ibv_create_flow man page

ibv_create_flow, ibv_destroy_flow — create or destroy flow steering rules


#include <infiniband/verbs.h>

struct ibv_flow *ibv_create_flow(struct ibv_qp *qp,
                                 struct ibv_flow_attr *flow_attr);
int ibv_destroy_flow(struct ibv_flow *flow_id);



allows a user application QP qp to be attached into a specified flow flow which is defined in <infiniband/verbs.h>

struct ibv_flow_attr {
uint32_t comp_mask;                     /* Future extendibility */
enum ibv_flow_attr_type type;           /* Rule type - see below */
uint16_t size;                          /* Size of command */
uint16_t priority;                      /* Rule priority - see below */
uint8_t num_of_specs;                   /* Number of ibv_flow_spec_xxx */
uint8_t port;                           /* The uplink port number */
uint32_t flags;                         /* Extra flags for rule - see below */
/* Following are the optional layers according to user request
 * struct ibv_flow_spec_xxx
 * struct ibv_flow_spec_yyy

enum ibv_flow_attr_type {
IBV_FLOW_ATTR_NORMAL		= 0x0,		/* Steering according to rule specifications */
IBV_FLOW_ATTR_ALL_DEFAULT	= 0x1,		/* Default unicast and multicast rule - receive all Eth traffic which isn't steered to any QP */
IBV_FLOW_ATTR_MC_DEFAULT 	= 0x2,		/* Default multicast rule - receive all Eth multicast traffic which isn't steered to any QP */

enum ibv_flow_flags {
IBV_FLOW_ATTR_FLAGS_ALLOW_LOOP_BACK = 1 << 0,	/* Apply the rules on packets that were sent from the attached QP through loopback */
IBV_FLOW_ATTR_FLAGS_DONT_TRAP       = 1 << 1,	/* Rule doesn't trap received packets, allowing them to match lower prioritized rules */

Each spec struct holds the relevant network layer parameters for matching. To enforce the match, the user sets a mask for each parameter.
If the bit is set in the mask, the corresponding bit in the value should be matched.
Note that most vendors support either full mask (all "1"s) or zero mask (all "0"s).
Network parameters in the relevant network structs should be given in network order (big endian).

Flow domains and priority

Flow steering defines the concept of domain and priority. Each domain represents an application that can attach a flow. Domains are prioritized. A higher priority domain will always supersede a lower priority domain when their flow specifications overlap.
IB verbs have the higher priority domain.
In addition to the domain, there is priority within each of the domains. A lower priority numeric value (higher priority) takes precedence over matching rules with higher numeric priority value (lower priority). It is important to note that the priority value of a flow spec is used not only to establish the precedence of conflicting flow matches but also as a way to abstract the order on which flow specs are tested for matches. Flows with higher priorities will be tested before flows with lower priorities.


destroys the flow flow_id.

Return Value

ibv_create_flow() returns a pointer to the flow, or NULL if the request fails. In case of an error, errno is updated.

ibv_destroy_flow() returns 0 on success, or the value of errno on failure (which indicates the failure reason).



ibv_create_flow() flow specification, QP or priority are invalid

ibv_destroy_flow() flow_id is invalid


Couldn't create/destroy flow, not enough memory


Device managed flow steering isn't currently supported


No permissions to add the flow steering rule


These verbs are available only for devices supporting


Below flow_attr defines a rule in priority 0 to match a destination mac address and a source ipv4 address. For that, L2 and L3 specs are used.
If there is a hit on this rule, means the received packet has destination mac: 66:11:22:33:44:55 and source ip: 0x0B86C806, the packet is steered to its attached qp.

struct raw_eth_flow_attr {
struct ibv_flow_attr            attr;
struct ibv_flow_spec_eth        spec_eth;
struct ibv_flow_spec_ipv4       spec_ipv4;
} __attribute__((packed));

struct raw_eth_flow_attr flow_attr = {
        .attr = {
                .comp_mask      = 0,
                .type           = IBV_FLOW_ATTR_NORMAL,
                .size           = sizeof(flow_attr),
                .priority       = 0,
                .num_of_specs   = 2,
                .port           = 1,
                .flags          = 0,
        .spec_eth = {
                .type   = IBV_FLOW_SPEC_ETH,
                .size   = sizeof(struct ibv_flow_spec_eth),
                .val = {
                        .dst_mac = {0x66, 0x11, 0x22, 0x33, 0x44, 0x55},
                        .src_mac = { 0x00, 0x00, 0x00, 0x00, 0x00, 0x00},
                        .ether_type = 0,
                        .vlan_tag = 0,
                .mask = {
                        .dst_mac = { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF},
                        .src_mac = { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF},
                        .ether_type = 0,
                        .vlan_tag = 0,
        .spec_ipv4 = {
                .type   = IBV_FLOW_SPEC_IPV4,
                .size   = sizeof(struct ibv_flow_spec_ipv4),
                .val = {
                        .src_ip = 0x0B86C806,
                        .dst_ip = 0,
                .mask = {
                        .src_ip = 0xFFFFFFFF,
                        .dst_ip = 0,


Hadar Hen Zion <hadarh@mellanox.com>

Matan Barak <matanb@mellanox.com>

Yishai Hadas <yishaih@mellanox.com>


2016-03-15 libibverbs Libibverbs Programmer's Manual