[Sheepdog] [PATCH 1/3] collie: write node list log when epoch is incremented

[MORITA Kazutaka]

The past node lists are used to look up objects efficiently

Signed-off-by: MORITA Kazutaka <morita.kazutaka at lab.ntt.co.jp>
---
 collie/collie.h |    6 +++++
 collie/group.c  |   13 ++++++++++++
 collie/store.c  |   56 +++++++++++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 75 insertions(+), 0 deletions(-)

diff --git a/collie/collie.h b/collie/collie.h
index 3b2d871..f5dc30d 100644
--- a/collie/collie.h
+++ b/collie/collie.h
@@ -74,6 +74,12 @@ int create_listen_port(int port, void *data);
 
 int init_store(char *dir);
 
+int read_node_list(uint32_t epoch, struct sheepdog_node_list_entry *entries,
+		   int nr_nodes);
+
+int write_node_list(uint32_t epoch, struct sheepdog_node_list_entry *entries,
+		    int nr_nodes);
+
 int add_vdi(struct cluster_info *cluster,
 	    char *name, int len, uint64_t size, uint64_t * added_oid,
 	    uint64_t base_oid, uint32_t tag);
diff --git a/collie/group.c b/collie/group.c
index ced0a5e..01f959e 100644
--- a/collie/group.c
+++ b/collie/group.c
@@ -246,6 +246,8 @@ static void print_node_list(struct cluster_info *ci)
 static void add_node(struct cluster_info *ci, uint32_t nodeid, uint32_t pid,
 		     struct sheepdog_node_list_entry *sd_ent)
 {
+	int nr_nodes;
+	struct sheepdog_node_list_entry entries[SD_MAX_NODES];
 	struct node *node;
 
 	node = zalloc(sizeof(*node));
@@ -257,6 +259,10 @@ static void add_node(struct cluster_info *ci, uint32_t nodeid, uint32_t pid,
 	node->pid = pid;
 	node->ent = *sd_ent;
 	list_add_tail(&node->list, &ci->node_list);
+
+	nr_nodes = build_node_list(&ci->node_list, entries);
+	write_node_list(ci->epoch + 1, entries, nr_nodes);
+
 	ci->epoch++;
 }
 
@@ -528,12 +534,19 @@ static void __sd_confch(struct work *work, int idx)
 
 	for (i = 0; i < left_list_entries; i++) {
 		list_for_each_entry_safe(node, e, &ci->node_list, list) {
+			int nr_nodes;
+			struct sheepdog_node_list_entry entries[SD_MAX_NODES];
+
 			if (node->nodeid != left_list[i].nodeid ||
 			    node->pid != left_list[i].pid)
 				continue;
 
 			list_del(&node->list);
 			free(node);
+
+			nr_nodes = build_node_list(&ci->node_list, entries);
+			write_node_list(ci->epoch + 1, entries, nr_nodes);
+
 			ci->epoch++;
 		}
 	}
diff --git a/collie/store.c b/collie/store.c
index 4e95469..aadddc5 100644
--- a/collie/store.c
+++ b/collie/store.c
@@ -622,6 +622,7 @@ int init_store(char *dir)
 	struct mntent *mnt;
 	struct stat s, ms;
 	FILE *fp;
+	char path[PATH_MAX];
 
 	ret = stat(dir, &s);
 	if (ret) {
@@ -650,6 +651,13 @@ int init_store(char *dir)
 
 	obj_dir = dir;
 
+	strcpy(path, dir);
+	strcat(path, "/epoch");
+	ret = mkdir(path, S_IRUSR | S_IWUSR | S_IXUSR | S_IRGRP |
+		    S_IWGRP | S_IXGRP);
+	if (ret && errno != EEXIST)
+		return 1;
+
 	fp = setmntent(MOUNTED, "r");
 	if (!fp)
 		return 1;
@@ -673,3 +681,51 @@ int init_store(char *dir)
 
 	return ret;
 }
+
+int read_node_list(uint32_t epoch, struct sheepdog_node_list_entry *entries,
+		    int nr_nodes)
+{
+	int ret;
+	int fd;
+	char path[PATH_MAX];
+
+	if (epoch <= 0)
+		return -1;
+
+	snprintf(path, sizeof(path), "%s/epoch/%x", obj_dir, epoch);
+	fd = open(path, O_RDONLY, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
+	if (!fd)
+		return -1;
+
+	ret = pread64(fd, entries,
+		      nr_nodes * sizeof(struct sheepdog_node_list_entry), 0);
+
+	close(fd);
+
+	return (ret / sizeof(struct sheepdog_node_list_entry));
+}
+
+int write_node_list(uint32_t epoch, struct sheepdog_node_list_entry *entries,
+		    int nr_nodes)
+{
+	int ret;
+	int fd;
+	char path[PATH_MAX];
+
+	if (epoch <= 0)
+		return -1;
+
+	snprintf(path, sizeof(path), "%s/epoch/%x", obj_dir, epoch);
+	fd = open(path, O_WRONLY | O_CREAT, S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP);
+	if (!fd)
+		return -1;
+
+	ret = pwrite64(fd, entries,
+		       nr_nodes * sizeof(struct sheepdog_node_list_entry), 0);
+	if (ret != nr_nodes * sizeof(struct sheepdog_node_list_entry))
+		return -1;
+
+	close(fd);
+
+	return nr_nodes;
+}
-- 
1.6.5