[sheepdog] [PATCH v4 04/10] sheep: add vdi_copy tree to store copies number for every VDI
levin li
levin108 at gmail.com
Thu Aug 9 07:27:39 CEST 2012
From: levin li <xingke.lwp at taobao.com>
The normal requests from QEMU include the copies number in its header,
QEMU stores the nr_copies the first time it reads the inode data, but
many other local requests such as read_copy_from_replica and
recover_object_from_replica don't know the copies number for every object,
so this tree is necessary to keep the copies number
Signed-off-by: levin li <xingke.lwp at taobao.com>
---
sheep/ops.c | 4 +-
sheep/sheep_priv.h | 5 ++
sheep/vdi.c | 118 ++++++++++++++++++++++++++++++++++++++++++++++++++++
3 files changed, 126 insertions(+), 1 deletions(-)
diff --git a/sheep/ops.c b/sheep/ops.c
index 062a7b4..efaf979 100644
--- a/sheep/ops.c
+++ b/sheep/ops.c
@@ -141,8 +141,10 @@ static int post_cluster_new_vdi(const struct sd_req *req, struct sd_rsp *rsp,
int ret = rsp->result;
vprintf(SDOG_INFO, "done %d %ld\n", ret, nr);
- if (ret == SD_RES_SUCCESS)
+ if (ret == SD_RES_SUCCESS) {
set_bit(nr, sys->vdi_inuse);
+ add_vdi_copies(nr, rsp->vdi.copies);
+ }
return ret;
}
diff --git a/sheep/sheep_priv.h b/sheep/sheep_priv.h
index 8aeed83..86c4c56 100644
--- a/sheep/sheep_priv.h
+++ b/sheep/sheep_priv.h
@@ -200,6 +200,11 @@ int create_listen_port(int port, void *data);
int init_store(const char *dir, int enable_write_cache);
int init_base_path(const char *dir);
+int get_vdi_copy_number(uint32_t vid);
+int get_obj_copy_number(uint64_t oid);
+int get_max_copy_number(void);
+int get_req_copy_number(struct request *req);
+int add_vdi_copies(uint32_t vid, int nr_copies);
int vdi_exist(uint32_t vid);
int add_vdi(struct vdi_iocb *iocb, uint32_t *new_vid);
diff --git a/sheep/vdi.c b/sheep/vdi.c
index 48aa9e6..25c4449 100644
--- a/sheep/vdi.c
+++ b/sheep/vdi.c
@@ -10,11 +10,129 @@
*/
#include <stdio.h>
#include <stdlib.h>
+#include <pthread.h>
#include <sys/time.h>
#include "sheepdog_proto.h"
#include "sheep_priv.h"
+struct vdi_copy_entry {
+ uint32_t vid;
+ unsigned int nr_copies;
+ struct rb_node node;
+};
+
+uint32_t max_copies = 0;
+static struct rb_root vdi_copy_root = RB_ROOT;
+pthread_rwlock_t vdi_copy_lock = PTHREAD_RWLOCK_INITIALIZER;
+
+static struct vdi_copy_entry *vdi_copy_search(struct rb_root *root,
+ uint32_t vid)
+{
+ struct rb_node *n = root->rb_node;
+ struct vdi_copy_entry *t;
+
+ while (n) {
+ t = rb_entry(n, struct vdi_copy_entry, node);
+
+ if (vid < t->vid)
+ n = n->rb_left;
+ else if (vid > t->vid)
+ n = n->rb_right;
+ else
+ return t;
+ }
+
+ return NULL;
+}
+
+static struct vdi_copy_entry *vdi_copy_insert(struct rb_root *root,
+ struct vdi_copy_entry *new)
+{
+ struct rb_node **p = &root->rb_node;
+ struct rb_node *parent = NULL;
+ struct vdi_copy_entry *entry;
+
+ while (*p) {
+ parent = *p;
+ entry = rb_entry(parent, struct vdi_copy_entry, node);
+
+ if (new->vid < entry->vid)
+ p = &(*p)->rb_left;
+ else if (new->vid > entry->vid)
+ p = &(*p)->rb_right;
+ else
+ return entry; /* already has this entry */
+ }
+ rb_link_node(&new->node, parent, p);
+ rb_insert_color(&new->node, root);
+
+ return NULL; /* insert successfully */
+}
+
+int get_vdi_copy_number(uint32_t vid)
+{
+ struct vdi_copy_entry *entry;
+
+ pthread_rwlock_rdlock(&vdi_copy_lock);
+ entry = vdi_copy_search(&vdi_copy_root, vid);
+ pthread_rwlock_unlock(&vdi_copy_lock);
+
+ if (!entry)
+ panic("No VDI copy entry for %" PRIx32 " found\n", vid);
+
+ return entry->nr_copies;
+}
+
+int get_obj_copy_number(uint64_t oid)
+{
+ uint32_t vid = oid_to_vid(oid);
+
+ return get_vdi_copy_number(vid);
+}
+
+int get_req_copy_number(struct request *req)
+{
+ int nr_copies;
+
+ nr_copies = req->rq.obj.copies;
+ if (!nr_copies)
+ nr_copies = get_obj_copy_number(req->rq.obj.oid);
+
+ return nr_copies;
+}
+
+int get_max_copy_number(void)
+{
+ return uatomic_read(&max_copies);
+}
+
+int add_vdi_copies(uint32_t vid, int nr_copies)
+{
+ struct vdi_copy_entry *entry, *old;
+
+ entry = xzalloc(sizeof(*entry));
+ entry->vid = vid;
+ entry->nr_copies = nr_copies;
+
+ dprintf("%" PRIx32 ", %d\n", vid, nr_copies);
+
+ pthread_rwlock_wrlock(&vdi_copy_lock);
+ old = vdi_copy_insert(&vdi_copy_root, entry);
+ if (old) {
+ free(entry);
+ entry = old;
+ entry->nr_copies = nr_copies;
+ }
+
+ if (uatomic_read(&max_copies) == 0 ||
+ nr_copies > uatomic_read(&max_copies))
+ uatomic_set(&max_copies, nr_copies);
+ pthread_rwlock_unlock(&vdi_copy_lock);
+
+ return SD_RES_SUCCESS;
+}
+
int vdi_exist(uint32_t vid)
{
struct sheepdog_inode *inode;
--
1.7.1
More information about the sheepdog
mailing list