Skip to contentSkip to conversation

assorted fixes

PATCH In review Pablo Neira Ayuso 2 participants 7 messages Jan 10, 2024 – Jan 11, 2024 netfilter

lore
Share & tools
lore

Pablo Neira Ayuso

author maintainer
#0 lore raw

cover assorted fixes

Hi Florian,

This in an alternative path to address the issue described here:

https://patchwork.ozlabs.org/project/netfilter-devel/patch/[email protected]/

(I am partially integrating this patch into 3/4 in this series).

Patch #1 fixes a bug in the set optimization with single elements, which
results in strange ruleset listings. Concatenations are only supported with sets, so let's just skip this.

Patch #2 do not fetch next key in case runaway flag is set on with
concatenations. I could not crash nftables with this, but I found it when reviewing this code.

Patch #3 iterate over the anonymous set in set_evaluate() to validate
consistency of elements as concat expressions. Otherwise, bail out with:

ruleset.nft:3:46-53: Error: expression is not a concatenation

               ip protocol . th dport vmap { tcp / 22 : accept, tcp . 80 : drop}
                                             ^^^^^^^^

         I extended tests to cover maps too.

I needed special error handling when set_evaluate() fails to release sets so ASAN does not complain with incorrect memory handling, this chunk:

@@ -118,7 +119,15 @@ static struct expr *implicit_set_declaration(struct eval_ctx *ctx,

                list_add_tail(&cmd->list, &ctx->cmd->list);
        }
-       set_evaluate(ctx, set);
+       err = set_evaluate(ctx, set);
+       if (err < 0) {

+ list_del(&cmd->list);
+ if (set->flags & NFT_SET_MAP)
+ cmd->set->init = NULL;
+
+ cmd_free(cmd);
+ return NULL;
+ }

        return set_ref_expr_alloc(&expr->location, set);

}

Patch #4 revert a recent late sanity check which is already covered by this
patchset.

In general the idea is to make stricter validations from evaluation phase to avoid propagating errors any further.

This passing tests/shell and tests/py, it might be a good idea to add more bogons tests for concatenations with imbalanced number of components / runaway number of components, I will follow up with a patch to extend tests infrastructure with this.

Pablo Neira Ayuso (4):
evaluate: skip anonymous set optimization for concatenations
evaluate: do not fetch next expression on runaway number of concatenation components
evaluate: bail out if anonymous concat set defines a non concat expression Revert "datatype: do not assert when value exceeds expected width"

 src/datatype.c                                |  6 +-
 src/evaluate.c                                | 59 +++++++++++++++----
 .../bogons/nft-f/unhandled_key_type_13_assert |  5 ++
 .../nft-f/unhandled_key_type_13_assert_map    |  5 ++
 .../nft-f/unhandled_key_type_13_assert_vmap   |  5 ++
 5 files changed, 64 insertions(+), 16 deletions(-)
 create mode 100644 tests/shell/testcases/bogons/nft-f/unhandled_key_type_13_assert
 create mode 100644 tests/shell/testcases/bogons/nft-f/unhandled_key_type_13_assert_map
 create mode 100644 tests/shell/testcases/bogons/nft-f/unhandled_key_type_13_assert_vmap

--
2.30.2

author maintainer
#1 lore raw

PATCH 1/4 evaluate: skip anonymous set optimization for concatenations

Concatenation is only supported with sets. Moreover, stripping of the set leads to broken ruleset listing, therefore, skip this optimization for the concatenations.

Fixes fa17b17ea74a ("evaluate: revisit anonymous set with single element optimization")
---
 src/evaluate.c | 20 +++++++++++---------
 1 file changed, 11 insertions(+), 9 deletions(-)
src/evaluate.c +11 -9
index 41524eef12b7..6405d55647fa 100644
@@ -2548,15 +2548,17 @@ static int expr_evaluate_relational(struct eval_ctx *ctx, struct expr **expr)
 		return expr_binary_error(ctx->msgs, right, left,
 					 "Cannot be used with right hand side constant value");
 
-	switch (rel->op) {
-	case OP_EQ:
-	case OP_IMPLICIT:
-	case OP_NEQ:
-		if (right->etype == EXPR_SET && right->size == 1)
-			optimize_singleton_set(rel, &right);
-		break;
-	default:
-		break;
+	if (left->etype != EXPR_CONCAT) {
+		switch (rel->op) {
+		case OP_EQ:
+		case OP_IMPLICIT:
+		case OP_NEQ:
+			if (right->etype == EXPR_SET && right->size == 1)
+				optimize_singleton_set(rel, &right);
+			break;
+		default:
+			break;
+		}
 	}
 
 	switch (rel->op) {
Signature · 1 line
2.30.2
author maintainer
#2 lore raw

PATCH 2/4 evaluate: do not fetch next expression on runaway number of concatenation components

In reply to Pablo Neira Ayuso

If this is the last expression, then the runaway flag is set on and evaluation bails in the next iteration, do not fetch next list element which refers to the list head.

I found this by code inspection, I could not trigger any crash with this one.

Fixes ae1d54d1343f ("evaluate: do not crash on runaway number of concatenation components")
---
 src/evaluate.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
src/evaluate.c +2 -2
index 6405d55647fa..8ef1b5e39bdc 100644
@@ -1621,8 +1621,8 @@ static int expr_evaluate_concat(struct eval_ctx *ctx, struct expr **expr)
 		if (key && expressions) {
 			if (list_is_last(&key->list, expressions))
 				runaway = true;
-
-			key = list_next_entry(key, list);
+			else
+				key = list_next_entry(key, list);
 		}
 
 		ctx->inner_desc = NULL;
Signature · 1 line
2.30.2
author maintainer
#3 lore raw

PATCH 3/4 evaluate: bail out if anonymous concat set defines a non concat expression

In reply to Pablo Neira Ayuso

Iterate over the element list in the anonymous set to validate that all expressions are concatenations, otherwise bail out.

ruleset.nft:3:46-53: Error: expression is not a concatenation

               ip protocol . th dport vmap { tcp / 22 : accept, tcp . 80 : drop}
                                             ^^^^^^^^

This is based on a patch from Florian Westphal.

---
 src/evaluate.c                                | 35 ++++++++++++++++++-
 .../bogons/nft-f/unhandled_key_type_13_assert |  5 +++
 .../nft-f/unhandled_key_type_13_assert_map    |  5 +++
 .../nft-f/unhandled_key_type_13_assert_vmap   |  5 +++
 4 files changed, 49 insertions(+), 1 deletion(-)
 create mode 100644 tests/shell/testcases/bogons/nft-f/unhandled_key_type_13_assert
 create mode 100644 tests/shell/testcases/bogons/nft-f/unhandled_key_type_13_assert_map
 create mode 100644 tests/shell/testcases/bogons/nft-f/unhandled_key_type_13_assert_vmap
src/evaluate.c +34 -1
index 8ef1b5e39bdc..b79268e82e4f 100644
@@ -94,6 +94,7 @@ static struct expr *implicit_set_declaration(struct eval_ctx *ctx,
 	struct cmd *cmd;
 	struct set *set;
 	struct handle h;
+	int err;
 
 	if (set_is_datamap(expr->set_flags))
 		key_fix_dtype_byteorder(key);
@@ -118,7 +119,15 @@ static struct expr *implicit_set_declaration(struct eval_ctx *ctx,
 		list_add_tail(&cmd->list, &ctx->cmd->list);
 	}
 
-	set_evaluate(ctx, set);
+	err = set_evaluate(ctx, set);
+	if (err < 0) {
+		list_del(&cmd->list);
+		if (set->flags & NFT_SET_MAP)
+			cmd->set->init = NULL;
+
+		cmd_free(cmd);
+		return NULL;
+	}
 
 	return set_ref_expr_alloc(&expr->location, set);
 }
@@ -2038,6 +2047,8 @@ static int expr_evaluate_map(struct eval_ctx *ctx, struct expr **expr)
 		mappings = implicit_set_declaration(ctx, "__map%d",
 						    key, data,
 						    mappings);
+		if (!mappings)
+			return -1;
 
 		if (ectx.len && mappings->set->data->len != ectx.len)
 			BUG("%d vs %d\n", mappings->set->data->len, ectx.len);
@@ -2609,6 +2620,9 @@ static int expr_evaluate_relational(struct eval_ctx *ctx, struct expr **expr)
 				implicit_set_declaration(ctx, "__set%d",
 							 expr_get(left), NULL,
 							 right);
+			if (!right)
+				return -1;
+
 			/* fall through */
 		case EXPR_SET_REF:
 			if (rel->left->etype == EXPR_CT &&
@@ -3253,6 +3267,8 @@ static int stmt_evaluate_meter(struct eval_ctx *ctx, struct stmt *stmt)
 
 	setref = implicit_set_declaration(ctx, stmt->meter.name,
 					  expr_get(key), NULL, set);
+	if (!setref)
+		return -1;
 
 	setref->set->desc.size = stmt->meter.size;
 	stmt->meter.set = setref;
@@ -4532,6 +4548,8 @@ static int stmt_evaluate_objref_map(struct eval_ctx *ctx, struct stmt *stmt)
 
 		mappings = implicit_set_declaration(ctx, "__objmap%d",
 						    key, NULL, mappings);
+		if (!mappings)
+			return -1;
 		mappings->set->objtype  = stmt->objref.type;
 
 		map->mappings = mappings;
@@ -4865,6 +4883,21 @@ static int set_evaluate(struct eval_ctx *ctx, struct set *set)
 		set->flags |= NFT_SET_CONCAT;
 	}
 
+	if (set_is_anonymous(set->flags) && set->key->etype == EXPR_CONCAT) {
+		struct expr *i;
+
+		list_for_each_entry(i, &set->init->expressions, list) {
+			if ((i->etype == EXPR_SET_ELEM &&
+			     i->key->etype != EXPR_CONCAT &&
+			     i->key->etype != EXPR_SET_ELEM_CATCHALL) ||
+			    (i->etype == EXPR_MAPPING &&
+			     i->left->etype == EXPR_SET_ELEM &&
+			     i->left->key->etype != EXPR_CONCAT &&
+			     i->left->key->etype != EXPR_SET_ELEM_CATCHALL))
+				return expr_error(ctx->msgs, i, "expression is not a concatenation");
+		}
+	}
+
 	if (set_is_datamap(set->flags)) {
 		if (set->data == NULL)
 			return set_error(ctx, set, "map definition does not "
tests/shell/testcases/bogons/nft-f/unhandled_key_type_13_assert new file +5 -0
new file mode 100644
index 000000000000..35eecf607230
@@ -0,0 +1,5 @@
+table ip x {
+	chain y {
+		ip protocol . th dport { tcp / 22, udp . 67 }
+	}
+}
tests/shell/testcases/bogons/nft-f/unhandled_key_type_13_assert_map new file +5 -0
new file mode 100644
index 000000000000..3da16ce15886
@@ -0,0 +1,5 @@
+table ip x {
+	chain y {
+		meta mark set ip protocol . th dport map { tcp / 22 : 1234, udp . 67 : 1234 }
+	}
+}
tests/shell/testcases/bogons/nft-f/unhandled_key_type_13_assert_vmap new file +5 -0
new file mode 100644
index 000000000000..f4dc273fdb85
@@ -0,0 +1,5 @@
+table ip x {
+	chain y {
+		ip protocol . th dport vmap { tcp / 22 : accept, udp . 67 : drop }
+	}
+}
Signature · 1 line
2.30.2
author maintainer
#4 lore raw

PATCH 4/4 Revert "datatype: do not assert when value exceeds expected width"

In reply to Pablo Neira Ayuso
 # nft -f ruleset.nft
 ruleset.nft:3:28-35: Error: expression is not a concatenation
                ip protocol . th dport { tcp / 22,  }
                                         ^^^^^^^^

Therefore, a852022d719e ("datatype: do not assert when value exceeds expected width") not needed anymore after two previous fixes.

---
 src/datatype.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
src/datatype.c +2 -4
index 099e7580bd6c..3b19ae8ef52d 100644
@@ -715,8 +715,7 @@ const struct datatype ip6addr_type = {
 static void inet_protocol_type_print(const struct expr *expr,
 				      struct output_ctx *octx)
 {
-	if (!nft_output_numeric_proto(octx) &&
-	    mpz_cmp_ui(expr->value, UINT8_MAX) <= 0) {
+	if (!nft_output_numeric_proto(octx)) {
 		char name[NFT_PROTONAME_MAXSIZE];
 
 		if (nft_getprotobynumber(mpz_get_uint8(expr->value), name, sizeof(name))) {
@@ -797,8 +796,7 @@ static void inet_service_print(const struct expr *expr, struct output_ctx *octx)
 
 void inet_service_type_print(const struct expr *expr, struct output_ctx *octx)
 {
-	if (nft_output_service(octx) &&
-	    mpz_cmp_ui(expr->value, UINT16_MAX) <= 0) {
+	if (nft_output_service(octx)) {
 		inet_service_print(expr, octx);
 		return;
 	}
Signature · 1 line
2.30.2
maintainer
#5 lore raw

Re: [PATCH nft 4/4] Revert "datatype: do not assert when value exceeds expected width"

Quoted 7 lines · Pablo Neira Ayuso
 # nft -f ruleset.nft
 ruleset.nft:3:28-35: Error: expression is not a concatenation
                ip protocol . th dport { tcp / 22,  }
                                         ^^^^^^^^

Therefore, a852022d719e ("datatype: do not assert when value exceeds expected width") not needed anymore after two previous fixes.

We can't rely on the expression soup coming from nftables.

author maintainer
#6 lore raw

Re: [PATCH nft 4/4] Revert "datatype: do not assert when value exceeds expected width"

Quoted 10 lines · Florian Westphal
Quoted 7 lines · Pablo Neira Ayuso
 # nft -f ruleset.nft
 ruleset.nft:3:28-35: Error: expression is not a concatenation
                ip protocol . th dport { tcp / 22,  }
                                         ^^^^^^^^

Therefore, a852022d719e ("datatype: do not assert when value exceeds expected width") not needed anymore after two previous fixes.

We can't rely on the expression soup coming from nftables.

I can keep this patch then, no problem.

Or you mean this series is botched? Please elaborate :)