Merge branch 'cjd3000-dog-force-alter-copy'

INSTALL

@@ -12,11 +12,11 @@ Run-time dependencies
 Compile-time dependencies
 -------------------------
 * GNU Autotools
-* corosync devel package
-* liburcu devel package
-* git
+* corosync devel package (needed for building cluster using corosync)
+* liburcu (userspace rcu) devel package
 * optional:fuse-devel (for sheepfs)
-* optional:libzookeeper-mt-dev (for zookeeper support)
+* optional:libzookeeper-mt-dev (needed for building cluster using zookeeper)
+* optional:curl, fcgi (for http request service)
 
 Installing from source
 ------------------------------------

MAINTAINERS

@@ -76,3 +76,9 @@ Debian related stuff
 M: Alexander Guy <alexander@andern.org>
 F: debian/*
 L: sheepdog@lists.wpkg.org
+
+Intel Intelligent Storage Acceleration Library
+------------------------------
+M: hongzhou zhang <hongzhou.h.zhang@intel.com>
+F: lib/isa-l/*
+L: sheepdog@lists.wpkg.org

configure.ac

@@ -88,6 +88,25 @@ AC_CHECK_PROGS([GROFF], [groff])
 
 AM_MISSING_PROG(AUTOM4TE, autom4te, $missing_dir)
 
+# Check for programs
+if [[[ $host = *x86_64* ]]]; then
+  AC_CHECK_PROG(HAVE_YASM, yasm, yes, no)
+  if test "$HAVE_YASM" = "no"; then
+    AC_MSG_ERROR([yasm not found as required.])
+  fi
+  AC_MSG_CHECKING([checking yasm version])
+  AC_LANG_CONFTEST([AC_LANG_SOURCE([[vextracti128 xmm0, ymm0, 0;]])])
+  if yasm -f elf64 -i lib/isa-l/include lib/isa-l/erasure_code/gf_vect_dot_prod_avx2.asm -o /dev/null 2> /dev/null ; then
+    yasm_avx2_support=yes
+    AC_DEFINE_UNQUOTED([ENABLE_ISAL], 1, [enable isal])
+    AC_MSG_RESULT([yes])
+  else
+    yasm_avx2_support=no
+    AC_MSG_NOTICE([avx2 requires yasm 1.2.0 or higher])
+  fi
+fi
+AM_CONDITIONAL(YASM_AVX2_SUPPORT, test "$yasm_avx2_support" = "yes")
+
 # Checks for libraries.
 AC_CHECK_LIB([socket], [socket])
 AC_CHECK_LIB([rt], [clock_gettime], , AC_MSG_ERROR(librt not found))
@@ -250,27 +269,27 @@ AC_ARG_ENABLE([sheepfs],
 AM_CONDITIONAL(BUILD_SHEEPFS, test x$enable_sheepfs = xyes)
 
 AC_ARG_ENABLE([http],
-	[ --enable-http : enable http request service (default no) ],,
+	[  --enable-http            : enable http request service (default no) ],,
 	[ enable_http="no" ],)
 AM_CONDITIONAL(BUILD_HTTP, test x$enable_http = xyes)
 
 AC_ARG_ENABLE([nfs],
-	[ --enable-nfs : enable nfs server service (default no) ],,
+	[  --enable-nfs             : enable nfs server service (default no) ],,
 	[ enable_nfs="no" ],)
 AM_CONDITIONAL(BUILD_NFS, test x$enable_nfs = xyes)
 
 AC_ARG_ENABLE([diskvnodes],
-	[ --enable-diskvnodes : enable disk as vnodes (default no) ],,
+	[  --enable-diskvnodes      : enable disk as vnodes (default no) ],,
 	[ enable_diskvnodes="no" ],)
 AM_CONDITIONAL(BUILD_DISKVNODES, test x$enable_diskvnodes = xyes)
 
 AC_ARG_ENABLE([earthquake],
-	[ --enable-earthquake : enable earthquake debugger (default no) ],,
+	[  --enable-earthquake      : enable earthquake debugger (default no) ],,
 	[ enable_earthquake="no" ],)
 AM_CONDITIONAL(BUILD_EARTHQUAKE, test x$enable_earthquake = xyes)
 
 AC_ARG_ENABLE([lttng-ust],
-	[ --enable-lttng-ust : enable tracing feature based on LTTng UST (default no) ],,
+	[  --enable-lttng-ust       : enable tracing feature based on LTTng UST (default no) ],,
 	[ enable_lttng_ust="no" ],)
 AM_CONDITIONAL(BUILD_LTTNG_UST, test x$enable_lttng_ust = xyes)
 
@@ -509,7 +528,7 @@ fi
 if test "x${enable_trace}" = xyes && \
 		cc_supports_flag -pg ; then
 	AC_MSG_NOTICE([Enabling trace (-pg)])
-	TRACE_CFLAGS="-pg"
+	TRACE_CFLAGS="-pg -nopie"
 else
 	TRACE_CFLAGS=""
 fi

dog/cluster.c

@@ -833,7 +833,7 @@ static int cluster_alter_copy(int argc, char **argv)
 			   "use %d as default.\n", SD_DEFAULT_COPIES);
 	}
 
-	if (cluster_cmd_data.copies > sd_nodes_nr) {
+	if (!cluster_cmd_data.force && (cluster_cmd_data.copies > sd_nodes_nr)) {
 		char info[1024];
 		snprintf(info, sizeof(info), "Number of copies (%d) is larger "
 			 "than number of nodes (%d).\n"
@@ -865,7 +865,8 @@ static int cluster_alter_copy(int argc, char **argv)
 		goto failure;
 	}
 
-	confirm(ALTER_CLUSTER_COPY_PRINT);
+	if (!cluster_cmd_data.force)
+		confirm(ALTER_CLUSTER_COPY_PRINT);
 
 	sd_init_req(&hdr, SD_OP_ALTER_CLUSTER_COPY);
 	hdr.cluster.copies = cluster_cmd_data.copies;
@@ -908,7 +909,7 @@ static struct subcommand cluster_cmd[] = {
 	 cluster_reweight, cluster_options},
 	{"check", NULL, "aphT", "check and repair cluster", NULL,
 	 CMD_NEED_ROOT|CMD_NEED_NODELIST, cluster_check, cluster_options},
-	{"alter-copy", NULL, "aphTc", "set the cluster's redundancy level",
+	{"alter-copy", NULL, "aphTcf", "set the cluster's redundancy level",
 	 NULL, CMD_NEED_ROOT|CMD_NEED_NODELIST, cluster_alter_copy, cluster_options},
 	{NULL,},
 };

dog/dog.h

@@ -18,6 +18,8 @@
 #include <unistd.h>
 #include <sys/ioctl.h>
 #include <sys/types.h>
+#include <sys/stat.h>
+#include <fcntl.h>
 
 #include "sheepdog_proto.h"
 #include "sheep.h"

dog/node.c

@@ -829,6 +829,67 @@ static int node_vnodes(int argc, char **argv)
 	return do_generic_subcommand(node_vnodes_cmd, argc, argv);
 }
 
+static int node_format(int argc, char **argv)
+{
+	char *dir = argv[optind++], *store_name;
+	char config_path[PATH_MAX];
+	struct stat st;
+	int ret, fd;
+	struct node_info info;
+
+	/*
+	 * TODO: for more safe operation, this command should acquire lock which
+	 * is shared with sheep process
+	 */
+
+	if (optind == argc) {
+		sd_err("specify store format");
+		return EXIT_USAGE;
+	}
+	store_name = argv[optind];
+
+	if (strcmp(store_name, "plain") && strcmp(store_name, "tree")) {
+		/*
+		 * FIXME: store names should be macro defined in somewhere
+		 * suitable
+		 */
+		sd_err("expected store format: plain, tree");
+		return EXIT_SYSFAIL;
+	}
+
+	memset(config_path, 0, sizeof(config_path));
+	snprintf(config_path, PATH_MAX, "%s/" NODE_CONFIG_PATH, dir);
+
+	ret = stat(config_path, &st);
+	if (ret < 0) {
+		if (errno != ENOENT) {
+			sd_err("failed to check existence of config (%s): %m",
+			       config_path);
+			return EXIT_SYSFAIL;
+		}
+	} else {
+		/* TODO: dialogue and force option? */
+		sd_err("config file already exists: %s", config_path);
+		return EXIT_SYSFAIL;
+	}
+
+	fd = open(config_path, O_CREAT | O_WRONLY, S_IRUSR | S_IWUSR);
+	if (fd < 0) {
+		sd_err("failed to create a config file (%s): %m", config_path);
+		return EXIT_SYSFAIL;
+	}
+
+	memset(&info, 0, sizeof(info));
+	memcpy(info.store, store_name, STORE_LEN);
+	ret = xwrite(fd, &info, sizeof(info));
+	if (ret != sizeof(info)) {
+		sd_err("failed to write a config file (%s): %m", config_path);
+		return EXIT_SYSFAIL;
+	}
+
+	return 0;
+}
+
 static struct subcommand node_cmd[] = {
 	{"kill", "<node id>", "aprhlT", "kill node", NULL,
 	 CMD_NEED_ROOT|CMD_NEED_NODELIST, node_kill, node_options},
@@ -847,6 +908,9 @@ static struct subcommand node_cmd[] = {
 	 CMD_NEED_ROOT|CMD_NEED_ARG, node_log},
 	{"vnodes", "<num of vnodes>", "aph", "set new vnodes", node_vnodes_cmd,
 	 CMD_NEED_ROOT|CMD_NEED_ARG, node_vnodes},
+	{"format", "<directory of sheep> <a name of store format>",
+	 "aphT", "initialize store format of the node", NULL, CMD_NEED_ARG,
+	 node_format},
 	{NULL,},
 };
 

dog/vdi.c

@@ -2930,7 +2930,7 @@ static int vdi_alter_copy(int argc, char **argv)
 			   "use %d as default.\n", SD_DEFAULT_COPIES);
 	}
 
-	if (vdi_cmd_data.nr_copies > sd_nodes_nr) {
+	if (!vdi_cmd_data.force && (vdi_cmd_data.nr_copies > sd_nodes_nr)) {
 		char info[1024];
 		snprintf(info, sizeof(info), "Number of copies (%d) is larger "
 			 "than number of nodes (%d).\n"
@@ -2966,7 +2966,8 @@ static int vdi_alter_copy(int argc, char **argv)
 		return EXIT_FAILURE;
 	}
 
-	confirm(ALTER_VDI_COPY_PRINT);
+	if (!vdi_cmd_data.force)
+		confirm(ALTER_VDI_COPY_PRINT);
 
 	inode->nr_copies = vdi_cmd_data.nr_copies;
 	ret = dog_write_object(vid_to_vdi_oid(vid), 0, inode,
@@ -3179,7 +3180,7 @@ static struct subcommand vdi_cmd[] = {
 	 "Run 'dog vdi cache' for more information",
 	 vdi_cache_cmd, CMD_NEED_ROOT|CMD_NEED_ARG,
 	 vdi_cache, vdi_options},
-	{"alter-copy", "<vdiname>", "caphT", "set the vdi's redundancy level",
+	{"alter-copy", "<vdiname>", "caphTf", "set the vdi's redundancy level",
 	 NULL, CMD_NEED_ROOT|CMD_NEED_ARG|CMD_NEED_NODELIST, vdi_alter_copy, vdi_options},
 	{"lock", NULL, "aphT", "See 'dog vdi lock' for more information",
 	 vdi_lock_cmd, CMD_NEED_ROOT|CMD_NEED_ARG, vdi_lock, vdi_options},

include/fec.h

@@ -101,8 +101,10 @@ void fec_decode_buffer(struct fec *ctx, uint8_t *input[], const int in_idx[],
 
 /* for isa-l */
 
+#if defined __x86_64__ && defined(ENABLE_ISAL)
 void isa_decode_buffer(struct fec *ctx, uint8_t *input[], const int in_idx[],
 		       char *buf, int idx, uint32_t object_size);
+#endif
 
 /*
  * @param inpkts an array of packets (size k); If a primary block, i, is present
@@ -172,12 +174,15 @@ static inline void ec_encode(struct fec *ctx, const uint8_t *ds[],
 			     uint8_t *ps[])
 {
 	int p = ctx->dp - ctx->d;
+
+#if !defined __x86_64__ || !defined(ENABLE_ISAL)
 	int pidx[p];
 
 	for (int i = 0; i < p; i++)
 		pidx[i] = ctx->d + i;
+#endif
 
-#ifdef __x86_64__
+#if defined __x86_64__ && defined(ENABLE_ISAL)
 		ec_encode_data(SD_EC_DATA_STRIPE_SIZE / ctx->d, ctx->d, p,
 				   ctx->ec_tbl, (unsigned char **)ds, ps);
 #else
@@ -209,7 +214,7 @@ static inline void ec_decode_buffer(struct fec *ctx, uint8_t *input[],
 				    const int in_idx[], char *buf,
 				    int idx, uint32_t object_size)
 {
-#ifdef __x86_64__
+#if defined __x86_64__ && defined(ENABLE_ISAL)
 		isa_decode_buffer(ctx, input, in_idx, buf, idx, object_size);
 #else
 		fec_decode_buffer(ctx, input, in_idx, buf, idx, object_size);

include/internal_proto.h

@@ -239,6 +239,18 @@ struct oid_entry {
 	int last;             /* idx to the last element of the oid array */
 };
 
+/*
+ * node_info: node specific configuration
+ * This is a config data initialized with "dog node format".
+ * Currently, it is very simple so doesn't have different expressions
+ * like cluster_info and sheepdog_config.
+ */
+struct node_info {
+	uint8_t store[STORE_LEN];
+};
+
+#define NODE_CONFIG_PATH "node_config"
+
 /*
  * A joining sheep multicasts the local cluster info.  Then, the existing nodes
  * reply the latest cluster info which is unique among all of the nodes.
@@ -255,7 +267,7 @@ struct cluster_info {
 	enum sd_status status : 8;
 	uint8_t block_size_shift;
 	uint8_t __pad[3];
-	uint8_t store[STORE_LEN];
+	uint8_t default_store[STORE_LEN];
 
 	/* Node list at cluster_info->epoch */
 	struct sd_node nodes[SD_MAX_NODES];

include/sheep.h

@@ -357,7 +357,7 @@ struct sheepdog_config {
 	uint64_t ctime;
 	uint16_t flags;
 	uint8_t copies;
-	uint8_t store[STORE_LEN];
+	uint8_t default_store[STORE_LEN];
 	uint8_t shutdown;
 	uint8_t copy_policy;
 	uint8_t block_size_shift;

lib/Makefile.am

@@ -26,18 +26,27 @@ libsheepdog_a_CPPFLAGS  = $(AM_CPPFLAGS) -DNO_SHEEPDOG_LOGGER
 
 libsd_a_DEPENDENCIES =
 
-noinst_LIBRARIES	= libisa.a libsd.a
+noinst_LIBRARIES	=
 
+if YASM_AVX2_SUPPORT
+noinst_LIBRARIES	+= libisa.a
+endif
+
+noinst_LIBRARIES	+= libsd.a
+
+if YASM_AVX2_SUPPORT
 libisa_a_SOURCES	= $(shell find isa-l/ -type f -regex ".*\.\(c\|h\|asm\)") \
 			  isa-l/erasure_code/Makefile \
 			  isa-l/Makefile \
 			  isa-l/Makefile.nmake \
 			  isa-l/make.inc
+endif
 
 libsd_a_SOURCES		= event.c logger.c net.c util.c rbtree.c strbuf.c \
 			  sha1.c option.c work.c sockfd_cache.c fec.c \
 			  sd_inode.c common.c
 
+if YASM_AVX2_SUPPORT
 libsd_a_LIBADD_		= isa-l/bin/ec_base.o \
 			  isa-l/bin/ec_highlevel_func.o \
 			  isa-l/bin/ec_multibinary.o \
@@ -53,18 +62,21 @@ libsd_a_LIBADD_		= isa-l/bin/ec_base.o \
 			  isa-l/bin/gf_5vect_dot_prod_avx.o \
 			  isa-l/bin/gf_6vect_dot_prod_avx.o \
 			  isa-l/bin/gf_vect_dot_prod_avx.o \
+			  isa-l/bin/gf_vect_mul_avx.o \
+			  isa-l/bin/gf_vect_mul_sse.o \
 			  isa-l/bin/gf_2vect_dot_prod_avx2.o \
 		  	  isa-l/bin/gf_3vect_dot_prod_avx2.o \
 		  	  isa-l/bin/gf_4vect_dot_prod_avx2.o \
 			  isa-l/bin/gf_5vect_dot_prod_avx2.o \
 			  isa-l/bin/gf_6vect_dot_prod_avx2.o \
-			  isa-l/bin/gf_vect_dot_prod_avx2.o \
-			  isa-l/bin/gf_vect_mul_avx.o \
-			  isa-l/bin/gf_vect_mul_sse.o
+			  isa-l/bin/gf_vect_dot_prod_avx2.o
+endif
 
+if YASM_AVX2_SUPPORT
 libsd_a_LIBADD_32	= isa-l/bin/ec_base.o \
 			  isa-l/bin/ec_highlevel_func.o \
 			  isa-l/bin/ec_multibinary.o
+endif
 
 if !X86_64
 arch = 32

lib/fec.c

@@ -455,7 +455,7 @@ void fec_free(struct fec *p)
 	sd_assert(p != NULL && p->magic == (((FEC_MAGIC ^ p->d) ^ p->dp) ^
 					 (unsigned long) (p->enc_matrix)));
 	free(p->enc_matrix);
-#ifdef __x86_64__
+#if defined __x86_64__ && defined(ENABLE_ISAL)
 		free(p->ec_tbl);
 #endif
 	free(p);
@@ -498,7 +498,7 @@ struct fec *fec_new(unsigned short d, unsigned short dp)
 		*p = 1;
 	free(tmp_m);
 
-#ifdef __x86_64__
+#if defined __x86_64__ && defined(ENABLE_ISAL)
 		retval->ec_tbl = xmalloc(dp * d * 32);
 		ec_init_tables(d, dp - d, retval->enc_matrix + (d * d),
 			       retval->ec_tbl);
@@ -716,6 +716,7 @@ void fec_decode_buffer(struct fec *ctx, uint8_t *input[], const int in_idx[],
 	}
 }
 
+#if defined __x86_64__ && defined(ENABLE_ISAL)
 void isa_decode_buffer(struct fec *ctx, uint8_t *input[], const int in_idx[],
 		       char *buf, int idx, uint32_t object_size)
 {
@@ -743,3 +744,4 @@ void isa_decode_buffer(struct fec *ctx, uint8_t *input[], const int in_idx[],
 	ec_init_tables(ed, 1, cm, ec_tbl);
 	ec_encode_data(len, ed, 1, ec_tbl, input, lost);
 }
+#endif