From 7864c667aed01a58b87ca518a631322cd0ac34c0 Mon Sep 17 00:00:00 2001 From: Florian Westphal Date: Tue, 3 Mar 2026 16:31:32 +0100 Subject: [PATCH] netfilter: nft_set_pipapo: split gc into unlink and reclaim phase commit 9df95785d3d8302f7c066050117b04cd3c2048c2 upstream. Yiming Qian reports Use-after-free in the pipapo set type: Under a large number of expired elements, commit-time GC can run for a very long time in a non-preemptible context, triggering soft lockup warnings and RCU stall reports (local denial of service). We must split GC in an unlink and a reclaim phase. We cannot queue elements for freeing until pointers have been swapped. Expired elements are still exposed to both the packet path and userspace dumpers via the live copy of the data structure. call_rcu() does not protect us: dump operations or element lookups starting after call_rcu has fired can still observe the free'd element, unless the commit phase has made enough progress to swap the clone and live pointers before any new reader has picked up the old version. This a similar approach as done recently for the rbtree backend in commit 35f83a75529a ("netfilter: nft_set_rbtree: don't gc elements on insert"). Fixes: 3c4287f62044 ("nf_tables: Add set type for arbitrary concatenation of ranges") Reported-by: Yiming Qian Signed-off-by: Florian Westphal Signed-off-by: Greg Kroah-Hartman --- include/net/netfilter/nf_tables.h | 5 +++ net/netfilter/nf_tables_api.c | 5 --- net/netfilter/nft_set_pipapo.c | 51 ++++++++++++++++++++++++++----- net/netfilter/nft_set_pipapo.h | 2 ++ 4 files changed, 50 insertions(+), 13 deletions(-) diff --git a/include/net/netfilter/nf_tables.h b/include/net/netfilter/nf_tables.h index 02a73037fae0..000ae2900f8c 100644 --- a/include/net/netfilter/nf_tables.h +++ b/include/net/netfilter/nf_tables.h @@ -1770,6 +1770,11 @@ struct nft_trans_gc { struct rcu_head rcu; }; +static inline int nft_trans_gc_space(const struct nft_trans_gc *trans) +{ + return NFT_TRANS_GC_BATCHCOUNT - trans->count; +} + struct nft_trans_gc *nft_trans_gc_alloc(struct nft_set *set, unsigned int gc_seq, gfp_t gfp); void nft_trans_gc_destroy(struct nft_trans_gc *trans); diff --git a/net/netfilter/nf_tables_api.c b/net/netfilter/nf_tables_api.c index a3f7c7ae55b8..b411abe9743b 100644 --- a/net/netfilter/nf_tables_api.c +++ b/net/netfilter/nf_tables_api.c @@ -9945,11 +9945,6 @@ static void nft_trans_gc_queue_work(struct nft_trans_gc *trans) schedule_work(&trans_gc_work); } -static int nft_trans_gc_space(struct nft_trans_gc *trans) -{ - return NFT_TRANS_GC_BATCHCOUNT - trans->count; -} - struct nft_trans_gc *nft_trans_gc_queue_async(struct nft_trans_gc *gc, unsigned int gc_seq, gfp_t gfp) { diff --git a/net/netfilter/nft_set_pipapo.c b/net/netfilter/nft_set_pipapo.c index f4d0bb6b377d..00a13c50ec8b 100644 --- a/net/netfilter/nft_set_pipapo.c +++ b/net/netfilter/nft_set_pipapo.c @@ -1615,11 +1615,11 @@ static void nft_pipapo_gc_deactivate(struct net *net, struct nft_set *set, } /** - * pipapo_gc() - Drop expired entries from set, destroy start and end elements + * pipapo_gc_scan() - Drop expired entries from set and link them to gc list * @set: nftables API set representation * @m: Matching data */ -static void pipapo_gc(struct nft_set *set, struct nft_pipapo_match *m) +static void pipapo_gc_scan(struct nft_set *set, struct nft_pipapo_match *m) { struct nft_pipapo *priv = nft_set_priv(set); struct net *net = read_pnet(&set->net); @@ -1632,6 +1632,8 @@ static void pipapo_gc(struct nft_set *set, struct nft_pipapo_match *m) if (!gc) return; + list_add(&gc->list, &priv->gc_head); + while ((rules_f0 = pipapo_rules_same_key(m->f, first_rule))) { union nft_pipapo_map_bucket rulemap[NFT_PIPAPO_MAX_FIELDS]; const struct nft_pipapo_field *f; @@ -1661,9 +1663,13 @@ static void pipapo_gc(struct nft_set *set, struct nft_pipapo_match *m) if (__nft_set_elem_expired(&e->ext, tstamp)) { priv->dirty = true; - gc = nft_trans_gc_queue_sync(gc, GFP_ATOMIC); - if (!gc) - return; + if (!nft_trans_gc_space(gc)) { + gc = nft_trans_gc_alloc(set, 0, GFP_KERNEL); + if (!gc) + return; + + list_add(&gc->list, &priv->gc_head); + } nft_pipapo_gc_deactivate(net, set, e); pipapo_drop(m, rulemap); @@ -1677,10 +1683,30 @@ static void pipapo_gc(struct nft_set *set, struct nft_pipapo_match *m) } } - gc = nft_trans_gc_catchall_sync(gc); + priv->last_gc = jiffies; +} + +/** + * pipapo_gc_queue() - Free expired elements + * @set: nftables API set representation + */ +static void pipapo_gc_queue(struct nft_set *set) +{ + struct nft_pipapo *priv = nft_set_priv(set); + struct nft_trans_gc *gc, *next; + + /* always do a catchall cycle: */ + gc = nft_trans_gc_alloc(set, 0, GFP_KERNEL); if (gc) { + gc = nft_trans_gc_catchall_sync(gc); + if (gc) + nft_trans_gc_queue_sync_done(gc); + } + + /* always purge queued gc elements. */ + list_for_each_entry_safe(gc, next, &priv->gc_head, list) { + list_del(&gc->list); nft_trans_gc_queue_sync_done(gc); - priv->last_gc = jiffies; } } @@ -1734,6 +1760,10 @@ static void pipapo_reclaim_match(struct rcu_head *rcu) * * We also need to create a new working copy for subsequent insertions and * deletions. + * + * After the live copy has been replaced by the clone, we can safely queue + * expired elements that have been collected by pipapo_gc_scan() for + * memory reclaim. */ static void nft_pipapo_commit(struct nft_set *set) { @@ -1741,7 +1771,7 @@ static void nft_pipapo_commit(struct nft_set *set) struct nft_pipapo_match *new_clone, *old; if (time_after_eq(jiffies, priv->last_gc + nft_set_gc_interval(set))) - pipapo_gc(set, priv->clone); + pipapo_gc_scan(set, priv->clone); if (!priv->dirty) return; @@ -1758,6 +1788,8 @@ static void nft_pipapo_commit(struct nft_set *set) call_rcu(&old->rcu, pipapo_reclaim_match); priv->clone = new_clone; + + pipapo_gc_queue(set); } static bool nft_pipapo_transaction_mutex_held(const struct nft_set *set) @@ -2229,6 +2261,7 @@ static int nft_pipapo_init(const struct nft_set *set, priv->dirty = false; + INIT_LIST_HEAD(&priv->gc_head); rcu_assign_pointer(priv->match, m); return 0; @@ -2281,6 +2314,8 @@ static void nft_pipapo_destroy(const struct nft_ctx *ctx, struct nft_pipapo_match *m; int cpu; + WARN_ON_ONCE(!list_empty(&priv->gc_head)); + m = rcu_dereference_protected(priv->match, true); if (m) { rcu_barrier(); diff --git a/net/netfilter/nft_set_pipapo.h b/net/netfilter/nft_set_pipapo.h index aad9130cc763..743ba7c153ff 100644 --- a/net/netfilter/nft_set_pipapo.h +++ b/net/netfilter/nft_set_pipapo.h @@ -165,6 +165,7 @@ struct nft_pipapo_match { * @width: Total bytes to be matched for one packet, including padding * @dirty: Working copy has pending insertions or deletions * @last_gc: Timestamp of last garbage collection run, jiffies + * @gc_head: list of nft_trans_gc to queue up for mem reclaim */ struct nft_pipapo { struct nft_pipapo_match __rcu *match; @@ -172,6 +173,7 @@ struct nft_pipapo { int width; bool dirty; unsigned long last_gc; + struct list_head gc_head; }; struct nft_pipapo_elem;