X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fdpif-linux.c;h=d84b5fa754ed1186c45d11213cad0c122406cf7a;hb=b2fda3effc787f265b5ad5dfa967ac00627bd075;hp=e5e66ff07099ee2f50f0bae24d405f8818fd2470;hpb=982b88105d0a82ab90b233bf68e4e51ac9cddd23;p=sliver-openvswitch.git diff --git a/lib/dpif-linux.c b/lib/dpif-linux.c index e5e66ff07..d84b5fa75 100644 --- a/lib/dpif-linux.c +++ b/lib/dpif-linux.c @@ -25,17 +25,17 @@ #include #include #include -#include #include #include #include #include -#include #include #include +#include "bitmap.h" #include "dpif-provider.h" #include "netdev.h" +#include "netdev-linux.h" #include "netdev-vport.h" #include "netlink-socket.h" #include "netlink.h" @@ -47,19 +47,23 @@ #include "rtnetlink.h" #include "rtnetlink-link.h" #include "shash.h" -#include "svec.h" +#include "sset.h" #include "unaligned.h" #include "util.h" #include "vlog.h" VLOG_DEFINE_THIS_MODULE(dpif_linux); +enum { LRU_MAX_PORTS = 1024 }; +enum { LRU_MASK = LRU_MAX_PORTS - 1}; +BUILD_ASSERT_DECL(IS_POW2(LRU_MAX_PORTS)); + struct dpif_linux_dp { - /* ioctl command argument. */ - int cmd; + /* Generic Netlink header. */ + uint8_t cmd; - /* struct odp_datapath header. */ - uint32_t dp_idx; + /* struct odp_header. */ + int dp_ifindex; /* Attributes. */ const char *name; /* ODP_DP_ATTR_NAME. */ @@ -70,6 +74,9 @@ struct dpif_linux_dp { }; static void dpif_linux_dp_init(struct dpif_linux_dp *); +static int dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *, + const struct ofpbuf *); +static void dpif_linux_dp_dump_start(struct nl_dump *); static int dpif_linux_dp_transact(const struct dpif_linux_dp *request, struct dpif_linux_dp *reply, struct ofpbuf **bufp); @@ -77,18 +84,21 @@ static int dpif_linux_dp_get(const struct dpif *, struct dpif_linux_dp *reply, struct ofpbuf **bufp); struct dpif_linux_flow { - /* ioctl command argument. */ - int cmd; + /* Generic Netlink header. */ + uint8_t cmd; - /* struct odp_flow header. */ + /* struct odp_header. */ unsigned int nlmsg_flags; - uint32_t dp_idx; + int dp_ifindex; /* Attributes. * - * The 'stats', 'used', and 'state' members point to 64-bit data that might - * only be aligned on 32-bit boundaries, so get_unaligned_u64() should be - * used to access their values. */ + * The 'stats' and 'used' members point to 64-bit data that might only be + * aligned on 32-bit boundaries, so get_unaligned_u64() should be used to + * access their values. + * + * If 'actions' is nonnull then ODP_FLOW_ATTR_ACTIONS will be included in + * the Netlink version of the command, even if actions_len is zero. */ const struct nlattr *key; /* ODP_FLOW_ATTR_KEY. */ size_t key_len; const struct nlattr *actions; /* ODP_FLOW_ATTR_ACTIONS. */ @@ -97,10 +107,13 @@ struct dpif_linux_flow { const uint8_t *tcp_flags; /* ODP_FLOW_ATTR_TCP_FLAGS. */ const uint64_t *used; /* ODP_FLOW_ATTR_USED. */ bool clear; /* ODP_FLOW_ATTR_CLEAR. */ - const uint64_t *state; /* ODP_FLOW_ATTR_STATE. */ }; static void dpif_linux_flow_init(struct dpif_linux_flow *); +static int dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *, + const struct ofpbuf *); +static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *, + struct ofpbuf *); static int dpif_linux_flow_transact(const struct dpif_linux_flow *request, struct dpif_linux_flow *reply, struct ofpbuf **bufp); @@ -110,42 +123,46 @@ static void dpif_linux_flow_get_stats(const struct dpif_linux_flow *, /* Datapath interface for the openvswitch Linux kernel module. */ struct dpif_linux { struct dpif dpif; - int fd; + int dp_ifindex; /* Multicast group messages. */ struct nl_sock *mc_sock; uint32_t mcgroups[DPIF_N_UC_TYPES]; unsigned int listen_mask; - /* Used by dpif_linux_get_all_names(). */ - char *local_ifname; - int minor; - /* Change notification. */ - int local_ifindex; /* Ifindex of local port. */ - struct shash changed_ports; /* Ports that have changed. */ + struct sset changed_ports; /* Ports that have changed. */ struct rtnetlink_notifier port_notifier; bool change_error; + + /* Queue of unused ports. */ + unsigned long *lru_bitmap; + uint16_t lru_ports[LRU_MAX_PORTS]; + size_t lru_head; + size_t lru_tail; }; static struct vlog_rate_limit error_rl = VLOG_RATE_LIMIT_INIT(9999, 5); /* Generic Netlink family numbers for ODP. */ +static int odp_datapath_family; +static int odp_vport_family; +static int odp_flow_family; static int odp_packet_family; /* Generic Netlink socket. */ static struct nl_sock *genl_sock; static int dpif_linux_init(void); -static int open_dpif(const struct dpif_linux_dp *, - const struct dpif_linux_vport *local_vport, - struct dpif **); -static int get_openvswitch_major(void); -static int open_minor(int minor, int *fdp); -static int make_openvswitch_device(int minor, char **fnp); +static int open_dpif(const struct dpif_linux_dp *, struct dpif **); static void dpif_linux_port_changed(const struct rtnetlink_link_change *, void *dpif); +static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *, + struct ofpbuf *); +static int dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *, + const struct ofpbuf *); + static struct dpif_linux * dpif_linux_cast(const struct dpif *dpif) { @@ -153,56 +170,58 @@ dpif_linux_cast(const struct dpif *dpif) return CONTAINER_OF(dpif, struct dpif_linux, dpif); } -static int -dpif_linux_enumerate(struct svec *all_dps) +static void +dpif_linux_push_port(struct dpif_linux *dp, uint16_t port) { - uint32_t dp_idx; - int major; - int err; - - err = dpif_linux_init(); - if (err) { - return err; + if (port < LRU_MAX_PORTS && !bitmap_is_set(dp->lru_bitmap, port)) { + bitmap_set1(dp->lru_bitmap, port); + dp->lru_ports[dp->lru_head++ & LRU_MASK] = port; } +} + +static uint32_t +dpif_linux_pop_port(struct dpif_linux *dp) +{ + uint16_t port; - /* Check that the Open vSwitch module is loaded. */ - major = get_openvswitch_major(); - if (major < 0) { - return -major; + if (dp->lru_head == dp->lru_tail) { + return UINT32_MAX; } - dp_idx = 0; - for (;;) { - struct dpif_linux_dp request, reply; - struct ofpbuf *buf; - char devname[16]; - int error; + port = dp->lru_ports[dp->lru_tail++ & LRU_MASK]; + bitmap_set0(dp->lru_bitmap, port); + return port; +} - dpif_linux_dp_init(&request); - request.dp_idx = dp_idx; - request.cmd = ODP_DP_DUMP; +static int +dpif_linux_enumerate(struct sset *all_dps) +{ + struct nl_dump dump; + struct ofpbuf msg; + int error; - error = dpif_linux_dp_transact(&request, &reply, &buf); - if (error) { - return error == ENODEV ? 0 : error; - } - ofpbuf_delete(buf); + error = dpif_linux_init(); + if (error) { + return error; + } - sprintf(devname, "dp%d", reply.dp_idx); - svec_add(all_dps, devname); + dpif_linux_dp_dump_start(&dump); + while (nl_dump_next(&dump, &msg)) { + struct dpif_linux_dp dp; - dp_idx = reply.dp_idx + 1; + if (!dpif_linux_dp_from_ofpbuf(&dp, &msg)) { + sset_add(all_dps, dp.name); + } } + return nl_dump_done(&dump); } static int dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, bool create, struct dpif **dpifp) { - struct dpif_linux_vport vport_request, vport; struct dpif_linux_dp dp_request, dp; struct ofpbuf *buf; - int minor; int error; error = dpif_linux_init(); @@ -210,58 +229,27 @@ dpif_linux_open(const struct dpif_class *class OVS_UNUSED, const char *name, return error; } - minor = !strncmp(name, "dp", 2) - && isdigit((unsigned char)name[2]) ? atoi(name + 2) : -1; - /* Create or look up datapath. */ dpif_linux_dp_init(&dp_request); - dp_request.cmd = create ? ODP_DP_NEW : ODP_DP_GET; - dp_request.dp_idx = minor; - dp_request.name = minor < 0 ? name : NULL; + dp_request.cmd = create ? ODP_DP_CMD_NEW : ODP_DP_CMD_GET; + dp_request.name = name; error = dpif_linux_dp_transact(&dp_request, &dp, &buf); if (error) { return error; } - ofpbuf_delete(buf); /* Pointers inside 'dp' are now invalid! */ - - /* Look up local port. */ - dpif_linux_vport_init(&vport_request); - vport_request.cmd = ODP_VPORT_GET; - vport_request.dp_idx = dp.dp_idx; - vport_request.port_no = ODPP_LOCAL; - vport_request.name = minor < 0 ? name : NULL; - error = dpif_linux_vport_transact(&vport_request, &vport, &buf); - if (error) { - return error; - } else if (vport.port_no != ODPP_LOCAL) { - /* This is an Open vSwitch device but not the local port. We - * intentionally support only using the name of the local port as the - * name of a datapath; otherwise, it would be too difficult to - * enumerate all the names of a datapath. */ - error = EOPNOTSUPP; - } else { - error = open_dpif(&dp, &vport, dpifp); - } + error = open_dpif(&dp, dpifp); ofpbuf_delete(buf); + return error; } static int -open_dpif(const struct dpif_linux_dp *dp, - const struct dpif_linux_vport *local_vport, struct dpif **dpifp) +open_dpif(const struct dpif_linux_dp *dp, struct dpif **dpifp) { - int dp_idx = local_vport->dp_idx; struct dpif_linux *dpif; - char *name; int error; - int fd; int i; - error = open_minor(dp_idx, &fd); - if (error) { - goto error; - } - dpif = xmalloc(sizeof *dpif); error = rtnetlink_link_notifier_register(&dpif->port_notifier, dpif_linux_port_changed, dpif); @@ -269,29 +257,29 @@ open_dpif(const struct dpif_linux_dp *dp, goto error_free; } - name = xasprintf("dp%d", dp_idx); - dpif_init(&dpif->dpif, &dpif_linux_class, name, dp_idx, dp_idx); - free(name); + dpif_init(&dpif->dpif, &dpif_linux_class, dp->name, + dp->dp_ifindex, dp->dp_ifindex); - dpif->fd = fd; dpif->mc_sock = NULL; for (i = 0; i < DPIF_N_UC_TYPES; i++) { dpif->mcgroups[i] = dp->mcgroups[i]; } dpif->listen_mask = 0; - dpif->local_ifname = xstrdup(local_vport->name); - dpif->local_ifindex = local_vport->ifindex; - dpif->minor = dp_idx; - shash_init(&dpif->changed_ports); + dpif->dp_ifindex = dp->dp_ifindex; + sset_init(&dpif->changed_ports); dpif->change_error = false; *dpifp = &dpif->dpif; + dpif->lru_head = dpif->lru_tail = 0; + dpif->lru_bitmap = bitmap_allocate(LRU_MAX_PORTS); + bitmap_set1(dpif->lru_bitmap, ODPP_LOCAL); + for (i = 1; i < LRU_MAX_PORTS; i++) { + dpif_linux_push_port(dpif, i); + } return 0; error_free: free(dpif); - close(fd); -error: return error; } @@ -300,22 +288,11 @@ dpif_linux_close(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); rtnetlink_link_notifier_unregister(&dpif->port_notifier); - shash_destroy(&dpif->changed_ports); - free(dpif->local_ifname); - close(dpif->fd); + sset_destroy(&dpif->changed_ports); + free(dpif->lru_bitmap); free(dpif); } -static int -dpif_linux_get_all_names(const struct dpif *dpif_, struct svec *all_names) -{ - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - - svec_add_nocopy(all_names, xasprintf("dp%d", dpif->minor)); - svec_add(all_names, dpif->local_ifname); - return 0; -} - static int dpif_linux_destroy(struct dpif *dpif_) { @@ -323,8 +300,8 @@ dpif_linux_destroy(struct dpif *dpif_) struct dpif_linux_dp dp; dpif_linux_dp_init(&dp); - dp.cmd = ODP_DP_DEL; - dp.dp_idx = dpif->minor; + dp.cmd = ODP_DP_CMD_DEL; + dp.dp_ifindex = dpif->dp_ifindex; return dpif_linux_dp_transact(&dp, NULL, NULL); } @@ -365,8 +342,8 @@ dpif_linux_set_drop_frags(struct dpif *dpif_, bool drop_frags) struct dpif_linux_dp dp; dpif_linux_dp_init(&dp); - dp.cmd = ODP_DP_SET; - dp.dp_idx = dpif->minor; + dp.cmd = ODP_DP_CMD_SET; + dp.dp_ifindex = dpif->dp_ifindex; dp.ipv4_frags = drop_frags ? ODP_DP_FRAG_DROP : ODP_DP_FRAG_ZERO; return dpif_linux_dp_transact(&dp, NULL, NULL); } @@ -384,8 +361,8 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, int error; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_NEW; - request.dp_idx = dpif->minor; + request.cmd = ODP_VPORT_CMD_NEW; + request.dp_ifindex = dpif->dp_ifindex; request.type = netdev_vport_get_vport_type(netdev); if (request.type == ODP_VPORT_TYPE_UNSPEC) { VLOG_WARN_RL(&error_rl, "%s: cannot create port `%s' because it has " @@ -401,11 +378,17 @@ dpif_linux_port_add(struct dpif *dpif_, struct netdev *netdev, request.options_len = options->size; } - error = dpif_linux_vport_transact(&request, &reply, &buf); - if (!error) { - *port_nop = reply.port_no; + /* Loop until we find a port that isn't used. */ + do { + request.port_no = dpif_linux_pop_port(dpif); + error = dpif_linux_vport_transact(&request, &reply, &buf); + + if (!error) { + *port_nop = reply.port_no; + } ofpbuf_delete(buf); - } + } while (request.port_no != UINT32_MAX + && (error == EBUSY || error == EFBIG)); return error; } @@ -415,12 +398,18 @@ dpif_linux_port_del(struct dpif *dpif_, uint16_t port_no) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_vport vport; + int error; dpif_linux_vport_init(&vport); - vport.cmd = ODP_VPORT_DEL; - vport.dp_idx = dpif->minor; + vport.cmd = ODP_VPORT_CMD_DEL; + vport.dp_ifindex = dpif->dp_ifindex; vport.port_no = port_no; - return dpif_linux_vport_transact(&vport, NULL, NULL); + error = dpif_linux_vport_transact(&vport, NULL, NULL); + + if (!error) { + dpif_linux_push_port(dpif, port_no); + } + return error; } static int @@ -433,8 +422,8 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, int error; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_GET; - request.dp_idx = dpif_linux_cast(dpif)->minor; + request.cmd = ODP_VPORT_CMD_GET; + request.dp_ifindex = dpif_linux_cast(dpif)->dp_ifindex; request.port_no = port_no; request.name = port_name; @@ -443,6 +432,12 @@ dpif_linux_port_query__(const struct dpif *dpif, uint32_t port_no, dpif_port->name = xstrdup(reply.name); dpif_port->type = xstrdup(netdev_vport_get_netdev_type(&reply)); dpif_port->port_no = reply.port_no; + if (reply.stats) { + netdev_stats_from_rtnl_link_stats64(&dpif_port->stats, + reply.stats); + } else { + memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats); + } ofpbuf_delete(buf); } return error; @@ -474,57 +469,98 @@ static int dpif_linux_flow_flush(struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - return ioctl(dpif->fd, ODP_FLOW_FLUSH, dpif->minor) ? errno : 0; + struct dpif_linux_flow flow; + + dpif_linux_flow_init(&flow); + flow.cmd = ODP_FLOW_CMD_DEL; + flow.dp_ifindex = dpif->dp_ifindex; + return dpif_linux_flow_transact(&flow, NULL, NULL); } struct dpif_linux_port_state { - struct ofpbuf *buf; - uint32_t next; + struct nl_dump dump; + unsigned long *port_bitmap; /* Ports in the datapath. */ + bool complete; /* Dump completed without error. */ }; static int -dpif_linux_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) +dpif_linux_port_dump_start(const struct dpif *dpif_, void **statep) { - *statep = xzalloc(sizeof(struct dpif_linux_port_state)); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_port_state *state; + struct dpif_linux_vport request; + struct ofpbuf *buf; + + *statep = state = xmalloc(sizeof *state); + state->port_bitmap = bitmap_allocate(LRU_MAX_PORTS); + state->complete = false; + + dpif_linux_vport_init(&request); + request.cmd = ODP_DP_CMD_GET; + request.dp_ifindex = dpif->dp_ifindex; + + buf = ofpbuf_new(1024); + dpif_linux_vport_to_ofpbuf(&request, buf); + nl_dump_start(&state->dump, genl_sock, buf); + ofpbuf_delete(buf); + return 0; } static int -dpif_linux_port_dump_next(const struct dpif *dpif, void *state_, +dpif_linux_port_dump_next(const struct dpif *dpif OVS_UNUSED, void *state_, struct dpif_port *dpif_port) { struct dpif_linux_port_state *state = state_; - struct dpif_linux_vport request, reply; - struct ofpbuf *buf; + struct dpif_linux_vport vport; + struct ofpbuf buf; int error; - ofpbuf_delete(state->buf); - state->buf = NULL; + if (!nl_dump_next(&state->dump, &buf)) { + state->complete = true; + return EOF; + } - dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_DUMP; - request.dp_idx = dpif_linux_cast(dpif)->minor; - request.port_no = state->next; - error = dpif_linux_vport_transact(&request, &reply, &buf); + error = dpif_linux_vport_from_ofpbuf(&vport, &buf); if (error) { - return error == ENODEV ? EOF : error; + return error; + } + + if (vport.port_no < LRU_MAX_PORTS) { + bitmap_set1(state->port_bitmap, vport.port_no); + } + + dpif_port->name = (char *) vport.name; + dpif_port->type = (char *) netdev_vport_get_netdev_type(&vport); + dpif_port->port_no = vport.port_no; + if (vport.stats) { + netdev_stats_from_rtnl_link_stats64(&dpif_port->stats, vport.stats); } else { - dpif_port->name = (char *) reply.name; - dpif_port->type = (char *) netdev_vport_get_netdev_type(&reply); - dpif_port->port_no = reply.port_no; - state->buf = buf; - state->next = reply.port_no + 1; - return 0; + memset(&dpif_port->stats, 0xff, sizeof dpif_port->stats); } + return 0; } static int -dpif_linux_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) +dpif_linux_port_dump_done(const struct dpif *dpif_, void *state_) { + struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_port_state *state = state_; - ofpbuf_delete(state->buf); + int error = nl_dump_done(&state->dump); + + if (state->complete) { + uint16_t i; + + for (i = 0; i < LRU_MAX_PORTS; i++) { + if (!bitmap_is_set(state->port_bitmap, i)) { + dpif_linux_push_port(dpif, i); + } + } + } + + free(state->port_bitmap); free(state); - return 0; + return error; } static int @@ -534,11 +570,10 @@ dpif_linux_port_poll(const struct dpif *dpif_, char **devnamep) if (dpif->change_error) { dpif->change_error = false; - shash_clear(&dpif->changed_ports); + sset_clear(&dpif->changed_ports); return ENOBUFS; - } else if (!shash_is_empty(&dpif->changed_ports)) { - struct shash_node *node = shash_first(&dpif->changed_ports); - *devnamep = shash_steal(&dpif->changed_ports, node); + } else if (!sset_is_empty(&dpif->changed_ports)) { + *devnamep = sset_pop(&dpif->changed_ports); return 0; } else { return EAGAIN; @@ -549,29 +584,39 @@ static void dpif_linux_port_poll_wait(const struct dpif *dpif_) { struct dpif_linux *dpif = dpif_linux_cast(dpif_); - if (!shash_is_empty(&dpif->changed_ports) || dpif->change_error) { + if (!sset_is_empty(&dpif->changed_ports) || dpif->change_error) { poll_immediate_wake(); } else { rtnetlink_link_notifier_wait(); } } +static int +dpif_linux_flow_get__(const struct dpif *dpif_, + const struct nlattr *key, size_t key_len, + struct dpif_linux_flow *reply, struct ofpbuf **bufp) +{ + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_flow request; + + dpif_linux_flow_init(&request); + request.cmd = ODP_FLOW_CMD_GET; + request.dp_ifindex = dpif->dp_ifindex; + request.key = key; + request.key_len = key_len; + return dpif_linux_flow_transact(&request, reply, bufp); +} + static int dpif_linux_flow_get(const struct dpif *dpif_, const struct nlattr *key, size_t key_len, struct ofpbuf **actionsp, struct dpif_flow_stats *stats) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); - struct dpif_linux_flow request, reply; + struct dpif_linux_flow reply; struct ofpbuf *buf; int error; - dpif_linux_flow_init(&request); - request.cmd = ODP_FLOW_GET; - request.dp_idx = dpif->minor; - request.key = key; - request.key_len = key_len; - error = dpif_linux_flow_transact(&request, &reply, &buf); + error = dpif_linux_flow_get__(dpif_, key, key_len, &reply, &buf); if (!error) { if (stats) { dpif_linux_flow_get_stats(&reply, stats); @@ -595,15 +640,17 @@ dpif_linux_flow_put(struct dpif *dpif_, enum dpif_flow_put_flags flags, { struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_flow request, reply; + struct nlattr dummy_action; struct ofpbuf *buf; int error; dpif_linux_flow_init(&request); - request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_NEW : ODP_FLOW_SET; - request.dp_idx = dpif->minor; + request.cmd = flags & DPIF_FP_CREATE ? ODP_FLOW_CMD_NEW : ODP_FLOW_CMD_SET; + request.dp_ifindex = dpif->dp_ifindex; request.key = key; request.key_len = key_len; - request.actions = actions; + /* Ensure that ODP_FLOW_ATTR_ACTIONS will always be included. */ + request.actions = actions ? actions : &dummy_action; request.actions_len = actions_len; if (flags & DPIF_FP_ZERO_STATS) { request.clear = true; @@ -630,8 +677,8 @@ dpif_linux_flow_del(struct dpif *dpif_, int error; dpif_linux_flow_init(&request); - request.cmd = ODP_FLOW_DEL; - request.dp_idx = dpif->minor; + request.cmd = ODP_FLOW_CMD_DEL; + request.dp_ifindex = dpif->dp_ifindex; request.key = key; request.key_len = key_len; error = dpif_linux_flow_transact(&request, @@ -645,62 +692,94 @@ dpif_linux_flow_del(struct dpif *dpif_, } struct dpif_linux_flow_state { + struct nl_dump dump; struct dpif_linux_flow flow; - struct ofpbuf *buf; struct dpif_flow_stats stats; + struct ofpbuf *buf; }; static int -dpif_linux_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep) +dpif_linux_flow_dump_start(const struct dpif *dpif_, void **statep) { - *statep = xzalloc(sizeof(struct dpif_linux_flow_state)); + struct dpif_linux *dpif = dpif_linux_cast(dpif_); + struct dpif_linux_flow_state *state; + struct dpif_linux_flow request; + struct ofpbuf *buf; + + *statep = state = xmalloc(sizeof *state); + + dpif_linux_flow_init(&request); + request.cmd = ODP_DP_CMD_GET; + request.dp_ifindex = dpif->dp_ifindex; + + buf = ofpbuf_new(1024); + dpif_linux_flow_to_ofpbuf(&request, buf); + nl_dump_start(&state->dump, genl_sock, buf); + ofpbuf_delete(buf); + + state->buf = NULL; + return 0; } static int -dpif_linux_flow_dump_next(const struct dpif *dpif_, void *state_, +dpif_linux_flow_dump_next(const struct dpif *dpif_ OVS_UNUSED, void *state_, const struct nlattr **key, size_t *key_len, const struct nlattr **actions, size_t *actions_len, const struct dpif_flow_stats **stats) { - struct dpif_linux *dpif = dpif_linux_cast(dpif_); struct dpif_linux_flow_state *state = state_; - struct ofpbuf *old_buf = state->buf; - struct dpif_linux_flow request; + struct ofpbuf buf; int error; - dpif_linux_flow_init(&request); - request.cmd = ODP_FLOW_DUMP; - request.dp_idx = dpif->minor; - request.state = state->flow.state; - error = dpif_linux_flow_transact(&request, &state->flow, &state->buf); - ofpbuf_delete(old_buf); + do { + ofpbuf_delete(state->buf); + state->buf = NULL; - if (!error) { - if (key) { - *key = state->flow.key; - *key_len = state->flow.key_len; + if (!nl_dump_next(&state->dump, &buf)) { + return EOF; } - if (actions) { - *actions = state->flow.actions; - *actions_len = state->flow.actions_len; + + error = dpif_linux_flow_from_ofpbuf(&state->flow, &buf); + if (error) { + return error; } - if (stats) { - dpif_linux_flow_get_stats(&state->flow, &state->stats); - *stats = &state->stats; + + if (actions && !state->flow.actions) { + error = dpif_linux_flow_get__(dpif_, state->flow.key, + state->flow.key_len, + &state->flow, &state->buf); + if (error == ENOENT) { + VLOG_DBG("dumped flow disappeared on get"); + } else if (error) { + VLOG_WARN("error fetching dumped flow: %s", strerror(error)); + } } + } while (error); + + if (actions) { + *actions = state->flow.actions; + *actions_len = state->flow.actions_len; + } + if (key) { + *key = state->flow.key; + *key_len = state->flow.key_len; } - return error == ENODEV ? EOF : error; + if (stats) { + dpif_linux_flow_get_stats(&state->flow, &state->stats); + *stats = &state->stats; + } + return error; } static int dpif_linux_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_) { struct dpif_linux_flow_state *state = state_; - + int error = nl_dump_done(&state->dump); ofpbuf_delete(state->buf); free(state); - return 0; + return error; } static int @@ -719,7 +798,7 @@ dpif_linux_execute(struct dpif *dpif_, ODP_PACKET_CMD_EXECUTE, 1); execute = ofpbuf_put_uninit(buf, sizeof *execute); - execute->dp_idx = dpif->minor; + execute->dp_ifindex = dpif->dp_ifindex; nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, packet->data, packet->size); nl_msg_put_unspec(buf, ODP_PACKET_ATTR_ACTIONS, actions, actions_len); @@ -806,8 +885,8 @@ dpif_linux_set_sflow_probability(struct dpif *dpif_, uint32_t probability) struct dpif_linux_dp dp; dpif_linux_dp_init(&dp); - dp.cmd = ODP_DP_SET; - dp.dp_idx = dpif->minor; + dp.cmd = ODP_DP_CMD_SET; + dp.dp_ifindex = dpif->dp_ifindex; dp.sampling = &probability; return dpif_linux_dp_transact(&dp, NULL, NULL); } @@ -826,7 +905,7 @@ dpif_linux_queue_to_priority(const struct dpif *dpif OVS_UNUSED, static int parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, - uint32_t *dp_idx) + int *dp_ifindex) { static const struct nl_policy odp_packet_policy[] = { /* Always present. */ @@ -847,6 +926,7 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, struct nlmsghdr *nlmsg; struct genlmsghdr *genl; struct ofpbuf b; + int type; ofpbuf_use_const(&b, buf->data, buf->size); @@ -854,18 +934,22 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, genl = ofpbuf_try_pull(&b, sizeof *genl); odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); if (!nlmsg || !genl || !odp_header + || nlmsg->nlmsg_type != odp_packet_family || !nl_policy_parse(&b, 0, odp_packet_policy, a, ARRAY_SIZE(odp_packet_policy))) { return EINVAL; } - memset(upcall, 0, sizeof *upcall); - - upcall->type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS - : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION - : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE - : -1); + type = (genl->cmd == ODP_PACKET_CMD_MISS ? DPIF_UC_MISS + : genl->cmd == ODP_PACKET_CMD_ACTION ? DPIF_UC_ACTION + : genl->cmd == ODP_PACKET_CMD_SAMPLE ? DPIF_UC_SAMPLE + : -1); + if (type < 0) { + return EINVAL; + } + memset(upcall, 0, sizeof *upcall); + upcall->type = type; upcall->packet = buf; upcall->packet->data = (void *) nl_attr_get(a[ODP_PACKET_ATTR_PACKET]); upcall->packet->size = nl_attr_get_size(a[ODP_PACKET_ATTR_PACKET]); @@ -882,7 +966,7 @@ parse_odp_packet(struct ofpbuf *buf, struct dpif_upcall *upcall, upcall->actions_len = nl_attr_get_size(a[ODP_PACKET_ATTR_ACTIONS]); } - *dp_idx = odp_header->dp_idx; + *dp_ifindex = odp_header->dp_ifindex; return 0; } @@ -900,16 +984,16 @@ dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall) } for (i = 0; i < 50; i++) { - uint32_t dp_idx; + int dp_ifindex; error = nl_sock_recv(dpif->mc_sock, &buf, false); if (error) { return error; } - error = parse_odp_packet(buf, upcall, &dp_idx); + error = parse_odp_packet(buf, upcall, &dp_ifindex); if (!error - && dp_idx == dpif->minor + && dp_ifindex == dpif->dp_ifindex && dpif->listen_mask & (1u << upcall->type)) { return 0; } @@ -944,13 +1028,12 @@ dpif_linux_recv_purge(struct dpif *dpif_) const struct dpif_class dpif_linux_class = { "system", - NULL, /* run */ - NULL, /* wait */ dpif_linux_enumerate, dpif_linux_open, dpif_linux_close, - dpif_linux_get_all_names, dpif_linux_destroy, + NULL, /* run */ + NULL, /* wait */ dpif_linux_get_stats, dpif_linux_get_drop_frags, dpif_linux_set_drop_frags, @@ -982,15 +1065,29 @@ const struct dpif_class dpif_linux_class = { dpif_linux_recv_purge, }; -static int get_major(const char *target); - static int dpif_linux_init(void) { static int error = -1; if (error < 0) { - error = nl_lookup_genl_family(ODP_PACKET_FAMILY, &odp_packet_family); + error = nl_lookup_genl_family(ODP_DATAPATH_FAMILY, + &odp_datapath_family); + if (error) { + VLOG_ERR("Generic Netlink family '%s' does not exist. " + "The Open vSwitch kernel module is probably not loaded.", + ODP_DATAPATH_FAMILY); + } + if (!error) { + error = nl_lookup_genl_family(ODP_VPORT_FAMILY, &odp_vport_family); + } + if (!error) { + error = nl_lookup_genl_family(ODP_FLOW_FAMILY, &odp_flow_family); + } + if (!error) { + error = nl_lookup_genl_family(ODP_PACKET_FAMILY, + &odp_packet_family); + } if (!error) { error = nl_sock_create(NETLINK_GENERIC, &genl_sock); } @@ -1009,7 +1106,7 @@ dpif_linux_is_internal_device(const char *name) error = dpif_linux_vport_get(name, &reply, &buf); if (!error) { ofpbuf_delete(buf); - } else if (error != ENODEV) { + } else if (error != ENODEV && error != ENOENT) { VLOG_WARN_RL(&error_rl, "%s: vport query failed (%s)", name, strerror(error)); } @@ -1017,144 +1114,32 @@ dpif_linux_is_internal_device(const char *name) return reply.type == ODP_VPORT_TYPE_INTERNAL; } -static int -make_openvswitch_device(int minor, char **fnp) -{ - const char dirname[] = "/dev/net"; - int major; - dev_t dev; - struct stat s; - char fn[128]; - - *fnp = NULL; - - major = get_openvswitch_major(); - if (major < 0) { - return -major; - } - dev = makedev(major, minor); - - sprintf(fn, "%s/dp%d", dirname, minor); - if (!stat(fn, &s)) { - if (!S_ISCHR(s.st_mode)) { - VLOG_WARN_RL(&error_rl, "%s is not a character device, fixing", - fn); - } else if (s.st_rdev != dev) { - VLOG_WARN_RL(&error_rl, - "%s is device %u:%u but should be %u:%u, fixing", - fn, major(s.st_rdev), minor(s.st_rdev), - major(dev), minor(dev)); - } else { - goto success; - } - if (unlink(fn)) { - VLOG_WARN_RL(&error_rl, "%s: unlink failed (%s)", - fn, strerror(errno)); - return errno; - } - } else if (errno == ENOENT) { - if (stat(dirname, &s)) { - if (errno == ENOENT) { - if (mkdir(dirname, 0755)) { - VLOG_WARN_RL(&error_rl, "%s: mkdir failed (%s)", - dirname, strerror(errno)); - return errno; - } - } else { - VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", - dirname, strerror(errno)); - return errno; - } - } - } else { - VLOG_WARN_RL(&error_rl, "%s: stat failed (%s)", fn, strerror(errno)); - return errno; - } - - /* The device needs to be created. */ - if (mknod(fn, S_IFCHR | 0700, dev)) { - VLOG_WARN_RL(&error_rl, - "%s: creating character device %u:%u failed (%s)", - fn, major(dev), minor(dev), strerror(errno)); - return errno; - } - -success: - *fnp = xstrdup(fn); - return 0; -} - -/* Return the major device number of the Open vSwitch device. If it - * cannot be determined, a negative errno is returned. */ -static int -get_openvswitch_major(void) +int +dpif_linux_vport_send(int dp_ifindex, uint32_t port_no, + const void *data, size_t size) { - static int openvswitch_major = -1; - if (openvswitch_major < 0) { - openvswitch_major = get_major("openvswitch"); - } - return openvswitch_major; -} + struct odp_header *execute; + struct ofpbuf *buf; + size_t actions_ofs; + int error; -static int -get_major(const char *target) -{ - const char fn[] = "/proc/devices"; - char line[128]; - FILE *file; - int ln; - - file = fopen(fn, "r"); - if (!file) { - VLOG_ERR("opening %s failed (%s)", fn, strerror(errno)); - return -errno; - } - - for (ln = 1; fgets(line, sizeof line, file); ln++) { - char name[64]; - int major; - - if (!strncmp(line, "Character", 9) || line[0] == '\0') { - /* Nothing to do. */ - } else if (!strncmp(line, "Block", 5)) { - /* We only want character devices, so skip the rest of the file. */ - break; - } else if (sscanf(line, "%d %63s", &major, name)) { - if (!strcmp(name, target)) { - fclose(file); - return major; - } - } else { - VLOG_WARN_ONCE("%s:%d: syntax error", fn, ln); - } - } + buf = ofpbuf_new(128 + size); - fclose(file); + nl_msg_put_genlmsghdr(buf, 0, odp_packet_family, NLM_F_REQUEST, + ODP_PACKET_CMD_EXECUTE, 1); - VLOG_ERR("%s: %s major not found (is the module loaded?)", fn, target); - return -ENODEV; -} + execute = ofpbuf_put_uninit(buf, sizeof *execute); + execute->dp_ifindex = dp_ifindex; -static int -open_minor(int minor, int *fdp) -{ - int error; - char *fn; + nl_msg_put_unspec(buf, ODP_PACKET_ATTR_PACKET, data, size); - error = make_openvswitch_device(minor, &fn); - if (error) { - return error; - } + actions_ofs = nl_msg_start_nested(buf, ODP_PACKET_ATTR_ACTIONS); + nl_msg_put_u32(buf, ODP_ACTION_ATTR_OUTPUT, port_no); + nl_msg_end_nested(buf, actions_ofs); - *fdp = open(fn, O_RDONLY | O_NONBLOCK); - if (*fdp < 0) { - error = errno; - VLOG_WARN("%s: open failed (%s)", fn, strerror(error)); - free(fn); - return error; - } - free(fn); - return 0; + error = nl_sock_transact(genl_sock, buf, NULL); + ofpbuf_delete(buf); + return error; } static void @@ -1164,38 +1149,20 @@ dpif_linux_port_changed(const struct rtnetlink_link_change *change, struct dpif_linux *dpif = dpif_; if (change) { - if (change->master_ifindex == dpif->local_ifindex + if (change->master_ifindex == dpif->dp_ifindex && (change->nlmsg_type == RTM_NEWLINK || change->nlmsg_type == RTM_DELLINK)) { /* Our datapath changed, either adding a new port or deleting an * existing one. */ - shash_add_once(&dpif->changed_ports, change->ifname, NULL); + sset_add(&dpif->changed_ports, change->ifname); } } else { dpif->change_error = true; } } - -static int -get_dp0_fd(int *dp0_fdp) -{ - static int dp0_fd = -1; - if (dp0_fd < 0) { - int error; - int fd; - - error = open_minor(0, &fd); - if (error) { - return error; - } - dp0_fd = fd; - } - *dp0_fdp = dp0_fd; - return 0; -} -/* Parses the contents of 'buf', which contains a "struct odp_vport" followed +/* Parses the contents of 'buf', which contains a "struct odp_header" followed * by Netlink attributes, into 'vport'. Returns 0 if successful, otherwise a * positive errno value. * @@ -1223,18 +1190,27 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, [ODP_VPORT_ATTR_IFLINK] = { .type = NL_A_U32, .optional = true }, }; - struct odp_vport *odp_vport; struct nlattr *a[ARRAY_SIZE(odp_vport_policy)]; + struct odp_header *odp_header; + struct nlmsghdr *nlmsg; + struct genlmsghdr *genl; + struct ofpbuf b; dpif_linux_vport_init(vport); - if (!nl_policy_parse(buf, sizeof *odp_vport, odp_vport_policy, - a, ARRAY_SIZE(odp_vport_policy))) { + ofpbuf_use_const(&b, buf->data, buf->size); + nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); + genl = ofpbuf_try_pull(&b, sizeof *genl); + odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); + if (!nlmsg || !genl || !odp_header + || nlmsg->nlmsg_type != odp_vport_family + || !nl_policy_parse(&b, 0, odp_vport_policy, a, + ARRAY_SIZE(odp_vport_policy))) { return EINVAL; } - odp_vport = buf->data; - vport->dp_idx = odp_vport->dp_idx; + vport->cmd = genl->cmd; + vport->dp_ifindex = odp_header->dp_ifindex; vport->port_no = nl_attr_get_u32(a[ODP_VPORT_ATTR_PORT_NO]); vport->type = nl_attr_get_u32(a[ODP_VPORT_ATTR_TYPE]); vport->name = nl_attr_get_string(a[ODP_VPORT_ATTR_NAME]); @@ -1246,6 +1222,8 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, } if (a[ODP_VPORT_ATTR_MTU]) { vport->mtu = nl_attr_get_u32(a[ODP_VPORT_ATTR_MTU]); + } else { + vport->mtu = INT_MAX; } if (a[ODP_VPORT_ATTR_OPTIONS]) { vport->options = nl_attr_get(a[ODP_VPORT_ATTR_OPTIONS]); @@ -1260,15 +1238,19 @@ dpif_linux_vport_from_ofpbuf(struct dpif_linux_vport *vport, return 0; } -/* Appends to 'buf' (which must initially be empty) a "struct odp_vport" +/* Appends to 'buf' (which must initially be empty) a "struct odp_header" * followed by Netlink attributes corresponding to 'vport'. */ static void dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, struct ofpbuf *buf) { - struct odp_vport *odp_vport; + struct odp_header *odp_header; + + nl_msg_put_genlmsghdr(buf, 0, odp_vport_family, NLM_F_REQUEST | NLM_F_ECHO, + vport->cmd, 1); - ofpbuf_reserve(buf, sizeof odp_vport); + odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); + odp_header->dp_ifindex = vport->dp_ifindex; if (vport->port_no != UINT32_MAX) { nl_msg_put_u32(buf, ODP_VPORT_ATTR_PORT_NO, vport->port_no); @@ -1292,7 +1274,7 @@ dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, vport->address, ETH_ADDR_LEN); } - if (vport->mtu) { + if (vport->mtu && vport->mtu != INT_MAX) { nl_msg_put_u32(buf, ODP_VPORT_ATTR_MTU, vport->mtu); } @@ -1308,11 +1290,6 @@ dpif_linux_vport_to_ofpbuf(const struct dpif_linux_vport *vport, if (vport->iflink) { nl_msg_put_u32(buf, ODP_VPORT_ATTR_IFLINK, vport->iflink); } - - odp_vport = ofpbuf_push_uninit(buf, sizeof *odp_vport); - odp_vport->dp_idx = vport->dp_idx; - odp_vport->len = buf->size; - odp_vport->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data; } /* Clears 'vport' to "empty" values. */ @@ -1320,7 +1297,6 @@ void dpif_linux_vport_init(struct dpif_linux_vport *vport) { memset(vport, 0, sizeof *vport); - vport->dp_idx = UINT32_MAX; vport->port_no = UINT32_MAX; } @@ -1335,42 +1311,34 @@ dpif_linux_vport_transact(const struct dpif_linux_vport *request, struct dpif_linux_vport *reply, struct ofpbuf **bufp) { - struct ofpbuf *buf = NULL; + struct ofpbuf *request_buf; int error; - int fd; assert((reply != NULL) == (bufp != NULL)); - error = get_dp0_fd(&fd); + error = dpif_linux_init(); if (error) { - goto error; + if (reply) { + *bufp = NULL; + dpif_linux_vport_init(reply); + } + return error; } - buf = ofpbuf_new(1024); - dpif_linux_vport_to_ofpbuf(request, buf); - - error = ioctl(fd, request->cmd, buf->data) ? errno : 0; - if (error) { - goto error; - } + request_buf = ofpbuf_new(1024); + dpif_linux_vport_to_ofpbuf(request, request_buf); + error = nl_sock_transact(genl_sock, request_buf, bufp); + ofpbuf_delete(request_buf); - if (bufp) { - buf->size = ((struct odp_vport *) buf->data)->len; - error = dpif_linux_vport_from_ofpbuf(reply, buf); + if (reply) { + if (!error) { + error = dpif_linux_vport_from_ofpbuf(reply, *bufp); + } if (error) { - goto error; + dpif_linux_vport_init(reply); + ofpbuf_delete(*bufp); + *bufp = NULL; } - *bufp = buf; - } else { - ofpbuf_delete(buf); - } - return 0; - -error: - ofpbuf_delete(buf); - if (bufp) { - memset(reply, 0, sizeof *reply); - *bufp = NULL; } return error; } @@ -1385,15 +1353,15 @@ dpif_linux_vport_get(const char *name, struct dpif_linux_vport *reply, struct dpif_linux_vport request; dpif_linux_vport_init(&request); - request.cmd = ODP_VPORT_GET; + request.cmd = ODP_VPORT_CMD_GET; request.name = name; return dpif_linux_vport_transact(&request, reply, bufp); } -/* Parses the contents of 'buf', which contains a "struct odp_datapath" - * followed by Netlink attributes, into 'dp'. Returns 0 if successful, - * otherwise a positive errno value. +/* Parses the contents of 'buf', which contains a "struct odp_header" followed + * by Netlink attributes, into 'dp'. Returns 0 if successful, otherwise a + * positive errno value. * * 'dp' will contain pointers into 'buf', so the caller should not free 'buf' * while 'dp' is still in use. */ @@ -1411,18 +1379,27 @@ dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) [ODP_DP_ATTR_MCGROUPS] = { .type = NL_A_NESTED, .optional = true }, }; - struct odp_datapath *odp_dp; struct nlattr *a[ARRAY_SIZE(odp_datapath_policy)]; + struct odp_header *odp_header; + struct nlmsghdr *nlmsg; + struct genlmsghdr *genl; + struct ofpbuf b; dpif_linux_dp_init(dp); - if (!nl_policy_parse(buf, sizeof *odp_dp, odp_datapath_policy, - a, ARRAY_SIZE(odp_datapath_policy))) { + ofpbuf_use_const(&b, buf->data, buf->size); + nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); + genl = ofpbuf_try_pull(&b, sizeof *genl); + odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); + if (!nlmsg || !genl || !odp_header + || nlmsg->nlmsg_type != odp_datapath_family + || !nl_policy_parse(&b, 0, odp_datapath_policy, a, + ARRAY_SIZE(odp_datapath_policy))) { return EINVAL; } - odp_dp = buf->data; - dp->dp_idx = odp_dp->dp_idx; + dp->cmd = genl->cmd; + dp->dp_ifindex = odp_header->dp_ifindex; dp->name = nl_attr_get_string(a[ODP_DP_ATTR_NAME]); if (a[ODP_DP_ATTR_STATS]) { /* Can't use structure assignment because Netlink doesn't ensure @@ -1468,14 +1445,17 @@ dpif_linux_dp_from_ofpbuf(struct dpif_linux_dp *dp, const struct ofpbuf *buf) return 0; } -/* Appends to 'buf' (which must initially be empty) a "struct odp_datapath" - * followed by Netlink attributes corresponding to 'dp'. */ +/* Appends to 'buf' the Generic Netlink message described by 'dp'. */ static void dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) { - struct odp_datapath *odp_dp; + struct odp_header *odp_header; + + nl_msg_put_genlmsghdr(buf, 0, odp_datapath_family, + NLM_F_REQUEST | NLM_F_ECHO, dp->cmd, 1); - ofpbuf_reserve(buf, sizeof odp_dp); + odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); + odp_header->dp_ifindex = dp->dp_ifindex; if (dp->name) { nl_msg_put_string(buf, ODP_DP_ATTR_NAME, dp->name); @@ -1490,68 +1470,59 @@ dpif_linux_dp_to_ofpbuf(const struct dpif_linux_dp *dp, struct ofpbuf *buf) if (dp->sampling) { nl_msg_put_u32(buf, ODP_DP_ATTR_SAMPLING, *dp->sampling); } - - odp_dp = ofpbuf_push_uninit(buf, sizeof *odp_dp); - odp_dp->dp_idx = dp->dp_idx; - odp_dp->len = buf->size; - odp_dp->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data; } /* Clears 'dp' to "empty" values. */ -void +static void dpif_linux_dp_init(struct dpif_linux_dp *dp) { memset(dp, 0, sizeof *dp); - dp->dp_idx = -1; +} + +static void +dpif_linux_dp_dump_start(struct nl_dump *dump) +{ + struct dpif_linux_dp request; + struct ofpbuf *buf; + + dpif_linux_dp_init(&request); + request.cmd = ODP_DP_CMD_GET; + + buf = ofpbuf_new(1024); + dpif_linux_dp_to_ofpbuf(&request, buf); + nl_dump_start(dump, genl_sock, buf); + ofpbuf_delete(buf); } /* Executes 'request' in the kernel datapath. If the command fails, returns a * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the - * result of the command is expected to be an odp_datapath also, which is - * decoded and stored in '*reply' and '*bufp'. The caller must free '*bufp' - * when the reply is no longer needed ('reply' will contain pointers into - * '*bufp'). */ -int + * result of the command is expected to be of the same form, which is decoded + * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the + * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ +static int dpif_linux_dp_transact(const struct dpif_linux_dp *request, struct dpif_linux_dp *reply, struct ofpbuf **bufp) { - struct ofpbuf *buf = NULL; + struct ofpbuf *request_buf; int error; - int fd; assert((reply != NULL) == (bufp != NULL)); - error = get_dp0_fd(&fd); - if (error) { - goto error; - } - - buf = ofpbuf_new(1024); - dpif_linux_dp_to_ofpbuf(request, buf); - - error = ioctl(fd, request->cmd, buf->data) ? errno : 0; - if (error) { - goto error; - } + request_buf = ofpbuf_new(1024); + dpif_linux_dp_to_ofpbuf(request, request_buf); + error = nl_sock_transact(genl_sock, request_buf, bufp); + ofpbuf_delete(request_buf); - if (bufp) { - buf->size = ((struct odp_datapath *) buf->data)->len; - error = dpif_linux_dp_from_ofpbuf(reply, buf); + if (reply) { + if (!error) { + error = dpif_linux_dp_from_ofpbuf(reply, *bufp); + } if (error) { - goto error; + dpif_linux_dp_init(reply); + ofpbuf_delete(*bufp); + *bufp = NULL; } - *bufp = buf; - } else { - ofpbuf_delete(buf); - } - return 0; - -error: - ofpbuf_delete(buf); - if (bufp) { - memset(reply, 0, sizeof *reply); - *bufp = NULL; } return error; } @@ -1559,7 +1530,7 @@ error: /* Obtains information about 'dpif_' and stores it into '*reply' and '*bufp'. * The caller must free '*bufp' when the reply is no longer needed ('reply' * will contain pointers into '*bufp'). */ -int +static int dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply, struct ofpbuf **bufp) { @@ -1567,14 +1538,14 @@ dpif_linux_dp_get(const struct dpif *dpif_, struct dpif_linux_dp *reply, struct dpif_linux_dp request; dpif_linux_dp_init(&request); - request.cmd = ODP_DP_GET; - request.dp_idx = dpif->minor; + request.cmd = ODP_DP_CMD_GET; + request.dp_ifindex = dpif->dp_ifindex; return dpif_linux_dp_transact(&request, reply, bufp); } -/* Parses the contents of 'buf', which contains a "struct odp_flow" followed by - * Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a +/* Parses the contents of 'buf', which contains a "struct odp_header" followed + * by Netlink attributes, into 'flow'. Returns 0 if successful, otherwise a * positive errno value. * * 'flow' will contain pointers into 'buf', so the caller should not free 'buf' @@ -1593,22 +1564,29 @@ dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, [ODP_FLOW_ATTR_TCP_FLAGS] = { .type = NL_A_U8, .optional = true }, [ODP_FLOW_ATTR_USED] = { .type = NL_A_U64, .optional = true }, /* The kernel never uses ODP_FLOW_ATTR_CLEAR. */ - [ODP_FLOW_ATTR_STATE] = { .type = NL_A_U64, .optional = true }, }; - struct odp_flow *odp_flow; struct nlattr *a[ARRAY_SIZE(odp_flow_policy)]; + struct odp_header *odp_header; + struct nlmsghdr *nlmsg; + struct genlmsghdr *genl; + struct ofpbuf b; dpif_linux_flow_init(flow); - if (!nl_policy_parse(buf, sizeof *odp_flow, odp_flow_policy, - a, ARRAY_SIZE(odp_flow_policy))) { + ofpbuf_use_const(&b, buf->data, buf->size); + nlmsg = ofpbuf_try_pull(&b, sizeof *nlmsg); + genl = ofpbuf_try_pull(&b, sizeof *genl); + odp_header = ofpbuf_try_pull(&b, sizeof *odp_header); + if (!nlmsg || !genl || !odp_header + || nlmsg->nlmsg_type != odp_flow_family + || !nl_policy_parse(&b, 0, odp_flow_policy, a, + ARRAY_SIZE(odp_flow_policy))) { return EINVAL; } - odp_flow = buf->data; - flow->nlmsg_flags = odp_flow->nlmsg_flags; - flow->dp_idx = odp_flow->dp_idx; + flow->nlmsg_flags = nlmsg->nlmsg_flags; + flow->dp_ifindex = odp_header->dp_ifindex; flow->key = nl_attr_get(a[ODP_FLOW_ATTR_KEY]); flow->key_len = nl_attr_get_size(a[ODP_FLOW_ATTR_KEY]); if (a[ODP_FLOW_ATTR_ACTIONS]) { @@ -1621,27 +1599,32 @@ dpif_linux_flow_from_ofpbuf(struct dpif_linux_flow *flow, if (a[ODP_FLOW_ATTR_TCP_FLAGS]) { flow->tcp_flags = nl_attr_get(a[ODP_FLOW_ATTR_TCP_FLAGS]); } - if (a[ODP_FLOW_ATTR_STATE]) { - flow->state = nl_attr_get(a[ODP_FLOW_ATTR_STATE]); + if (a[ODP_FLOW_ATTR_USED]) { + flow->used = nl_attr_get(a[ODP_FLOW_ATTR_USED]); } return 0; } -/* Appends to 'buf' (which must initially be empty) a "struct odp_flow" +/* Appends to 'buf' (which must initially be empty) a "struct odp_header" * followed by Netlink attributes corresponding to 'flow'. */ static void dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, struct ofpbuf *buf) { - struct odp_flow *odp_flow; + struct odp_header *odp_header; - ofpbuf_reserve(buf, sizeof odp_flow); + nl_msg_put_genlmsghdr(buf, 0, odp_flow_family, + NLM_F_REQUEST | NLM_F_ECHO | flow->nlmsg_flags, + flow->cmd, 1); + + odp_header = ofpbuf_put_uninit(buf, sizeof *odp_header); + odp_header->dp_ifindex = flow->dp_ifindex; if (flow->key_len) { nl_msg_put_unspec(buf, ODP_FLOW_ATTR_KEY, flow->key, flow->key_len); } - if (flow->actions_len) { + if (flow->actions || flow->actions_len) { nl_msg_put_unspec(buf, ODP_FLOW_ATTR_ACTIONS, flow->actions, flow->actions_len); } @@ -1654,21 +1637,10 @@ dpif_linux_flow_to_ofpbuf(const struct dpif_linux_flow *flow, if (flow->clear) { nl_msg_put_flag(buf, ODP_FLOW_ATTR_CLEAR); } - - if (flow->state) { - nl_msg_put_u64(buf, ODP_FLOW_ATTR_STATE, - get_unaligned_u64(flow->state)); - } - - odp_flow = ofpbuf_push_uninit(buf, sizeof *odp_flow); - odp_flow->nlmsg_flags = flow->nlmsg_flags; - odp_flow->dp_idx = flow->dp_idx; - odp_flow->len = buf->size; - odp_flow->total_len = (char *) ofpbuf_end(buf) - (char *) buf->data; } /* Clears 'flow' to "empty" values. */ -void +static void dpif_linux_flow_init(struct dpif_linux_flow *flow) { memset(flow, 0, sizeof *flow); @@ -1677,49 +1649,32 @@ dpif_linux_flow_init(struct dpif_linux_flow *flow) /* Executes 'request' in the kernel datapath. If the command fails, returns a * positive errno value. Otherwise, if 'reply' and 'bufp' are null, returns 0 * without doing anything else. If 'reply' and 'bufp' are nonnull, then the - * result of the command is expected to be an odp_flow also, which is decoded - * and stored in '*reply' and '*bufp'. The caller must free '*bufp' when the - * reply is no longer needed ('reply' will contain pointers into '*bufp'). */ -int + * result of the command is expected to be a flow also, which is decoded and + * stored in '*reply' and '*bufp'. The caller must free '*bufp' when the reply + * is no longer needed ('reply' will contain pointers into '*bufp'). */ +static int dpif_linux_flow_transact(const struct dpif_linux_flow *request, struct dpif_linux_flow *reply, struct ofpbuf **bufp) { - struct ofpbuf *buf = NULL; + struct ofpbuf *request_buf; int error; - int fd; assert((reply != NULL) == (bufp != NULL)); - error = get_dp0_fd(&fd); - if (error) { - goto error; - } - - buf = ofpbuf_new(1024); - dpif_linux_flow_to_ofpbuf(request, buf); - - error = ioctl(fd, request->cmd, buf->data) ? errno : 0; - if (error) { - goto error; - } + request_buf = ofpbuf_new(1024); + dpif_linux_flow_to_ofpbuf(request, request_buf); + error = nl_sock_transact(genl_sock, request_buf, bufp); + ofpbuf_delete(request_buf); - if (bufp) { - buf->size = ((struct odp_flow *) buf->data)->len; - error = dpif_linux_flow_from_ofpbuf(reply, buf); + if (reply) { + if (!error) { + error = dpif_linux_flow_from_ofpbuf(reply, *bufp); + } if (error) { - goto error; + dpif_linux_flow_init(reply); + ofpbuf_delete(*bufp); + *bufp = NULL; } - *bufp = buf; - } else { - ofpbuf_delete(buf); - } - return 0; - -error: - ofpbuf_delete(buf); - if (bufp) { - memset(reply, 0, sizeof *reply); - *bufp = NULL; } return error; }