Re: [PATCH net-next v2 2/3] bridge: Add a limit on learned FDB entries
From: Ido Schimmel <idosch@nvidia.com>
Date: 2023-06-19 15:34:51
Also in:
bridge
On Mon, Jun 19, 2023 at 09:14:42AM +0200, Johannes Nixdorf wrote:
A malicious actor behind one bridge port may spam the kernel with packets
with a random source MAC address, each of which will create an FDB entry,
each of which is a dynamic allocation in the kernel.
There are roughly 2^48 different MAC addresses, further limited by the
rhashtable they are stored in to 2^31. Each entry is of the type struct
net_bridge_fdb_entry, which is currently 128 bytes big. This means the
maximum amount of memory allocated for FDB entries is 2^31 * 128B =
256GiB, which is too much for most computers.
Mitigate this by adding a bridge netlink setting
IFLA_BR_FDB_MAX_LEARNED_ENTRIES, which, if nonzero, limits the amount
of learned entries to a user specified maximum.
For backwards compatibility the default setting of 0 disables the limit.
User-added entries by netlink or from bridge or bridge port addresses
are never blocked and do not count towards that limit.
All changes to fdb_n_entries are under br->hash_lock, which means we do
not need additional locking. The call paths are (✓ denotes that
br->hash_lock is taken around the next call):
- fdb_delete <-+- fdb_delete_local <-+- br_fdb_changeaddr ✓
| +- br_fdb_change_mac_address ✓
| +- br_fdb_delete_by_port ✓
+- br_fdb_find_delete_local ✓
+- fdb_add_local <-+- br_fdb_changeaddr ✓
| +- br_fdb_change_mac_address ✓
| +- br_fdb_add_local ✓
+- br_fdb_cleanup ✓
+- br_fdb_flush ✓
+- br_fdb_delete_by_port ✓
+- fdb_delete_by_addr_and_port <--- __br_fdb_delete ✓
+- br_fdb_external_learn_del ✓
- fdb_create <-+- fdb_add_local <-+- br_fdb_changeaddr ✓
| +- br_fdb_change_mac_address ✓
| +- br_fdb_add_local ✓
+- br_fdb_update ✓
+- fdb_add_entry <--- __br_fdb_add ✓
+- br_fdb_external_learn_add ✓
The flags that imply an entry does not come from learning
(BR_FDB_NOT_LEARNED_MASK) are now only set or cleared under br->hash_lock
as well, and when the boolean value of (fdb->flags &
BR_FDB_NOT_LEARNED_MASK) changes the accounting is updated.
This introduces one additional locked update in br_fdb_update if
BR_FDB_ADDED_BY_USER was set. This is only the case when creating a new
entry via netlink, and never in the packet handling fast path.
Signed-off-by: Johannes Nixdorf <redacted>
---
Changes since v1:
- Do not initialize fdb_*_entries to 0. (from review)
- Do not skip decrementing on 0. (from review)
- Moved the counters to a conditional hole in struct net_bridge to
avoid growing the struct. (from review, it still grows the struct as
there are 2 32-bit values)
- Add IFLA_BR_FDB_CUR_LEARNED_ENTRIES (from review)
- Fix br_get_size()
- Only limit learned entries, rename to
*_(CUR|MAX)_LEARNED_ENTRIES. (from review)
Obsolete v1 review comments:
- Return better errors to users: Due to limiting the limit to
automatically created entries, netlink fdb add requests and changing
bridge ports are never rejected, so they do not yet need a more
friendly error returned.
include/uapi/linux/if_link.h | 2 ++
net/bridge/br_fdb.c | 67 +++++++++++++++++++++++++++++++++---
net/bridge/br_netlink.c | 13 ++++++-
net/bridge/br_private.h | 6 ++++To minimize the number of changes per patch and make review easier, try to first maintain the count and the maximum and then in a separate patch expose them via netlink. See b57e8d870d52 and a1aee20d5db2, for example. Merge commit is cb3086cee656.
quoted hunk ↗ jump to hunk
4 files changed, 83 insertions(+), 5 deletions(-)diff --git a/include/uapi/linux/if_link.h b/include/uapi/linux/if_link.h index 4ac1000b0ef2..165b9014379b 100644 --- a/include/uapi/linux/if_link.h +++ b/include/uapi/linux/if_link.h@@ -510,6 +510,8 @@ enum { IFLA_BR_VLAN_STATS_PER_PORT, IFLA_BR_MULTI_BOOLOPT, IFLA_BR_MCAST_QUERIER_STATE, + IFLA_BR_FDB_CUR_LEARNED_ENTRIES, + IFLA_BR_FDB_MAX_LEARNED_ENTRIES, __IFLA_BR_MAX, };diff --git a/net/bridge/br_fdb.c b/net/bridge/br_fdb.c index ac1dc8723b9c..bc61d1fd5fcf 100644 --- a/net/bridge/br_fdb.c +++ b/net/bridge/br_fdb.c@@ -301,6 +301,38 @@ static void fdb_add_hw_addr(struct net_bridge *br, const unsigned char *addr) } } +/* Set a FDB flag that implies the entry was not learned, and account + * for changes in the learned status. + */ +static void __fdb_set_flag_not_learned(struct net_bridge *br, + struct net_bridge_fdb_entry *fdb, + long nr) +{ + WARN_ON_ONCE(!(BIT(nr) & BR_FDB_NOT_LEARNED_MASK)); + + /* learned before, but we set a flag that implies it's manually added */ + if (!(fdb->flags & BR_FDB_NOT_LEARNED_MASK)) + br->fdb_cur_learned_entries--; + set_bit(nr, &fdb->flags); +} + +/* Set a FDB flag that implies the entry was not learned, and account + * for changes in the learned status. + * + * This function takes a lock, so ensure it is not called in the fast + * path. + */ +static void fdb_set_flag_not_learned(struct net_bridge *br, + struct net_bridge_fdb_entry *fdb, + long nr) +{ + spin_lock_bh(&br->hash_lock); + + __fdb_set_flag_not_learned(br, fdb, nr); + + spin_unlock_bh(&br->hash_lock); +} + /* When a static FDB entry is deleted, the HW address from that entry is * also removed from the bridge private HW address list and updates all * the ports with needed information.@@ -321,6 +353,8 @@ static void fdb_del_hw_addr(struct net_bridge *br, const unsigned char *addr) static void fdb_delete(struct net_bridge *br, struct net_bridge_fdb_entry *f, bool swdev_notify) { + bool learned = !(f->flags & BR_FDB_NOT_LEARNED_MASK); + trace_fdb_delete(br, f); if (test_bit(BR_FDB_STATIC, &f->flags))@@ -329,11 +363,16 @@ static void fdb_delete(struct net_bridge *br, struct net_bridge_fdb_entry *f, hlist_del_init_rcu(&f->fdb_node); rhashtable_remove_fast(&br->fdb_hash_tbl, &f->rhnode, br_fdb_rht_params); + br->fdb_cur_learned_entries -= learned; fdb_notify(br, f, RTM_DELNEIGH, swdev_notify); call_rcu(&f->rcu, fdb_rcu_free); } -/* Delete a local entry if no other port had the same address. */ +/* Delete a local entry if no other port had the same address. + * + * This function should only be called on entries with BR_FDB_LOCAL set, + * so clear_bit never removes the last bit in BR_FDB_NOT_LEARNED_MASK. + */ static void fdb_delete_local(struct net_bridge *br, const struct net_bridge_port *p, struct net_bridge_fdb_entry *f)@@ -390,6 +429,11 @@ static struct net_bridge_fdb_entry *fdb_create(struct net_bridge *br, { struct net_bridge_fdb_entry *fdb; int err; + bool learned = !(flags & BR_FDB_NOT_LEARNED_MASK);
https://www.kernel.org/doc/html/latest/process/maintainer-netdev.html#local-variable-ordering-reverse-xmas-tree-rcs
+ + if (unlikely(learned && br->fdb_max_learned_entries && + br->fdb_cur_learned_entries >= br->fdb_max_learned_entries))
This function is not run with RTNL held and so 'fdb_max_learned_entries' can be updated concurrently from user space. You will need READ_ONCE() / WRITE_ONCE() to silence KCSAN.
quoted hunk ↗ jump to hunk
+ return NULL; fdb = kmem_cache_alloc(br_fdb_cache, GFP_ATOMIC); if (!fdb)@@ -409,6 +453,8 @@ static struct net_bridge_fdb_entry *fdb_create(struct net_bridge *br, hlist_add_head_rcu(&fdb->fdb_node, &br->fdb_list); + br->fdb_cur_learned_entries += learned;
Don't remember seeing bool being added to a u32...
quoted hunk ↗ jump to hunk
+ return fdb; }@@ -894,7 +940,7 @@ void br_fdb_update(struct net_bridge *br, struct net_bridge_port *source, } if (unlikely(test_bit(BR_FDB_ADDED_BY_USER, &flags))) - set_bit(BR_FDB_ADDED_BY_USER, &fdb->flags); + fdb_set_flag_not_learned(br, fdb, BR_FDB_ADDED_BY_USER); if (unlikely(fdb_modified)) { trace_br_fdb_update(br, source, addr, vid, flags); fdb_notify(br, fdb, RTM_NEWNEIGH, true);@@ -1070,6 +1116,8 @@ static int fdb_add_entry(struct net_bridge *br, struct net_bridge_port *source, modified = true; } + if (!(fdb->flags & BR_FDB_NOT_LEARNED_MASK)) + br->fdb_cur_learned_entries--; set_bit(BR_FDB_ADDED_BY_USER, &fdb->flags); }@@ -1440,10 +1488,10 @@ int br_fdb_external_learn_add(struct net_bridge *br, struct net_bridge_port *p, } if (swdev_notify) - set_bit(BR_FDB_ADDED_BY_USER, &fdb->flags); + __fdb_set_flag_not_learned(br, fdb, BR_FDB_ADDED_BY_USER); if (!p) - set_bit(BR_FDB_LOCAL, &fdb->flags); + __fdb_set_flag_not_learned(br, fdb, BR_FDB_LOCAL); if (modified) fdb_notify(br, fdb, RTM_NEWNEIGH, swdev_notify);@@ -1508,3 +1556,14 @@ void br_fdb_clear_offload(const struct net_device *dev, u16 vid) spin_unlock_bh(&p->br->hash_lock); } EXPORT_SYMBOL_GPL(br_fdb_clear_offload); + +u32 br_fdb_get_cur_learned_entries(struct net_bridge *br) +{ + u32 ret; + + spin_lock_bh(&br->hash_lock); + ret = br->fdb_cur_learned_entries;
I believe that for MDB Nik asked not to take a spin lock for each dump and we used READ_ONCE() / WRITE_ONCE() instead.
quoted hunk ↗ jump to hunk
+ spin_unlock_bh(&br->hash_lock); + + return ret; +}diff --git a/net/bridge/br_netlink.c b/net/bridge/br_netlink.c index 05c5863d2e20..954c468d52ec 100644 --- a/net/bridge/br_netlink.c +++ b/net/bridge/br_netlink.c@@ -1527,6 +1527,12 @@ static int br_changelink(struct net_device *brdev, struct nlattr *tb[], return err; } + if (data[IFLA_BR_FDB_MAX_LEARNED_ENTRIES]) {
Please add this attribute to 'br_policy' and in a separate patch convert the policy to use 'strict_start_type' so that future submissions will be forced to update the policy. See c00041cf1cb82 for example. I suggest writing a selftest to verify the expected behavior and to make sure this feature does not regress. There are a lot of examples under tools/testing/selftests/net/ and tools/testing/selftests/net/forwarding in particular. You can post another version before writing a test, but mark it as RFC so that it doesn't get applied by mistake. Thanks
quoted hunk ↗ jump to hunk
+ u32 val = nla_get_u32(data[IFLA_BR_FDB_MAX_LEARNED_ENTRIES]); + + br->fdb_max_learned_entries = val; + } + return 0; }@@ -1581,6 +1587,8 @@ static size_t br_get_size(const struct net_device *brdev) nla_total_size_64bit(sizeof(u64)) + /* IFLA_BR_TOPOLOGY_CHANGE_TIMER */ nla_total_size_64bit(sizeof(u64)) + /* IFLA_BR_GC_TIMER */ nla_total_size(ETH_ALEN) + /* IFLA_BR_GROUP_ADDR */ + nla_total_size(sizeof(u32)) + /* IFLA_BR_FDB_CUR_LEARNED_ENTRIES */ + nla_total_size(sizeof(u32)) + /* IFLA_BR_FDB_MAX_LEARNED_ENTRIES */ #ifdef CONFIG_BRIDGE_IGMP_SNOOPING nla_total_size(sizeof(u8)) + /* IFLA_BR_MCAST_ROUTER */ nla_total_size(sizeof(u8)) + /* IFLA_BR_MCAST_SNOOPING */@@ -1620,6 +1628,7 @@ static int br_fill_info(struct sk_buff *skb, const struct net_device *brdev) u32 stp_enabled = br->stp_enabled; u16 priority = (br->bridge_id.prio[0] << 8) | br->bridge_id.prio[1]; u8 vlan_enabled = br_vlan_enabled(br->dev); + u32 fdb_cur_learned_entries = br_fdb_get_cur_learned_entries(br); struct br_boolopt_multi bm; u64 clockval;@@ -1656,7 +1665,9 @@ static int br_fill_info(struct sk_buff *skb, const struct net_device *brdev) nla_put_u8(skb, IFLA_BR_TOPOLOGY_CHANGE_DETECTED, br->topology_change_detected) || nla_put(skb, IFLA_BR_GROUP_ADDR, ETH_ALEN, br->group_addr) || - nla_put(skb, IFLA_BR_MULTI_BOOLOPT, sizeof(bm), &bm)) + nla_put(skb, IFLA_BR_MULTI_BOOLOPT, sizeof(bm), &bm) || + nla_put_u32(skb, IFLA_BR_FDB_CUR_LEARNED_ENTRIES, fdb_cur_learned_entries) || + nla_put_u32(skb, IFLA_BR_FDB_MAX_LEARNED_ENTRIES, br->fdb_max_learned_entries)) return -EMSGSIZE; #ifdef CONFIG_BRIDGE_VLAN_FILTERINGdiff --git a/net/bridge/br_private.h b/net/bridge/br_private.h index 2119729ded2b..df079191479e 100644 --- a/net/bridge/br_private.h +++ b/net/bridge/br_private.h@@ -275,6 +275,8 @@ enum { BR_FDB_LOCKED, }; +#define BR_FDB_NOT_LEARNED_MASK (BIT(BR_FDB_LOCAL) | BIT(BR_FDB_ADDED_BY_USER)) + struct net_bridge_fdb_key { mac_addr addr; u16 vlan_id;@@ -553,6 +555,9 @@ struct net_bridge { struct kobject *ifobj; u32 auto_cnt; + u32 fdb_max_learned_entries; + u32 fdb_cur_learned_entries; + #ifdef CONFIG_NET_SWITCHDEV /* Counter used to make sure that hardware domains get unique * identifiers in case a bridge spans multiple switchdev instances.@@ -838,6 +843,7 @@ int br_fdb_external_learn_del(struct net_bridge *br, struct net_bridge_port *p, bool swdev_notify); void br_fdb_offloaded_set(struct net_bridge *br, struct net_bridge_port *p, const unsigned char *addr, u16 vid, bool offloaded); +u32 br_fdb_get_cur_learned_entries(struct net_bridge *br); /* br_forward.c */ enum br_pkt_type {-- 2.40.1