checksum_statement
 	|
+	compression_statement
+	|
 	timeout_statement
 	|
 	exec_statement
@@ -416,6 +428,30 @@ checksum_type:
 	SHA256		{ $$ = HAST_CHECKSUM_SHA256; }
 	;
 
+compression_statement:	COMPRESSION compression_type
+	{
+		switch (depth) {
+		case 0:
+			depth0_compression = $2;
+			break;
+		case 1:
+			if (curres != NULL)
+				curres->hr_compression = $2;
+			break;
+		default:
+			assert(!"compression at wrong depth level");
+		}
+	}
+	;
+
+compression_type:
+	NONE		{ $$ = HAST_COMPRESSION_NONE; }
+	|
+	HOLE		{ $$ = HAST_COMPRESSION_HOLE; }
+	|
+	LZF		{ $$ = HAST_COMPRESSION_LZF; }
+	;
+
 timeout_statement:	TIMEOUT NUM
 	{
 		switch (depth) {
@@ -609,6 +645,7 @@ resource_start:	STR
 		curres->hr_previous_role = HAST_ROLE_INIT;
 		curres->hr_replication = -1;
 		curres->hr_checksum = -1;
+		curres->hr_compression = -1;
 		curres->hr_timeout = -1;
 		curres->hr_exec[0] = '\0';
 		curres->hr_provname[0] = '\0';
@@ -629,6 +666,8 @@ resource_entry:
 	|
 	checksum_statement
 	|
+	compression_statement
+	|
 	timeout_statement
 	|
 	exec_statement
diff --git a/sbin/hastd/primary.c b/sbin/hastd/primary.c
index aa4d103..5691c0d 100644
--- a/sbin/hastd/primary.c
+++ b/sbin/hastd/primary.c
@@ -1910,6 +1910,7 @@ primary_config_reload(struct hast_resource *res, struct nv *nv)
 	nv_assert(nv, "remoteaddr");
 	nv_assert(nv, "replication");
 	nv_assert(nv, "checksum");
+	nv_assert(nv, "compression");
 	nv_assert(nv, "timeout");
 	nv_assert(nv, "exec");
 
@@ -1918,6 +1919,7 @@ primary_config_reload(struct hast_resource *res, struct nv *nv)
 #define MODIFIED_REMOTEADDR	0x01
 #define MODIFIED_REPLICATION	0x02
 #define MODIFIED_CHECKSUM	0x04
+#define MODIFIED_COMPRESSION	0x08
 #define MODIFIED_TIMEOUT	0x10
 #define MODIFIED_EXEC		0x20
 	modified = 0;
@@ -1941,6 +1943,11 @@ primary_config_reload(struct hast_resource *res, struct nv *nv)
 		gres->hr_checksum = vint;
 		modified |= MODIFIED_CHECKSUM;
 	}
+	vint = nv_get_int32(nv, "compression");
+	if (gres->hr_compression != vint) {
+		gres->hr_compression = vint;
+		modified |= MODIFIED_COMPRESSION;
+	}
 	vint = nv_get_int32(nv, "timeout");
 	if (gres->hr_timeout != vint) {
 		gres->hr_timeout = vint;
@@ -1994,6 +2001,7 @@ primary_config_reload(struct hast_resource *res, struct nv *nv)
 #undef	MODIFIED_REMOTEADDR
 #undef	MODIFIED_REPLICATION
 #undef	MODIFIED_CHECKSUM
+#undef	MODIFIED_COMPRESSION
 #undef	MODIFIED_TIMEOUT
 #undef	MODIFIED_EXEC
 
diff --git a/sbin/hastd/token.l b/sbin/hastd/token.l
index c42c130..1992b77 100644
--- a/sbin/hastd/token.l
+++ b/sbin/hastd/token.l
@@ -50,6 +50,7 @@ listen			{ DP; return LISTEN; }
 port			{ DP; return PORT; }
 replication		{ DP; return REPLICATION; }
 checksum		{ DP; return CHECKSUM; }
+compression		{ DP; return COMPRESSION; }
 timeout			{ DP; return TIMEOUT; }
 exec			{ DP; return EXEC; }
 resource		{ DP; return RESOURCE; }
@@ -63,6 +64,8 @@ async			{ DP; return ASYNC; }
 none			{ DP; return NONE; }
 crc32			{ DP; return CRC32; }
 sha256			{ DP; return SHA256; }
+hole			{ DP; return HOLE; }
+lzf			{ DP; return LZF; }
 [0-9]+			{ DP; yylval.num = atoi(yytext); return NUM; }
 [a-zA-Z0-9\.\-_/\:]+	{ DP; yylval.str = strdup(yytext); return STR; }
 \{			{ DP; depth++; return OB; }