1
1

Merge pull request #5938 from bwbarrett/bugfix/tcp-multi-address

Always use same IP address for module and modex
Этот коммит содержится в:
bosilca 2018-10-21 09:19:43 -07:00 коммит произвёл GitHub
родитель a66373454e 457f058e73
Коммит 690024917d
Не найден ключ, соответствующий данной подписи
Идентификатор ключа GPG: 4AEE18F83AFDEB23
5 изменённых файлов: 228 добавлений и 185 удалений

Просмотреть файл

@ -164,13 +164,11 @@ OPAL_MODULE_DECLSPEC extern mca_btl_tcp_component_t mca_btl_tcp_component;
struct mca_btl_tcp_module_t {
mca_btl_base_module_t super; /**< base BTL interface */
uint16_t tcp_ifkindex; /** <BTL kernel interface index */
#if 0
int tcp_ifindex; /**< BTL interface index */
#endif
struct sockaddr_storage tcp_ifaddr; /**< First IPv4 address discovered for this interface, bound as sending address for this BTL */
#if OPAL_ENABLE_IPV6
struct sockaddr_storage tcp_ifaddr_6; /**< First IPv6 address discovered for this interface, bound as sending address for this BTL */
#endif
struct sockaddr_storage tcp_ifaddr; /**< First address
discovered for this
interface, bound as
sending address for this
BTL */
uint32_t tcp_ifmask; /**< BTL interface netmask */
opal_mutex_t tcp_endpoints_mutex;

Просмотреть файл

@ -33,40 +33,52 @@
/**
* Structure used to publish TCP connection information to peers.
* Modex address structure.
*
* One of these structures will be sent for every btl module in use by
* the local BTL TCP component.
*/
struct mca_btl_tcp_modex_addr_t {
uint8_t addr[16]; /* endpoint address. for addr_family
of MCA_BTL_TCP_AF_INET, only the
first 4 bytes have meaning. */
uint32_t addr_ifkindex; /* endpoint kernel index */
uint16_t addr_port; /* endpoint listen port */
uint8_t addr_family; /* endpoint address family. Note that
this is
MCA_BTL_TCP_AF_{INET,INET6}, not
the traditional
AF_INET/AF_INET6. */
uint8_t padding[1]; /* padd out to an 8-byte word */
};
typedef struct mca_btl_tcp_modex_addr_t mca_btl_tcp_modex_addr_t;
/**
* Remote peer address structure
*
* One of these structures will be allocated for every remote endpoint
* associated with a remote proc. The data is pulled from the
* mca_btl_tcp_modex_addr_t structure, except for the addr_inuse
* field, which is local.
*/
struct mca_btl_tcp_addr_t {
/* the following information is exchanged between different
machines (read: byte order), so use network byte order
for everything and don't add padding
*/
union {
struct in_addr addr_inet; /* IPv6 listen address */
#if OPAL_ENABLE_IPV6
struct in6_addr addr_inet; /**< IPv4/IPv6 listen address > */
#else
/* Bug, FIXME: needs testing */
struct my_in6_addr {
union {
uint32_t u6_addr32[4];
struct _my_in6_addr {
struct in_addr _addr_inet;
uint32_t _pad[3];
} _addr__inet;
} _union_inet;
} addr_inet;
struct in6_addr addr_inet6; /* IPv6 listen address */
#endif
in_port_t addr_port; /**< listen port */
uint16_t addr_ifkindex; /**< remote interface index assigned with
};
in_port_t addr_port; /**< listen port */
int addr_ifkindex; /**< remote interface index assigned with
this address */
unsigned short addr_inuse; /**< local meaning only */
uint8_t addr_family; /**< AF_INET or AF_INET6 */
uint8_t addr_family; /**< AF_INET or AF_INET6 */
bool addr_inuse; /**< local meaning only */
};
typedef struct mca_btl_tcp_addr_t mca_btl_tcp_addr_t;
#define MCA_BTL_TCP_AF_INET 0
#if OPAL_ENABLE_IPV6
# define MCA_BTL_TCP_AF_INET6 1
#endif
#define MCA_BTL_TCP_AF_INET6 1
#endif

Просмотреть файл

@ -490,11 +490,59 @@ static int mca_btl_tcp_component_close(void)
* Create a btl instance and add to modules list.
*/
static int mca_btl_tcp_create(int if_kindex, const char* if_name)
static int mca_btl_tcp_create(const int if_kindex, const char* if_name)
{
struct mca_btl_tcp_module_t* btl;
char param[256];
int i;
struct sockaddr_storage addr;
bool found = false;
/*
* Look for an address on the given device (ie, kindex) which
* isn't disabled by the disable_family option. If there isn't
* one, skip creating the modules for this interface. We store
* the address on the module both to publish in the modex and to
* use as the source address of all packets sent by this module.
*
* This still isn't quite right. Better would be to pull apart
* split_and_resolve and pass the address used to select the
* device into mca_btl_tcp_create(). This is a cleanup of the
* logic that's been in use for years, but the case it doesn't
* cover is (say) only specifing mca_btl_if_include 10.0.0.0/16
* when the interface has addresses of both 10.0.0.1 and 10.1.0.1;
* there's absolutely nothing that keeps this code from picking
* 10.1.0.1 as the one that is published in the modex and used for
* connection.
*/
for (i = opal_ifbegin() ; i >= 0 ; i = opal_ifnext(i)) {
int ret;
if (if_kindex != opal_ifindextokindex(i)) {
continue;
}
ret = opal_ifindextoaddr(i, (struct sockaddr*)&addr,
sizeof(struct sockaddr_storage));
if (OPAL_SUCCESS != ret) {
return ret;
}
if (addr.ss_family == AF_INET &&
4 != mca_btl_tcp_component.tcp_disable_family) {
found = true;
break;
} else if (addr.ss_family == AF_INET6 &&
6 != mca_btl_tcp_component.tcp_disable_family) {
found = true;
break;
}
}
/* if we didn't find an address that works for us on this
interface, just move on. */
if (!found) {
return OPAL_SUCCESS;
}
for( i = 0; i < (int)mca_btl_tcp_component.tcp_num_links; i++ ) {
btl = (struct mca_btl_tcp_module_t *)malloc(sizeof(mca_btl_tcp_module_t));
@ -513,17 +561,8 @@ static int mca_btl_tcp_create(int if_kindex, const char* if_name)
btl->tcp_send_handler = 0;
#endif
struct sockaddr_storage addr;
opal_ifkindextoaddr(if_kindex, (struct sockaddr*) &addr,
sizeof (struct sockaddr_storage));
#if OPAL_ENABLE_IPV6
if (addr.ss_family == AF_INET6) {
btl->tcp_ifaddr_6 = addr;
}
#endif
if (addr.ss_family == AF_INET) {
btl->tcp_ifaddr = addr;
}
memcpy(&btl->tcp_ifaddr, &addr, sizeof(struct sockaddr_storage));
/* allow user to specify interface bandwidth */
sprintf(param, "bandwidth_%s", if_name);
mca_btl_tcp_param_register_uint(param, NULL, btl->super.btl_bandwidth, OPAL_INFO_LVL_5, &btl->super.btl_bandwidth);
@ -567,8 +606,8 @@ static int mca_btl_tcp_create(int if_kindex, const char* if_name)
opal_output_verbose(5, opal_btl_base_framework.framework_output,
"btl:tcp: %p: if %s kidx %d cnt %i addr %s %s bw %d lt %d\n",
(void*)btl, if_name, (int) btl->tcp_ifkindex, i,
opal_net_get_hostname((struct sockaddr*)&addr),
(addr.ss_family == AF_INET) ? "IPv4" : "IPv6",
opal_net_get_hostname((struct sockaddr*)&btl->tcp_ifaddr),
(btl->tcp_ifaddr.ss_family == AF_INET) ? "IPv4" : "IPv6",
btl->super.btl_bandwidth, btl->super.btl_latency);
}
return OPAL_SUCCESS;
@ -721,6 +760,7 @@ static int mca_btl_tcp_component_create_instances(void)
*/
for(if_index = opal_ifbegin(); if_index >= 0; if_index = opal_ifnext(if_index)){
int index = opal_ifindextokindex (if_index);
if (index > 0) {
bool want_this_if = true;
@ -1121,97 +1161,64 @@ static int mca_btl_tcp_component_create_listen(uint16_t af_family)
static int mca_btl_tcp_component_exchange(void)
{
int rc = 0, index;
size_t i = 0;
size_t size = mca_btl_tcp_component.tcp_addr_count *
mca_btl_tcp_component.tcp_num_links * sizeof(mca_btl_tcp_addr_t);
/* adi@2007-04-12:
*
* We'll need to explain things a bit here:
* 1. We normally have as many BTLs as physical NICs.
* 2. With num_links, we now have num_btl = num_links * #NICs
* 3. we might have more than one address per NIC
*/
size_t xfer_size = 0; /* real size to transfer (may differ from 'size') */
size_t current_addr = 0;
int rc;
size_t i;
size_t num_btls = mca_btl_tcp_component.tcp_num_btls;
size_t size = num_btls * sizeof(mca_btl_tcp_modex_addr_t);
mca_btl_tcp_modex_addr_t *addrs;
if(mca_btl_tcp_component.tcp_num_btls != 0) {
char ifn[32];
mca_btl_tcp_addr_t *addrs = (mca_btl_tcp_addr_t *)malloc(size);
memset(addrs, 0, size);
if (num_btls <= 0) {
return 0;
}
/* here we start populating our addresses */
for( i = 0; i < mca_btl_tcp_component.tcp_num_btls; i++ ) {
for (index = opal_ifbegin(); index >= 0;
index = opal_ifnext(index)) {
struct sockaddr_storage my_ss;
addrs = (mca_btl_tcp_modex_addr_t*)malloc(size);
if (NULL == addrs) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
memset(addrs, 0, size);
/* Look if the module's address belongs to this
* kernel IP interface. If not, go to next address.
*/
if (opal_ifindextokindex (index) !=
mca_btl_tcp_component.tcp_btls[i]->tcp_ifkindex) {
continue;
}
opal_ifindextoname(index, ifn, sizeof(ifn));
opal_output_verbose(30, opal_btl_base_framework.framework_output,
"btl: tcp: component_exchange: examining interface %s",
ifn);
if (OPAL_SUCCESS !=
opal_ifindextoaddr(index, (struct sockaddr*) &my_ss,
sizeof (my_ss))) {
opal_output (0,
"btl_tcp_component: problems getting address for index %i (kernel index %i)\n",
index, opal_ifindextokindex (index));
continue;
}
for (i = 0 ; i < num_btls ; i++) {
struct mca_btl_tcp_module_t *btl = mca_btl_tcp_component.tcp_btls[i];
struct sockaddr *addr = (struct sockaddr*)&(btl->tcp_ifaddr);
#if OPAL_ENABLE_IPV6
if ((AF_INET6 == my_ss.ss_family) &&
(6 != mca_btl_tcp_component.tcp_disable_family)) {
memcpy(&addrs[current_addr].addr_inet,
&((struct sockaddr_in6*)&my_ss)->sin6_addr,
sizeof(addrs[0].addr_inet));
addrs[current_addr].addr_port =
mca_btl_tcp_component.tcp6_listen_port;
addrs[current_addr].addr_family = MCA_BTL_TCP_AF_INET6;
xfer_size += sizeof (mca_btl_tcp_addr_t);
addrs[current_addr].addr_inuse = 0;
addrs[current_addr].addr_ifkindex =
opal_ifindextokindex (index);
current_addr++;
opal_output_verbose(30, opal_btl_base_framework.framework_output,
"btl: tcp: component_exchange: "
"%s IPv6 %s", ifn,
opal_net_get_hostname((struct sockaddr*)&my_ss));
} else
if (AF_INET6 == addr->sa_family) {
struct sockaddr_in6 *inaddr6 = (struct sockaddr_in6*)addr;
memcpy(&addrs[i].addr, &(inaddr6->sin6_addr),
sizeof(struct in6_addr));
addrs[i].addr_port = mca_btl_tcp_component.tcp6_listen_port;
addrs[i].addr_ifkindex = btl->tcp_ifkindex;
addrs[i].addr_family = MCA_BTL_TCP_AF_INET6;
opal_output_verbose(5, opal_btl_base_framework.framework_output,
"btl: tcp: exchange: %d %d IPv6 %s",
(int)i, btl->tcp_ifkindex,
opal_net_get_hostname(addr));
} else
#endif
if ((AF_INET == my_ss.ss_family) &&
(4 != mca_btl_tcp_component.tcp_disable_family)) {
memcpy(&addrs[current_addr].addr_inet,
&((struct sockaddr_in*)&my_ss)->sin_addr,
sizeof(struct in_addr));
addrs[current_addr].addr_port =
mca_btl_tcp_component.tcp_listen_port;
addrs[current_addr].addr_family = MCA_BTL_TCP_AF_INET;
xfer_size += sizeof (mca_btl_tcp_addr_t);
addrs[current_addr].addr_inuse = 0;
addrs[current_addr].addr_ifkindex =
opal_ifindextokindex (index);
current_addr++;
opal_output_verbose(30, opal_btl_base_framework.framework_output,
"btl: tcp: component_exchange: "
"%s IPv4 %s", ifn,
opal_net_get_hostname((struct sockaddr*)&my_ss));
}
} /* end of for opal_ifbegin() */
} /* end of for tcp_num_btls */
OPAL_MODEX_SEND(rc, OPAL_PMIX_GLOBAL,
&mca_btl_tcp_component.super.btl_version,
addrs, xfer_size);
free(addrs);
} /* end if */
if (AF_INET == addr->sa_family) {
struct sockaddr_in *inaddr = (struct sockaddr_in*)addr;
memcpy(&addrs[i].addr, &(inaddr->sin_addr),
sizeof(struct in_addr));
addrs[i].addr_port = mca_btl_tcp_component.tcp_listen_port;
addrs[i].addr_ifkindex = btl->tcp_ifkindex;
addrs[i].addr_family = MCA_BTL_TCP_AF_INET;
opal_output_verbose(5, opal_btl_base_framework.framework_output,
"btl: tcp: exchange: %d %d IPv4 %s",
(int)i, btl->tcp_ifkindex,
opal_net_get_hostname(addr));
} else {
BTL_ERROR(("Unexpected address family: %d", addr->sa_family));
return OPAL_ERR_BAD_PARAM;
}
}
OPAL_MODEX_SEND(rc, OPAL_PMIX_GLOBAL,
&mca_btl_tcp_component.super.btl_version,
addrs, size);
free(addrs);
return rc;
}

Просмотреть файл

@ -743,11 +743,11 @@ static int mca_btl_tcp_endpoint_start_connect(mca_btl_base_endpoint_t* btl_endpo
#if OPAL_ENABLE_IPV6
if (endpoint_addr.ss_family == AF_INET6) {
assert(NULL != &btl_endpoint->endpoint_btl->tcp_ifaddr_6);
if (bind(btl_endpoint->endpoint_sd, (struct sockaddr*) &btl_endpoint->endpoint_btl->tcp_ifaddr_6,
if (bind(btl_endpoint->endpoint_sd, (struct sockaddr*) &btl_endpoint->endpoint_btl->tcp_ifaddr,
sizeof(struct sockaddr_in6)) < 0) {
BTL_ERROR(("bind on local address (%s:%d) failed: %s (%d)",
opal_net_get_hostname((struct sockaddr*) &btl_endpoint->endpoint_btl->tcp_ifaddr),
htons(((struct sockaddr_in*)&btl_endpoint->endpoint_btl->tcp_ifaddr)->sin_port),
htons(((struct sockaddr_in6*)&btl_endpoint->endpoint_btl->tcp_ifaddr)->sin6_port),
strerror(opal_socket_errno), opal_socket_errno));
CLOSE_THE_SOCKET(btl_endpoint->endpoint_sd);

Просмотреть файл

@ -117,73 +117,99 @@ void mca_btl_tcp_proc_destruct(mca_btl_tcp_proc_t* tcp_proc)
mca_btl_tcp_proc_t* mca_btl_tcp_proc_create(opal_proc_t* proc)
{
mca_btl_tcp_proc_t* btl_proc;
size_t size;
int rc;
mca_btl_tcp_modex_addr_t *remote_addrs = NULL;
size_t i, size;
OPAL_THREAD_LOCK(&mca_btl_tcp_component.tcp_lock);
rc = opal_proc_table_get_value(&mca_btl_tcp_component.tcp_procs,
proc->proc_name, (void**)&btl_proc);
if(OPAL_SUCCESS == rc) {
if (OPAL_SUCCESS == rc) {
OPAL_THREAD_UNLOCK(&mca_btl_tcp_component.tcp_lock);
return btl_proc;
}
do { /* This loop is only necessary so that we can break out of the serial code */
btl_proc = OBJ_NEW(mca_btl_tcp_proc_t);
if(NULL == btl_proc) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
break;
/* proc was not found, so create one */
btl_proc = OBJ_NEW(mca_btl_tcp_proc_t);
if (NULL == btl_proc) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
goto cleanup;
}
/* Retain the proc, but don't store the ref into the btl_proc just yet. This
* provides a way to release the btl_proc in case of failure without having to
* unlock the mutex.
*/
OBJ_RETAIN(proc);
/* lookup tcp parameters exported by this proc */
OPAL_MODEX_RECV(rc, &mca_btl_tcp_component.super.btl_version,
&proc->proc_name, (uint8_t**)&remote_addrs, &size);
if (OPAL_SUCCESS != rc) {
if (OPAL_ERR_NOT_FOUND != rc) {
BTL_ERROR(("opal_modex_recv: failed with return value=%d", rc));
}
goto cleanup;
}
/* Retain the proc, but don't store the ref into the btl_proc just yet. This
* provides a way to release the btl_proc in case of failure without having to
* unlock the mutex.
*/
OBJ_RETAIN(proc);
if (0 != (size % sizeof(mca_btl_tcp_modex_addr_t))) {
BTL_ERROR(("opal_modex_recv: invalid size %lu: btl-size: %lu\n",
(unsigned long)size,
(unsigned long)sizeof(mca_btl_tcp_modex_addr_t)));
rc = OPAL_ERROR;
goto cleanup;
}
/* lookup tcp parameters exported by this proc */
OPAL_MODEX_RECV(rc, &mca_btl_tcp_component.super.btl_version,
&proc->proc_name, (uint8_t**)&btl_proc->proc_addrs, &size);
if(rc != OPAL_SUCCESS) {
if(OPAL_ERR_NOT_FOUND != rc)
BTL_ERROR(("opal_modex_recv: failed with return value=%d", rc));
break;
}
btl_proc->proc_addr_count = size / sizeof(mca_btl_tcp_modex_addr_t);
btl_proc->proc_addrs = malloc(btl_proc->proc_addr_count *
sizeof(mca_btl_tcp_addr_t));
if (NULL == btl_proc->proc_addrs) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
goto cleanup;
}
if(0 != (size % sizeof(mca_btl_tcp_addr_t))) {
BTL_ERROR(("opal_modex_recv: invalid size %lu: btl-size: %lu\n",
(unsigned long) size, (unsigned long)sizeof(mca_btl_tcp_addr_t)));
rc = OPAL_ERROR;
break;
}
btl_proc->proc_addr_count = size / sizeof(mca_btl_tcp_addr_t);
/* allocate space for endpoint array - one for each exported address */
btl_proc->proc_endpoints = (mca_btl_base_endpoint_t**)
malloc((1 + btl_proc->proc_addr_count) *
sizeof(mca_btl_base_endpoint_t*));
if(NULL == btl_proc->proc_endpoints) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
break;
}
/* convert the OPAL addr_family field to OS constants,
* so we can check for AF_INET (or AF_INET6) and don't have
* to deal with byte ordering anymore.
*/
for (unsigned int i = 0; i < btl_proc->proc_addr_count; i++) {
if (MCA_BTL_TCP_AF_INET == btl_proc->proc_addrs[i].addr_family) {
btl_proc->proc_addrs[i].addr_family = AF_INET;
}
/* the modex and proc structures differ slightly, so copy the
fields needed in the proc version */
for (i = 0 ; i < btl_proc->proc_addr_count ; i++) {
if (MCA_BTL_TCP_AF_INET == remote_addrs[i].addr_family) {
memcpy(&btl_proc->proc_addrs[i].addr_inet,
remote_addrs[i].addr, sizeof(struct in_addr));
btl_proc->proc_addrs[i].addr_port = remote_addrs[i].addr_port;
btl_proc->proc_addrs[i].addr_ifkindex = remote_addrs[i].addr_ifkindex;
btl_proc->proc_addrs[i].addr_family = AF_INET;
btl_proc->proc_addrs[i].addr_inuse = false;
} else if (MCA_BTL_TCP_AF_INET6 == remote_addrs[i].addr_family) {
#if OPAL_ENABLE_IPV6
if (MCA_BTL_TCP_AF_INET6 == btl_proc->proc_addrs[i].addr_family) {
btl_proc->proc_addrs[i].addr_family = AF_INET6;
}
memcpy(&btl_proc->proc_addrs[i].addr_inet6,
remote_addrs[i].addr, sizeof(struct in6_addr));
btl_proc->proc_addrs[i].addr_port = remote_addrs[i].addr_port;
btl_proc->proc_addrs[i].addr_ifkindex = remote_addrs[i].addr_ifkindex;
btl_proc->proc_addrs[i].addr_family = AF_INET6;
btl_proc->proc_addrs[i].addr_inuse = false;
#else
rc = OPAL_ERR_NOT_SUPPORTED;
goto cleanup;
#endif
} else {
BTL_ERROR(("Unexpected address family %d",
(int)remote_addrs[i].addr_family));
rc = OPAL_ERR_BAD_PARAM;
goto cleanup;
}
} while (0);
}
free(remote_addrs);
/* allocate space for endpoint array - one for each exported address */
btl_proc->proc_endpoints = (mca_btl_base_endpoint_t**)
malloc((1 + btl_proc->proc_addr_count) *
sizeof(mca_btl_base_endpoint_t*));
if (NULL == btl_proc->proc_endpoints) {
rc = OPAL_ERR_OUT_OF_RESOURCE;
goto cleanup;
}
cleanup:
if (OPAL_SUCCESS == rc) {
btl_proc->proc_opal = proc; /* link with the proc */
/* add to hash table of all proc instance. */
@ -669,7 +695,7 @@ int mca_btl_tcp_proc_insert( mca_btl_tcp_proc_t* btl_proc,
}
peer_interfaces[best]->inuse++;
btl_endpoint->endpoint_addr = proc_data->best_addr[i][best];
btl_endpoint->endpoint_addr->addr_inuse++;
btl_endpoint->endpoint_addr->addr_inuse = true;
rc = OPAL_SUCCESS;
break;
}
@ -695,7 +721,7 @@ int mca_btl_tcp_proc_insert( mca_btl_tcp_proc_t* btl_proc,
if (CQ_NO_CONNECTION != max) {
peer_interfaces[j_max]->inuse++;
btl_endpoint->endpoint_addr = proc_data->best_addr[i_max][j_max];
btl_endpoint->endpoint_addr->addr_inuse++;
btl_endpoint->endpoint_addr->addr_inuse = true;
rc = OPAL_SUCCESS;
}
}
@ -772,7 +798,7 @@ int mca_btl_tcp_proc_remove(mca_btl_tcp_proc_t* btl_proc, mca_btl_base_endpoint_
being removed early in the wireup sequence (e.g., if it
is unreachable by all other procs) */
if (NULL != btl_endpoint->endpoint_addr) {
btl_endpoint->endpoint_addr->addr_inuse--;
btl_endpoint->endpoint_addr->addr_inuse = false;
}
break;
}