vitalif
/
vitastor
5
4
Fork 9
Code Issues 1 Pull Requests 3 Actions Releases 81 Wiki Activity

Compare commits

...

3 Commits
1e3f143892 ... 4b48b60ce1

Author SHA1 Message Date
Vitaliy Filippov 4b48b60ce1 Implement support of RDMA via RDMA-CM 
Required for iWARP and, possibly, Infiniband cards
 2025-03-31 13:02:38 +03:00
Vitaliy Filippov 5075cb27c8 Support multiple RDMA networks 2025-03-31 13:02:36 +03:00
Vitaliy Filippov 7808d07d26 Support multiple OSD networks and separate OSD cluster network 2025-03-31 13:02:18 +03:00
20 changed files with 1065 additions and 344 deletions

1
src/client/CMakeLists.txt
View File

@ -28,6 +28,7 @@ target_link_libraries(vitastor_client
vitastor_cli
${LIBURING_LIBRARIES}
${IBVERBS_LIBRARIES}
${RDMACM_LIBRARIES}
)
set_target_properties(vitastor_client PROPERTIES VERSION ${VITASTOR_VERSION} SOVERSION 0)
configure_file(vitastor.pc.in vitastor.pc @ONLY)

2
src/client/etcd_state_client.cpp
View File

@ -182,7 +182,7 @@ void etcd_state_client_t::add_etcd_url(std::string addr)
exit(1);
}
if (!local_ips.size())
local_ips = getifaddr_list(std::vector<std::string>(), true);
local_ips = getifaddr_list(std::vector<addr_mask_t>(), true);
std::string check_addr;
int pos = addr.find('/');
int pos2 = addr.find(':');

238
src/client/messenger.cpp
View File

@ -117,29 +117,56 @@ void msgr_iothread_t::run()
void osd_messenger_t::init()
{
#ifdef WITH_RDMACM
if (use_rdmacm)
{
// RDMA-CM only requires the event channel. All the remaining work is done separately
rdmacm_evch = rdma_create_event_channel();
if (!rdmacm_evch)
{
fprintf(stderr, "Failed to initialize RDMA-CM event channel: %s (code %d)\n", strerror(errno), errno);
}
else
{
fcntl(rdmacm_evch->fd, F_SETFL, fcntl(rdmacm_evch->fd, F_GETFL, 0) | O_NONBLOCK);
tfd->set_fd_handler(rdmacm_evch->fd, false, [this](int rdmacm_eventfd, int epoll_events)
{
handle_rdmacm_events();
});
}
}
else
#endif
#ifdef WITH_RDMA
if (use_rdma)
{
rdma_context = msgr_rdma_context_t::create(
osd_networks, rdma_device != "" ? rdma_device.c_str() : NULL,
rdma_contexts = msgr_rdma_context_t::create_all(
osd_num && osd_cluster_network_masks.size() ? osd_cluster_network_masks : osd_network_masks,
rdma_device != "" ? rdma_device.c_str() : NULL,
rdma_port_num, rdma_gid_index, rdma_mtu, rdma_odp, log_level
);
if (!rdma_context)
if (!rdma_contexts.size())
{
if (force_rdma)
{
fprintf(stderr, "[OSD %ju] Couldn't initialize RDMA, force_rdma is enabled, exiting\n", osd_num);
exit(1);
}
if (log_level > 0)
fprintf(stderr, "[OSD %ju] Couldn't initialize RDMA, proceeding with TCP only\n", osd_num);
}
else
{
rdma_max_sge = rdma_max_sge < rdma_context->attrx.orig_attr.max_sge
? rdma_max_sge : rdma_context->attrx.orig_attr.max_sge;
fprintf(stderr, "[OSD %ju] RDMA initialized successfully\n", osd_num);
fcntl(rdma_context->channel->fd, F_SETFL, fcntl(rdma_context->channel->fd, F_GETFL, 0) | O_NONBLOCK);
tfd->set_fd_handler(rdma_context->channel->fd, false, [this](int notify_fd, int epoll_events)
for (msgr_rdma_context_t* rdma_context: rdma_contexts)
{
handle_rdma_events();
});
handle_rdma_events();
fcntl(rdma_context->channel->fd, F_SETFL, fcntl(rdma_context->channel->fd, F_GETFL, 0) | O_NONBLOCK);
tfd->set_fd_handler(rdma_context->channel->fd, false, [this, rdma_context](int notify_fd, int epoll_events)
{
handle_rdma_events(rdma_context);
});
handle_rdma_events(rdma_context);
}
}
}
#endif
@ -247,10 +274,19 @@ osd_messenger_t::~osd_messenger_t()
iothreads.clear();
}
#ifdef WITH_RDMA
if (rdma_context)
for (auto rdma_context: rdma_contexts)
{
delete rdma_context;
}
rdma_contexts.clear();
#endif
#ifdef WITH_RDMACM
if (rdmacm_evch)
{
tfd->set_fd_handler(rdmacm_evch->fd, false, NULL);
rdma_destroy_event_channel(rdmacm_evch);
rdmacm_evch = NULL;
}
#endif
}
@ -262,10 +298,18 @@ void osd_messenger_t::parse_config(const json11::Json & config)
// RDMA is on by default in RDMA-enabled builds
this->use_rdma = config["use_rdma"].bool_value() || config["use_rdma"].uint64_value() != 0;
}
#ifdef WITH_RDMACM
// Use RDMA CM? (required for iWARP and may be useful for IB)
// FIXME: Only parse during start
this->use_rdmacm = config["use_rdmacm"].bool_value() || config["use_rdmacm"].uint64_value() != 0;
this->disable_tcp = this->use_rdmacm && (config["disable_tcp"].bool_value() || config["disable_tcp"].uint64_value() != 0);
#endif
if (!config["force_rdma"].is_null())
{
this->force_rdma = config["force_rdma"].bool_value() || config["force_rdma"].uint64_value() != 0;
}
this->rdma_device = config["rdma_device"].string_value();
this->rdma_port_num = (uint8_t)config["rdma_port_num"].uint64_value();
if (!this->rdma_port_num)
this->rdma_port_num = 1;
if (!config["rdma_gid_index"].is_null())
this->rdma_gid_index = (uint8_t)config["rdma_gid_index"].uint64_value();
this->rdma_mtu = (uint32_t)config["rdma_mtu"].uint64_value();
@ -282,15 +326,6 @@ void osd_messenger_t::parse_config(const json11::Json & config)
if (!this->rdma_max_msg || this->rdma_max_msg > 128*1024*1024)
this->rdma_max_msg = 129*1024;
this->rdma_odp = config["rdma_odp"].bool_value();
std::vector<std::string> mask;
if (config["bind_address"].is_string())
mask.push_back(config["bind_address"].string_value());
else if (config["osd_network"].is_string())
mask.push_back(config["osd_network"].string_value());
else
for (auto v: config["osd_network"].array_items())
mask.push_back(v.string_value());
this->osd_networks = mask;
#endif
if (!osd_num)
this->iothread_count = (uint32_t)config["client_iothread_count"].uint64_value();
@ -314,23 +349,88 @@ void osd_messenger_t::parse_config(const json11::Json & config)
if (!this->osd_ping_timeout)
this->osd_ping_timeout = 5;
this->log_level = config["log_level"].uint64_value();
// OSD public & cluster networks
this->osd_networks.clear();
if (config["bind_address"].is_string())
this->osd_networks.push_back(config["bind_address"].string_value());
else if (config["osd_network"].is_string())
this->osd_networks.push_back(config["osd_network"].string_value());
else
for (auto v: config["osd_network"].array_items())
this->osd_networks.push_back(v.string_value());
this->osd_cluster_networks.clear();
if (config["osd_cluster_network"].is_string())
this->osd_cluster_networks.push_back(config["osd_cluster_network"].string_value());
else
for (auto v: config["osd_cluster_network"].array_items())
this->osd_cluster_networks.push_back(v.string_value());
if (this->osd_cluster_networks.size())
for (auto & net: this->osd_cluster_networks)
for (int i = this->osd_networks.size()-1; i >= 0; i--)
if (this->osd_networks[i] == net)
this->osd_networks.erase(this->osd_networks.begin()+i, this->osd_networks.begin()+i+1);
this->osd_network_masks.clear();
for (auto & netstr: this->osd_networks)
this->osd_network_masks.push_back(cidr_parse(netstr));
this->osd_cluster_network_masks.clear();
for (auto & netstr: this->osd_cluster_networks)
this->osd_cluster_network_masks.push_back(cidr_parse(netstr));
this->all_osd_networks.clear();
this->all_osd_networks.insert(this->all_osd_networks.end(), this->osd_networks.begin(), this->osd_networks.end());
this->all_osd_networks.insert(this->all_osd_networks.end(), this->osd_cluster_networks.begin(), this->osd_cluster_networks.end());
this->all_osd_network_masks.clear();
this->all_osd_network_masks.insert(this->all_osd_network_masks.end(), this->osd_network_masks.begin(), this->osd_network_masks.end());
this->all_osd_network_masks.insert(this->all_osd_network_masks.end(), this->osd_cluster_network_masks.begin(), this->osd_cluster_network_masks.end());
if (!this->osd_networks.size())
{
this->osd_networks = this->osd_cluster_networks;
this->osd_network_masks = this->osd_cluster_network_masks;
}
}
void osd_messenger_t::connect_peer(uint64_t peer_osd, json11::Json peer_state)
{
if (wanted_peers.find(peer_osd) == wanted_peers.end())
if (wanted_peers[peer_osd].raw_address_list != peer_state["addresses"])
{
wanted_peers[peer_osd] = (osd_wanted_peer_t){
.address_list = peer_state["addresses"],
.port = (int)peer_state["port"].int64_value(),
};
wanted_peers[peer_osd].raw_address_list = peer_state["addresses"];
if (osd_cluster_networks.size())
{
json11::Json::array address_list, cluster_address_list;
for (auto json_addr: peer_state["addresses"].array_items())
{
struct sockaddr_storage addr;
auto ok = string_to_addr(json_addr.string_value(), false, 0, &addr);
if (ok)
{
bool is_cluster = false;
for (auto & mask: osd_cluster_network_masks)
{
if (cidr_sockaddr_match(addr, mask))
{
is_cluster = true;
break;
}
}
if (is_cluster)
cluster_address_list.push_back(json_addr);
else
address_list.push_back(json_addr);
}
}
auto n_cluster = this->osd_num ? cluster_address_list.size() : 0;
if (this->osd_num)
address_list.insert(address_list.begin(), cluster_address_list.begin(), cluster_address_list.end());
wanted_peers[peer_osd].address_list = address_list;
wanted_peers[peer_osd].n_cluster_addr = n_cluster;
}
else
wanted_peers[peer_osd].address_list = peer_state["addresses"];
wanted_peers[peer_osd].address_changed = true;
}
else
{
wanted_peers[peer_osd].address_list = peer_state["addresses"];
wanted_peers[peer_osd].port = (int)peer_state["port"].int64_value();
}
wanted_peers[peer_osd].address_changed = true;
#ifdef WITH_RDMACM
wanted_peers[peer_osd].peer_rdmacm = peer_state["rdmacm"].bool_value();
#endif
wanted_peers[peer_osd].port = (int)peer_state["port"].int64_value();
try_connect_peer(peer_osd);
}
@ -355,12 +455,24 @@ void osd_messenger_t::try_connect_peer(uint64_t peer_osd)
wp.cur_addr = wp.address_list[wp.address_index].string_value();
wp.cur_port = wp.port;
wp.connecting = true;
try_connect_peer_addr(peer_osd, wp.cur_addr.c_str(), wp.cur_port);
#ifdef WITH_RDMACM
if (use_rdmacm && wp.peer_rdmacm)
rdmacm_try_connect_peer(peer_osd, wp.cur_addr.c_str(), wp.cur_port);
else
#endif
try_connect_peer_tcp(peer_osd, wp.cur_addr.c_str(), wp.cur_port);
}
void osd_messenger_t::try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port)
void osd_messenger_t::try_connect_peer_tcp(osd_num_t peer_osd, const char *peer_host, int peer_port)
{
assert(peer_osd != this->osd_num);
#ifdef WITH_RDMACM
if (disable_tcp)
{
on_connect_peer(peer_osd, -EINVAL);
return;
}
#endif
struct sockaddr_storage addr;
if (!string_to_addr(peer_host, 0, peer_port, &addr))
{
@ -524,20 +636,30 @@ void osd_messenger_t::check_peer_config(osd_client_t *cl)
},
};
#ifdef WITH_RDMA
if (rdma_context)
if (rdma_contexts.size())
{
cl->rdma_conn = msgr_rdma_connection_t::create(rdma_context, rdma_max_send, rdma_max_recv, rdma_max_sge, rdma_max_msg);
if (cl->rdma_conn)
// Choose the right context for the selected network
msgr_rdma_context_t *selected_ctx = choose_rdma_context(cl);
if (!selected_ctx)
{
json11::Json payload = json11::Json::object {
{ "connect_rdma", cl->rdma_conn->addr.to_string() },
{ "rdma_max_msg", cl->rdma_conn->max_msg },
};
std::string payload_str = payload.dump();
op->req.show_conf.json_len = payload_str.size();
op->buf = malloc_or_die(payload_str.size());
op->iov.push_back(op->buf, payload_str.size());
memcpy(op->buf, payload_str.c_str(), payload_str.size());
if (log_level > 0)
fprintf(stderr, "No RDMA context for OSD %ju connection (peer %d), using only TCP\n", cl->osd_num, cl->peer_fd);
}
else
{
cl->rdma_conn = msgr_rdma_connection_t::create(selected_ctx, rdma_max_send, rdma_max_recv, rdma_max_sge, rdma_max_msg);
if (cl->rdma_conn)
{
json11::Json payload = json11::Json::object {
{ "connect_rdma", cl->rdma_conn->addr.to_string() },
{ "rdma_max_msg", cl->rdma_conn->max_msg },
};
std::string payload_str = payload.dump();
op->req.show_conf.json_len = payload_str.size();
op->buf = malloc_or_die(payload_str.size());
op->iov.push_back(op->buf, payload_str.size());
memcpy(op->buf, payload_str.c_str(), payload_str.size());
}
}
}
#endif
@ -667,9 +789,29 @@ void osd_messenger_t::accept_connections(int listen_fd)
}
#ifdef WITH_RDMA
msgr_rdma_context_t* osd_messenger_t::choose_rdma_context(osd_client_t *cl)
{
// Choose the right context for the selected network
msgr_rdma_context_t *selected_ctx = NULL;
for (auto rdma_ctx: rdma_contexts)
{
if (!rdma_ctx->net_mask.family && !selected_ctx ||
rdma_ctx->net_mask.family && cidr_sockaddr_match(cl->peer_addr, rdma_ctx->net_mask))
{
selected_ctx = rdma_ctx;
}
}
return selected_ctx;
}
bool osd_messenger_t::is_rdma_enabled()
{
return rdma_context != NULL;
return rdma_contexts.size() > 0;
}
bool osd_messenger_t::is_use_rdmacm()
{
return use_rdmacm;
}
#endif

68
src/client/messenger.h
View File

@ -16,6 +16,7 @@
#include "json11/json11.hpp"
#include "msgr_op.h"
#include "timerfd_manager.h"
#include "addr_util.h"
#include <ringloop.h>
#define CL_READ_HDR 1
@ -49,10 +50,10 @@ struct osd_client_t
{
int refs = 0;
sockaddr_storage peer_addr;
int peer_port;
sockaddr_storage peer_addr = {};
int peer_port = 0;
int peer_fd = -1;
int peer_state;
int peer_state = 0;
int connect_timeout_id = -1;
int ping_time_remaining = 0;
int idle_time_remaining = 0;
@ -93,13 +94,16 @@ struct osd_client_t
struct osd_wanted_peer_t
{
json11::Json raw_address_list;
json11::Json address_list;
int port;
time_t last_connect_attempt;
bool connecting, address_changed;
int address_index;
bool peer_rdmacm = false;
int n_cluster_addr = 0;
int port = 0;
time_t last_connect_attempt = 0;
bool connecting = false, address_changed = false;
int address_index = 0;
std::string cur_addr;
int cur_port;
int cur_port = 0;
};
struct osd_op_stats_t
@ -149,6 +153,15 @@ public:
};
#endif
#ifdef WITH_RDMA
struct rdma_event_channel;
struct rdma_cm_id;
struct rdma_cm_event;
struct ibv_context;
struct osd_messenger_t;
struct rdmacm_connecting_t;
#endif
struct osd_messenger_t
{
protected:
@ -165,14 +178,20 @@ protected:
#ifdef WITH_RDMA
bool use_rdma = true;
std::vector<std::string> osd_networks;
bool use_rdmacm = false;
bool disable_tcp = false;
bool force_rdma = false;
std::string rdma_device;
uint64_t rdma_port_num = 1, rdma_mtu = 0;
uint64_t rdma_port_num = 1;
int rdma_mtu = 0;
int rdma_gid_index = -1;
msgr_rdma_context_t *rdma_context = NULL;
std::vector<msgr_rdma_context_t *> rdma_contexts;
uint64_t rdma_max_sge = 0, rdma_max_send = 0, rdma_max_recv = 0;
uint64_t rdma_max_msg = 0;
bool rdma_odp = false;
rdma_event_channel *rdmacm_evch = NULL;
std::map<rdma_cm_id*, osd_client_t*> rdmacm_connections;
std::map<rdma_cm_id*, rdmacm_connecting_t*> rdmacm_connecting;
#endif
std::vector<msgr_iothread_t*> iothreads;
@ -190,6 +209,12 @@ public:
std::map<int, osd_client_t*> clients;
std::map<osd_num_t, osd_wanted_peer_t> wanted_peers;
std::map<uint64_t, int> osd_peer_fds;
std::vector<std::string> osd_networks;
std::vector<addr_mask_t> osd_network_masks;
std::vector<std::string> osd_cluster_networks;
std::vector<addr_mask_t> osd_cluster_network_masks;
std::vector<std::string> all_osd_networks;
std::vector<addr_mask_t> all_osd_network_masks;
// op statistics
osd_op_stats_t stats, recovery_stats;
@ -216,13 +241,18 @@ public:
bool is_rdma_enabled();
bool connect_rdma(int peer_fd, std::string rdma_address, uint64_t client_max_msg);
#endif
#ifdef WITH_RDMACM
bool is_use_rdmacm();
rdma_cm_id *rdmacm_listen(const std::string & bind_address, int rdmacm_port, int *bound_port, int log_level);
void rdmacm_destroy_listener(rdma_cm_id *listener);
#endif
void inc_op_stats(osd_op_stats_t & stats, uint64_t opcode, timespec & tv_begin, timespec & tv_end, uint64_t len);
void measure_exec(osd_op_t *cur_op);
protected:
void try_connect_peer(uint64_t osd_num);
void try_connect_peer_addr(osd_num_t peer_osd, const char *peer_host, int peer_port);
void try_connect_peer_tcp(osd_num_t peer_osd, const char *peer_host, int peer_port);
void handle_peer_epoll(int peer_fd, int epoll_events);
void handle_connect_epoll(int peer_fd);
void on_connect_peer(osd_num_t peer_osd, int peer_fd);
@ -247,6 +277,18 @@ protected:
void try_send_rdma_odp(osd_client_t *cl);
void try_send_rdma_nodp(osd_client_t *cl);
bool try_recv_rdma(osd_client_t *cl);
void handle_rdma_events();
void handle_rdma_events(msgr_rdma_context_t *rdma_context);
msgr_rdma_context_t* choose_rdma_context(osd_client_t *cl);
#endif
#ifdef WITH_RDMACM
void handle_rdmacm_events();
msgr_rdma_context_t* rdmacm_get_context(ibv_context *verbs);
msgr_rdma_context_t* rdmacm_create_qp(rdma_cm_id *cmid);
void rdmacm_accept(rdma_cm_event *ev);
void rdmacm_try_connect_peer(uint64_t peer_osd, const std::string & addr, int peer_port);
void rdmacm_on_connect_peer_error(rdma_cm_id *cmid, int res);
void rdmacm_address_resolved(rdma_cm_event *ev);
void rdmacm_route_resolved(rdma_cm_event *ev);
void rdmacm_established(rdma_cm_event *ev);
#endif
};

927
src/client/msgr_rdma.cpp
View File

File diff suppressed because it is too large Load Diff

21
src/client/msgr_rdma.h
View File

@ -2,9 +2,13 @@
// License: VNPL-1.1 or GNU GPL-2.0+ (see README.md for details)
#pragma once
#ifdef WITH_RDMACM
#include <rdma/rdma_cma.h>
#endif
#include <infiniband/verbs.h>
#include <string>
#include <vector>
#include "addr_util.h"
struct msgr_rdma_address_t
{
@ -20,7 +24,6 @@ struct msgr_rdma_address_t
struct msgr_rdma_context_t
{
ibv_context *context = NULL;
ibv_device *dev = NULL;
ibv_device_attr_ex attrx;
ibv_pd *pd = NULL;
bool odp = false;
@ -35,8 +38,17 @@ struct msgr_rdma_context_t
uint32_t mtu;
int max_cqe = 0;
int used_max_cqe = 0;
addr_mask_t net_mask = {};
bool is_cm = false;
int cm_refs = 0;
static std::vector<msgr_rdma_context_t*> create_all(const std::vector<addr_mask_t> & osd_network_masks,
const char *sel_dev_name, int sel_port_num, int sel_gid_index, uint32_t sel_mtu, bool odp, int log_level);
static msgr_rdma_context_t *create(ibv_device *dev, ibv_port_attr & portinfo,
int ib_port, int gid_index, uint32_t mtu, bool odp, int log_level);
static msgr_rdma_context_t* create_cm(ibv_context *ctx);
bool reserve_cqe(int n);
static msgr_rdma_context_t *create(std::vector<std::string> osd_networks, const char *ib_devname, uint8_t ib_port, int gid_index, uint32_t mtu, bool odp, int log_level);
~msgr_rdma_context_t();
};
@ -50,11 +62,14 @@ struct msgr_rdma_connection_t
{
msgr_rdma_context_t *ctx = NULL;
ibv_qp *qp = NULL;
#ifdef WITH_RDMACM
rdma_cm_id *cmid = NULL;
#endif
msgr_rdma_address_t addr;
int max_send = 0, max_recv = 0, max_sge = 0;
int cur_send = 0, cur_recv = 0;
uint64_t max_msg = 0;
int cur_send = 0, cur_recv = 0;
int send_pos = 0, send_buf_pos = 0;
int next_recv_buf = 0;
std::vector<msgr_rdma_buf_t> recv_buffers;

2
src/cmd/cli.h
View File

@ -98,5 +98,3 @@ std::string format_lat(uint64_t lat);
std::string format_q(double depth);
bool stupid_glob(const std::string str, const std::string glob);
std::string implode(const std::string & sep, json11::Json array);

1
src/cmd/cli_osd_tree.cpp
View File

@ -7,6 +7,7 @@
#include "epoll_manager.h"
#include "pg_states.h"
#include "str_util.h"
#include "json_util.h"
struct placement_osd_t
{

1
src/cmd/cli_pg_ls.cpp
View File

@ -5,6 +5,7 @@
#include "cluster_client.h"
#include "pg_states.h"
#include "str_util.h"
#include "json_util.h"
struct pg_lister_t
{

1
src/cmd/cli_pool_create.cpp
View File

@ -10,6 +10,7 @@
#include "epoll_manager.h"
#include "pg_states.h"
#include "str_util.h"
#include "json_util.h"
struct pool_creator_t
{

17
src/cmd/cli_pool_ls.cpp
View File

@ -5,6 +5,7 @@
#include "cli.h"
#include "cluster_client.h"
#include "str_util.h"
#include "json_util.h"
#include "pg_states.h"
// List pools with space statistics
@ -665,19 +666,3 @@ std::function<bool(cli_result_t &)> cli_tool_t::start_pool_ls(json11::Json cfg)
return false;
};
}
std::string implode(const std::string & sep, json11::Json array)
{
if (array.is_number() || array.is_bool() || array.is_string())
{
return array.as_string();
}
std::string res;
bool first = true;
for (auto & item: array.array_items())
{
res += (first ? item.as_string() : sep+item.as_string());
first = false;
}
return res;
}

1
src/osd/CMakeLists.txt
View File

@ -14,6 +14,7 @@ target_link_libraries(vitastor-osd
Jerasure
${ISAL_LIBRARIES}
${IBVERBS_LIBRARIES}
${RDMACM_LIBRARIES}
)
# osd_rmw_test

76
src/osd/osd.cpp
View File

@ -120,7 +120,14 @@ osd_t::~osd_t()
delete epmgr;
if (bs)
delete bs;
close(listen_fd);
#ifdef WITH_RDMACM
for (rdma_cm_id *listener: rdmacm_listeners)
msgr.rdmacm_destroy_listener(listener);
rdmacm_listeners.clear();
#endif
for (auto listen_fd: listen_fds)
close(listen_fd);
listen_fds.clear();
free(zero_buffer);
}
@ -162,12 +169,14 @@ void osd_t::parse_config(bool init)
else
immediate_commit = IMMEDIATE_NONE;
// Bind address
bind_address = config["bind_address"].string_value();
if (bind_address == "")
bind_address = "0.0.0.0";
bind_port = config["bind_port"].uint64_value();
if (bind_port <= 0 || bind_port > 65535)
bind_port = 0;
#ifdef WITH_RDMACM
// Use RDMA CM? (required for iWARP and may be useful for IB)
this->use_rdmacm = config["use_rdmacm"].bool_value() || config["use_rdmacm"].uint64_value() != 0;
this->disable_tcp = this->use_rdmacm && (config["disable_tcp"].bool_value() || config["disable_tcp"].uint64_value() != 0);
#endif
// OSD configuration
etcd_report_interval = config["etcd_report_interval"].uint64_value();
if (etcd_report_interval <= 0)
@ -322,41 +331,44 @@ void osd_t::parse_config(bool init)
void osd_t::bind_socket()
{
if (config["osd_network"].is_string() ||
config["osd_network"].is_array())
if (msgr.all_osd_network_masks.size())
{
std::vector<std::string> mask;
if (config["osd_network"].is_string())
mask.push_back(config["osd_network"].string_value());
else
for (auto v: config["osd_network"].array_items())
mask.push_back(v.string_value());
auto matched_addrs = getifaddr_list(mask);
if (matched_addrs.size() > 1)
bind_addresses = getifaddr_list(msgr.all_osd_network_masks);
if (!bind_addresses.size())
{
fprintf(stderr, "More than 1 address matches requested network(s): %s\n", json11::Json(matched_addrs).dump().c_str());
force_stop(1);
}
if (!matched_addrs.size())
{
std::string nets;
for (auto v: mask)
nets += (nets == "" ? v : ","+v);
auto nets = implode(", ", msgr.all_osd_networks);
fprintf(stderr, "Addresses matching osd_network(s) %s not found\n", nets.c_str());
force_stop(1);
}
bind_address = matched_addrs[0];
}
// FIXME Support multiple listening sockets
listen_fd = create_and_bind_socket(bind_address, bind_port, listen_backlog, &listening_port);
fcntl(listen_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
epmgr->set_fd_handler(listen_fd, false, [this](int fd, int events)
else
{
msgr.accept_connections(listen_fd);
});
bind_addresses.push_back("0.0.0.0");
}
if (!disable_tcp)
{
for (auto & bind_address: bind_addresses)
{
int listen_fd = create_and_bind_socket(bind_address, listening_port ? listening_port : bind_port, listen_backlog, &listening_port);
fcntl(listen_fd, F_SETFL, fcntl(listen_fd, F_GETFL, 0) | O_NONBLOCK);
epmgr->set_fd_handler(listen_fd, false, [this](int fd, int events)
{
msgr.accept_connections(fd);
});
listen_fds.push_back(listen_fd);
}
}
#ifdef WITH_RDMACM
if (use_rdmacm)
{
for (auto & bind_address: bind_addresses)
{
auto listener = msgr.rdmacm_listen(bind_address, listening_port, &listening_port, log_level);
if (listener)
rdmacm_listeners.push_back(listener);
}
}
#endif
}
bool osd_t::shutdown()

9
src/osd/osd.h
View File

@ -107,8 +107,9 @@ class osd_t
bool no_recovery = false;
bool no_scrub = false;
bool allow_net_split = false;
std::string bind_address;
int bind_port, listen_backlog = 128;
bool use_rdmacm = false;
bool disable_tcp = false;
// FIXME: Implement client queue depth limit
int client_queue_depth = 128;
bool allow_test_ops = false;
@ -200,7 +201,11 @@ class osd_t
epoll_manager_t *epmgr = NULL;
int listening_port = 0;
int listen_fd = 0;
std::vector<std::string> bind_addresses;
std::vector<int> listen_fds;
#ifdef WITH_RDMACM
std::vector<rdma_cm_id *> rdmacm_listeners;
#endif
ring_consumer_t consumer;
// op statistics

8
src/osd/osd_cluster.cpp
View File

@ -165,13 +165,17 @@ json11::Json osd_t::get_osd_state()
hostname.resize(strnlen(hostname.data(), hostname.size()));
json11::Json::object st;
st["state"] = "up";
if (bind_address != "0.0.0.0")
st["addresses"] = json11::Json::array { bind_address };
if (bind_addresses.size() != 1 || bind_addresses[0] != "0.0.0.0")
st["addresses"] = bind_addresses;
else
st["addresses"] = getifaddr_list();
st["host"] = std::string(hostname.data(), hostname.size());
st["version"] = VITASTOR_VERSION;
st["port"] = listening_port;
#ifdef WITH_RDMACM
if (rdmacm_listeners.size())
st["rdmacm"] = true;
#endif
st["primary_enabled"] = run_primary;
st["blockstore_enabled"] = bs ? true : false;
return st;

1
src/test/CMakeLists.txt
View File

@ -25,6 +25,7 @@ target_link_libraries(stub_uring_osd
vitastor_common
${LIBURING_LIBRARIES}
${IBVERBS_LIBRARIES}
${RDMACM_LIBRARIES}
tcmalloc_minimal
)

15
src/util/addr_util.cpp
View File

@ -93,6 +93,13 @@ bool cidr6_match(const in6_addr &address, const in6_addr &network, uint8_t bits)
return true;
}
bool cidr_sockaddr_match(const sockaddr_storage &addr, const addr_mask_t &mask)
{
return mask.family == addr.ss_family && (mask.family == AF_INET
? cidr_match(*(in_addr*)&addr, mask.ipv4, mask.bits)
: cidr6_match(*(in6_addr*)&addr, mask.ipv6, mask.bits));
}
addr_mask_t cidr_parse(std::string mask)
{
unsigned bits = 255;
@ -126,13 +133,11 @@ addr_mask_t cidr_parse(std::string mask)
}
}
std::vector<std::string> getifaddr_list(std::vector<std::string> mask_cfg, bool include_v6)
std::vector<std::string> getifaddr_list(const std::vector<addr_mask_t> & masks, bool include_v6)
{
std::vector<addr_mask_t> masks;
for (auto mask: mask_cfg)
for (auto & mask: masks)
{
masks.push_back(cidr_parse(mask));
if (masks[masks.size()-1].family == AF_INET6)
if (mask.family == AF_INET6)
{
// Auto-enable IPv6 addresses
include_v6 = true;

3
src/util/addr_util.h
View File

@ -18,5 +18,6 @@ std::string addr_to_string(const sockaddr_storage &addr);
addr_mask_t cidr_parse(std::string mask);
bool cidr_match(const in_addr &address, const in_addr &network, uint8_t bits);
bool cidr6_match(const in6_addr &address, const in6_addr &network, uint8_t bits);
std::vector<std::string> getifaddr_list(std::vector<std::string> mask_cfg = std::vector<std::string>(), bool include_v6 = false);
bool cidr_sockaddr_match(const sockaddr_storage &addr, const addr_mask_t &mask);
std::vector<std::string> getifaddr_list(const std::vector<addr_mask_t> & masks = std::vector<addr_mask_t>(), bool include_v6 = false);
int create_and_bind_socket(std::string bind_address, int bind_port, int listen_backlog, int *listening_port);

16
src/util/json_util.cpp
View File

@ -33,3 +33,19 @@ bool json_is_false(const json11::Json & val)
return !val.bool_value();
return false;
}
std::string implode(const std::string & sep, json11::Json array)
{
if (array.is_number() || array.is_bool() || array.is_string())
{
return array.as_string();
}
std::string res;
bool first = true;
for (auto & item: array.array_items())
{
res += (first ? item.as_string() : sep+item.as_string());
first = false;
}
return res;
}

1
src/util/json_util.h
View File

@ -11,3 +11,4 @@
std::map<std::string, std::string> json_to_string_map(const json11::Json::object & config);
bool json_is_true(const json11::Json & val);
bool json_is_false(const json11::Json & val);
std::string implode(const std::string & sep, json11::Json array);