X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=datapath%2Fdatapath.c;h=4117ba9d8c7d7ea9ad8d3266f7644f034809b915;hb=6e11a6a15519561d932a018b2ca19dc3acee87c2;hp=98685c0c5d08c9438e70e736f28f6c15bde19a97;hpb=1b29ebe576bd4cf1354d1c2dac2a77f6e753c932;p=openvswitch diff --git a/datapath/datapath.c b/datapath/datapath.c index 98685c0c..4117ba9d 100644 --- a/datapath/datapath.c +++ b/datapath/datapath.c @@ -67,7 +67,7 @@ EXPORT_SYMBOL(dp_ioctl_hook); * It is safe to access the datapath and vport structures with just * dp_mutex. */ -static struct datapath *dps[ODP_MAX]; +static struct datapath __rcu *dps[ODP_MAX]; static DEFINE_MUTEX(dp_mutex); static int new_vport(struct datapath *, struct odp_port *, int port_no); @@ -77,7 +77,8 @@ struct datapath *get_dp(int dp_idx) { if (dp_idx < 0 || dp_idx >= ODP_MAX) return NULL; - return rcu_dereference(dps[dp_idx]); + return rcu_dereference_check(dps[dp_idx], rcu_read_lock_held() || + lockdep_is_held(&dp_mutex)); } EXPORT_SYMBOL_GPL(get_dp); @@ -93,10 +94,22 @@ static struct datapath *get_dp_locked(int dp_idx) return dp; } +static struct tbl *get_table_protected(struct datapath *dp) +{ + return rcu_dereference_protected(dp->table, + lockdep_is_held(&dp->mutex)); +} + +static struct vport *get_vport_protected(struct datapath *dp, u16 port_no) +{ + return rcu_dereference_protected(dp->ports[port_no], + lockdep_is_held(&dp->mutex)); +} + /* Must be called with rcu_read_lock or RTNL lock. */ const char *dp_name(const struct datapath *dp) { - return vport_get_name(dp->ports[ODPP_LOCAL]); + return vport_get_name(rcu_dereference_rtnl(dp->ports[ODPP_LOCAL])); } static inline size_t br_nlmsg_size(void) @@ -114,7 +127,7 @@ static int dp_fill_ifinfo(struct sk_buff *skb, const struct vport *port, int event, unsigned int flags) { - const struct datapath *dp = port->dp; + struct datapath *dp = port->dp; int ifindex = vport_get_ifindex(port); int iflink = vport_get_iflink(port); struct ifinfomsg *hdr; @@ -139,7 +152,8 @@ static int dp_fill_ifinfo(struct sk_buff *skb, hdr->ifi_change = 0; NLA_PUT_STRING(skb, IFLA_IFNAME, vport_get_name(port)); - NLA_PUT_U32(skb, IFLA_MASTER, vport_get_ifindex(dp->ports[ODPP_LOCAL])); + NLA_PUT_U32(skb, IFLA_MASTER, + vport_get_ifindex(get_vport_protected(dp, ODPP_LOCAL))); NLA_PUT_U32(skb, IFLA_MTU, vport_get_mtu(port)); #ifdef IFLA_OPERSTATE NLA_PUT_U8(skb, IFLA_OPERSTATE, @@ -233,6 +247,7 @@ static int create_dp(int dp_idx, const char __user *devnamep) goto err_put_module; INIT_LIST_HEAD(&dp->port_list); mutex_init(&dp->mutex); + mutex_lock(&dp->mutex); dp->dp_idx = dp_idx; for (i = 0; i < DP_N_QUEUES; i++) skb_queue_head_init(&dp->queues[i]); @@ -245,7 +260,7 @@ static int create_dp(int dp_idx, const char __user *devnamep) /* Allocate table. */ err = -ENOMEM; - rcu_assign_pointer(dp->table, tbl_create(0)); + rcu_assign_pointer(dp->table, tbl_create(TBL_MIN_BUCKETS)); if (!dp->table) goto err_free_dp; @@ -263,22 +278,26 @@ static int create_dp(int dp_idx, const char __user *devnamep) dp->drop_frags = 0; dp->stats_percpu = alloc_percpu(struct dp_stats_percpu); - if (!dp->stats_percpu) + if (!dp->stats_percpu) { + err = -ENOMEM; goto err_destroy_local_port; + } rcu_assign_pointer(dps[dp_idx], dp); dp_sysfs_add_dp(dp); + mutex_unlock(&dp->mutex); mutex_unlock(&dp_mutex); rtnl_unlock(); return 0; err_destroy_local_port: - dp_detach_port(dp->ports[ODPP_LOCAL]); + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); err_destroy_table: - tbl_destroy(dp->table, NULL); + tbl_destroy(get_table_protected(dp), NULL); err_free_dp: + mutex_unlock(&dp->mutex); kfree(dp); err_put_module: module_put(THIS_MODULE); @@ -289,52 +308,54 @@ err: return err; } -static void do_destroy_dp(struct datapath *dp) +static void destroy_dp_rcu(struct rcu_head *rcu) { - struct vport *p, *n; + struct datapath *dp = container_of(rcu, struct datapath, rcu); int i; - list_for_each_entry_safe (p, n, &dp->port_list, node) - if (p->port_no != ODPP_LOCAL) - dp_detach_port(p); - - dp_sysfs_del_dp(dp); - - rcu_assign_pointer(dps[dp->dp_idx], NULL); - - dp_detach_port(dp->ports[ODPP_LOCAL]); - - tbl_destroy(dp->table, flow_free_tbl); - for (i = 0; i < DP_N_QUEUES; i++) skb_queue_purge(&dp->queues[i]); + + tbl_destroy((struct tbl __force *)dp->table, flow_free_tbl); free_percpu(dp->stats_percpu); kobject_put(&dp->ifobj); - module_put(THIS_MODULE); } static int destroy_dp(int dp_idx) { struct datapath *dp; - int err; + int err = 0; + struct vport *p, *n; rtnl_lock(); mutex_lock(&dp_mutex); dp = get_dp(dp_idx); - err = -ENODEV; - if (!dp) - goto err_unlock; + if (!dp) { + err = -ENODEV; + goto out; + } - do_destroy_dp(dp); - err = 0; + mutex_lock(&dp->mutex); -err_unlock: + list_for_each_entry_safe (p, n, &dp->port_list, node) + if (p->port_no != ODPP_LOCAL) + dp_detach_port(p); + + dp_sysfs_del_dp(dp); + rcu_assign_pointer(dps[dp->dp_idx], NULL); + dp_detach_port(get_vport_protected(dp, ODPP_LOCAL)); + + mutex_unlock(&dp->mutex); + call_rcu(&dp->rcu, destroy_dp_rcu); + module_put(THIS_MODULE); + +out: mutex_unlock(&dp_mutex); rtnl_unlock(); return err; } -/* Called with RTNL lock and dp_mutex. */ +/* Called with RTNL lock and dp->mutex. */ static int new_vport(struct datapath *dp, struct odp_port *odp_port, int port_no) { struct vport_parms parms; @@ -393,7 +414,7 @@ got_port_no: goto out_unlock_dp; set_internal_devs_mtu(dp); - dp_sysfs_add_if(dp->ports[port_no]); + dp_sysfs_add_if(get_vport_protected(dp, port_no)); err = put_user(port_no, &portp->port); @@ -444,7 +465,7 @@ static int detach_port(int dp_idx, int port_no) if (!dp) goto out_unlock_rtnl; - p = dp->ports[port_no]; + p = get_vport_protected(dp, port_no); err = -ENOENT; if (!p) goto out_unlock_dp; @@ -493,7 +514,8 @@ void dp_process_received_packet(struct vport *p, struct sk_buff *skb) flow_node = tbl_lookup(rcu_dereference(dp->table), &key, flow_hash(&key), flow_cmp); if (unlikely(!flow_node)) { - dp_output_control(dp, skb, _ODPL_MISS_NR, OVS_CB(skb)->tun_id); + dp_output_control(dp, skb, _ODPL_MISS_NR, + (__force u64)OVS_CB(skb)->tun_id); stats_counter_off = offsetof(struct dp_stats_percpu, n_missed); goto out; } @@ -640,10 +662,10 @@ err: static int flush_flows(struct datapath *dp) { - struct tbl *old_table = rcu_dereference(dp->table); + struct tbl *old_table = get_table_protected(dp); struct tbl *new_table; - new_table = tbl_create(0); + new_table = tbl_create(TBL_MIN_BUCKETS); if (!new_table) return -ENOMEM; @@ -738,7 +760,7 @@ static struct sw_flow_actions *get_actions(const struct odp_flow *flow) error = -EFAULT; if (copy_from_user(actions->actions, - (struct nlattr __user *)flow->actions, + (struct nlattr __user __force *)flow->actions, flow->actions_len)) goto error_free_actions; error = validate_actions(actions->actions, actions->actions_len); @@ -787,7 +809,7 @@ static void clear_stats(struct sw_flow *flow) static int expand_table(struct datapath *dp) { - struct tbl *old_table = rcu_dereference(dp->table); + struct tbl *old_table = get_table_protected(dp); struct tbl *new_table; new_table = tbl_expand(old_table); @@ -806,14 +828,15 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, struct tbl_node *flow_node; struct sw_flow *flow; struct tbl *table; + struct sw_flow_actions *acts = NULL; int error; + u32 hash; - table = rcu_dereference(dp->table); - flow_node = tbl_lookup(table, &uf->flow.key, flow_hash(&uf->flow.key), flow_cmp); + hash = flow_hash(&uf->flow.key); + table = get_table_protected(dp); + flow_node = tbl_lookup(table, &uf->flow.key, hash, flow_cmp); if (!flow_node) { /* No such flow. */ - struct sw_flow_actions *acts; - error = -ENOENT; if (!(uf->flags & ODPPF_CREATE)) goto error; @@ -823,7 +846,7 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, error = expand_table(dp); if (error) goto error; - table = rcu_dereference(dp->table); + table = get_table_protected(dp); } /* Allocate flow. */ @@ -843,7 +866,7 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, rcu_assign_pointer(flow->sf_acts, acts); /* Put flow in bucket. */ - error = tbl_insert(table, &flow->tbl_node, flow_hash(&flow->key)); + error = tbl_insert(table, &flow->tbl_node, hash); if (error) goto error_free_flow_acts; @@ -864,7 +887,9 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, error = PTR_ERR(new_acts); if (IS_ERR(new_acts)) goto error; - old_acts = rcu_dereference(flow->sf_acts); + + old_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_is_held(&dp->mutex)); if (old_acts->actions_len != new_acts->actions_len || memcmp(old_acts->actions, new_acts->actions, old_acts->actions_len)) { @@ -885,7 +910,7 @@ static int do_put_flow(struct datapath *dp, struct odp_flow_put *uf, return 0; error_free_flow_acts: - kfree(flow->sf_acts); + kfree(acts); error_free_flow: flow->sf_acts = NULL; flow_put(flow); @@ -913,7 +938,8 @@ static int put_flow(struct datapath *dp, struct odp_flow_put __user *ufp) return 0; } -static int do_answer_query(struct sw_flow *flow, u32 query_flags, +static int do_answer_query(struct datapath *dp, struct sw_flow *flow, + u32 query_flags, struct odp_flow_stats __user *ustats, struct nlattr __user *actions, u32 __user *actions_lenp) @@ -936,7 +962,8 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags, if (!actions_len) return 0; - sf_acts = rcu_dereference(flow->sf_acts); + sf_acts = rcu_dereference_protected(flow->sf_acts, + lockdep_is_held(&dp->mutex)); if (put_user(sf_acts->actions_len, actions_lenp) || (actions && copy_to_user(actions, sf_acts->actions, min(sf_acts->actions_len, actions_len)))) @@ -945,21 +972,21 @@ static int do_answer_query(struct sw_flow *flow, u32 query_flags, return 0; } -static int answer_query(struct sw_flow *flow, u32 query_flags, - struct odp_flow __user *ufp) +static int answer_query(struct datapath *dp, struct sw_flow *flow, + u32 query_flags, struct odp_flow __user *ufp) { struct nlattr __user *actions; if (get_user(actions, (struct nlattr __user * __user *)&ufp->actions)) return -EFAULT; - return do_answer_query(flow, query_flags, + return do_answer_query(dp, flow, query_flags, &ufp->stats, actions, &ufp->actions_len); } static struct sw_flow *do_del_flow(struct datapath *dp, struct odp_flow_key *key) { - struct tbl *table = rcu_dereference(dp->table); + struct tbl *table = get_table_protected(dp); struct tbl_node *flow_node; int error; @@ -991,18 +1018,18 @@ 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(dp, flow, 0, ufp); flow_deferred_free(flow); return error; } static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec) { - struct tbl *table = rcu_dereference(dp->table); + struct tbl *table = get_table_protected(dp); u32 i; for (i = 0; i < flowvec->n_flows; i++) { - struct odp_flow __user *ufp = (struct odp_flow __user *)&flowvec->flows[i]; + struct odp_flow __user *ufp = (struct odp_flow __user __force *)&flowvec->flows[i]; struct odp_flow uf; struct tbl_node *flow_node; int error; @@ -1014,7 +1041,7 @@ 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(dp, flow_cast(flow_node), uf.flags, ufp); if (error) return -EFAULT; } @@ -1022,6 +1049,7 @@ static int do_query_flows(struct datapath *dp, const struct odp_flowvec *flowvec } struct list_flows_cbdata { + struct datapath *dp; struct odp_flow __user *uflows; u32 n_flows; u32 listed_flows; @@ -1036,7 +1064,7 @@ 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(cbdata->dp, flow, 0, ufp); if (error) return error; @@ -1053,11 +1081,12 @@ static int do_list_flows(struct datapath *dp, const struct odp_flowvec *flowvec) if (!flowvec->n_flows) return 0; - cbdata.uflows = (struct odp_flow __user *)flowvec->flows; + cbdata.dp = dp; + cbdata.uflows = (struct odp_flow __user __force*)flowvec->flows; cbdata.n_flows = flowvec->n_flows; cbdata.listed_flows = 0; - error = tbl_foreach(rcu_dereference(dp->table), list_flow, &cbdata); + error = tbl_foreach(get_table_protected(dp), list_flow, &cbdata); return error ? error : cbdata.listed_flows; } @@ -1103,7 +1132,7 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute) err = -EFAULT; if (copy_from_user(actions->actions, - (struct nlattr __user *)execute->actions, execute->actions_len)) + (struct nlattr __user __force *)execute->actions, execute->actions_len)) goto error_free_actions; err = validate_actions(actions->actions, execute->actions_len); @@ -1117,7 +1146,7 @@ static int do_execute(struct datapath *dp, const struct odp_execute *execute) err = -EFAULT; if (copy_from_user(skb_put(skb, execute->length), - (const void __user *)execute->data, + (const void __user __force *)execute->data, execute->length)) goto error_free_skb; @@ -1163,7 +1192,7 @@ static int execute_packet(struct datapath *dp, const struct odp_execute __user * static int get_dp_stats(struct datapath *dp, struct odp_stats __user *statsp) { - struct tbl *table = rcu_dereference(dp->table); + struct tbl *table = get_table_protected(dp); struct odp_stats stats; int i; @@ -1245,6 +1274,7 @@ static int put_port(const struct vport *p, struct odp_port __user *uop) rcu_read_lock(); strncpy(op.devname, vport_get_name(p), sizeof op.devname); strncpy(op.type, vport_get_type(p), sizeof op.type); + vport_get_config(p, op.config); rcu_read_unlock(); op.port = p->port_no; @@ -1255,45 +1285,32 @@ static int put_port(const struct vport *p, struct odp_port __user *uop) static int query_port(struct datapath *dp, struct odp_port __user *uport) { struct odp_port port; + struct vport *vport; if (copy_from_user(&port, uport, sizeof port)) return -EFAULT; if (port.devname[0]) { - struct vport *vport; - int err = 0; - port.devname[IFNAMSIZ - 1] = '\0'; vport_lock(); - rcu_read_lock(); - vport = vport_locate(port.devname); - if (!vport) { - err = -ENODEV; - goto error_unlock; - } - if (vport->dp != dp) { - err = -ENOENT; - goto error_unlock; - } - - port.port = vport->port_no; - -error_unlock: - rcu_read_unlock(); vport_unlock(); - if (err) - return err; + if (!vport) + return -ENODEV; + if (vport->dp != dp) + return -ENOENT; } else { if (port.port >= DP_MAX_PORTS) return -EINVAL; - if (!dp->ports[port.port]) + + vport = get_vport_protected(dp, port.port); + if (!vport) return -ENOENT; } - return put_port(dp->ports[port.port], uport); + return put_port(vport, uport); } static int do_list_ports(struct datapath *dp, struct odp_port __user *uports, @@ -1321,7 +1338,7 @@ static int list_ports(struct datapath *dp, struct odp_portvec __user *upv) if (copy_from_user(&pv, upv, sizeof pv)) return -EFAULT; - retval = do_list_ports(dp, (struct odp_port __user *)pv.ports, + retval = do_list_ports(dp, (struct odp_port __user __force *)pv.ports, pv.n_ports); if (retval < 0) return retval; @@ -1551,7 +1568,8 @@ static int compat_put_flow(struct datapath *dp, struct compat_odp_flow_put __use return 0; } -static int compat_answer_query(struct sw_flow *flow, u32 query_flags, +static int compat_answer_query(struct datapath *dp, struct sw_flow *flow, + u32 query_flags, struct compat_odp_flow __user *ufp) { compat_uptr_t actions; @@ -1559,7 +1577,7 @@ 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(dp, flow, query_flags, &ufp->stats, compat_ptr(actions), &ufp->actions_len); } @@ -1576,7 +1594,7 @@ 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(dp, flow, 0, ufp); flow_deferred_free(flow); return error; } @@ -1585,7 +1603,7 @@ static int compat_query_flows(struct datapath *dp, struct compat_odp_flow __user *flows, u32 n_flows) { - struct tbl *table = rcu_dereference(dp->table); + struct tbl *table = get_table_protected(dp); u32 i; for (i = 0; i < n_flows; i++) { @@ -1601,7 +1619,8 @@ static int compat_query_flows(struct datapath *dp, 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(dp, flow_cast(flow_node), + uf.flags, ufp); if (error) return -EFAULT; } @@ -1609,6 +1628,7 @@ static int compat_query_flows(struct datapath *dp, } struct compat_list_flows_cbdata { + struct datapath *dp; struct compat_odp_flow __user *uflows; u32 n_flows; u32 listed_flows; @@ -1623,7 +1643,7 @@ 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(cbdata->dp, flow, 0, ufp); if (error) return error; @@ -1641,11 +1661,12 @@ static int compat_list_flows(struct datapath *dp, if (!n_flows) return 0; + cbdata.dp = dp; cbdata.uflows = flows; cbdata.n_flows = n_flows; cbdata.listed_flows = 0; - error = tbl_foreach(rcu_dereference(dp->table), compat_list_flow, &cbdata); + error = tbl_foreach(get_table_protected(dp), compat_list_flow, &cbdata); return error ? error : cbdata.listed_flows; } @@ -1918,6 +1939,8 @@ success: u16 csum_start, csum_offset; get_skb_csum_pointers(skb, &csum_start, &csum_offset); + csum_start -= skb_headroom(skb); + BUG_ON(csum_start >= skb_headlen(skb)); retval = skb_copy_and_csum_datagram(skb, csum_start, buf + csum_start, copy_bytes - csum_start, &csum); @@ -1973,6 +1996,7 @@ static unsigned int openvswitch_poll(struct file *file, poll_table *wait) } static struct file_operations openvswitch_fops = { + .owner = THIS_MODULE, .read = openvswitch_read, .poll = openvswitch_poll, .unlocked_ioctl = openvswitch_ioctl,