int sd_read_object(uint64_t oid, void *data, unsigned int datalen,
uint64_t offset, bool direct)
{
struct sd_req hdr;
struct sd_rsp *rsp = (struct sd_rsp *)&hdr;
int ret;
sd_init_req(&hdr, SD_OP_READ_OBJ);
hdr.data_length = datalen;
hdr.obj.oid = oid;
hdr.obj.offset = offset;
if (direct)
hdr.flags |= SD_FLAG_CMD_DIRECT;
ret = dog_exec_req(sdhost, sdport, &hdr, data);
if (ret < 0) {
sd_err("Failed to read object %" PRIx64, oid);
return SD_RES_EIO;
}
if (rsp->result != SD_RES_SUCCESS) {
sd_err("Failed to read object %" PRIx64 " %s", oid,
sd_strerror(rsp->result));
return rsp->result;
}
untrim_zero_blocks(data, rsp->obj.offset, rsp->data_length, datalen);
return SD_RES_SUCCESS;
}
int read_backend_object(uint64_t oid, char *data, unsigned int datalen,
uint64_t offset)
{
struct sd_req hdr;
struct sd_rsp *rsp = (struct sd_rsp *)&hdr;
int ret;
sd_init_req(&hdr, SD_OP_READ_OBJ);
hdr.data_length = datalen;
hdr.obj.oid = oid;
hdr.obj.offset = offset;
hdr.obj.copies = get_vdi_copy_number(oid_to_vid(oid));
ret = exec_local_req(&hdr, data);
if (ret != SD_RES_SUCCESS)
sd_eprintf("failed to read object %" PRIx64 ", %s", oid,
sd_strerror(ret));
untrim_zero_blocks(data, rsp->obj.offset, rsp->data_length, datalen);
return ret;
}
/*
* Try our best to read one copy and read local first.
*
* Return success if any read succeed. We don't call gateway_forward_request()
* because we only read once.
*/
int gateway_read_obj(struct request *req)
{
int i, ret = SD_RES_SUCCESS;
struct sd_req fwd_hdr;
struct sd_rsp *rsp = (struct sd_rsp *)&fwd_hdr;
const struct sd_vnode *v;
const struct sd_vnode *my_v;
const struct sd_vnode *obj_vnodes[SD_MAX_COPIES];
const struct sd_vnode *my_obj_vnodes[SD_MAX_COPIES];
uint64_t oid = req->rq.obj.oid;
int nr_copies, j;
//PRINT_TO_LOG("WYH\n");
if (sys->enable_object_cache && !req->local &&
!bypass_object_cache(req)) {
ret = object_cache_handle_request(req);
goto out;
}
nr_copies = get_req_copy_number(req);
if (nr_copies == 0) {
sd_debug("there is no living nodes");
return SD_RES_HALT;
}
oid_to_vnodes(req->vinfo->vnodes, req->vinfo->nr_vnodes, oid,
nr_copies, obj_vnodes);
//PRINT_TO_LOG("out:%s\n", addr_to_str(obj_vnodes[0]->nid.addr,obj_vnodes[0]->nid.port));
oid_to_vnodes(req->vinfo->my_vnodes, req->vinfo->my_nr_vnodes, oid, nr_copies, my_obj_vnodes);
my_v = my_obj_vnodes[0];
//PRINT_TO_LOG("out:%s\n", addr_to_str(my_v->nid.addr,my_v->nid.port));
for (i = 0; i < nr_copies; i++) {
v = obj_vnodes[0];
my_v = my_obj_vnodes[0];
//PRINT_TO_LOG("phy:%s\n", addr_to_str(v->nid.addr,v->nid.port));
//PRINT_TO_LOG("vir:%s\n", addr_to_str(my_v->nid.addr, my_v->nid.port));
//PRINT_TO_LOG("%d,%d\n", req->vinfo->nr_vnodes, req->vinfo->my_nr_vnodes);
if (!vnode_is_local(v))
continue;
ret = peer_read_obj(req);
if (ret == SD_RES_SUCCESS)
goto out;
sd_err("local read %"PRIx64" failed, %s", oid,
sd_strerror(ret));
break;
}
/*
* Read random copy from cluster for better load balance, useful for
* reading base VM's COW objects
*/
j = random();
for (i = 0; i < nr_copies; i++) {
int idx = (i + j) % nr_copies;
v = obj_vnodes[idx];
if (vnode_is_local(v))
continue;
/*
* We need to re-init it because rsp and req share the same
* structure.
*/
gateway_init_fwd_hdr(&fwd_hdr, &req->rq);
ret = sheep_exec_req(&v->nid, &fwd_hdr, req->data);
if (ret != SD_RES_SUCCESS)
continue;
/* Read success */
memcpy(&req->rp, rsp, sizeof(*rsp));
break;
}
out:
if (ret == SD_RES_SUCCESS &&
req->rq.proto_ver < SD_PROTO_VER_TRIM_ZERO_SECTORS) {
/* the client doesn't support trimming zero bytes */
untrim_zero_blocks(req->data, req->rp.obj.offset,
req->rp.data_length, req->rq.data_length);
req->rp.data_length = req->rq.data_length;
req->rp.obj.offset = 0;
}
return ret;
}
