netfilter updates for net
Subsystems (3)
Florian Westphal
maintainercover netfilter updates for net
Hello,
This PR contains nf_tables updates for your *net* tree.
First patch fixes table validation, I broke this in 6.4 when tracking validation state per table, reported by Pablo, fixup from myself.
Second patch makes sure objects waiting for memory release have been released, this was broken in 6.1, patch from Pablo Neira Ayuso.
Patch three is a fix-for-fix from previous PR: In case a transaction gets aborted, gc sequence counter needs to be incremented so pending gc requests are invalidated, from Pablo.
Same for patch 4: gc list needs to use gc list lock, not destroy lock, also from Pablo.
Patch 5 fixes a UaF in a set backend, but this should only occur when failslab is enabled for GFP_KERNEL allocations, broken since feature was added in 5.6, from myself.
Patch 6 fixes a double-free bug that was also added via previous PR:
We must not schedule gc work if the previous batch is still queued.
The following changes since commit bfedba3b2c7793ce127680bc8f70711e05ec7a17:
ibmveth: Use dcbf rather than dcbfl (2023-08-23 11:51:16 +0100)
are available in the Git repository at:
ssh://[email protected]/pub/scm/linux/kernel/git/netfilter/nf tags/nf-23-08-23
for you to fetch changes up to 8e51830e29e12670b4c10df070a4ea4c9593e961:
netfilter: nf_tables: defer gc run if previous batch is still pending (2023-08-23 16:12:59 +0200)
----------------------------------------------------------------
netfilter pull request 2023-08-23
----------------------------------------------------------------
Florian Westphal (3):
netfilter: nf_tables: validate all pending tables
netfilter: nf_tables: fix out of memory error handling
netfilter: nf_tables: defer gc run if previous batch is still pending
Pablo Neira Ayuso (3):
netfilter: nf_tables: flush pending destroy work before netlink notifier
netfilter: nf_tables: GC transaction race with abort path
netfilter: nf_tables: use correct lock to protect gc_list
include/net/netfilter/nf_tables.h | 6 ++++++ net/netfilter/nf_tables_api.c | 23 +++++++++++++++-------- net/netfilter/nft_set_hash.c | 3 +++ net/netfilter/nft_set_pipapo.c | 13 ++++++++++--- net/netfilter/nft_set_rbtree.c | 3 +++ 5 files changed, 37 insertions(+), 11 deletions(-)
Florian Westphal
maintainerPATCH 1/6 netfilter: nf_tables: validate all pending tables
We have to validate all tables in the transaction that are in VALIDATE_DO state, the blamed commit below did not move the break statement to its right location so we only validate one table.
Moreover, we can't init table->validate to _SKIP when a table object is allocated.
If we do, then if a transcaction creates a new table and then fails the transaction, nfnetlink will loop and nft will hang until user cancels the command.
Add back the pernet state as a place to stash the last state encountered. This is either _DO (we hit an error during commit validation) or _SKIP (transaction passed all checks).
--- include/net/netfilter/nf_tables.h | 1 + net/netfilter/nf_tables_api.c | 11 +++++++---- 2 files changed, 8 insertions(+), 4 deletions(-)
include/net/netfilter/nf_tables.h +1 -0
@@ -1729,6 +1729,7 @@ struct nftables_pernet {u64 table_handle;unsigned int base_seq;unsigned int gc_seq; +u8 validate_state; };extern unsigned int nf_tables_net_id;
net/netfilter/nf_tables_api.c +7 -4
@@ -1373,7 +1373,7 @@ static int nf_tables_newtable(struct sk_buff *skb, const struct nfnl_info *info,if (table ==NULL ) goto err_kzalloc; - table->validate_state =NFT_VALIDATE_SKIP ; + table->validate_state = nft_net->validate_state; table->name =nla_strdup (attr,GFP_KERNEL_ACCOUNT );if (table->name ==NULL ) goto err_strdup;
@@ -9051,9 +9051,8 @@ static int nf_tables_validate(struct net *net)return -EAGAIN ;nft_validate_state_update (table,NFT_VALIDATE_SKIP ); +break ; } - -break ; }return 0 ;
@@ -9799,8 +9798,10 @@ static int nf_tables_commit(struct net *net, struct sk_buff *skb) }/* 0. Validate ruleset, otherwise roll back for error reporting. */ -if (nf_tables_validate (net) <0 ) +if (nf_tables_validate (net) <0 ) { + nft_net->validate_state =NFT_VALIDATE_DO ;return -EAGAIN ; + } err =nft_flow_rule_offload_commit (net);if (err <0 )
@@ -10059,6 +10060,7 @@ static int nf_tables_commit(struct net *net, struct sk_buff *skb)nf_tables_commit_audit_log (&adl, nft_net->base_seq);nft_gc_seq_end (nft_net, gc_seq); + nft_net->validate_state =NFT_VALIDATE_SKIP ;nf_tables_commit_release (net);return 0 ;
@@ -11115,6 +11117,7 @@ static int __net_init nf_tables_init_net(struct net *net)mutex_init (&nft_net->commit_mutex); nft_net->base_seq =1 ; nft_net->gc_seq =0 ; + nft_net->validate_state =NFT_VALIDATE_SKIP ;return 0 ; }
Signature · 1 line
2.41.0
Re: [PATCH net 1/6] netfilter: nf_tables: validate all pending tables
This series was applied to netdev/net.git (main) by Florian Westphal <[email protected]>:
Hello:
This series was applied to netdev/net.git (main)
by Florian Westphal <[email protected]>:
Quoted 8 lines · On Wed, 23 Aug 2023 17:26:49 +0200 you wrote:
We have to validate all tables in the transaction that are in
VALIDATE_DOstate, the blamed commit below did not move the break statement to its right location so we only validate one table.Moreover, we can't init
table->validateto_SKIPwhen a table object is allocated.[...]
Here is the summary with links:
- [net,1/6] netfilter: nf_tables: validate all pending tables
https://git.kernel.org/netdev/net/c/4b80ced971b0 - [net,2/6] netfilter: nf_tables: flush pending destroy work before netlink notifier https://git.kernel.org/netdev/net/c/2c9f0293280e - [net,3/6] netfilter: nf_tables: GC transaction race with abort path https://git.kernel.org/netdev/net/c/720344340fb9 - [net,4/6] netfilter: nf_tables: use correct lock to protect gc_list https://git.kernel.org/netdev/net/c/8357bc946a2a - [net,5/6] netfilter: nf_tables: fix out of memory error handling https://git.kernel.org/netdev/net/c/5e1be4cdc98c - [net,6/6] netfilter: nf_tables: defer gc run if previous batch is still pending https://git.kernel.org/netdev/net/c/8e51830e29e1
You are awesome, thank you!
Signature · 2 lines
Deet-doot-dot, I am a bot. https://korg.docs.kernel.org/patchwork/pwbot.html
Florian Westphal
maintainerPATCH 2/6 netfilter: nf_tables: flush pending destroy work before netlink notifier
In reply to Florian WestphalFrom: Pablo Neira Ayuso <[email protected]>
Destroy work waits for the RCU grace period then it releases the objects with no mutex held. All releases objects follow this path for transactions, therefore, order is guaranteed and references to top-level objects in the hierarchy remain valid.
However, netlink notifier might interfer with pending destroy work. rcu_barrier() is not correct because objects are not release via RCU callback. Flush destroy work before releasing objects from netlink notifier path.
--- net/netfilter/nf_tables_api.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
net/netfilter/nf_tables_api.c +1 -1
@@ -11073,7 +11073,7 @@ static int nft_rcv_nl_event(struct notifier_block *this, unsigned long event, gc_seq =nft_gc_seq_begin (nft_net);if (!list_empty (&nf_tables_destroy_list)) -rcu_barrier (); +nf_tables_trans_destroy_flush_work (); again:list_for_each_entry (table, &nft_net->tables, list) { if (nft_table_has_owner(table) &&
Signature · 1 line
2.41.0
Florian Westphal
maintainerPATCH 3/6 netfilter: nf_tables: GC transaction race with abort path
In reply to Florian WestphalFrom: Pablo Neira Ayuso <[email protected]>
Abort path is missing a synchronization point with GC transactions. Add GC sequence number hence any GC transaction losing race will be discarded.
--- net/netfilter/nf_tables_api.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)
net/netfilter/nf_tables_api.c +5 -1
@@ -10337,8 +10337,12 @@ static int nf_tables_abort(struct net *net, struct sk_buff *skb, enum nfnl_abort_action action) {struct nftables_pernet *nft_net =nft_pernet (net); -int ret =__nf_tables_abort (net, action); +unsigned int gc_seq; +int ret; + gc_seq =nft_gc_seq_begin (nft_net); + ret =__nf_tables_abort (net, action); +nft_gc_seq_end (nft_net, gc_seq);mutex_unlock (&nft_net->commit_mutex);return ret;
Signature · 1 line
2.41.0
Florian Westphal
maintainerPATCH 4/6 netfilter: nf_tables: use correct lock to protect gc_list
In reply to Florian WestphalFrom: Pablo Neira Ayuso <[email protected]>
Use nf_tables_gc_list_lock spinlock, not nf_tables_destroy_list_lock to protect the gc list.
--- net/netfilter/nf_tables_api.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
net/netfilter/nf_tables_api.c +2 -2
@@ -9456,9 +9456,9 @@ static void nft_trans_gc_work(struct work_struct *work)struct nft_trans_gc *trans, *next;LIST_HEAD (trans_gc_list); -spin_lock (&nf_tables_destroy_list_lock); +spin_lock (&nf_tables_gc_list_lock);list_splice_init (&nf_tables_gc_list, &trans_gc_list); -spin_unlock (&nf_tables_destroy_list_lock); +spin_unlock (&nf_tables_gc_list_lock);list_for_each_entry_safe (trans, next, &trans_gc_list, list) {list_del (&trans->list);
Signature · 1 line
2.41.0
Florian Westphal
maintainerPATCH 5/6 netfilter: nf_tables: fix out of memory error handling
In reply to Florian WestphalSeveral instances of pipapo_resize() don't propagate allocation failures, this causes a crash when fault injection is enabled for gfp_kernel slabs.
--- net/netfilter/nft_set_pipapo.c | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-)
net/netfilter/nft_set_pipapo.c +10 -3
@@ -902,12 +902,14 @@ static void pipapo_lt_bits_adjust(struct nft_pipapo_field *f)static int pipapo_insert (struct nft_pipapo_field *f,const uint8_t *k,int mask_bits) { -int rule = f->rules++, group, ret, bit_offset =0 ; +int rule = f->rules, group, ret, bit_offset =0 ; - ret =pipapo_resize (f, f->rules -1 , f->rules); + ret =pipapo_resize (f, f->rules, f->rules +1 );if (ret)return ret; + f->rules++; +for (group =0 ; group < f->groups; group++) {int i, v;u8 mask;
@@ -1052,7 +1054,9 @@ static int pipapo_expand(struct nft_pipapo_field *f, step++;if (step >= len) {if (!masks) { -pipapo_insert (f, base,0 ); + err =pipapo_insert (f, base,0 ); +if (err <0 ) +return err; masks =1 ; } goto out;
@@ -1235,6 +1239,9 @@ static int nft_pipapo_insert(const struct net *net, const struct nft_set *set,else ret =pipapo_expand (f, start, end, f->groups * f->bb); +if (ret <0 ) +return ret; +if (f->bsize > bsize_max) bsize_max = f->bsize;
Signature · 1 line
2.41.0
Florian Westphal
maintainerPATCH 6/6 netfilter: nf_tables: defer gc run if previous batch is still pending
In reply to Florian WestphalDon't queue more gc work, else we may queue the same elements multiple times.
If an element is flagged as dead, this can mean that either the previous gc request was invalidated/discarded by a transaction or that the previous request is still pending in the system work queue.
The latter will happen if the gc interval is set to a very low value, e.g. 1ms, and system work queue is backlogged.
The sets refcount is 1 if no previous gc requeusts are queued, so add a helper for this and skip gc run if old requests are pending.
Add a helper for this and skip the gc run in this case.
--- include/net/netfilter/nf_tables.h | 5 +++++ net/netfilter/nft_set_hash.c | 3 +++ net/netfilter/nft_set_rbtree.c | 3 +++ 3 files changed, 11 insertions(+)
include/net/netfilter/nf_tables.h +5 -0
@@ -587,6 +587,11 @@ static inline void *nft_set_priv(const struct nft_set *set)return (void *)set->data; } +static inline bool nft_set_gc_is_pending (const struct nft_set *s) +{ +return refcount_read (&s->refs) !=1 ; +} +static inline struct nft_set *nft_set_container_of (const void *priv) {return (void *)priv - offsetof(struct nft_set , data);
net/netfilter/nft_set_hash.c +3 -0
@@ -326,6 +326,9 @@ static void nft_rhash_gc(struct work_struct *work) nft_net =nft_pernet (net); gc_seq =READ_ONCE (nft_net->gc_seq); +if (nft_set_gc_is_pending (set)) + goto done; + gc =nft_trans_gc_alloc (set, gc_seq,GFP_KERNEL );if (!gc) goto done;
net/netfilter/nft_set_rbtree.c +3 -0
@@ -611,6 +611,9 @@ static void nft_rbtree_gc(struct work_struct *work) nft_net =nft_pernet (net); gc_seq =READ_ONCE (nft_net->gc_seq); +if (nft_set_gc_is_pending (set)) + goto done; + gc =nft_trans_gc_alloc (set, gc_seq,GFP_KERNEL );if (!gc) goto done;
Signature · 1 line
2.41.0