From 8c63ff09bddf944ab0033fea97aacfadfffa76de Mon Sep 17 00:00:00 2001 From: Joe Perches Date: Tue, 18 Feb 2014 11:15:45 -0800 Subject: openvswitch: Use ether_addr_copy It's slightly smaller/faster for some architectures. Signed-off-by: Joe Perches Signed-off-by: Jesse Gross --- net/openvswitch/flow.c | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) (limited to 'net/openvswitch/flow.c') diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c index 2998989e76d..332aa01492f 100644 --- a/net/openvswitch/flow.c +++ b/net/openvswitch/flow.c @@ -372,14 +372,14 @@ static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key, && opt_len == 8) { if (unlikely(!is_zero_ether_addr(key->ipv6.nd.sll))) goto invalid; - memcpy(key->ipv6.nd.sll, - &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN); + ether_addr_copy(key->ipv6.nd.sll, + &nd->opt[offset+sizeof(*nd_opt)]); } else if (nd_opt->nd_opt_type == ND_OPT_TARGET_LL_ADDR && opt_len == 8) { if (unlikely(!is_zero_ether_addr(key->ipv6.nd.tll))) goto invalid; - memcpy(key->ipv6.nd.tll, - &nd->opt[offset+sizeof(*nd_opt)], ETH_ALEN); + ether_addr_copy(key->ipv6.nd.tll, + &nd->opt[offset+sizeof(*nd_opt)]); } icmp_len -= opt_len; @@ -439,8 +439,8 @@ int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key) * header in the linear data area. */ eth = eth_hdr(skb); - memcpy(key->eth.src, eth->h_source, ETH_ALEN); - memcpy(key->eth.dst, eth->h_dest, ETH_ALEN); + ether_addr_copy(key->eth.src, eth->h_source); + ether_addr_copy(key->eth.dst, eth->h_dest); __skb_pull(skb, 2 * ETH_ALEN); /* We are going to push all headers that we pull, so no need to @@ -538,8 +538,8 @@ int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key) key->ip.proto = ntohs(arp->ar_op); memcpy(&key->ipv4.addr.src, arp->ar_sip, sizeof(key->ipv4.addr.src)); memcpy(&key->ipv4.addr.dst, arp->ar_tip, sizeof(key->ipv4.addr.dst)); - memcpy(key->ipv4.arp.sha, arp->ar_sha, ETH_ALEN); - memcpy(key->ipv4.arp.tha, arp->ar_tha, ETH_ALEN); + ether_addr_copy(key->ipv4.arp.sha, arp->ar_sha); + ether_addr_copy(key->ipv4.arp.tha, arp->ar_tha); } } else if (key->eth.type == htons(ETH_P_IPV6)) { int nh_len; /* IPv6 Header + Extensions */ -- cgit v1.2.3-70-g09d2 From 23dabf88abb48a866fdb19ee08ebcf1ddd9b1840 Mon Sep 17 00:00:00 2001 From: Jarno Rajahalme Date: Thu, 27 Mar 2014 12:35:23 -0700 Subject: openvswitch: Remove 5-tuple optimization. The 5-tuple optimization becomes unnecessary with a later per-NUMA node stats patch. Remove it first to make the changes easier to grasp. Signed-off-by: Jarno Rajahalme Signed-off-by: Jesse Gross --- net/openvswitch/datapath.c | 11 ++++---- net/openvswitch/flow.c | 32 +++++++++-------------- net/openvswitch/flow.h | 10 +------- net/openvswitch/flow_netlink.c | 58 +++--------------------------------------- net/openvswitch/flow_netlink.h | 1 - net/openvswitch/flow_table.c | 31 +++++++--------------- net/openvswitch/flow_table.h | 2 +- 7 files changed, 32 insertions(+), 113 deletions(-) (limited to 'net/openvswitch/flow.c') diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c index a3276e3c4fe..8867d7e2d65 100644 --- a/net/openvswitch/datapath.c +++ b/net/openvswitch/datapath.c @@ -524,7 +524,7 @@ static int ovs_packet_cmd_execute(struct sk_buff *skb, struct genl_info *info) packet->protocol = htons(ETH_P_802_2); /* Build an sw_flow for sending this packet. */ - flow = ovs_flow_alloc(false); + flow = ovs_flow_alloc(); err = PTR_ERR(flow); if (IS_ERR(flow)) goto err_kfree_skb; @@ -782,7 +782,6 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) struct datapath *dp; struct sw_flow_actions *acts = NULL; struct sw_flow_match match; - bool exact_5tuple; int error; /* Extract key. */ @@ -791,7 +790,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) goto error; ovs_match_init(&match, &key, &mask); - error = ovs_nla_get_match(&match, &exact_5tuple, + error = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], a[OVS_FLOW_ATTR_MASK]); if (error) goto error; @@ -830,7 +829,7 @@ static int ovs_flow_cmd_new_or_set(struct sk_buff *skb, struct genl_info *info) goto err_unlock_ovs; /* Allocate flow. */ - flow = ovs_flow_alloc(!exact_5tuple); + flow = ovs_flow_alloc(); if (IS_ERR(flow)) { error = PTR_ERR(flow); goto err_unlock_ovs; @@ -914,7 +913,7 @@ static int ovs_flow_cmd_get(struct sk_buff *skb, struct genl_info *info) } ovs_match_init(&match, &key, NULL); - err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL); + err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL); if (err) return err; @@ -968,7 +967,7 @@ static int ovs_flow_cmd_del(struct sk_buff *skb, struct genl_info *info) } ovs_match_init(&match, &key, NULL); - err = ovs_nla_get_match(&match, NULL, a[OVS_FLOW_ATTR_KEY], NULL); + err = ovs_nla_get_match(&match, a[OVS_FLOW_ATTR_KEY], NULL); if (err) goto unlock; diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c index 332aa01492f..aad7a8da70b 100644 --- a/net/openvswitch/flow.c +++ b/net/openvswitch/flow.c @@ -66,10 +66,7 @@ void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb) struct flow_stats *stats; __be16 tcp_flags = 0; - if (!flow->stats.is_percpu) - stats = flow->stats.stat; - else - stats = this_cpu_ptr(flow->stats.cpu_stats); + stats = this_cpu_ptr(flow->stats); if ((flow->key.eth.type == htons(ETH_P_IP) || flow->key.eth.type == htons(ETH_P_IPV6)) && @@ -110,16 +107,14 @@ void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats, memset(ovs_stats, 0, sizeof(*ovs_stats)); local_bh_disable(); - if (!flow->stats.is_percpu) { - stats_read(flow->stats.stat, ovs_stats, used, tcp_flags); - } else { - for_each_possible_cpu(cpu) { - struct flow_stats *stats; - - stats = per_cpu_ptr(flow->stats.cpu_stats, cpu); - stats_read(stats, ovs_stats, used, tcp_flags); - } + + for_each_possible_cpu(cpu) { + struct flow_stats *stats; + + stats = per_cpu_ptr(flow->stats.cpu_stats, cpu); + stats_read(stats, ovs_stats, used, tcp_flags); } + local_bh_enable(); } @@ -138,13 +133,10 @@ void ovs_flow_stats_clear(struct sw_flow *flow) int cpu; local_bh_disable(); - if (!flow->stats.is_percpu) { - stats_reset(flow->stats.stat); - } else { - for_each_possible_cpu(cpu) { - stats_reset(per_cpu_ptr(flow->stats.cpu_stats, cpu)); - } - } + + for_each_possible_cpu(cpu) + stats_reset(per_cpu_ptr(flow->stats, cpu)); + local_bh_enable(); } diff --git a/net/openvswitch/flow.h b/net/openvswitch/flow.h index 2d770e28a3a..9c0dd8aa311 100644 --- a/net/openvswitch/flow.h +++ b/net/openvswitch/flow.h @@ -155,14 +155,6 @@ struct flow_stats { __be16 tcp_flags; /* Union of seen TCP flags. */ }; -struct sw_flow_stats { - bool is_percpu; - union { - struct flow_stats *stat; - struct flow_stats __percpu *cpu_stats; - }; -}; - struct sw_flow { struct rcu_head rcu; struct hlist_node hash_node[2]; @@ -172,7 +164,7 @@ struct sw_flow { struct sw_flow_key unmasked_key; struct sw_flow_mask *mask; struct sw_flow_actions __rcu *sf_acts; - struct sw_flow_stats stats; + struct flow_stats __percpu *stats; }; struct arp_eth_header { diff --git a/net/openvswitch/flow_netlink.c b/net/openvswitch/flow_netlink.c index 5511ad18be5..84caa99b376 100644 --- a/net/openvswitch/flow_netlink.c +++ b/net/openvswitch/flow_netlink.c @@ -268,20 +268,6 @@ static bool is_all_zero(const u8 *fp, size_t size) return true; } -static bool is_all_set(const u8 *fp, size_t size) -{ - int i; - - if (!fp) - return false; - - for (i = 0; i < size; i++) - if (fp[i] != 0xff) - return false; - - return true; -} - static int __parse_flow_nlattrs(const struct nlattr *attr, const struct nlattr *a[], u64 *attrsp, bool nz) @@ -503,9 +489,8 @@ static int metadata_from_nlattrs(struct sw_flow_match *match, u64 *attrs, return 0; } -static int ovs_key_from_nlattrs(struct sw_flow_match *match, bool *exact_5tuple, - u64 attrs, const struct nlattr **a, - bool is_mask) +static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs, + const struct nlattr **a, bool is_mask) { int err; u64 orig_attrs = attrs; @@ -562,11 +547,6 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, bool *exact_5tuple SW_FLOW_KEY_PUT(match, eth.type, htons(ETH_P_802_2), is_mask); } - if (is_mask && exact_5tuple) { - if (match->mask->key.eth.type != htons(0xffff)) - *exact_5tuple = false; - } - if (attrs & (1 << OVS_KEY_ATTR_IPV4)) { const struct ovs_key_ipv4 *ipv4_key; @@ -589,13 +569,6 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, bool *exact_5tuple SW_FLOW_KEY_PUT(match, ipv4.addr.dst, ipv4_key->ipv4_dst, is_mask); attrs &= ~(1 << OVS_KEY_ATTR_IPV4); - - if (is_mask && exact_5tuple && *exact_5tuple) { - if (ipv4_key->ipv4_proto != 0xff || - ipv4_key->ipv4_src != htonl(0xffffffff) || - ipv4_key->ipv4_dst != htonl(0xffffffff)) - *exact_5tuple = false; - } } if (attrs & (1 << OVS_KEY_ATTR_IPV6)) { @@ -627,15 +600,6 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, bool *exact_5tuple is_mask); attrs &= ~(1 << OVS_KEY_ATTR_IPV6); - - if (is_mask && exact_5tuple && *exact_5tuple) { - if (ipv6_key->ipv6_proto != 0xff || - !is_all_set((const u8 *)ipv6_key->ipv6_src, - sizeof(match->key->ipv6.addr.src)) || - !is_all_set((const u8 *)ipv6_key->ipv6_dst, - sizeof(match->key->ipv6.addr.dst))) - *exact_5tuple = false; - } } if (attrs & (1 << OVS_KEY_ATTR_ARP)) { @@ -678,11 +642,6 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, bool *exact_5tuple tcp_key->tcp_dst, is_mask); } attrs &= ~(1 << OVS_KEY_ATTR_TCP); - - if (is_mask && exact_5tuple && *exact_5tuple && - (tcp_key->tcp_src != htons(0xffff) || - tcp_key->tcp_dst != htons(0xffff))) - *exact_5tuple = false; } if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) { @@ -714,11 +673,6 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, bool *exact_5tuple udp_key->udp_dst, is_mask); } attrs &= ~(1 << OVS_KEY_ATTR_UDP); - - if (is_mask && exact_5tuple && *exact_5tuple && - (udp_key->udp_src != htons(0xffff) || - udp_key->udp_dst != htons(0xffff))) - *exact_5tuple = false; } if (attrs & (1 << OVS_KEY_ATTR_SCTP)) { @@ -804,7 +758,6 @@ static void sw_flow_mask_set(struct sw_flow_mask *mask, * attribute specifies the mask field of the wildcarded flow. */ int ovs_nla_get_match(struct sw_flow_match *match, - bool *exact_5tuple, const struct nlattr *key, const struct nlattr *mask) { @@ -852,13 +805,10 @@ int ovs_nla_get_match(struct sw_flow_match *match, } } - err = ovs_key_from_nlattrs(match, NULL, key_attrs, a, false); + err = ovs_key_from_nlattrs(match, key_attrs, a, false); if (err) return err; - if (exact_5tuple) - *exact_5tuple = true; - if (mask) { err = parse_flow_mask_nlattrs(mask, a, &mask_attrs); if (err) @@ -896,7 +846,7 @@ int ovs_nla_get_match(struct sw_flow_match *match, } } - err = ovs_key_from_nlattrs(match, exact_5tuple, mask_attrs, a, true); + err = ovs_key_from_nlattrs(match, mask_attrs, a, true); if (err) return err; } else { diff --git a/net/openvswitch/flow_netlink.h b/net/openvswitch/flow_netlink.h index b31fbe28bc7..440151045d3 100644 --- a/net/openvswitch/flow_netlink.h +++ b/net/openvswitch/flow_netlink.h @@ -45,7 +45,6 @@ int ovs_nla_put_flow(const struct sw_flow_key *, int ovs_nla_get_flow_metadata(struct sw_flow *flow, const struct nlattr *attr); int ovs_nla_get_match(struct sw_flow_match *match, - bool *exact_5tuple, const struct nlattr *, const struct nlattr *); diff --git a/net/openvswitch/flow_table.c b/net/openvswitch/flow_table.c index 1ba1e0b8ade..aa92da23053 100644 --- a/net/openvswitch/flow_table.c +++ b/net/openvswitch/flow_table.c @@ -72,7 +72,7 @@ void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src, *d++ = *s++ & *m++; } -struct sw_flow *ovs_flow_alloc(bool percpu_stats) +struct sw_flow *ovs_flow_alloc(void) { struct sw_flow *flow; int cpu; @@ -84,25 +84,15 @@ struct sw_flow *ovs_flow_alloc(bool percpu_stats) flow->sf_acts = NULL; flow->mask = NULL; - flow->stats.is_percpu = percpu_stats; + flow->stats = alloc_percpu(struct flow_stats); + if (!flow->stats) + goto err; - if (!percpu_stats) { - flow->stats.stat = kzalloc(sizeof(*flow->stats.stat), GFP_KERNEL); - if (!flow->stats.stat) - goto err; + for_each_possible_cpu(cpu) { + struct flow_stats *cpu_stats; - spin_lock_init(&flow->stats.stat->lock); - } else { - flow->stats.cpu_stats = alloc_percpu(struct flow_stats); - if (!flow->stats.cpu_stats) - goto err; - - for_each_possible_cpu(cpu) { - struct flow_stats *cpu_stats; - - cpu_stats = per_cpu_ptr(flow->stats.cpu_stats, cpu); - spin_lock_init(&cpu_stats->lock); - } + cpu_stats = per_cpu_ptr(flow->stats, cpu); + spin_lock_init(&cpu_stats->lock); } return flow; err: @@ -141,10 +131,7 @@ static struct flex_array *alloc_buckets(unsigned int n_buckets) static void flow_free(struct sw_flow *flow) { kfree((struct sf_flow_acts __force *)flow->sf_acts); - if (flow->stats.is_percpu) - free_percpu(flow->stats.cpu_stats); - else - kfree(flow->stats.stat); + free_percpu(flow->stats); kmem_cache_free(flow_cache, flow); } diff --git a/net/openvswitch/flow_table.h b/net/openvswitch/flow_table.h index baaeb101924..c26c59a7ab5 100644 --- a/net/openvswitch/flow_table.h +++ b/net/openvswitch/flow_table.h @@ -55,7 +55,7 @@ struct flow_table { int ovs_flow_init(void); void ovs_flow_exit(void); -struct sw_flow *ovs_flow_alloc(bool percpu_stats); +struct sw_flow *ovs_flow_alloc(void); void ovs_flow_free(struct sw_flow *, bool deferred); int ovs_flow_tbl_init(struct flow_table *); -- cgit v1.2.3-70-g09d2 From 63e7959c4b9bd6f791061c460a22d9ee32ae2240 Mon Sep 17 00:00:00 2001 From: Jarno Rajahalme Date: Thu, 27 Mar 2014 12:42:54 -0700 Subject: openvswitch: Per NUMA node flow stats. Keep kernel flow stats for each NUMA node rather than each (logical) CPU. This avoids using the per-CPU allocator and removes most of the kernel-side OVS locking overhead otherwise on the top of perf reports and allows OVS to scale better with higher number of threads. With 9 handlers and 4 revalidators netperf TCP_CRR test flow setup rate doubles on a server with two hyper-threaded physical CPUs (16 logical cores each) compared to the current OVS master. Tested with non-trivial flow table with a TCP port match rule forcing all new connections with unique port numbers to OVS userspace. The IP addresses are still wildcarded, so the kernel flows are not considered as exact match 5-tuple flows. This type of flows can be expected to appear in large numbers as the result of more effective wildcarding made possible by improvements in OVS userspace flow classifier. Perf results for this test (master): Events: 305K cycles + 8.43% ovs-vswitchd [kernel.kallsyms] [k] mutex_spin_on_owner + 5.64% ovs-vswitchd [kernel.kallsyms] [k] __ticket_spin_lock + 4.75% ovs-vswitchd ovs-vswitchd [.] find_match_wc + 3.32% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_lock + 2.61% ovs-vswitchd [kernel.kallsyms] [k] pcpu_alloc_area + 2.19% ovs-vswitchd ovs-vswitchd [.] flow_hash_in_minimask_range + 2.03% swapper [kernel.kallsyms] [k] intel_idle + 1.84% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_unlock + 1.64% ovs-vswitchd ovs-vswitchd [.] classifier_lookup + 1.58% ovs-vswitchd libc-2.15.so [.] 0x7f4e6 + 1.07% ovs-vswitchd [kernel.kallsyms] [k] memset + 1.03% netperf [kernel.kallsyms] [k] __ticket_spin_lock + 0.92% swapper [kernel.kallsyms] [k] __ticket_spin_lock ... And after this patch: Events: 356K cycles + 6.85% ovs-vswitchd ovs-vswitchd [.] find_match_wc + 4.63% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_lock + 3.06% ovs-vswitchd [kernel.kallsyms] [k] __ticket_spin_lock + 2.81% ovs-vswitchd ovs-vswitchd [.] flow_hash_in_minimask_range + 2.51% ovs-vswitchd libpthread-2.15.so [.] pthread_mutex_unlock + 2.27% ovs-vswitchd ovs-vswitchd [.] classifier_lookup + 1.84% ovs-vswitchd libc-2.15.so [.] 0x15d30f + 1.74% ovs-vswitchd [kernel.kallsyms] [k] mutex_spin_on_owner + 1.47% swapper [kernel.kallsyms] [k] intel_idle + 1.34% ovs-vswitchd ovs-vswitchd [.] flow_hash_in_minimask + 1.33% ovs-vswitchd ovs-vswitchd [.] rule_actions_unref + 1.16% ovs-vswitchd ovs-vswitchd [.] hindex_node_with_hash + 1.16% ovs-vswitchd ovs-vswitchd [.] do_xlate_actions + 1.09% ovs-vswitchd ovs-vswitchd [.] ofproto_rule_ref + 1.01% netperf [kernel.kallsyms] [k] __ticket_spin_lock ... There is a small increase in kernel spinlock overhead due to the same spinlock being shared between multiple cores of the same physical CPU, but that is barely visible in the netperf TCP_CRR test performance (maybe ~1% performance drop, hard to tell exactly due to variance in the test results), when testing for kernel module throughput (with no userspace activity, handful of kernel flows). On flow setup, a single stats instance is allocated (for the NUMA node 0). As CPUs from multiple NUMA nodes start updating stats, new NUMA-node specific stats instances are allocated. This allocation on the packet processing code path is made to never block or look for emergency memory pools, minimizing the allocation latency. If the allocation fails, the existing preallocated stats instance is used. Also, if only CPUs from one NUMA-node are updating the preallocated stats instance, no additional stats instances are allocated. This eliminates the need to pre-allocate stats instances that will not be used, also relieving the stats reader from the burden of reading stats that are never used. Signed-off-by: Jarno Rajahalme Acked-by: Pravin B Shelar Signed-off-by: Jesse Gross --- net/openvswitch/flow.c | 119 ++++++++++++++++++++++++++++--------------- net/openvswitch/flow.h | 10 +++- net/openvswitch/flow_table.c | 46 +++++++++++++---- net/openvswitch/flow_table.h | 2 + 4 files changed, 122 insertions(+), 55 deletions(-) (limited to 'net/openvswitch/flow.c') diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c index aad7a8da70b..432f04d5c89 100644 --- a/net/openvswitch/flow.c +++ b/net/openvswitch/flow.c @@ -65,8 +65,9 @@ void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb) { struct flow_stats *stats; __be16 tcp_flags = 0; + int node = numa_node_id(); - stats = this_cpu_ptr(flow->stats); + stats = rcu_dereference(flow->stats[node]); if ((flow->key.eth.type == htons(ETH_P_IP) || flow->key.eth.type == htons(ETH_P_IPV6)) && @@ -76,68 +77,102 @@ void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb) tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb)); } - spin_lock(&stats->lock); + /* Check if already have node-specific stats. */ + if (likely(stats)) { + spin_lock(&stats->lock); + /* Mark if we write on the pre-allocated stats. */ + if (node == 0 && unlikely(flow->stats_last_writer != node)) + flow->stats_last_writer = node; + } else { + stats = rcu_dereference(flow->stats[0]); /* Pre-allocated. */ + spin_lock(&stats->lock); + + /* If the current NUMA-node is the only writer on the + * pre-allocated stats keep using them. + */ + if (unlikely(flow->stats_last_writer != node)) { + /* A previous locker may have already allocated the + * stats, so we need to check again. If node-specific + * stats were already allocated, we update the pre- + * allocated stats as we have already locked them. + */ + if (likely(flow->stats_last_writer != NUMA_NO_NODE) + && likely(!rcu_dereference(flow->stats[node]))) { + /* Try to allocate node-specific stats. */ + struct flow_stats *new_stats; + + new_stats = + kmem_cache_alloc_node(flow_stats_cache, + GFP_THISNODE | + __GFP_NOMEMALLOC, + node); + if (likely(new_stats)) { + new_stats->used = jiffies; + new_stats->packet_count = 1; + new_stats->byte_count = skb->len; + new_stats->tcp_flags = tcp_flags; + spin_lock_init(&new_stats->lock); + + rcu_assign_pointer(flow->stats[node], + new_stats); + goto unlock; + } + } + flow->stats_last_writer = node; + } + } + stats->used = jiffies; stats->packet_count++; stats->byte_count += skb->len; stats->tcp_flags |= tcp_flags; - spin_unlock(&stats->lock); -} - -static void stats_read(struct flow_stats *stats, - struct ovs_flow_stats *ovs_stats, - unsigned long *used, __be16 *tcp_flags) -{ - spin_lock(&stats->lock); - if (!*used || time_after(stats->used, *used)) - *used = stats->used; - *tcp_flags |= stats->tcp_flags; - ovs_stats->n_packets += stats->packet_count; - ovs_stats->n_bytes += stats->byte_count; +unlock: spin_unlock(&stats->lock); } void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats, unsigned long *used, __be16 *tcp_flags) { - int cpu; + int node; *used = 0; *tcp_flags = 0; memset(ovs_stats, 0, sizeof(*ovs_stats)); - local_bh_disable(); - - for_each_possible_cpu(cpu) { - struct flow_stats *stats; + for_each_node(node) { + struct flow_stats *stats = rcu_dereference(flow->stats[node]); - stats = per_cpu_ptr(flow->stats.cpu_stats, cpu); - stats_read(stats, ovs_stats, used, tcp_flags); + if (stats) { + /* Local CPU may write on non-local stats, so we must + * block bottom-halves here. + */ + spin_lock_bh(&stats->lock); + if (!*used || time_after(stats->used, *used)) + *used = stats->used; + *tcp_flags |= stats->tcp_flags; + ovs_stats->n_packets += stats->packet_count; + ovs_stats->n_bytes += stats->byte_count; + spin_unlock_bh(&stats->lock); + } } - - local_bh_enable(); -} - -static void stats_reset(struct flow_stats *stats) -{ - spin_lock(&stats->lock); - stats->used = 0; - stats->packet_count = 0; - stats->byte_count = 0; - stats->tcp_flags = 0; - spin_unlock(&stats->lock); } void ovs_flow_stats_clear(struct sw_flow *flow) { - int cpu; - - local_bh_disable(); - - for_each_possible_cpu(cpu) - stats_reset(per_cpu_ptr(flow->stats, cpu)); - - local_bh_enable(); + int node; + + for_each_node(node) { + struct flow_stats *stats = rcu_dereference(flow->stats[node]); + + if (stats) { + spin_lock_bh(&stats->lock); + stats->used = 0; + stats->packet_count = 0; + stats->byte_count = 0; + stats->tcp_flags = 0; + spin_unlock_bh(&stats->lock); + } + } } static int check_header(struct sk_buff *skb, int len) diff --git a/net/openvswitch/flow.h b/net/openvswitch/flow.h index 9c0dd8aa311..ddcebc53224 100644 --- a/net/openvswitch/flow.h +++ b/net/openvswitch/flow.h @@ -159,12 +159,18 @@ struct sw_flow { struct rcu_head rcu; struct hlist_node hash_node[2]; u32 hash; - + int stats_last_writer; /* NUMA-node id of the last writer on + * 'stats[0]'. + */ struct sw_flow_key key; struct sw_flow_key unmasked_key; struct sw_flow_mask *mask; struct sw_flow_actions __rcu *sf_acts; - struct flow_stats __percpu *stats; + struct flow_stats __rcu *stats[]; /* One for each NUMA node. First one + * is allocated at flow creation time, + * the rest are allocated on demand + * while holding the 'stats[0].lock'. + */ }; struct arp_eth_header { diff --git a/net/openvswitch/flow_table.c b/net/openvswitch/flow_table.c index aa92da23053..d8ef37b937b 100644 --- a/net/openvswitch/flow_table.c +++ b/net/openvswitch/flow_table.c @@ -48,6 +48,7 @@ #define REHASH_INTERVAL (10 * 60 * HZ) static struct kmem_cache *flow_cache; +struct kmem_cache *flow_stats_cache __read_mostly; static u16 range_n_bytes(const struct sw_flow_key_range *range) { @@ -75,7 +76,8 @@ void ovs_flow_mask_key(struct sw_flow_key *dst, const struct sw_flow_key *src, struct sw_flow *ovs_flow_alloc(void) { struct sw_flow *flow; - int cpu; + struct flow_stats *stats; + int node; flow = kmem_cache_alloc(flow_cache, GFP_KERNEL); if (!flow) @@ -83,17 +85,22 @@ struct sw_flow *ovs_flow_alloc(void) flow->sf_acts = NULL; flow->mask = NULL; + flow->stats_last_writer = NUMA_NO_NODE; - flow->stats = alloc_percpu(struct flow_stats); - if (!flow->stats) + /* Initialize the default stat node. */ + stats = kmem_cache_alloc_node(flow_stats_cache, + GFP_KERNEL | __GFP_ZERO, 0); + if (!stats) goto err; - for_each_possible_cpu(cpu) { - struct flow_stats *cpu_stats; + spin_lock_init(&stats->lock); + + RCU_INIT_POINTER(flow->stats[0], stats); + + for_each_node(node) + if (node != 0) + RCU_INIT_POINTER(flow->stats[node], NULL); - cpu_stats = per_cpu_ptr(flow->stats, cpu); - spin_lock_init(&cpu_stats->lock); - } return flow; err: kmem_cache_free(flow_cache, flow); @@ -130,8 +137,13 @@ static struct flex_array *alloc_buckets(unsigned int n_buckets) static void flow_free(struct sw_flow *flow) { + int node; + kfree((struct sf_flow_acts __force *)flow->sf_acts); - free_percpu(flow->stats); + for_each_node(node) + if (flow->stats[node]) + kmem_cache_free(flow_stats_cache, + (struct flow_stats __force *)flow->stats[node]); kmem_cache_free(flow_cache, flow); } @@ -586,16 +598,28 @@ int ovs_flow_init(void) BUILD_BUG_ON(__alignof__(struct sw_flow_key) % __alignof__(long)); BUILD_BUG_ON(sizeof(struct sw_flow_key) % sizeof(long)); - flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow), 0, - 0, NULL); + flow_cache = kmem_cache_create("sw_flow", sizeof(struct sw_flow) + + (num_possible_nodes() + * sizeof(struct flow_stats *)), + 0, 0, NULL); if (flow_cache == NULL) return -ENOMEM; + flow_stats_cache + = kmem_cache_create("sw_flow_stats", sizeof(struct flow_stats), + 0, SLAB_HWCACHE_ALIGN, NULL); + if (flow_stats_cache == NULL) { + kmem_cache_destroy(flow_cache); + flow_cache = NULL; + return -ENOMEM; + } + return 0; } /* Uninitializes the flow module. */ void ovs_flow_exit(void) { + kmem_cache_destroy(flow_stats_cache); kmem_cache_destroy(flow_cache); } diff --git a/net/openvswitch/flow_table.h b/net/openvswitch/flow_table.h index c26c59a7ab5..ca8a5820f61 100644 --- a/net/openvswitch/flow_table.h +++ b/net/openvswitch/flow_table.h @@ -52,6 +52,8 @@ struct flow_table { unsigned int count; }; +extern struct kmem_cache *flow_stats_cache; + int ovs_flow_init(void); void ovs_flow_exit(void); -- cgit v1.2.3-70-g09d2 From 88d73f6c411ac2f057829b93b3cf202ee551f6cb Mon Sep 17 00:00:00 2001 From: Jarno Rajahalme Date: Thu, 27 Mar 2014 12:51:49 -0700 Subject: openvswitch: Use TCP flags in the flow key for stats. We already extract the TCP flags for the key, might as well use that for stats. Signed-off-by: Jarno Rajahalme Acked-by: Pravin B Shelar Signed-off-by: Jesse Gross --- net/openvswitch/flow.c | 12 +++++------- 1 file changed, 5 insertions(+), 7 deletions(-) (limited to 'net/openvswitch/flow.c') diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c index 432f04d5c89..e0fc12bbeeb 100644 --- a/net/openvswitch/flow.c +++ b/net/openvswitch/flow.c @@ -69,14 +69,12 @@ void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb) stats = rcu_dereference(flow->stats[node]); - if ((flow->key.eth.type == htons(ETH_P_IP) || - flow->key.eth.type == htons(ETH_P_IPV6)) && - flow->key.ip.frag != OVS_FRAG_TYPE_LATER && - flow->key.ip.proto == IPPROTO_TCP && - likely(skb->len >= skb_transport_offset(skb) + sizeof(struct tcphdr))) { - tcp_flags = TCP_FLAGS_BE16(tcp_hdr(skb)); + if (likely(flow->key.ip.proto == IPPROTO_TCP)) { + if (likely(flow->key.eth.type == htons(ETH_P_IP))) + tcp_flags = flow->key.ipv4.tp.flags; + else if (likely(flow->key.eth.type == htons(ETH_P_IPV6))) + tcp_flags = flow->key.ipv6.tp.flags; } - /* Check if already have node-specific stats. */ if (likely(stats)) { spin_lock(&stats->lock); -- cgit v1.2.3-70-g09d2 From 1139e241ec436b9e9610c7a33ac5c6657f87fda1 Mon Sep 17 00:00:00 2001 From: Jarno Rajahalme Date: Mon, 5 May 2014 09:54:49 -0700 Subject: openvswitch: Compact sw_flow_key. Minimize padding in sw_flow_key and move 'tp' top the main struct. These changes simplify code when accessing the transport port numbers and the tcp flags, and makes the sw_flow_key 8 bytes smaller on 64-bit systems (128->120 bytes). These changes also make the keys for IPv4 packets to fit in one cache line. There is a valid concern for safety of packing the struct ovs_key_ipv4_tunnel, as it would be possible to take the address of the tun_id member as a __be64 * which could result in unaligned access in some systems. However: - sw_flow_key itself is 64-bit aligned, so the tun_id within is always 64-bit aligned. - We never make arrays of ovs_key_ipv4_tunnel (which would force every second tun_key to be misaligned). - We never take the address of the tun_id in to a __be64 *. - Whereever we use struct ovs_key_ipv4_tunnel outside the sw_flow_key, it is in stack (on tunnel input functions), where compiler has full control of the alignment. Signed-off-by: Jarno Rajahalme Signed-off-by: Pravin B Shelar --- net/openvswitch/flow.c | 44 +++++++--------- net/openvswitch/flow.h | 29 ++++------- net/openvswitch/flow_netlink.c | 112 ++++++++++++----------------------------- 3 files changed, 62 insertions(+), 123 deletions(-) (limited to 'net/openvswitch/flow.c') diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c index e0fc12bbeeb..6d8d2da0a8e 100644 --- a/net/openvswitch/flow.c +++ b/net/openvswitch/flow.c @@ -64,17 +64,11 @@ u64 ovs_flow_used_time(unsigned long flow_jiffies) void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb) { struct flow_stats *stats; - __be16 tcp_flags = 0; + __be16 tcp_flags = flow->key.tp.flags; int node = numa_node_id(); stats = rcu_dereference(flow->stats[node]); - if (likely(flow->key.ip.proto == IPPROTO_TCP)) { - if (likely(flow->key.eth.type == htons(ETH_P_IP))) - tcp_flags = flow->key.ipv4.tp.flags; - else if (likely(flow->key.eth.type == htons(ETH_P_IPV6))) - tcp_flags = flow->key.ipv6.tp.flags; - } /* Check if already have node-specific stats. */ if (likely(stats)) { spin_lock(&stats->lock); @@ -357,8 +351,8 @@ static int parse_icmpv6(struct sk_buff *skb, struct sw_flow_key *key, /* The ICMPv6 type and code fields use the 16-bit transport port * fields, so we need to store them in 16-bit network byte order. */ - key->ipv6.tp.src = htons(icmp->icmp6_type); - key->ipv6.tp.dst = htons(icmp->icmp6_code); + key->tp.src = htons(icmp->icmp6_type); + key->tp.dst = htons(icmp->icmp6_code); if (icmp->icmp6_code == 0 && (icmp->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION || @@ -520,21 +514,21 @@ int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key) if (key->ip.proto == IPPROTO_TCP) { if (tcphdr_ok(skb)) { struct tcphdr *tcp = tcp_hdr(skb); - key->ipv4.tp.src = tcp->source; - key->ipv4.tp.dst = tcp->dest; - key->ipv4.tp.flags = TCP_FLAGS_BE16(tcp); + key->tp.src = tcp->source; + key->tp.dst = tcp->dest; + key->tp.flags = TCP_FLAGS_BE16(tcp); } } else if (key->ip.proto == IPPROTO_UDP) { if (udphdr_ok(skb)) { struct udphdr *udp = udp_hdr(skb); - key->ipv4.tp.src = udp->source; - key->ipv4.tp.dst = udp->dest; + key->tp.src = udp->source; + key->tp.dst = udp->dest; } } else if (key->ip.proto == IPPROTO_SCTP) { if (sctphdr_ok(skb)) { struct sctphdr *sctp = sctp_hdr(skb); - key->ipv4.tp.src = sctp->source; - key->ipv4.tp.dst = sctp->dest; + key->tp.src = sctp->source; + key->tp.dst = sctp->dest; } } else if (key->ip.proto == IPPROTO_ICMP) { if (icmphdr_ok(skb)) { @@ -542,8 +536,8 @@ int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key) /* The ICMP type and code fields use the 16-bit * transport port fields, so we need to store * them in 16-bit network byte order. */ - key->ipv4.tp.src = htons(icmp->type); - key->ipv4.tp.dst = htons(icmp->code); + key->tp.src = htons(icmp->type); + key->tp.dst = htons(icmp->code); } } @@ -589,21 +583,21 @@ int ovs_flow_extract(struct sk_buff *skb, u16 in_port, struct sw_flow_key *key) if (key->ip.proto == NEXTHDR_TCP) { if (tcphdr_ok(skb)) { struct tcphdr *tcp = tcp_hdr(skb); - key->ipv6.tp.src = tcp->source; - key->ipv6.tp.dst = tcp->dest; - key->ipv6.tp.flags = TCP_FLAGS_BE16(tcp); + key->tp.src = tcp->source; + key->tp.dst = tcp->dest; + key->tp.flags = TCP_FLAGS_BE16(tcp); } } else if (key->ip.proto == NEXTHDR_UDP) { if (udphdr_ok(skb)) { struct udphdr *udp = udp_hdr(skb); - key->ipv6.tp.src = udp->source; - key->ipv6.tp.dst = udp->dest; + key->tp.src = udp->source; + key->tp.dst = udp->dest; } } else if (key->ip.proto == NEXTHDR_SCTP) { if (sctphdr_ok(skb)) { struct sctphdr *sctp = sctp_hdr(skb); - key->ipv6.tp.src = sctp->source; - key->ipv6.tp.dst = sctp->dest; + key->tp.src = sctp->source; + key->tp.dst = sctp->dest; } } else if (key->ip.proto == NEXTHDR_ICMP) { if (icmp6hdr_ok(skb)) { diff --git a/net/openvswitch/flow.h b/net/openvswitch/flow.h index ddcebc53224..a292bf8ad75 100644 --- a/net/openvswitch/flow.h +++ b/net/openvswitch/flow.h @@ -47,7 +47,7 @@ struct ovs_key_ipv4_tunnel { __be16 tun_flags; u8 ipv4_tos; u8 ipv4_ttl; -}; +} __packed __aligned(4); /* Minimize padding. */ static inline void ovs_flow_tun_key_init(struct ovs_key_ipv4_tunnel *tun_key, const struct iphdr *iph, __be64 tun_id, @@ -71,7 +71,7 @@ struct sw_flow_key { u32 priority; /* Packet QoS priority. */ u32 skb_mark; /* SKB mark. */ u16 in_port; /* Input switch port (or DP_MAX_PORTS). */ - } phy; + } __packed phy; /* Safe when right after 'tun_key'. */ struct { u8 src[ETH_ALEN]; /* Ethernet source address. */ u8 dst[ETH_ALEN]; /* Ethernet destination address. */ @@ -84,23 +84,21 @@ struct sw_flow_key { u8 ttl; /* IP TTL/hop limit. */ u8 frag; /* One of OVS_FRAG_TYPE_*. */ } ip; + struct { + __be16 src; /* TCP/UDP/SCTP source port. */ + __be16 dst; /* TCP/UDP/SCTP destination port. */ + __be16 flags; /* TCP flags. */ + } tp; union { struct { struct { __be32 src; /* IP source address. */ __be32 dst; /* IP destination address. */ } addr; - union { - struct { - __be16 src; /* TCP/UDP/SCTP source port. */ - __be16 dst; /* TCP/UDP/SCTP destination port. */ - __be16 flags; /* TCP flags. */ - } tp; - struct { - u8 sha[ETH_ALEN]; /* ARP source hardware address. */ - u8 tha[ETH_ALEN]; /* ARP target hardware address. */ - } arp; - }; + struct { + u8 sha[ETH_ALEN]; /* ARP source hardware address. */ + u8 tha[ETH_ALEN]; /* ARP target hardware address. */ + } arp; } ipv4; struct { struct { @@ -108,11 +106,6 @@ struct sw_flow_key { struct in6_addr dst; /* IPv6 destination address. */ } addr; __be32 label; /* IPv6 flow label. */ - struct { - __be16 src; /* TCP/UDP/SCTP source port. */ - __be16 dst; /* TCP/UDP/SCTP destination port. */ - __be16 flags; /* TCP flags. */ - } tp; struct { struct in6_addr target; /* ND target address. */ u8 sll[ETH_ALEN]; /* ND source link layer address. */ diff --git a/net/openvswitch/flow_netlink.c b/net/openvswitch/flow_netlink.c index 32a725cfeb0..d757848da89 100644 --- a/net/openvswitch/flow_netlink.c +++ b/net/openvswitch/flow_netlink.c @@ -204,11 +204,11 @@ static bool match_validate(const struct sw_flow_match *match, if (match->mask && (match->mask->key.ip.proto == 0xff)) mask_allowed |= 1 << OVS_KEY_ATTR_ICMPV6; - if (match->key->ipv6.tp.src == + if (match->key->tp.src == htons(NDISC_NEIGHBOUR_SOLICITATION) || - match->key->ipv6.tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) { + match->key->tp.src == htons(NDISC_NEIGHBOUR_ADVERTISEMENT)) { key_expected |= 1 << OVS_KEY_ATTR_ND; - if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff))) + if (match->mask && (match->mask->key.tp.src == htons(0xffff))) mask_allowed |= 1 << OVS_KEY_ATTR_ND; } } @@ -630,27 +630,18 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs, const struct ovs_key_tcp *tcp_key; tcp_key = nla_data(a[OVS_KEY_ATTR_TCP]); - if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) { - SW_FLOW_KEY_PUT(match, ipv4.tp.src, - tcp_key->tcp_src, is_mask); - SW_FLOW_KEY_PUT(match, ipv4.tp.dst, - tcp_key->tcp_dst, is_mask); - } else { - SW_FLOW_KEY_PUT(match, ipv6.tp.src, - tcp_key->tcp_src, is_mask); - SW_FLOW_KEY_PUT(match, ipv6.tp.dst, - tcp_key->tcp_dst, is_mask); - } + SW_FLOW_KEY_PUT(match, tp.src, tcp_key->tcp_src, is_mask); + SW_FLOW_KEY_PUT(match, tp.dst, tcp_key->tcp_dst, is_mask); attrs &= ~(1 << OVS_KEY_ATTR_TCP); } if (attrs & (1 << OVS_KEY_ATTR_TCP_FLAGS)) { if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) { - SW_FLOW_KEY_PUT(match, ipv4.tp.flags, + SW_FLOW_KEY_PUT(match, tp.flags, nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]), is_mask); } else { - SW_FLOW_KEY_PUT(match, ipv6.tp.flags, + SW_FLOW_KEY_PUT(match, tp.flags, nla_get_be16(a[OVS_KEY_ATTR_TCP_FLAGS]), is_mask); } @@ -661,17 +652,8 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs, const struct ovs_key_udp *udp_key; udp_key = nla_data(a[OVS_KEY_ATTR_UDP]); - if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) { - SW_FLOW_KEY_PUT(match, ipv4.tp.src, - udp_key->udp_src, is_mask); - SW_FLOW_KEY_PUT(match, ipv4.tp.dst, - udp_key->udp_dst, is_mask); - } else { - SW_FLOW_KEY_PUT(match, ipv6.tp.src, - udp_key->udp_src, is_mask); - SW_FLOW_KEY_PUT(match, ipv6.tp.dst, - udp_key->udp_dst, is_mask); - } + SW_FLOW_KEY_PUT(match, tp.src, udp_key->udp_src, is_mask); + SW_FLOW_KEY_PUT(match, tp.dst, udp_key->udp_dst, is_mask); attrs &= ~(1 << OVS_KEY_ATTR_UDP); } @@ -679,17 +661,8 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs, const struct ovs_key_sctp *sctp_key; sctp_key = nla_data(a[OVS_KEY_ATTR_SCTP]); - if (orig_attrs & (1 << OVS_KEY_ATTR_IPV4)) { - SW_FLOW_KEY_PUT(match, ipv4.tp.src, - sctp_key->sctp_src, is_mask); - SW_FLOW_KEY_PUT(match, ipv4.tp.dst, - sctp_key->sctp_dst, is_mask); - } else { - SW_FLOW_KEY_PUT(match, ipv6.tp.src, - sctp_key->sctp_src, is_mask); - SW_FLOW_KEY_PUT(match, ipv6.tp.dst, - sctp_key->sctp_dst, is_mask); - } + SW_FLOW_KEY_PUT(match, tp.src, sctp_key->sctp_src, is_mask); + SW_FLOW_KEY_PUT(match, tp.dst, sctp_key->sctp_dst, is_mask); attrs &= ~(1 << OVS_KEY_ATTR_SCTP); } @@ -697,9 +670,9 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs, const struct ovs_key_icmp *icmp_key; icmp_key = nla_data(a[OVS_KEY_ATTR_ICMP]); - SW_FLOW_KEY_PUT(match, ipv4.tp.src, + SW_FLOW_KEY_PUT(match, tp.src, htons(icmp_key->icmp_type), is_mask); - SW_FLOW_KEY_PUT(match, ipv4.tp.dst, + SW_FLOW_KEY_PUT(match, tp.dst, htons(icmp_key->icmp_code), is_mask); attrs &= ~(1 << OVS_KEY_ATTR_ICMP); } @@ -708,9 +681,9 @@ static int ovs_key_from_nlattrs(struct sw_flow_match *match, u64 attrs, const struct ovs_key_icmpv6 *icmpv6_key; icmpv6_key = nla_data(a[OVS_KEY_ATTR_ICMPV6]); - SW_FLOW_KEY_PUT(match, ipv6.tp.src, + SW_FLOW_KEY_PUT(match, tp.src, htons(icmpv6_key->icmpv6_type), is_mask); - SW_FLOW_KEY_PUT(match, ipv6.tp.dst, + SW_FLOW_KEY_PUT(match, tp.dst, htons(icmpv6_key->icmpv6_code), is_mask); attrs &= ~(1 << OVS_KEY_ATTR_ICMPV6); } @@ -1024,19 +997,11 @@ int ovs_nla_put_flow(const struct sw_flow_key *swkey, if (!nla) goto nla_put_failure; tcp_key = nla_data(nla); - if (swkey->eth.type == htons(ETH_P_IP)) { - tcp_key->tcp_src = output->ipv4.tp.src; - tcp_key->tcp_dst = output->ipv4.tp.dst; - if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS, - output->ipv4.tp.flags)) - goto nla_put_failure; - } else if (swkey->eth.type == htons(ETH_P_IPV6)) { - tcp_key->tcp_src = output->ipv6.tp.src; - tcp_key->tcp_dst = output->ipv6.tp.dst; - if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS, - output->ipv6.tp.flags)) - goto nla_put_failure; - } + tcp_key->tcp_src = output->tp.src; + tcp_key->tcp_dst = output->tp.dst; + if (nla_put_be16(skb, OVS_KEY_ATTR_TCP_FLAGS, + output->tp.flags)) + goto nla_put_failure; } else if (swkey->ip.proto == IPPROTO_UDP) { struct ovs_key_udp *udp_key; @@ -1044,13 +1009,8 @@ int ovs_nla_put_flow(const struct sw_flow_key *swkey, if (!nla) goto nla_put_failure; udp_key = nla_data(nla); - if (swkey->eth.type == htons(ETH_P_IP)) { - udp_key->udp_src = output->ipv4.tp.src; - udp_key->udp_dst = output->ipv4.tp.dst; - } else if (swkey->eth.type == htons(ETH_P_IPV6)) { - udp_key->udp_src = output->ipv6.tp.src; - udp_key->udp_dst = output->ipv6.tp.dst; - } + udp_key->udp_src = output->tp.src; + udp_key->udp_dst = output->tp.dst; } else if (swkey->ip.proto == IPPROTO_SCTP) { struct ovs_key_sctp *sctp_key; @@ -1058,13 +1018,8 @@ int ovs_nla_put_flow(const struct sw_flow_key *swkey, if (!nla) goto nla_put_failure; sctp_key = nla_data(nla); - if (swkey->eth.type == htons(ETH_P_IP)) { - sctp_key->sctp_src = output->ipv4.tp.src; - sctp_key->sctp_dst = output->ipv4.tp.dst; - } else if (swkey->eth.type == htons(ETH_P_IPV6)) { - sctp_key->sctp_src = output->ipv6.tp.src; - sctp_key->sctp_dst = output->ipv6.tp.dst; - } + sctp_key->sctp_src = output->tp.src; + sctp_key->sctp_dst = output->tp.dst; } else if (swkey->eth.type == htons(ETH_P_IP) && swkey->ip.proto == IPPROTO_ICMP) { struct ovs_key_icmp *icmp_key; @@ -1073,8 +1028,8 @@ int ovs_nla_put_flow(const struct sw_flow_key *swkey, if (!nla) goto nla_put_failure; icmp_key = nla_data(nla); - icmp_key->icmp_type = ntohs(output->ipv4.tp.src); - icmp_key->icmp_code = ntohs(output->ipv4.tp.dst); + icmp_key->icmp_type = ntohs(output->tp.src); + icmp_key->icmp_code = ntohs(output->tp.dst); } else if (swkey->eth.type == htons(ETH_P_IPV6) && swkey->ip.proto == IPPROTO_ICMPV6) { struct ovs_key_icmpv6 *icmpv6_key; @@ -1084,8 +1039,8 @@ int ovs_nla_put_flow(const struct sw_flow_key *swkey, if (!nla) goto nla_put_failure; icmpv6_key = nla_data(nla); - icmpv6_key->icmpv6_type = ntohs(output->ipv6.tp.src); - icmpv6_key->icmpv6_code = ntohs(output->ipv6.tp.dst); + icmpv6_key->icmpv6_type = ntohs(output->tp.src); + icmpv6_key->icmpv6_code = ntohs(output->tp.dst); if (icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_SOLICITATION || icmpv6_key->icmpv6_type == NDISC_NEIGHBOUR_ADVERTISEMENT) { @@ -1263,13 +1218,10 @@ static int validate_and_copy_sample(const struct nlattr *attr, static int validate_tp_port(const struct sw_flow_key *flow_key) { - if (flow_key->eth.type == htons(ETH_P_IP)) { - if (flow_key->ipv4.tp.src || flow_key->ipv4.tp.dst) - return 0; - } else if (flow_key->eth.type == htons(ETH_P_IPV6)) { - if (flow_key->ipv6.tp.src || flow_key->ipv6.tp.dst) - return 0; - } + if ((flow_key->eth.type == htons(ETH_P_IP) || + flow_key->eth.type == htons(ETH_P_IPV6)) && + (flow_key->tp.src || flow_key->tp.dst)) + return 0; return -EINVAL; } -- cgit v1.2.3-70-g09d2 From bb6f9a708d4067713afae2e9eb2637f6b4c01ecb Mon Sep 17 00:00:00 2001 From: Jarno Rajahalme Date: Mon, 5 May 2014 11:32:17 -0700 Subject: openvswitch: Clarify locking. Remove unnecessary locking from functions that are always called with appropriate locking. Signed-off-by: Jarno Rajahalme Signed-off-by: Thomas Graf Signed-off-by: Pravin B Shelar --- net/openvswitch/datapath.c | 13 +++++++------ net/openvswitch/flow.c | 3 ++- 2 files changed, 9 insertions(+), 7 deletions(-) (limited to 'net/openvswitch/flow.c') diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c index 90a1e5e6628..138ea0c9e1b 100644 --- a/net/openvswitch/datapath.c +++ b/net/openvswitch/datapath.c @@ -173,6 +173,7 @@ static struct hlist_head *vport_hash_bucket(const struct datapath *dp, return &dp->ports[port_no & (DP_VPORT_HASH_BUCKETS - 1)]; } +/* Called with ovs_mutex or RCU read lock. */ struct vport *ovs_lookup_vport(const struct datapath *dp, u16 port_no) { struct vport *vport; @@ -652,7 +653,7 @@ static size_t ovs_flow_cmd_msg_size(const struct sw_flow_actions *acts) + nla_total_size(acts->actions_len); /* OVS_FLOW_ATTR_ACTIONS */ } -/* Called with ovs_mutex. */ +/* Called with ovs_mutex or RCU read lock. */ static int ovs_flow_cmd_fill_info(struct sw_flow *flow, struct datapath *dp, struct sk_buff *skb, u32 portid, u32 seq, u32 flags, u8 cmd) @@ -743,6 +744,7 @@ error: return err; } +/* Must be called with ovs_mutex. */ static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow, struct genl_info *info) { @@ -753,6 +755,7 @@ static struct sk_buff *ovs_flow_cmd_alloc_info(struct sw_flow *flow, return genlmsg_new_unicast(len, info, GFP_KERNEL); } +/* Must be called with ovs_mutex. */ static struct sk_buff *ovs_flow_cmd_build_info(struct sw_flow *flow, struct datapath *dp, struct genl_info *info, @@ -1094,6 +1097,7 @@ static size_t ovs_dp_cmd_msg_size(void) return msgsize; } +/* Called with ovs_mutex or RCU read lock. */ static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, u32 portid, u32 seq, u32 flags, u8 cmd) { @@ -1109,9 +1113,7 @@ static int ovs_dp_cmd_fill_info(struct datapath *dp, struct sk_buff *skb, ovs_header->dp_ifindex = get_dpifindex(dp); - rcu_read_lock(); err = nla_put_string(skb, OVS_DP_ATTR_NAME, ovs_dp_name(dp)); - rcu_read_unlock(); if (err) goto nla_put_failure; @@ -1136,6 +1138,7 @@ error: return -EMSGSIZE; } +/* Must be called with ovs_mutex. */ static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, struct genl_info *info, u8 cmd) { @@ -1154,7 +1157,7 @@ static struct sk_buff *ovs_dp_cmd_build_info(struct datapath *dp, return skb; } -/* Called with ovs_mutex. */ +/* Called with rcu_read_lock or ovs_mutex. */ static struct datapath *lookup_datapath(struct net *net, struct ovs_header *ovs_header, struct nlattr *a[OVS_DP_ATTR_MAX + 1]) @@ -1166,10 +1169,8 @@ static struct datapath *lookup_datapath(struct net *net, else { struct vport *vport; - rcu_read_lock(); vport = ovs_vport_locate(net, nla_data(a[OVS_DP_ATTR_NAME])); dp = vport && vport->port_no == OVSP_LOCAL ? vport->dp : NULL; - rcu_read_unlock(); } return dp ? dp : ERR_PTR(-ENODEV); } diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c index 6d8d2da0a8e..1019fc1db06 100644 --- a/net/openvswitch/flow.c +++ b/net/openvswitch/flow.c @@ -122,6 +122,7 @@ unlock: spin_unlock(&stats->lock); } +/* Called with ovs_mutex. */ void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats, unsigned long *used, __be16 *tcp_flags) { @@ -132,7 +133,7 @@ void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats, memset(ovs_stats, 0, sizeof(*ovs_stats)); for_each_node(node) { - struct flow_stats *stats = rcu_dereference(flow->stats[node]); + struct flow_stats *stats = ovsl_dereference(flow->stats[node]); if (stats) { /* Local CPU may write on non-local stats, so we must -- cgit v1.2.3-70-g09d2 From 86ec8dbae27e5fa2b5d54f10f77286d9ef55732a Mon Sep 17 00:00:00 2001 From: Jarno Rajahalme Date: Mon, 5 May 2014 14:17:28 -0700 Subject: openvswitch: Fix ovs_flow_stats_get/clear RCU dereference. For ovs_flow_stats_get() using ovsl_dereference() was wrong, since flow dumps call this with RCU read lock. ovs_flow_stats_clear() is always called with ovs_mutex, so can use ovsl_dereference(). Also, make the ovs_flow_stats_get() 'flow' argument const to make later patches cleaner. Signed-off-by: Jarno Rajahalme Signed-off-by: Pravin B Shelar --- net/openvswitch/flow.c | 10 ++++++---- net/openvswitch/flow.h | 6 +++--- 2 files changed, 9 insertions(+), 7 deletions(-) (limited to 'net/openvswitch/flow.c') diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c index 1019fc1db06..334751cb152 100644 --- a/net/openvswitch/flow.c +++ b/net/openvswitch/flow.c @@ -122,8 +122,9 @@ unlock: spin_unlock(&stats->lock); } -/* Called with ovs_mutex. */ -void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats, +/* Must be called with rcu_read_lock or ovs_mutex. */ +void ovs_flow_stats_get(const struct sw_flow *flow, + struct ovs_flow_stats *ovs_stats, unsigned long *used, __be16 *tcp_flags) { int node; @@ -133,7 +134,7 @@ void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats, memset(ovs_stats, 0, sizeof(*ovs_stats)); for_each_node(node) { - struct flow_stats *stats = ovsl_dereference(flow->stats[node]); + struct flow_stats *stats = rcu_dereference_ovsl(flow->stats[node]); if (stats) { /* Local CPU may write on non-local stats, so we must @@ -150,12 +151,13 @@ void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *ovs_stats, } } +/* Called with ovs_mutex. */ void ovs_flow_stats_clear(struct sw_flow *flow) { int node; for_each_node(node) { - struct flow_stats *stats = rcu_dereference(flow->stats[node]); + struct flow_stats *stats = ovsl_dereference(flow->stats[node]); if (stats) { spin_lock_bh(&stats->lock); diff --git a/net/openvswitch/flow.h b/net/openvswitch/flow.h index a292bf8ad75..ac395d2cd82 100644 --- a/net/openvswitch/flow.h +++ b/net/openvswitch/flow.h @@ -180,10 +180,10 @@ struct arp_eth_header { unsigned char ar_tip[4]; /* target IP address */ } __packed; -void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb); -void ovs_flow_stats_get(struct sw_flow *flow, struct ovs_flow_stats *stats, +void ovs_flow_stats_update(struct sw_flow *, struct sk_buff *); +void ovs_flow_stats_get(const struct sw_flow *, struct ovs_flow_stats *, unsigned long *used, __be16 *tcp_flags); -void ovs_flow_stats_clear(struct sw_flow *flow); +void ovs_flow_stats_clear(struct sw_flow *); u64 ovs_flow_used_time(unsigned long flow_jiffies); int ovs_flow_extract(struct sk_buff *, u16 in_port, struct sw_flow_key *); -- cgit v1.2.3-70-g09d2 From ad55200734c65a3ec5d0c39d6ea904008baea536 Mon Sep 17 00:00:00 2001 From: Ben Pfaff Date: Tue, 6 May 2014 16:48:38 -0700 Subject: openvswitch: Fix tracking of flags seen in TCP flows. Flow statistics need to take into account the TCP flags from the packet currently being processed (in 'key'), not the TCP flags matched by the flow found in the kernel flow table (in 'flow'). This bug made the Open vSwitch userspace fin_timeout action have no effect in many cases. This bug is introduced by commit 88d73f6c411ac2f0578 (openvswitch: Use TCP flags in the flow key for stats.) Reported-by: Len Gao Signed-off-by: Ben Pfaff Acked-by: Jarno Rajahalme Acked-by: Jesse Gross Signed-off-by: Pravin B Shelar --- net/openvswitch/datapath.c | 4 ++-- net/openvswitch/flow.c | 4 ++-- net/openvswitch/flow.h | 5 +++-- 3 files changed, 7 insertions(+), 6 deletions(-) (limited to 'net/openvswitch/flow.c') diff --git a/net/openvswitch/datapath.c b/net/openvswitch/datapath.c index 0d407bca81e..a863678c50a 100644 --- a/net/openvswitch/datapath.c +++ b/net/openvswitch/datapath.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2007-2013 Nicira, Inc. + * Copyright (c) 2007-2014 Nicira, Inc. * * This program is free software; you can redistribute it and/or * modify it under the terms of version 2 of the GNU General Public @@ -276,7 +276,7 @@ void ovs_dp_process_received_packet(struct vport *p, struct sk_buff *skb) OVS_CB(skb)->flow = flow; OVS_CB(skb)->pkt_key = &key; - ovs_flow_stats_update(OVS_CB(skb)->flow, skb); + ovs_flow_stats_update(OVS_CB(skb)->flow, key.tp.flags, skb); ovs_execute_actions(dp, skb); stats_counter = &stats->n_hit; diff --git a/net/openvswitch/flow.c b/net/openvswitch/flow.c index 334751cb152..d07ab538fc9 100644 --- a/net/openvswitch/flow.c +++ b/net/openvswitch/flow.c @@ -61,10 +61,10 @@ u64 ovs_flow_used_time(unsigned long flow_jiffies) #define TCP_FLAGS_BE16(tp) (*(__be16 *)&tcp_flag_word(tp) & htons(0x0FFF)) -void ovs_flow_stats_update(struct sw_flow *flow, struct sk_buff *skb) +void ovs_flow_stats_update(struct sw_flow *flow, __be16 tcp_flags, + struct sk_buff *skb) { struct flow_stats *stats; - __be16 tcp_flags = flow->key.tp.flags; int node = numa_node_id(); stats = rcu_dereference(flow->stats[node]); diff --git a/net/openvswitch/flow.h b/net/openvswitch/flow.h index ac395d2cd82..5e5aaed3a85 100644 --- a/net/openvswitch/flow.h +++ b/net/openvswitch/flow.h @@ -1,5 +1,5 @@ /* - * Copyright (c) 2007-2013 Nicira, Inc. + * Copyright (c) 2007-2014 Nicira, Inc. * * This program is free software; you can redistribute it and/or * modify it under the terms of version 2 of the GNU General Public @@ -180,7 +180,8 @@ struct arp_eth_header { unsigned char ar_tip[4]; /* target IP address */ } __packed; -void ovs_flow_stats_update(struct sw_flow *, struct sk_buff *); +void ovs_flow_stats_update(struct sw_flow *, __be16 tcp_flags, + struct sk_buff *); void ovs_flow_stats_get(const struct sw_flow *, struct ovs_flow_stats *, unsigned long *used, __be16 *tcp_flags); void ovs_flow_stats_clear(struct sw_flow *); -- cgit v1.2.3-70-g09d2