[ceph.git] / ceph / src / common / pick_address.cc

// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
// vim: ts=8 sw=2 smarttab
/*
 * Ceph - scalable distributed file system
 *
 * Copyright (C) 2004-2012 Inktank
 *
 * This is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License version 2.1, as published by the Free Software
 * Foundation.  See file COPYING.
 *
 */

#include "common/pick_address.h"
#include "include/ipaddr.h"
#include "include/str_list.h"
#include "common/ceph_context.h"
#ifndef WITH_SEASTAR
#include "common/config.h"
#include "common/config_obs.h"
#endif
#include "common/debug.h"
#include "common/errno.h"
#include "common/numa.h"

#include <netdb.h>

#define dout_subsys ceph_subsys_

const struct sockaddr *find_ip_in_subnet_list(
  CephContext *cct,
  const struct ifaddrs *ifa,
  unsigned ipv,
  const std::string &networks,
  const std::string &interfaces,
  int numa_node)
{
  std::list<string> nets;
  get_str_list(networks, nets);
  std::list<string> ifs;
  get_str_list(interfaces, ifs);

  // filter interfaces by name
  const struct ifaddrs *filtered = nullptr;
  if (ifs.empty()) {
    filtered = ifa;
  } else {
    if (nets.empty()) {
      lderr(cct) << "interface names specified but not network names" << dendl;
      exit(1);
    }
    const struct ifaddrs *t = ifa;
    struct ifaddrs *head = 0;
    while (t) {
      bool match = false;
      for (auto& i : ifs) {
	if (strcmp(i.c_str(), t->ifa_name) == 0) {
	  match = true;
	  break;
	}
      }
      if (match) {
	struct ifaddrs *n = new ifaddrs;
	memcpy(n, t, sizeof(*t));
	n->ifa_next = head;
	head = n;
      }
      t = t->ifa_next;
    }
    if (!head) {
      lderr(cct) << "no interfaces matching " << ifs << dendl;
      exit(1);
    }
    filtered = head;
  }

  struct sockaddr *r = nullptr;
  for (auto& s : nets) {
    struct sockaddr_storage net;
    unsigned int prefix_len;

    if (!parse_network(s.c_str(), &net, &prefix_len)) {
      lderr(cct) << "unable to parse network: " << s << dendl;
      exit(1);
    }

    switch (net.ss_family) {
    case AF_INET:
      if (!(ipv & CEPH_PICK_ADDRESS_IPV4)) {
	continue;
      }
      break;
    case AF_INET6:
      if (!(ipv & CEPH_PICK_ADDRESS_IPV6)) {
	continue;
      }
      break;
    }

    const struct ifaddrs *found = find_ip_in_subnet(
      filtered,
      (struct sockaddr *) &net, prefix_len, numa_node);
    if (found) {
      r = found->ifa_addr;
      break;
    }
  }

  if (filtered != ifa) {
    while (filtered) {
      struct ifaddrs *t = filtered->ifa_next;
      delete filtered;
      filtered = t;
    }
  }

  return r;
}

#ifndef WITH_SEASTAR
// observe this change
struct Observer : public md_config_obs_t {
  const char *keys[2];
  explicit Observer(const char *c) {
    keys[0] = c;
    keys[1] = NULL;
  }

  const char** get_tracked_conf_keys() const override {
    return (const char **)keys;
  }
  void handle_conf_change(const ConfigProxy& conf,
			  const std::set <std::string> &changed) override {
    // do nothing.
  }
};

static void fill_in_one_address(CephContext *cct,
				const struct ifaddrs *ifa,
				const string networks,
				const string interfaces,
				const char *conf_var,
				int numa_node = -1)
{
  const struct sockaddr *found = find_ip_in_subnet_list(
    cct,
    ifa,
    CEPH_PICK_ADDRESS_IPV4|CEPH_PICK_ADDRESS_IPV6,
    networks,
    interfaces,
    numa_node);
  if (!found) {
    lderr(cct) << "unable to find any IP address in networks '" << networks
	       << "' interfaces '" << interfaces << "'" << dendl;
    exit(1);
  }

  char buf[INET6_ADDRSTRLEN];
  int err;

  err = getnameinfo(found,
		    (found->sa_family == AF_INET)
		    ? sizeof(struct sockaddr_in)
		    : sizeof(struct sockaddr_in6),

		    buf, sizeof(buf),
		    nullptr, 0,
		    NI_NUMERICHOST);
  if (err != 0) {
    lderr(cct) << "unable to convert chosen address to string: " << gai_strerror(err) << dendl;
    exit(1);
  }

  Observer obs(conf_var);

  cct->_conf.add_observer(&obs);

  cct->_conf.set_val_or_die(conf_var, buf);
  cct->_conf.apply_changes(nullptr);

  cct->_conf.remove_observer(&obs);
}

void pick_addresses(CephContext *cct, int needs)
{
  struct ifaddrs *ifa;
  int r = getifaddrs(&ifa);
  auto public_addr = cct->_conf.get_val<entity_addr_t>("public_addr");
  auto public_network = cct->_conf.get_val<std::string>("public_network");
  auto public_network_interface =
    cct->_conf.get_val<std::string>("public_network_interface");
  auto cluster_addr = cct->_conf.get_val<entity_addr_t>("cluster_addr");
  auto cluster_network = cct->_conf.get_val<std::string>("cluster_network");
  auto cluster_network_interface =
    cct->_conf.get_val<std::string>("cluster_network_interface");

  if (r < 0) {
    string err = cpp_strerror(errno);
    lderr(cct) << "unable to fetch interfaces and addresses: " << err << dendl;
    exit(1);
  }

  if ((needs & CEPH_PICK_ADDRESS_PUBLIC) &&
    public_addr.is_blank_ip() && !public_network.empty()) {
    fill_in_one_address(cct, ifa, public_network, public_network_interface,
      "public_addr");
  }

  if ((needs & CEPH_PICK_ADDRESS_CLUSTER) && cluster_addr.is_blank_ip()) {
    if (!cluster_network.empty()) {
      fill_in_one_address(cct, ifa, cluster_network, cluster_network_interface,
	"cluster_addr");
    } else {
      if (!public_network.empty()) {
        lderr(cct) << "Public network was set, but cluster network was not set " << dendl;
        lderr(cct) << "    Using public network also for cluster network" << dendl;
        fill_in_one_address(cct, ifa, public_network, public_network_interface,
          "cluster_addr");
      }
    }
  }

  freeifaddrs(ifa);
}
#endif	// !WITH_SEASTAR

static int fill_in_one_address(
  CephContext *cct,
  const struct ifaddrs *ifa,
  unsigned ipv,
  const string networks,
  const string interfaces,
  entity_addrvec_t *addrs,
  int numa_node = -1)
{
  const struct sockaddr *found = find_ip_in_subnet_list(cct, ifa, ipv, networks,
							interfaces, numa_node);
  if (!found) {
    std::string ip_type = "";
    if ((ipv & CEPH_PICK_ADDRESS_IPV4) && (ipv & CEPH_PICK_ADDRESS_IPV6)) {
      ip_type = "IPv4 or IPv6";
    } else if (ipv & CEPH_PICK_ADDRESS_IPV4) {
      ip_type = "IPv4";
    } else {
      ip_type = "IPv6";
    }
    lderr(cct) << "unable to find any " << ip_type << " address in networks '"
               << networks << "' interfaces '" << interfaces << "'" << dendl;
    return -1;
  }

  char buf[INET6_ADDRSTRLEN];
  int err;

  err = getnameinfo(found,
		    (found->sa_family == AF_INET)
		    ? sizeof(struct sockaddr_in)
		    : sizeof(struct sockaddr_in6),

		    buf, sizeof(buf),
		    nullptr, 0,
		    NI_NUMERICHOST);
  if (err != 0) {
    lderr(cct) << "unable to convert chosen address to string: " << gai_strerror(err) << dendl;
    return -1;
  }

  entity_addr_t addr;
  const char *end = 0;
  bool r = addr.parse(buf, &end);
  if (!r) {
    return -1;
  }
  addrs->v.push_back(addr);
  return 0;
}

int pick_addresses(
  CephContext *cct,
  unsigned flags,
  struct ifaddrs *ifa,
  entity_addrvec_t *addrs,
  int preferred_numa_node)
{
  addrs->v.clear();

  unsigned addrt = (flags & (CEPH_PICK_ADDRESS_PUBLIC |
			     CEPH_PICK_ADDRESS_CLUSTER));
  if (addrt == 0 ||
      addrt == (CEPH_PICK_ADDRESS_PUBLIC |
		CEPH_PICK_ADDRESS_CLUSTER)) {
    return -EINVAL;
  }
  unsigned msgrv = flags & (CEPH_PICK_ADDRESS_MSGR1 |
			    CEPH_PICK_ADDRESS_MSGR2);
  if (msgrv == 0) {
    if (cct->_conf.get_val<bool>("ms_bind_msgr1")) {
      msgrv |= CEPH_PICK_ADDRESS_MSGR1;
    }
    if (cct->_conf.get_val<bool>("ms_bind_msgr2")) {
      msgrv |= CEPH_PICK_ADDRESS_MSGR2;
    }
    if (msgrv == 0) {
      return -EINVAL;
    }
  }
  unsigned ipv = flags & (CEPH_PICK_ADDRESS_IPV4 |
			  CEPH_PICK_ADDRESS_IPV6);
  if (ipv == 0) {
    if (cct->_conf.get_val<bool>("ms_bind_ipv4")) {
      ipv |= CEPH_PICK_ADDRESS_IPV4;
    }
    if (cct->_conf.get_val<bool>("ms_bind_ipv6")) {
      ipv |= CEPH_PICK_ADDRESS_IPV6;
    }
    if (ipv == 0) {
      return -EINVAL;
    }
    if (cct->_conf.get_val<bool>("ms_bind_prefer_ipv4")) {
      flags |= CEPH_PICK_ADDRESS_PREFER_IPV4;
    } else {
      flags &= ~CEPH_PICK_ADDRESS_PREFER_IPV4;
    }
  }

  entity_addr_t addr;
  string networks;
  string interfaces;
  if (addrt & CEPH_PICK_ADDRESS_PUBLIC) {
    addr = cct->_conf.get_val<entity_addr_t>("public_addr");
    networks = cct->_conf.get_val<std::string>("public_network");
    interfaces =
      cct->_conf.get_val<std::string>("public_network_interface");
  } else {
    addr = cct->_conf.get_val<entity_addr_t>("cluster_addr");
    networks = cct->_conf.get_val<std::string>("cluster_network");
    interfaces =
      cct->_conf.get_val<std::string>("cluster_network_interface");
    if (networks.empty()) {
      lderr(cct) << "Falling back to public interface" << dendl;
      // fall back to public_ network and interface if cluster is not set
      networks = cct->_conf.get_val<std::string>("public_network");
      interfaces =
	cct->_conf.get_val<std::string>("public_network_interface");
    }
  }
  if (addr.is_blank_ip() &&
      !networks.empty()) {
    int ipv4_r = !(ipv & CEPH_PICK_ADDRESS_IPV4) ? 0 : -1;
    int ipv6_r = !(ipv & CEPH_PICK_ADDRESS_IPV6) ? 0 : -1;
    // first try on preferred numa node (if >= 0), then anywhere.
    while (true) {
      // note: pass in ipv to filter the matching addresses
      if ((ipv & CEPH_PICK_ADDRESS_IPV4) &&
	  (flags & CEPH_PICK_ADDRESS_PREFER_IPV4)) {
	ipv4_r = fill_in_one_address(cct, ifa, CEPH_PICK_ADDRESS_IPV4,
                                     networks, interfaces, addrs,
                                     preferred_numa_node);
      }
      if (ipv & CEPH_PICK_ADDRESS_IPV6) {
	ipv6_r = fill_in_one_address(cct, ifa, CEPH_PICK_ADDRESS_IPV6,
                                     networks, interfaces, addrs,
                                     preferred_numa_node);
      }
      if ((ipv & CEPH_PICK_ADDRESS_IPV4) &&
	  !(flags & CEPH_PICK_ADDRESS_PREFER_IPV4)) {
	ipv4_r = fill_in_one_address(cct, ifa, CEPH_PICK_ADDRESS_IPV4,
                                     networks, interfaces, addrs,
                                     preferred_numa_node);
      }
      if (ipv4_r >= 0 && ipv6_r >= 0) {
	break;
      }
      if (preferred_numa_node < 0) {
	return ipv4_r >= 0 && ipv6_r >= 0 ? 0 : -1;
      }
      preferred_numa_node = -1;      // try any numa node
    }
  }

  // note: we may have a blank addr here

  // ipv4 and/or ipv6?
  if (addrs->v.empty()) {
    addr.set_type(entity_addr_t::TYPE_MSGR2);
    if ((ipv & CEPH_PICK_ADDRESS_IPV4) &&
	(flags & CEPH_PICK_ADDRESS_PREFER_IPV4)) {
      addr.set_family(AF_INET);
      addrs->v.push_back(addr);
    }
    if (ipv & CEPH_PICK_ADDRESS_IPV6) {
      addr.set_family(AF_INET6);
      addrs->v.push_back(addr);
    }
    if ((ipv & CEPH_PICK_ADDRESS_IPV4) &&
	!(flags & CEPH_PICK_ADDRESS_PREFER_IPV4)) {
      addr.set_family(AF_INET);
      addrs->v.push_back(addr);
    }
  }

  // msgr2 or legacy or both?
  if (msgrv == (CEPH_PICK_ADDRESS_MSGR1 | CEPH_PICK_ADDRESS_MSGR2)) {
    vector<entity_addr_t> v;
    v.swap(addrs->v);
    for (auto a : v) {
      a.set_type(entity_addr_t::TYPE_MSGR2);
      if (flags & CEPH_PICK_ADDRESS_DEFAULT_MON_PORTS) {
	a.set_port(CEPH_MON_PORT_IANA);
      }
      addrs->v.push_back(a);
      a.set_type(entity_addr_t::TYPE_LEGACY);
      if (flags & CEPH_PICK_ADDRESS_DEFAULT_MON_PORTS) {
	a.set_port(CEPH_MON_PORT_LEGACY);
      }
      addrs->v.push_back(a);
    }
  } else if (msgrv == CEPH_PICK_ADDRESS_MSGR1) {
    for (auto& a : addrs->v) {
      a.set_type(entity_addr_t::TYPE_LEGACY);
    }
  } else {
    for (auto& a : addrs->v) {
      a.set_type(entity_addr_t::TYPE_MSGR2);
    }
  }

  return 0;
}

int pick_addresses(
  CephContext *cct,
  unsigned flags,
  entity_addrvec_t *addrs,
  int preferred_numa_node)
{
  struct ifaddrs *ifa;
  int r = getifaddrs(&ifa);
  if (r < 0) {
    r = -errno;
    string err = cpp_strerror(r);
    lderr(cct) << "unable to fetch interfaces and addresses: "
	       <<  cpp_strerror(r) << dendl;
    return r;
  }
  r = pick_addresses(cct, flags, ifa, addrs, preferred_numa_node);
  freeifaddrs(ifa);
  return r;
}

std::string pick_iface(CephContext *cct, const struct sockaddr_storage &network)
{
  struct ifaddrs *ifa;
  int r = getifaddrs(&ifa);
  if (r < 0) {
    string err = cpp_strerror(errno);
    lderr(cct) << "unable to fetch interfaces and addresses: " << err << dendl;
    return {};
  }

  const unsigned int prefix_len = max(sizeof(in_addr::s_addr), sizeof(in6_addr::s6_addr)) * CHAR_BIT;
  const struct ifaddrs *found = find_ip_in_subnet(
    ifa,
    (const struct sockaddr *) &network, prefix_len);

  std::string result;
  if (found) {
    result = found->ifa_name;
  }

  freeifaddrs(ifa);

  return result;
}


bool have_local_addr(CephContext *cct, const list<entity_addr_t>& ls, entity_addr_t *match)
{
  struct ifaddrs *ifa;
  int r = getifaddrs(&ifa);
  if (r < 0) {
    lderr(cct) << "unable to fetch interfaces and addresses: " << cpp_strerror(errno) << dendl;
    exit(1);
  }

  bool found = false;
  for (struct ifaddrs *addrs = ifa; addrs != nullptr; addrs = addrs->ifa_next) {
    if (addrs->ifa_addr) {
      entity_addr_t a;
      a.set_sockaddr(addrs->ifa_addr);
      for (auto& p : ls) {
        if (a.is_same_host(p)) {
          *match = p;
          found = true;
          goto out;
        }
      }
    }
  }

 out:
  freeifaddrs(ifa);
  return found;
}

int get_iface_numa_node(
  const std::string& iface,
  int *node)
{
  string fn = std::string("/sys/class/net/") + iface + "/device/numa_node";

  int r = 0;
  char buf[1024];
  char *endptr = 0;
  int fd = ::open(fn.c_str(), O_RDONLY);
  if (fd < 0) {
    return -errno;
  }
  r = safe_read(fd, &buf, sizeof(buf));
  if (r < 0) {
    goto out;
  }
  buf[r] = 0;
  while (r > 0 && ::isspace(buf[--r])) {
    buf[r] = 0;
  }
  *node = strtoll(buf, &endptr, 10);
  if (endptr != buf + strlen(buf)) {
    r = -EINVAL;
    goto out;
  }
  r = 0;
 out:
  ::close(fd);
  return r;
}
Commit	Line	Data
7c673cae FG	1	// -- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t --
	2	// vim: ts=8 sw=2 smarttab
	3	/*
	4	* Ceph - scalable distributed file system
	5	*
	6	* Copyright (C) 2004-2012 Inktank
	7	*
	8	* This is free software; you can redistribute it and/or
	9	* modify it under the terms of the GNU Lesser General Public
	10	* License version 2.1, as published by the Free Software
	11	* Foundation. See file COPYING.
	12	*
	13	*/
	14
	15	#include "common/pick_address.h"
7c673cae FG	16	#include "include/ipaddr.h"
7c673cae FG	17	#include "include/str_list.h"
11fdf7f2 TL	18	#include "common/ceph_context.h"
	19	#ifndef WITH_SEASTAR
	20	#include "common/config.h"
	21	#include "common/config_obs.h"
	22	#endif
7c673cae FG	23	#include "common/debug.h"
7c673cae FG	24	#include "common/errno.h"
11fdf7f2	25	#include "common/numa.h"
7c673cae	26
31f18b77 FG	27	#include <netdb.h>
31f18b77 FG	28
7c673cae FG	29	#define dout_subsys ceph_subsys_
7c673cae FG	30
3efd9988 FG	31	const struct sockaddr *find_ip_in_subnet_list(
	32	CephContext *cct,
	33	const struct ifaddrs *ifa,
11fdf7f2	34	unsigned ipv,
3efd9988	35	const std::string &networks,
11fdf7f2 TL	36	const std::string &interfaces,
11fdf7f2 TL	37	int numa_node)
7c673cae FG	38	{
	39	std::list<string> nets;
	40	get_str_list(networks, nets);
3efd9988 FG	41	std::list<string> ifs;
	42	get_str_list(interfaces, ifs);
	43
	44	// filter interfaces by name
11fdf7f2	45	const struct ifaddrs *filtered = nullptr;
3efd9988 FG	46	if (ifs.empty()) {
	47	filtered = ifa;
	48	} else {
	49	if (nets.empty()) {
	50	lderr(cct) << "interface names specified but not network names" << dendl;
	51	exit(1);
	52	}
	53	const struct ifaddrs *t = ifa;
	54	struct ifaddrs *head = 0;
11fdf7f2	55	while (t) {
3efd9988 FG	56	bool match = false;
	57	for (auto& i : ifs) {
	58	if (strcmp(i.c_str(), t->ifa_name) == 0) {
	59	match = true;
	60	break;
	61	}
	62	}
	63	if (match) {
	64	struct ifaddrs *n = new ifaddrs;
	65	memcpy(n, t, sizeof(*t));
	66	n->ifa_next = head;
	67	head = n;
	68	}
	69	t = t->ifa_next;
	70	}
11fdf7f2	71	if (!head) {
3efd9988 FG	72	lderr(cct) << "no interfaces matching " << ifs << dendl;
	73	exit(1);
	74	}
	75	filtered = head;
	76	}
7c673cae	77
11fdf7f2 TL	78	struct sockaddr *r = nullptr;
11fdf7f2 TL	79	for (auto& s : nets) {
3efd9988 FG	80	struct sockaddr_storage net;
3efd9988 FG	81	unsigned int prefix_len;
7c673cae	82
11fdf7f2 TL	83	if (!parse_network(s.c_str(), &net, &prefix_len)) {
11fdf7f2 TL	84	lderr(cct) << "unable to parse network: " << s << dendl;
3efd9988 FG	85	exit(1);
	86	}
	87
11fdf7f2 TL	88	switch (net.ss_family) {
	89	case AF_INET:
	90	if (!(ipv & CEPH_PICK_ADDRESS_IPV4)) {
	91	continue;
	92	}
	93	break;
	94	case AF_INET6:
	95	if (!(ipv & CEPH_PICK_ADDRESS_IPV6)) {
	96	continue;
	97	}
	98	break;
	99	}
	100
3efd9988 FG	101	const struct ifaddrs *found = find_ip_in_subnet(
3efd9988 FG	102	filtered,
11fdf7f2	103	(struct sockaddr *) &net, prefix_len, numa_node);
3efd9988 FG	104	if (found) {
	105	r = found->ifa_addr;
	106	break;
	107	}
	108	}
7c673cae	109
3efd9988 FG	110	if (filtered != ifa) {
	111	while (filtered) {
	112	struct ifaddrs *t = filtered->ifa_next;
	113	delete filtered;
	114	filtered = t;
7c673cae	115	}
3efd9988	116	}
7c673cae	117
3efd9988	118	return r;
7c673cae FG	119	}
7c673cae FG	120
11fdf7f2	121	#ifndef WITH_SEASTAR
7c673cae FG	122	// observe this change
	123	struct Observer : public md_config_obs_t {
	124	const char *keys[2];
	125	explicit Observer(const char *c) {
	126	keys[0] = c;
	127	keys[1] = NULL;
	128	}
	129
	130	const char** get_tracked_conf_keys() const override {
	131	return (const char **)keys;
	132	}
11fdf7f2	133	void handle_conf_change(const ConfigProxy& conf,
7c673cae FG	134	const std::set <std::string> &changed) override {
	135	// do nothing.
	136	}
	137	};
	138
	139	static void fill_in_one_address(CephContext *cct,
	140	const struct ifaddrs *ifa,
	141	const string networks,
3efd9988	142	const string interfaces,
11fdf7f2 TL	143	const char *conf_var,
11fdf7f2 TL	144	int numa_node = -1)
7c673cae	145	{
11fdf7f2 TL	146	const struct sockaddr *found = find_ip_in_subnet_list(
	147	cct,
	148	ifa,
	149	CEPH_PICK_ADDRESS_IPV4\|CEPH_PICK_ADDRESS_IPV6,
	150	networks,
	151	interfaces,
	152	numa_node);
7c673cae	153	if (!found) {
3efd9988 FG	154	lderr(cct) << "unable to find any IP address in networks '" << networks
3efd9988 FG	155	<< "' interfaces '" << interfaces << "'" << dendl;
7c673cae FG	156	exit(1);
	157	}
	158
	159	char buf[INET6_ADDRSTRLEN];
	160	int err;
	161
	162	err = getnameinfo(found,
	163	(found->sa_family == AF_INET)
	164	? sizeof(struct sockaddr_in)
	165	: sizeof(struct sockaddr_in6),
	166
	167	buf, sizeof(buf),
11fdf7f2	168	nullptr, 0,
7c673cae FG	169	NI_NUMERICHOST);
	170	if (err != 0) {
	171	lderr(cct) << "unable to convert chosen address to string: " << gai_strerror(err) << dendl;
	172	exit(1);
	173	}
	174
	175	Observer obs(conf_var);
	176
11fdf7f2	177	cct->_conf.add_observer(&obs);
7c673cae	178
11fdf7f2 TL	179	cct->_conf.set_val_or_die(conf_var, buf);
11fdf7f2 TL	180	cct->_conf.apply_changes(nullptr);
7c673cae	181
11fdf7f2	182	cct->_conf.remove_observer(&obs);
7c673cae FG	183	}
	184
	185	void pick_addresses(CephContext *cct, int needs)
	186	{
	187	struct ifaddrs *ifa;
	188	int r = getifaddrs(&ifa);
11fdf7f2 TL	189	auto public_addr = cct->_conf.get_val<entity_addr_t>("public_addr");
	190	auto public_network = cct->_conf.get_val<std::string>("public_network");
	191	auto public_network_interface =
	192	cct->_conf.get_val<std::string>("public_network_interface");
	193	auto cluster_addr = cct->_conf.get_val<entity_addr_t>("cluster_addr");
	194	auto cluster_network = cct->_conf.get_val<std::string>("cluster_network");
	195	auto cluster_network_interface =
	196	cct->_conf.get_val<std::string>("cluster_network_interface");
	197
	198	if (r < 0) {
7c673cae FG	199	string err = cpp_strerror(errno);
	200	lderr(cct) << "unable to fetch interfaces and addresses: " << err << dendl;
	201	exit(1);
	202	}
	203
11fdf7f2 TL	204	if ((needs & CEPH_PICK_ADDRESS_PUBLIC) &&
	205	public_addr.is_blank_ip() && !public_network.empty()) {
	206	fill_in_one_address(cct, ifa, public_network, public_network_interface,
	207	"public_addr");
7c673cae FG	208	}
7c673cae FG	209
11fdf7f2 TL	210	if ((needs & CEPH_PICK_ADDRESS_CLUSTER) && cluster_addr.is_blank_ip()) {
	211	if (!cluster_network.empty()) {
	212	fill_in_one_address(cct, ifa, cluster_network, cluster_network_interface,
3efd9988	213	"cluster_addr");
7c673cae	214	} else {
11fdf7f2	215	if (!public_network.empty()) {
7c673cae FG	216	lderr(cct) << "Public network was set, but cluster network was not set " << dendl;
7c673cae FG	217	lderr(cct) << " Using public network also for cluster network" << dendl;
11fdf7f2 TL	218	fill_in_one_address(cct, ifa, public_network, public_network_interface,
11fdf7f2 TL	219	"cluster_addr");
7c673cae FG	220	}
	221	}
	222	}
	223
	224	freeifaddrs(ifa);
	225	}
11fdf7f2 TL	226	#endif // !WITH_SEASTAR
	227
	228	static int fill_in_one_address(
	229	CephContext *cct,
	230	const struct ifaddrs *ifa,
	231	unsigned ipv,
	232	const string networks,
	233	const string interfaces,
	234	entity_addrvec_t *addrs,
	235	int numa_node = -1)
	236	{
	237	const struct sockaddr *found = find_ip_in_subnet_list(cct, ifa, ipv, networks,
	238	interfaces, numa_node);
	239	if (!found) {
	240	std::string ip_type = "";
	241	if ((ipv & CEPH_PICK_ADDRESS_IPV4) && (ipv & CEPH_PICK_ADDRESS_IPV6)) {
	242	ip_type = "IPv4 or IPv6";
	243	} else if (ipv & CEPH_PICK_ADDRESS_IPV4) {
	244	ip_type = "IPv4";
	245	} else {
	246	ip_type = "IPv6";
	247	}
	248	lderr(cct) << "unable to find any " << ip_type << " address in networks '"
	249	<< networks << "' interfaces '" << interfaces << "'" << dendl;
	250	return -1;
	251	}
	252
	253	char buf[INET6_ADDRSTRLEN];
	254	int err;
7c673cae	255
11fdf7f2 TL	256	err = getnameinfo(found,
	257	(found->sa_family == AF_INET)
	258	? sizeof(struct sockaddr_in)
	259	: sizeof(struct sockaddr_in6),
	260
	261	buf, sizeof(buf),
	262	nullptr, 0,
	263	NI_NUMERICHOST);
	264	if (err != 0) {
	265	lderr(cct) << "unable to convert chosen address to string: " << gai_strerror(err) << dendl;
	266	return -1;
	267	}
	268
	269	entity_addr_t addr;
	270	const char *end = 0;
	271	bool r = addr.parse(buf, &end);
	272	if (!r) {
	273	return -1;
	274	}
	275	addrs->v.push_back(addr);
	276	return 0;
	277	}
	278
	279	int pick_addresses(
	280	CephContext *cct,
	281	unsigned flags,
	282	struct ifaddrs *ifa,
	283	entity_addrvec_t *addrs,
	284	int preferred_numa_node)
	285	{
	286	addrs->v.clear();
	287
	288	unsigned addrt = (flags & (CEPH_PICK_ADDRESS_PUBLIC \|
	289	CEPH_PICK_ADDRESS_CLUSTER));
	290	if (addrt == 0 \|\|
	291	addrt == (CEPH_PICK_ADDRESS_PUBLIC \|
	292	CEPH_PICK_ADDRESS_CLUSTER)) {
	293	return -EINVAL;
	294	}
	295	unsigned msgrv = flags & (CEPH_PICK_ADDRESS_MSGR1 \|
	296	CEPH_PICK_ADDRESS_MSGR2);
	297	if (msgrv == 0) {
	298	if (cct->_conf.get_val<bool>("ms_bind_msgr1")) {
	299	msgrv \|= CEPH_PICK_ADDRESS_MSGR1;
	300	}
	301	if (cct->_conf.get_val<bool>("ms_bind_msgr2")) {
	302	msgrv \|= CEPH_PICK_ADDRESS_MSGR2;
	303	}
	304	if (msgrv == 0) {
	305	return -EINVAL;
	306	}
	307	}
	308	unsigned ipv = flags & (CEPH_PICK_ADDRESS_IPV4 \|
	309	CEPH_PICK_ADDRESS_IPV6);
	310	if (ipv == 0) {
	311	if (cct->_conf.get_val<bool>("ms_bind_ipv4")) {
	312	ipv \|= CEPH_PICK_ADDRESS_IPV4;
	313	}
	314	if (cct->_conf.get_val<bool>("ms_bind_ipv6")) {
	315	ipv \|= CEPH_PICK_ADDRESS_IPV6;
	316	}
	317	if (ipv == 0) {
	318	return -EINVAL;
	319	}
320	if (cct->_conf.get_val<bool>("ms_bind_prefer_ipv4")) {
321	flags \|= CEPH_PICK_ADDRESS_PREFER_IPV4;
322	} else {
323	flags &= ~CEPH_PICK_ADDRESS_PREFER_IPV4;
324	}
325	}
326
327	entity_addr_t addr;
328	string networks;
329	string interfaces;
330	if (addrt & CEPH_PICK_ADDRESS_PUBLIC) {
331	addr = cct->_conf.get_val<entity_addr_t>("public_addr");
332	networks = cct->_conf.get_val<std::string>("public_network");
333	interfaces =
334	cct->_conf.get_val<std::string>("public_network_interface");
335	} else {
336	addr = cct->_conf.get_val<entity_addr_t>("cluster_addr");
337	networks = cct->_conf.get_val<std::string>("cluster_network");
338	interfaces =
339	cct->_conf.get_val<std::string>("cluster_network_interface");
340	if (networks.empty()) {
341	lderr(cct) << "Falling back to public interface" << dendl;
342	// fall back to public_ network and interface if cluster is not set
343	networks = cct->_conf.get_val<std::string>("public_network");
344	interfaces =
345	cct->_conf.get_val<std::string>("public_network_interface");
346	}
347	}
348	if (addr.is_blank_ip() &&
349	!networks.empty()) {
350	int ipv4_r = !(ipv & CEPH_PICK_ADDRESS_IPV4) ? 0 : -1;
351	int ipv6_r = !(ipv & CEPH_PICK_ADDRESS_IPV6) ? 0 : -1;
352	// first try on preferred numa node (if >= 0), then anywhere.
353	while (true) {
354	// note: pass in ipv to filter the matching addresses
355	if ((ipv & CEPH_PICK_ADDRESS_IPV4) &&
356	(flags & CEPH_PICK_ADDRESS_PREFER_IPV4)) {
357	ipv4_r = fill_in_one_address(cct, ifa, CEPH_PICK_ADDRESS_IPV4,
358	networks, interfaces, addrs,
359	preferred_numa_node);
360	}
361	if (ipv & CEPH_PICK_ADDRESS_IPV6) {
362	ipv6_r = fill_in_one_address(cct, ifa, CEPH_PICK_ADDRESS_IPV6,
363	networks, interfaces, addrs,
364	preferred_numa_node);
365	}
366	if ((ipv & CEPH_PICK_ADDRESS_IPV4) &&
367	!(flags & CEPH_PICK_ADDRESS_PREFER_IPV4)) {
368	ipv4_r = fill_in_one_address(cct, ifa, CEPH_PICK_ADDRESS_IPV4,
369	networks, interfaces, addrs,
370	preferred_numa_node);
371	}
372	if (ipv4_r >= 0 && ipv6_r >= 0) {
373	break;
374	}
375	if (preferred_numa_node < 0) {
376	return ipv4_r >= 0 && ipv6_r >= 0 ? 0 : -1;
377	}
378	preferred_numa_node = -1; // try any numa node
379	}
380	}
381
382	// note: we may have a blank addr here
383
384	// ipv4 and/or ipv6?
385	if (addrs->v.empty()) {
11fdf7f2 TL	386	addr.set_type(entity_addr_t::TYPE_MSGR2);
	387	if ((ipv & CEPH_PICK_ADDRESS_IPV4) &&
	388	(flags & CEPH_PICK_ADDRESS_PREFER_IPV4)) {
	389	addr.set_family(AF_INET);
	390	addrs->v.push_back(addr);
	391	}
	392	if (ipv & CEPH_PICK_ADDRESS_IPV6) {
	393	addr.set_family(AF_INET6);
	394	addrs->v.push_back(addr);
	395	}
	396	if ((ipv & CEPH_PICK_ADDRESS_IPV4) &&
	397	!(flags & CEPH_PICK_ADDRESS_PREFER_IPV4)) {
	398	addr.set_family(AF_INET);
	399	addrs->v.push_back(addr);
	400	}
	401	}
	402
	403	// msgr2 or legacy or both?
	404	if (msgrv == (CEPH_PICK_ADDRESS_MSGR1 \| CEPH_PICK_ADDRESS_MSGR2)) {
	405	vector<entity_addr_t> v;
	406	v.swap(addrs->v);
	407	for (auto a : v) {
	408	a.set_type(entity_addr_t::TYPE_MSGR2);
	409	if (flags & CEPH_PICK_ADDRESS_DEFAULT_MON_PORTS) {
	410	a.set_port(CEPH_MON_PORT_IANA);
	411	}
	412	addrs->v.push_back(a);
	413	a.set_type(entity_addr_t::TYPE_LEGACY);
	414	if (flags & CEPH_PICK_ADDRESS_DEFAULT_MON_PORTS) {
	415	a.set_port(CEPH_MON_PORT_LEGACY);
	416	}
	417	addrs->v.push_back(a);
	418	}
	419	} else if (msgrv == CEPH_PICK_ADDRESS_MSGR1) {
	420	for (auto& a : addrs->v) {
	421	a.set_type(entity_addr_t::TYPE_LEGACY);
	422	}
	423	} else {
	424	for (auto& a : addrs->v) {
	425	a.set_type(entity_addr_t::TYPE_MSGR2);
	426	}
	427	}
	428
	429	return 0;
	430	}
	431
	432	int pick_addresses(
	433	CephContext *cct,
	434	unsigned flags,
	435	entity_addrvec_t *addrs,
	436	int preferred_numa_node)
	437	{
	438	struct ifaddrs *ifa;
	439	int r = getifaddrs(&ifa);
	440	if (r < 0) {
	441	r = -errno;
	442	string err = cpp_strerror(r);
	443	lderr(cct) << "unable to fetch interfaces and addresses: "
	444	<< cpp_strerror(r) << dendl;
	445	return r;
	446	}
	447	r = pick_addresses(cct, flags, ifa, addrs, preferred_numa_node);
	448	freeifaddrs(ifa);
	449	return r;
450	}
b5b8bbf5 FG	451
	452	std::string pick_iface(CephContext *cct, const struct sockaddr_storage &network)
	453	{
	454	struct ifaddrs *ifa;
	455	int r = getifaddrs(&ifa);
	456	if (r < 0) {
	457	string err = cpp_strerror(errno);
	458	lderr(cct) << "unable to fetch interfaces and addresses: " << err << dendl;
	459	return {};
	460	}
	461
b32b8144	462	const unsigned int prefix_len = max(sizeof(in_addr::s_addr), sizeof(in6_addr::s6_addr)) * CHAR_BIT;
11fdf7f2 TL	463	const struct ifaddrs *found = find_ip_in_subnet(
	464	ifa,
	465	(const struct sockaddr *) &network, prefix_len);
b5b8bbf5 FG	466
	467	std::string result;
	468	if (found) {
	469	result = found->ifa_name;
	470	}
	471
	472	freeifaddrs(ifa);
	473
	474	return result;
	475	}
	476
	477
7c673cae FG	478	bool have_local_addr(CephContext cct, const list<entity_addr_t>& ls, entity_addr_t match)
	479	{
	480	struct ifaddrs *ifa;
	481	int r = getifaddrs(&ifa);
	482	if (r < 0) {
	483	lderr(cct) << "unable to fetch interfaces and addresses: " << cpp_strerror(errno) << dendl;
	484	exit(1);
	485	}
	486
	487	bool found = false;
11fdf7f2	488	for (struct ifaddrs *addrs = ifa; addrs != nullptr; addrs = addrs->ifa_next) {
7c673cae FG	489	if (addrs->ifa_addr) {
	490	entity_addr_t a;
	491	a.set_sockaddr(addrs->ifa_addr);
11fdf7f2 TL	492	for (auto& p : ls) {
	493	if (a.is_same_host(p)) {
	494	*match = p;
7c673cae FG	495	found = true;
	496	goto out;
	497	}
	498	}
	499	}
	500	}
	501
	502	out:
	503	freeifaddrs(ifa);
	504	return found;
	505	}
11fdf7f2 TL	506
	507	int get_iface_numa_node(
	508	const std::string& iface,
	509	int *node)
	510	{
	511	string fn = std::string("/sys/class/net/") + iface + "/device/numa_node";
	512
	513	int r = 0;
	514	char buf[1024];
	515	char *endptr = 0;
	516	int fd = ::open(fn.c_str(), O_RDONLY);
	517	if (fd < 0) {
	518	return -errno;
	519	}
	520	r = safe_read(fd, &buf, sizeof(buf));
	521	if (r < 0) {
	522	goto out;
	523	}
	524	buf[r] = 0;
	525	while (r > 0 && ::isspace(buf[--r])) {
	526	buf[r] = 0;
	527	}
	528	*node = strtoll(buf, &endptr, 10);
	529	if (endptr != buf + strlen(buf)) {
	530	r = -EINVAL;
	531	goto out;
	532	}
	533	r = 0;
	534	out:
	535	::close(fd);
	536	return r;
	537	}