X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;ds=sidebyside;f=datapath%2Fdatapath.c;h=fe37ec1ed4c1c7a1bd6ebc9b7d482d32fcc8f15d;hb=refs%2Fheads%2Flts-1.0;hp=39bc190bcc9dcd816c9585c52c54094b6116d233;hpb=780e620781c5237ab6c26d5edbc7edb5b4953c70;p=sliver-openvswitch.git diff --git a/datapath/datapath.c b/datapath/datapath.c index 39bc190bc..fe37ec1ed 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -8,6 +8,8 @@ /* Functions for managing the dp interface/device. */ +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + #include #include #include @@ -32,6 +34,7 @@ #include #include #include +#include #include #include #include @@ -68,8 +71,22 @@ EXPORT_SYMBOL(dp_ioctl_hook); static struct datapath *dps[ODP_MAX]; static DEFINE_MUTEX(dp_mutex); -/* Number of milliseconds between runs of the maintenance thread. */ -#define MAINT_SLEEP_MSECS 1000 +/* We limit the number of times that we pass into dp_process_received_packet() + * to avoid blowing out the stack in the event that we have a loop. */ +struct loop_counter { + int count; /* Count. */ + bool looping; /* Loop detected? */ +}; + +#define DP_MAX_LOOPS 5 + +/* We use a separate counter for each CPU for both interrupt and non-interrupt + * context in order to keep the limit deterministic for a given packet. */ +struct percpu_loop_counters { + struct loop_counter counters[2]; +}; + +static DEFINE_PER_CPU(struct percpu_loop_counters, dp_loop_counters); static int new_dp_port(struct datapath *, struct odp_port *, int port_no); @@ -379,6 +396,7 @@ static int new_dp_port(struct datapath *dp, struct odp_port *odp_port, int port_ p->port_no = port_no; p->dp = dp; + p->vport = vport; atomic_set(&p->sflow_pool, 0); err = vport_attach(vport, p); @@ -512,6 +530,14 @@ out: return err; } +static void suppress_loop(struct datapath *dp, struct sw_flow_actions *actions) +{ + if (net_ratelimit()) + pr_warn("%s: flow looped %d times, dropping\n", + dp_name(dp), DP_MAX_LOOPS); + actions->n_actions = 0; +} + /* Must be called with rcu_read_lock. */ void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb) { @@ -520,37 +546,71 @@ void dp_process_received_packet(struct dp_port *p, struct sk_buff *skb) int stats_counter_off; struct odp_flow_key key; struct tbl_node *flow_node; - - WARN_ON_ONCE(skb_shared(skb)); - skb_warn_if_lro(skb); + struct sw_flow *flow; + struct sw_flow_actions *acts; + struct loop_counter *loop; + int error; OVS_CB(skb)->dp_port = p; - if (flow_extract(skb, p ? p->port_no : ODPP_NONE, &key)) { - if (dp->drop_frags) { - kfree_skb(skb); - stats_counter_off = offsetof(struct dp_stats_percpu, n_frags); - goto out; - } + /* Extract flow from 'skb' into 'key'. */ + error = flow_extract(skb, p ? p->port_no : ODPP_NONE, &key); + if (unlikely(error)) { + kfree_skb(skb); + return; + } + + if (OVS_CB(skb)->is_frag && dp->drop_frags) { + kfree_skb(skb); + stats_counter_off = offsetof(struct dp_stats_percpu, n_frags); + goto out; } + /* Look up flow. */ flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp); - if (flow_node) { - struct sw_flow *flow = flow_cast(flow_node); - struct sw_flow_actions *acts = rcu_dereference(flow->sf_acts); - flow_used(flow, skb); - execute_actions(dp, skb, &key, acts->actions, acts->n_actions, - GFP_ATOMIC); - stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); - } else { - stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); + if (unlikely(!flow_node)) { dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); + stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); + goto out; } + flow = flow_cast(flow_node); + flow_used(flow, skb); + + acts = rcu_dereference(flow->sf_acts); + + /* Check whether we've looped too much. */ + loop = &get_cpu_var(dp_loop_counters).counters[!!in_interrupt()]; + if (unlikely(++loop->count > DP_MAX_LOOPS)) + loop->looping = true; + if (unlikely(loop->looping)) { + suppress_loop(dp, acts); + goto out_loop; + } + + /* Execute actions. */ + execute_actions(dp, skb, &key, acts->actions, acts->n_actions, GFP_ATOMIC); + stats_counter_off = offsetof(struct dp_stats_percpu, n_hit); + + /* Check whether sub-actions looped too much. */ + if (unlikely(loop->looping)) + suppress_loop(dp, acts); + +out_loop: + /* Decrement loop counter. */ + if (!--loop->count) + loop->looping = false; + put_cpu_var(dp_loop_counters); + out: + /* Update datapath statistics. */ local_bh_disable(); stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); + + write_seqcount_begin(&stats->seqlock); (*(u64 *)((u8 *)stats + stats_counter_off))++; + write_seqcount_end(&stats->seqlock); + local_bh_enable(); } @@ -587,9 +647,9 @@ int vswitch_skb_checksum_setup(struct sk_buff *skb) break; default: if (net_ratelimit()) - printk(KERN_ERR "Attempting to checksum a non-" - "TCP/UDP packet, dropping a protocol" - " %d packet", iph->protocol); + pr_err("Attempting to checksum a non-TCP/UDP packet, " + "dropping a protocol %d packet", + iph->protocol); goto out; } @@ -655,8 +715,7 @@ out: * be reverified). If we receive a packet with CHECKSUM_HW that really means * CHECKSUM_PARTIAL, it will be sent with the wrong checksum. However, there * shouldn't be any devices that do this with bridging. */ -void -compute_ip_summed(struct sk_buff *skb, bool xmit) +void compute_ip_summed(struct sk_buff *skb, bool xmit) { /* For our convenience these defines change repeatedly between kernel * versions, so we can't just copy them over... */ @@ -693,11 +752,10 @@ compute_ip_summed(struct sk_buff *skb, bool xmit) break; #endif default: - printk(KERN_ERR "openvswitch: unknown checksum type %d\n", - skb->ip_summed); + pr_err("unknown checksum type %d\n", skb->ip_summed); /* None seems the safest... */ OVS_CB(skb)->ip_summed = OVS_CSUM_NONE; - } + } #if defined(CONFIG_XEN) && defined(HAVE_PROTO_DATA_VALID) /* Xen has a special way of representing CHECKSUM_PARTIAL on older @@ -713,8 +771,7 @@ compute_ip_summed(struct sk_buff *skb, bool xmit) * is slightly different because we are only concerned with bridging and not * other types of forwarding and can get away with slightly more optimal * behavior.*/ -void -forward_ip_summed(struct sk_buff *skb) +void forward_ip_summed(struct sk_buff *skb) { #ifdef CHECKSUM_HW if (OVS_CB(skb)->ip_summed == OVS_CSUM_COMPLETE) @@ -724,9 +781,8 @@ forward_ip_summed(struct sk_buff *skb) /* Append each packet in 'skb' list to 'queue'. There will be only one packet * unless we broke up a GSO packet. */ -static int -queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue, - int queue_no, u32 arg) +static int queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue, + int queue_no, u32 arg) { struct sk_buff *nskb; int port_no; @@ -743,31 +799,6 @@ queue_control_packets(struct sk_buff *skb, struct sk_buff_head *queue, nskb = skb->next; skb->next = NULL; - /* If a checksum-deferred packet is forwarded to the - * controller, correct the pointers and checksum. - */ - err = vswitch_skb_checksum_setup(skb); - if (err) - goto err_kfree_skbs; - - if (skb->ip_summed == CHECKSUM_PARTIAL) { - -#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) - /* Until 2.6.22, the start of the transport header was - * also the start of data to be checksummed. Linux - * 2.6.22 introduced the csum_start field for this - * purpose, but we should point the transport header to - * it anyway for backward compatibility, as - * dev_queue_xmit() does even in 2.6.28. */ - skb_set_transport_header(skb, skb->csum_start - - skb_headroom(skb)); -#endif - - err = skb_checksum_help(skb); - if (err) - goto err_kfree_skbs; - } - err = skb_cow(skb, sizeof *header); if (err) goto err_kfree_skbs; @@ -793,9 +824,8 @@ err_kfree_skbs: return err; } -int -dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, - u32 arg) +int dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, + u32 arg) { struct dp_stats_percpu *stats; struct sk_buff_head *queue; @@ -810,10 +840,14 @@ dp_output_control(struct datapath *dp, struct sk_buff *skb, int queue_no, forward_ip_summed(skb); + err = vswitch_skb_checksum_setup(skb); + if (err) + goto err_kfree_skb; + /* Break apart GSO packets into their component pieces. Otherwise * userspace may try to stuff a 64kB packet into a 1500-byte MTU. */ if (skb_is_gso(skb)) { - struct sk_buff *nskb = skb_gso_segment(skb, 0); + struct sk_buff *nskb = skb_gso_segment(skb, NETIF_F_SG | NETIF_F_HW_CSUM); if (nskb) { kfree_skb(skb); skb = nskb; @@ -836,7 +870,11 @@ err_kfree_skb: err: local_bh_disable(); stats = per_cpu_ptr(dp->stats_percpu, smp_processor_id()); + + write_seqcount_begin(&stats->seqlock); stats->n_lost++; + write_seqcount_end(&stats->seqlock); + local_bh_enable(); return err; @@ -927,27 +965,43 @@ error: return ERR_PTR(error); } -static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats) +static struct timespec get_time_offset(void) { - if (flow->used.tv_sec) { - stats->used_sec = flow->used.tv_sec; - stats->used_nsec = flow->used.tv_nsec; + struct timespec now_mono, now_jiffies; + + ktime_get_ts(&now_mono); + jiffies_to_timespec(jiffies, &now_jiffies); + return timespec_sub(now_mono, now_jiffies); +} + +static void get_stats(struct sw_flow *flow, struct odp_flow_stats *stats, + struct timespec time_offset) +{ + if (flow->used) { + struct timespec flow_ts, used; + + jiffies_to_timespec(flow->used, &flow_ts); + set_normalized_timespec(&used, flow_ts.tv_sec + time_offset.tv_sec, + flow_ts.tv_nsec + time_offset.tv_nsec); + + stats->used_sec = used.tv_sec; + stats->used_nsec = used.tv_nsec; } else { stats->used_sec = 0; stats->used_nsec = 0; } + stats->n_packets = flow->packet_count; stats->n_bytes = flow->byte_count; - stats->ip_tos = flow->ip_tos; + stats->reserved = 0; stats->tcp_flags = flow->tcp_flags; stats->error = 0; } static void clear_stats(struct sw_flow *flow) { - flow->used.tv_sec = flow->used.tv_nsec = 0; + flow->used = 0; flow->tcp_flags = 0; - flow->ip_tos = 0; flow->packet_count = 0; flow->byte_count = 0; } @@ -1045,7 +1099,7 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, /* Fetch stats, then clear them if necessary. */ spin_lock_bh(&flow->lock); - get_stats(flow, stats); + get_stats(flow, stats, get_time_offset()); if (uf->flags & ODPPF_ZERO_STATS) clear_stats(flow); spin_unlock_bh(&flow->lock); @@ -1082,6 +1136,7 @@ static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) } static int do_answer_query(struct sw_flow *flow, u32 query_flags, + struct timespec time_offset, struct odp_flow_stats __user *ustats, union odp_action __user *actions, u32 __user *n_actionsp) @@ -1091,7 +1146,7 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags, u32 n_actions; spin_lock_bh(&flow->lock); - get_stats(flow, &stats); + get_stats(flow, &stats, time_offset); if (query_flags & ODPFF_ZERO_TCP_FLAGS) flow->tcp_flags = 0; @@ -1115,6 +1170,7 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags, } static int answer_query(struct sw_flow *flow, u32 query_flags, + struct timespec time_offset, struct odp_flow __user *ufp) { union odp_action *actions; @@ -1122,7 +1178,7 @@ static int answer_query(struct sw_flow *flow, u32 query_flags, if (get_user(actions, &ufp->actions)) return -EFAULT; - return do_answer_query(flow, query_flags, + return do_answer_query(flow, query_flags, time_offset, &ufp->stats, actions, &ufp->n_actions); } @@ -1161,7 +1217,7 @@ static int del_flow(struct datapath *dp, struct odp_flow __user *ufp) if (IS_ERR(flow)) return PTR_ERR(flow); - error = answer_query(flow, 0, ufp); + error = answer_query(flow, 0, get_time_offset(), ufp); flow_deferred_free(flow); return error; } @@ -1169,8 +1225,11 @@ static int del_flow(struct datapath *dp, struct odp_flow __user *ufp) static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec) { struct tbl *table = rcu_dereference(dp->table); + struct timespec time_offset; u32 i; + time_offset = get_time_offset(); + for (i = 0; i < flowvec->n_flows; i++) { struct odp_flow __user *ufp = &flowvec->flows[i]; struct odp_flow uf; @@ -1185,7 +1244,7 @@ static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec if (!flow_node) error = put_user(ENOENT, &ufp->stats.error); else - error = answer_query(flow_cast(flow_node), uf.flags, ufp); + error = answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp); if (error) return -EFAULT; } @@ -1196,6 +1255,7 @@ struct list_flows_cbdata { struct odp_flow __user *uflows; u32 n_flows; u32 listed_flows; + struct timespec time_offset; }; static int list_flow(struct tbl_node *node, void *cbdata_) @@ -1207,7 +1267,7 @@ static int list_flow(struct tbl_node *node, void *cbdata_) if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) return -EFAULT; - error = answer_query(flow, 0, ufp); + error = answer_query(flow, 0, cbdata->time_offset, ufp); if (error) return error; @@ -1227,6 +1287,8 @@ static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec) cbdata.uflows = flowvec->flows; cbdata.n_flows = flowvec->n_flows; cbdata.listed_flows = 0; + cbdata.time_offset = get_time_offset(); + error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata); return error ? error : cbdata.listed_flows; } @@ -1264,10 +1326,11 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute) if (execute->length < ETH_HLEN || execute->length > 65535) goto error; - err = -ENOMEM; actions = flow_actions_alloc(execute->n_actions); - if (!actions) + if (IS_ERR(actions)) { + err = PTR_ERR(actions); goto error; + } err = -EFAULT; if (copy_from_user(actions->actions, execute->actions, @@ -1304,7 +1367,9 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute) else skb->protocol = htons(ETH_P_802_2); - flow_extract(skb, execute->in_port, &key); + err = flow_extract(skb, execute->in_port, &key); + if (err) + goto error_free_skb; rcu_read_lock(); err = execute_actions(dp, skb, &key, actions->actions, @@ -1346,12 +1411,21 @@ static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) stats.max_groups = DP_MAX_GROUPS; stats.n_frags = stats.n_hit = stats.n_missed = stats.n_lost = 0; for_each_possible_cpu(i) { - const struct dp_stats_percpu *s; - s = per_cpu_ptr(dp->stats_percpu, i); - stats.n_frags += s->n_frags; - stats.n_hit += s->n_hit; - stats.n_missed += s->n_missed; - stats.n_lost += s->n_lost; + const struct dp_stats_percpu *percpu_stats; + struct dp_stats_percpu local_stats; + unsigned seqcount; + + percpu_stats = per_cpu_ptr(dp->stats_percpu, i); + + do { + seqcount = read_seqcount_begin(&percpu_stats->seqlock); + local_stats = *percpu_stats; + } while (read_seqcount_retry(&percpu_stats->seqlock, seqcount)); + + stats.n_frags += local_stats.n_frags; + stats.n_hit += local_stats.n_hit; + stats.n_missed += local_stats.n_missed; + stats.n_lost += local_stats.n_lost; } stats.max_miss_queue = DP_MAX_QUEUE_LEN; stats.max_action_queue = DP_MAX_QUEUE_LEN; @@ -1399,8 +1473,7 @@ void set_internal_devs_mtu(const struct datapath *dp) } } -static int -put_port(const struct dp_port *p, struct odp_port __user *uop) +static int put_port(const struct dp_port *p, struct odp_port __user *uop) { struct odp_port op; @@ -1416,8 +1489,7 @@ put_port(const struct dp_port *p, struct odp_port __user *uop) return copy_to_user(uop, &op, sizeof op) ? -EFAULT : 0; } -static int -query_port(struct datapath *dp, struct odp_port __user *uport) +static int query_port(struct datapath *dp, struct odp_port __user *uport) { struct odp_port port; @@ -1464,8 +1536,8 @@ error_unlock: return put_port(dp->ports[port.port], uport); } -static int -do_list_ports(struct datapath *dp, struct odp_port __user *uports, int n_ports) +static int do_list_ports(struct datapath *dp, struct odp_port __user *uports, + int n_ports) { int idx = 0; if (n_ports) { @@ -1481,8 +1553,7 @@ do_list_ports(struct datapath *dp, struct odp_port __user *uports, int n_ports) return idx; } -static int -list_ports(struct datapath *dp, struct odp_portvec __user *upv) +static int list_ports(struct datapath *dp, struct odp_portvec __user *upv) { struct odp_portvec pv; int retval; @@ -1504,8 +1575,8 @@ static void free_port_group(struct rcu_head *rcu) kfree(g); } -static int -do_set_port_group(struct datapath *dp, u16 __user *ports, int n_ports, int group) +static int do_set_port_group(struct datapath *dp, u16 __user *ports, + int n_ports, int group) { struct dp_port_group *new_group, *old_group; int error; @@ -1536,8 +1607,8 @@ error: return error; } -static int -set_port_group(struct datapath *dp, const struct odp_port_group __user *upg) +static int set_port_group(struct datapath *dp, + const struct odp_port_group __user *upg) { struct odp_port_group pg; @@ -1547,10 +1618,9 @@ set_port_group(struct datapath *dp, const struct odp_port_group __user *upg) return do_set_port_group(dp, pg.ports, pg.n_ports, pg.group); } -static int -do_get_port_group(struct datapath *dp, - u16 __user *ports, int n_ports, int group, - u16 __user *n_portsp) +static int do_get_port_group(struct datapath *dp, + u16 __user *ports, int n_ports, int group, + u16 __user *n_portsp) { struct dp_port_group *g; u16 n_copy; @@ -1839,6 +1909,7 @@ static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __use } static int compat_answer_query(struct sw_flow *flow, u32 query_flags, + struct timespec time_offset, struct compat_odp_flow __user *ufp) { compat_uptr_t actions; @@ -1846,7 +1917,7 @@ static int compat_answer_query(struct sw_flow *flow, u32 query_flags, if (get_user(actions, &ufp->actions)) return -EFAULT; - return do_answer_query(flow, query_flags, &ufp->stats, + return do_answer_query(flow, query_flags, time_offset, &ufp->stats, compat_ptr(actions), &ufp->n_actions); } @@ -1863,7 +1934,7 @@ static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *u if (IS_ERR(flow)) return PTR_ERR(flow); - error = compat_answer_query(flow, 0, ufp); + error = compat_answer_query(flow, 0, get_time_offset(), ufp); flow_deferred_free(flow); return error; } @@ -1871,8 +1942,11 @@ static int compat_del_flow(struct datapath *dp, struct compat_odp_flow __user *u static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows) { struct tbl *table = rcu_dereference(dp->table); + struct timespec time_offset; u32 i; + time_offset = get_time_offset(); + for (i = 0; i < n_flows; i++) { struct compat_odp_flow __user *ufp = &flows[i]; struct odp_flow uf; @@ -1887,7 +1961,7 @@ static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows if (!flow_node) error = put_user(ENOENT, &ufp->stats.error); else - error = compat_answer_query(flow_cast(flow_node), uf.flags, ufp); + error = compat_answer_query(flow_cast(flow_node), uf.flags, time_offset, ufp); if (error) return -EFAULT; } @@ -1898,6 +1972,7 @@ struct compat_list_flows_cbdata { struct compat_odp_flow __user *uflows; u32 n_flows; u32 listed_flows; + struct timespec time_offset; }; static int compat_list_flow(struct tbl_node *node, void *cbdata_) @@ -1909,7 +1984,7 @@ static int compat_list_flow(struct tbl_node *node, void *cbdata_) if (copy_to_user(&ufp->key, &flow->key, sizeof flow->key)) return -EFAULT; - error = compat_answer_query(flow, 0, ufp); + error = compat_answer_query(flow, 0, cbdata->time_offset, ufp); if (error) return error; @@ -1929,6 +2004,8 @@ static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, cbdata.uflows = flows; cbdata.n_flows = n_flows; cbdata.listed_flows = 0; + cbdata.time_offset = get_time_offset(); + error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata); return error ? error : cbdata.listed_flows; } @@ -2070,6 +2147,100 @@ exit: } #endif +/* Unfortunately this function is not exported so this is a verbatim copy + * from net/core/datagram.c in 2.6.30. */ +static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, + u8 __user *to, int len, + __wsum *csump) +{ + int start = skb_headlen(skb); + int pos = 0; + int i, copy = start - offset; + + /* Copy header. */ + if (copy > 0) { + int err = 0; + if (copy > len) + copy = len; + *csump = csum_and_copy_to_user(skb->data + offset, to, copy, + *csump, &err); + if (err) + goto fault; + if ((len -= copy) == 0) + return 0; + offset += copy; + to += copy; + pos = copy; + } + + for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { + int end; + + WARN_ON(start > offset + len); + + end = start + skb_shinfo(skb)->frags[i].size; + if ((copy = end - offset) > 0) { + __wsum csum2; + int err = 0; + u8 *vaddr; + skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; + struct page *page = frag->page; + + if (copy > len) + copy = len; + vaddr = kmap(page); + csum2 = csum_and_copy_to_user(vaddr + + frag->page_offset + + offset - start, + to, copy, 0, &err); + kunmap(page); + if (err) + goto fault; + *csump = csum_block_add(*csump, csum2, pos); + if (!(len -= copy)) + return 0; + offset += copy; + to += copy; + pos += copy; + } + start = end; + } + + if (skb_shinfo(skb)->frag_list) { + struct sk_buff *list = skb_shinfo(skb)->frag_list; + + for (; list; list=list->next) { + int end; + + WARN_ON(start > offset + len); + + end = start + list->len; + if ((copy = end - offset) > 0) { + __wsum csum2 = 0; + if (copy > len) + copy = len; + if (skb_copy_and_csum_datagram(list, + offset - start, + to, copy, + &csum2)) + goto fault; + *csump = csum_block_add(*csump, csum2, pos); + if ((len -= copy) == 0) + return 0; + offset += copy; + to += copy; + pos += copy; + } + start = end; + } + } + if (!len) + return 0; + +fault: + return -EFAULT; +} + ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, loff_t *ppos) { @@ -2078,8 +2249,7 @@ ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, int dp_idx = iminor(f->f_dentry->d_inode); struct datapath *dp = get_dp(dp_idx); struct sk_buff *skb; - struct iovec __user iov; - size_t copy_bytes; + size_t copy_bytes, tot_copy_bytes; int retval; if (!dp) @@ -2114,12 +2284,48 @@ ssize_t openvswitch_read(struct file *f, char __user *buf, size_t nbytes, } } success: - copy_bytes = min_t(size_t, skb->len, nbytes); - iov.iov_base = buf; - iov.iov_len = copy_bytes; - retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); + copy_bytes = tot_copy_bytes = min_t(size_t, skb->len, nbytes); + + retval = 0; + if (skb->ip_summed == CHECKSUM_PARTIAL) { + if (copy_bytes == skb->len) { + __wsum csum = 0; + unsigned int csum_start, csum_offset; + +#if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,22) + csum_start = skb->csum_start - skb_headroom(skb); + csum_offset = skb->csum_offset; +#else + csum_start = skb_transport_header(skb) - skb->data; + csum_offset = skb->csum; +#endif + BUG_ON(csum_start >= skb_headlen(skb)); + retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start, + copy_bytes - csum_start, &csum); + if (!retval) { + __sum16 __user *csump; + + copy_bytes = csum_start; + csump = (__sum16 __user *)(buf + csum_start + csum_offset); + + BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes); + put_user(csum_fold(csum), csump); + } + } else + retval = skb_checksum_help(skb); + } + + if (!retval) { + struct iovec __user iov; + + iov.iov_base = buf; + iov.iov_len = copy_bytes; + retval = skb_copy_datagram_iovec(skb, 0, &iov, iov.iov_len); + } + if (!retval) - retval = copy_bytes; + retval = tot_copy_bytes; + kfree_skb(skb); error: