/*
- * Copyright (c) 2007, 2008, 2009, 2010 Nicira Networks.
+ * Copyright (c) 2007, 2008, 2009, 2010, 2011 Nicira Networks.
* Distributed under the terms of the GNU GPL version 2.
*
* Significant portions of this file may be copied from parts of the Linux
#include "table.h"
#include "vport-internal_dev.h"
-#include "compat.h"
-
int (*dp_ioctl_hook)(struct net_device *dev, struct ifreq *rq, int cmd);
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);
{
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);
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)
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;
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,
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]);
/* 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;
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);
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;
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);
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;
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;
}
OVS_CB(skb)->flow = flow_cast(flow_node);
}
+ stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
flow_used(OVS_CB(skb)->flow, skb);
acts = rcu_dereference(OVS_CB(skb)->flow->sf_acts);
loop->looping = true;
if (unlikely(loop->looping)) {
loop_suppress(dp, acts);
+ kfree_skb(skb);
goto out_loop;
}
/* Execute actions. */
execute_actions(dp, skb, &OVS_CB(skb)->flow->key, acts->actions,
acts->actions_len);
- stats_counter_off = offsetof(struct dp_stats_percpu, n_hit);
/* Check whether sub-actions looped too much. */
if (unlikely(loop->looping))
kfree_skb(skb);
skb = nskb;
- if (unlikely(IS_ERR(skb))) {
+ if (IS_ERR(skb)) {
err = PTR_ERR(skb);
goto 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;
goto error;
error = -EFAULT;
- if (copy_from_user(actions->actions, flow->actions, flow->actions_len))
+ if (copy_from_user(actions->actions,
+ (struct nlattr __user __force *)flow->actions,
+ flow->actions_len))
goto error_free_actions;
error = validate_actions(actions->actions, actions->actions_len);
if (error)
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);
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;
error = expand_table(dp);
if (error)
goto error;
- table = rcu_dereference(dp->table);
+ table = get_table_protected(dp);
}
/* Allocate flow. */
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;
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)) {
return 0;
error_free_flow_acts:
- kfree(flow->sf_acts);
+ kfree(acts);
error_free_flow:
flow->sf_acts = NULL;
flow_put(flow);
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)
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))))
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 *actions;
+ struct nlattr __user *actions;
- if (get_user(actions, &ufp->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;
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 = &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;
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;
}
}
struct list_flows_cbdata {
+ struct datapath *dp;
struct odp_flow __user *uflows;
u32 n_flows;
u32 listed_flows;
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;
if (!flowvec->n_flows)
return 0;
- cbdata.uflows = 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;
}
}
err = -EFAULT;
- if (copy_from_user(actions->actions, execute->actions, execute->actions_len))
+ if (copy_from_user(actions->actions,
+ (struct nlattr __user __force *)execute->actions, execute->actions_len))
goto error_free_actions;
err = validate_actions(actions->actions, execute->actions_len);
goto error_free_actions;
err = -EFAULT;
- if (copy_from_user(skb_put(skb, execute->length), execute->data,
+ if (copy_from_user(skb_put(skb, execute->length),
+ (const void __user __force *)execute->data,
execute->length))
goto error_free_skb;
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;
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;
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,
if (copy_from_user(&pv, upv, sizeof pv))
return -EFAULT;
- retval = do_list_ports(dp, pv.ports, pv.n_ports);
+ retval = do_list_ports(dp, (struct odp_port __user __force *)pv.ports,
+ pv.n_ports);
if (retval < 0)
return retval;
__get_user(flow->flags, &compat->flags))
return -EFAULT;
- flow->actions = compat_ptr(actions);
+ flow->actions = (struct nlattr __force *)compat_ptr(actions);
return 0;
}
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;
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);
}
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;
}
-static int compat_query_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
+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++) {
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;
}
}
struct compat_list_flows_cbdata {
+ struct datapath *dp;
struct compat_odp_flow __user *uflows;
u32 n_flows;
u32 listed_flows;
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;
return 0;
}
-static int compat_list_flows(struct datapath *dp, struct compat_odp_flow *flows, u32 n_flows)
+static int compat_list_flows(struct datapath *dp,
+ struct compat_odp_flow __user *flows, u32 n_flows)
{
struct compat_list_flows_cbdata cbdata;
int error;
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;
}
static int compat_flowvec_ioctl(struct datapath *dp, unsigned long argp,
int (*function)(struct datapath *,
- struct compat_odp_flow *,
+ struct compat_odp_flow __user *,
u32 n_flows))
{
struct compat_odp_flowvec __user *uflowvec;
__get_user(execute.length, &uexecute->length))
return -EFAULT;
- execute.actions = compat_ptr(actions);
- execute.data = compat_ptr(data);
+ execute.actions = (struct nlattr __force *)compat_ptr(actions);
+ execute.data = (const void __force *)compat_ptr(data);
return do_execute(dp, &execute);
}
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);
copy_bytes = csum_start;
csump = (__sum16 __user *)(buf + csum_start + csum_offset);
- BUG_ON((char *)csump + sizeof(__sum16) > buf + nbytes);
+ BUG_ON((char __user *)csump + sizeof(__sum16) >
+ buf + nbytes);
put_user(csum_fold(csum), csump);
}
} else
}
if (!retval) {
- struct iovec __user iov;
+ struct iovec iov;
iov.iov_base = buf;
iov.iov_len = copy_bytes;
}
static struct file_operations openvswitch_fops = {
+ .owner = THIS_MODULE,
.read = openvswitch_read,
.poll = openvswitch_poll,
.unlocked_ioctl = openvswitch_ioctl,