Skip to contentSkip to conversation

A bunch of covscan detected fixes

PATCH Applied Phil Sutter 2 participants 16 messages Aug 11, 2016 – Aug 12, 2016 netfilter

lore
Share & tools
lore

Phil Sutter

author reviewer
#0 lore raw

cover A bunch of covscan detected fixes

The following series aims at fixing a number of issues identified by Coverity tool. Due to limited familiarity with the whole code layout, I am not sure all of them are really valid, but I tried my best to verify the concerns are legitimate and worth fixing.

Phil Sutter (7):
set: prevent memleak in nftnl_jansson_parse_set_info()
ruleset: Prevent memleak in nftnl_ruleset_snprintf_*() functions expr/ct: prevent array index overrun in ctkey2str() expr/limit: Drop unreachable code in limit_to_type()
common: Avoid integer overflow in nftnl_batch_is_supported() Avoid returning uninitialized data
ruleset: Initialize ctx.flags before calling nftnl_ruleset_ctx_set()

 src/chain.c      |  2 +-
 src/common.c     |  3 +++
 src/expr/ct.c    |  2 +-
 src/expr/limit.c |  1 -
 src/rule.c       |  2 +-
 src/ruleset.c    | 18 ++++++++++++++++++
 src/set.c        | 12 ++++++------
 src/table.c      |  2 +-
 8 files changed, 31 insertions(+), 11 deletions(-)
Signature · 1 line
2.8.2
author reviewer
#1 lore raw

PATCH 1/7 set: prevent memleak in nftnl_jansson_parse_set_info()

From: Phil Sutter <[email protected]>

During list populating, in error case the function returns without freeing the newly allocated 'elem' object, thereby losing any references to it.

---
 src/set.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
src/set.c +5 -5
index 5f7245855db53..7a41e8caeb3c0 100644
@@ -610,12 +610,12 @@ static int nftnl_jansson_parse_set_info(struct nftnl_set *s, json_t *tree,
 				return -1;
 
 			json_elem = json_array_get(array, i);
-			if (json_elem == NULL)
-				return -1;
-
-			if (nftnl_jansson_set_elem_parse(elem,
-						       json_elem, err) < 0)
+			if (json_elem == NULL ||
+			    nftnl_jansson_set_elem_parse(elem,
+							 json_elem, err) < 0) {
+				free(elem);
 				return -1;
+			}
 
 			list_add_tail(&elem->head, &s->element_list);
 		}
Signature · 1 line
2.8.2
author reviewer
#2 lore raw

PATCH 2/7 ruleset: Prevent memleak in nftnl_ruleset_snprintf_*() functions

In reply to Phil Sutter

From: Phil Sutter <[email protected]>

This is an ugly aspect of the SNPRINTF_BUFFER_SIZE() macro: it contains a return statement and if that triggers, the function returns without freeing the iterator object. Therefore duplicate the 'ret < 0' check before calling it, freeing the iterator knowing that we will bail out immediately afterwards anyway.

Cc Arturo Borrero <[email protected]>
---
 src/ruleset.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
src/ruleset.c +16 -0
index 666bcc7a246b6..93cf95ab61e15 100644
@@ -888,12 +888,16 @@ nftnl_ruleset_snprintf_table(char *buf, size_t size,
 	t = nftnl_table_list_iter_next(ti);
 	while (t != NULL) {
 		ret = nftnl_table_snprintf(buf+offset, len, t, type, flags);
+		if (ret < 0)
+			nftnl_table_list_iter_destroy(ti);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);
 
 		t = nftnl_table_list_iter_next(ti);
 
 		ret = snprintf(buf+offset, len, "%s",
 			       nftnl_ruleset_o_separator(t, type));
+		if (ret < 0)
+			nftnl_table_list_iter_destroy(ti);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);
 	}
 	nftnl_table_list_iter_destroy(ti);
@@ -917,12 +921,16 @@ nftnl_ruleset_snprintf_chain(char *buf, size_t size,
 	c = nftnl_chain_list_iter_next(ci);
 	while (c != NULL) {
 		ret = nftnl_chain_snprintf(buf+offset, len, c, type, flags);
+		if (ret < 0)
+			nftnl_chain_list_iter_destroy(ci);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);
 
 		c = nftnl_chain_list_iter_next(ci);
 
 		ret = snprintf(buf+offset, len, "%s",
 			       nftnl_ruleset_o_separator(c, type));
+		if (ret < 0)
+			nftnl_chain_list_iter_destroy(ci);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);
 	}
 	nftnl_chain_list_iter_destroy(ci);
@@ -946,12 +954,16 @@ nftnl_ruleset_snprintf_set(char *buf, size_t size,
 	s = nftnl_set_list_iter_next(si);
 	while (s != NULL) {
 		ret = nftnl_set_snprintf(buf+offset, len, s, type, flags);
+		if (ret < 0)
+			nftnl_set_list_iter_destroy(si);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);
 
 		s = nftnl_set_list_iter_next(si);
 
 		ret = snprintf(buf+offset, len, "%s",
 			       nftnl_ruleset_o_separator(s, type));
+		if (ret < 0)
+			nftnl_set_list_iter_destroy(si);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);
 	}
 	nftnl_set_list_iter_destroy(si);
@@ -975,12 +987,16 @@ nftnl_ruleset_snprintf_rule(char *buf, size_t size,
 	r = nftnl_rule_list_iter_next(ri);
 	while (r != NULL) {
 		ret = nftnl_rule_snprintf(buf+offset, len, r, type, flags);
+		if (ret < 0)
+			nftnl_rule_list_iter_destroy(ri);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);
 
 		r = nftnl_rule_list_iter_next(ri);
 
 		ret = snprintf(buf+offset, len, "%s",
 			       nftnl_ruleset_o_separator(r, type));
+		if (ret < 0)
+			nftnl_rule_list_iter_destroy(ri);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);
 	}
 	nftnl_rule_list_iter_destroy(ri);
Signature · 1 line
2.8.2
#3 lore raw

Re: [libnftnl PATCH 2/7] ruleset: Prevent memleak in nftnl_ruleset_snprintf_*() functions

Quoted 19 lines · Phil Sutter

From: Phil Sutter <[email protected]>

This is an ugly aspect of the SNPRINTF_BUFFER_SIZE() macro: it contains a return statement and if that triggers, the function returns without freeing the iterator object. Therefore duplicate the 'ret < 0' check before calling it, freeing the iterator knowing that we will bail out immediately afterwards anyway.

Cc Arturo Borrero <[email protected]>
---
 src/ruleset.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
src/ruleset.c +0 -0
index 666bcc7a246b6..93cf95ab61e15 100644

src/ruleset.c nftnl_ruleset_snprintf_table(char *buf, size_t size, · line 893

Quoted 6 lines
src/ruleset.c +2 -0
@@ -888,12 +888,16 @@ nftnl_ruleset_snprintf_table(char *buf, size_t size,
 	t = nftnl_table_list_iter_next(ti);
 	while (t != NULL) {
 		ret = nftnl_table_snprintf(buf+offset, len, t, type, flags);
+		if (ret < 0)
+			nftnl_table_list_iter_destroy(ti);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);

Better get rid of the obscure if (ret < 0) hidden in SNPRINT_BUFFER_SIZE.

Or simply set:

        if (ret < 0)
                ret = 0;

in SNPRINT_BUFFER_SIZE.

author reviewer
#4 lore raw

Re: [libnftnl PATCH 2/7] ruleset: Prevent memleak in nftnl_ruleset_snprintf_*() functions

Quoted 30 lines · Pablo Neira Ayuso
Quoted 19 lines · Phil Sutter

From: Phil Sutter <[email protected]>

This is an ugly aspect of the SNPRINTF_BUFFER_SIZE() macro: it contains a return statement and if that triggers, the function returns without freeing the iterator object. Therefore duplicate the 'ret < 0' check before calling it, freeing the iterator knowing that we will bail out immediately afterwards anyway.

Cc Arturo Borrero <[email protected]>
---
 src/ruleset.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
src/ruleset.c +0 -0
index 666bcc7a246b6..93cf95ab61e15 100644

src/ruleset.c nftnl_ruleset_snprintf_table(char *buf, size_t size, · line 893

Quoted 6 lines
src/ruleset.c +2 -0
@@ -888,12 +888,16 @@ nftnl_ruleset_snprintf_table(char *buf, size_t size,
 	t = nftnl_table_list_iter_next(ti);
 	while (t != NULL) {
 		ret = nftnl_table_snprintf(buf+offset, len, t, type, flags);
+		if (ret < 0)
+			nftnl_table_list_iter_destroy(ti);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);

Better get rid of the obscure if (ret < 0) hidden in SNPRINT_BUFFER_SIZE.

Quoted 6 lines

Or simply set:

        if (ret < 0)
                ret = 0;

in SNPRINT_BUFFER_SIZE.

Hmm. This means we will lose error propagation. Given how widely this macro is being used (grep says 200 calls), this needs a good second thought.

Thanks, Phil

#5 lore raw

Re: [libnftnl PATCH 2/7] ruleset: Prevent memleak in nftnl_ruleset_snprintf_*() functions

Quoted 33 lines · Phil Sutter
Quoted 32 lines · Pablo Neira Ayuso
Quoted 19 lines · Phil Sutter

From: Phil Sutter <[email protected]>

This is an ugly aspect of the SNPRINTF_BUFFER_SIZE() macro: it contains a return statement and if that triggers, the function returns without freeing the iterator object. Therefore duplicate the 'ret < 0' check before calling it, freeing the iterator knowing that we will bail out immediately afterwards anyway.

Cc Arturo Borrero <[email protected]>
---
 src/ruleset.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
src/ruleset.c +0 -0
index 666bcc7a246b6..93cf95ab61e15 100644

src/ruleset.c nftnl_ruleset_snprintf_table(char *buf, size_t size, · line 893

Quoted 6 lines
src/ruleset.c +2 -0
@@ -888,12 +888,16 @@ nftnl_ruleset_snprintf_table(char *buf, size_t size,
 	t = nftnl_table_list_iter_next(ti);
 	while (t != NULL) {
 		ret = nftnl_table_snprintf(buf+offset, len, t, type, flags);
+		if (ret < 0)
+			nftnl_table_list_iter_destroy(ti);
 		SNPRINTF_BUFFER_SIZE(ret, size, len, offset);

Better get rid of the obscure if (ret < 0) hidden in SNPRINT_BUFFER_SIZE.

Or simply set:

Quoted 8 lines
Quoted 4 lines
        if (ret < 0)
                ret = 0;

in SNPRINT_BUFFER_SIZE.

Hmm. This means we will lose error propagation. Given how widely this macro is being used (grep says 200 calls), this needs a good second thought.

The usual suggested idiom to deal with this looks like:

    snprintf(cp, blen, "AF=%d ", sau->soa.sa_family)
    blen -= strlen(cp);
    cp += strlen(cp);

See: https://goo.gl/AUGE5m

No need to second thought, we can just ignore the -1 case.

author reviewer
#6 lore raw

PATCH 3/7 expr/ct: prevent array index overrun in ctkey2str()

In reply to Phil Sutter

The array has NFT_CT_MAX fields, so indices must be less than that number.

Fixes 977b7a1dbe1bd ("ct: xml: use key names instead of numbers")
Cc Arturo Borrero Gonzalez <[email protected]>
---
 src/expr/ct.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
src/expr/ct.c +1 -1
index 7d96df4e1d5b0..1a53b49fde0ad 100644
@@ -173,7 +173,7 @@ static const char *ctkey2str_array[NFT_CT_MAX] = {
 
 static const char *ctkey2str(uint32_t ctkey)
 {
-	if (ctkey > NFT_CT_MAX)
+	if (ctkey >= NFT_CT_MAX)
 		return "unknown";
 
 	return ctkey2str_array[ctkey];
Signature · 1 line
2.8.2
author reviewer
#7 lore raw

PATCH 4/7 expr/limit: Drop unreachable code in limit_to_type()

In reply to Phil Sutter

The function returns from inside the switch() in any case, so the final return statement is never reached.

Fixes 7769cbd9dfe69 ("expr: limit: add per-byte limiting support")
---
 src/expr/limit.c | 1 -
 1 file changed, 1 deletion(-)
src/expr/limit.c +0 -1
index 4bd096e8e921c..cdff81d620caa 100644
@@ -259,7 +259,6 @@ static const char *limit_to_type(enum nft_limit_type type)
 	case NFT_LIMIT_PKT_BYTES:
 		return "bytes";
 	}
-	return "unknown";
 }
 
 static int nftnl_expr_limit_snprintf_default(char *buf, size_t len,
Signature · 1 line
2.8.2
author reviewer
#8 lore raw

PATCH 5/7 common: Avoid integer overflow in nftnl_batch_is_supported()

In reply to Phil Sutter

time() may return -1 which is then assigned to an unsigned integer type and used as sequence number. The following code increments that number multiple times, so it may overflow and get libmnl confused. To avoid this, fall back to a starting sequence number of zero in case the call to time() failed.

---
 src/common.c | 3 +++
 1 file changed, 3 insertions(+)
src/common.c +3 -0
index bf4176ceb76eb..2189cc8a3e882 100644
@@ -192,6 +192,9 @@ int nftnl_batch_is_supported(void)
 	uint32_t seq = time(NULL), req_seq;
 	int ret;
 
+	if (seq == (uint32_t)-1)
+		seq = 0;
+
 	nl = mnl_socket_open(NETLINK_NETFILTER);
 	if (nl == NULL)
 		return -1;
Signature · 1 line
2.8.2
#9 lore raw

Re: [libnftnl PATCH 5/7] common: Avoid integer overflow in nftnl_batch_is_supported()

Quoted 5 lines · Phil Sutter

time() may return -1 which is then assigned to an unsigned integer type and used as sequence number. The following code increments that number multiple times, so it may overflow and get libmnl confused. To avoid this, fall back to a starting sequence number of zero in case the call to time() failed.

(uint32_t)-1 should be fine for netlink as sequence number.

author reviewer
#10 lore raw

PATCH 6/7 Avoid returning uninitialized data

In reply to Phil Sutter

Although the 'err' pointer should be interesting for users only if the parser returned non-zero, having it point to uninitialized data is generally a bad thing.

---
 src/chain.c | 2 +-
 src/rule.c  | 2 +-
 src/set.c   | 2 +-
 src/table.c | 2 +-
 4 files changed, 4 insertions(+), 4 deletions(-)
src/chain.c +1 -1
index ff59f7f705783..dfe0cf23b0f51 100644
@@ -816,7 +816,7 @@ static int nftnl_chain_do_parse(struct nftnl_chain *c, enum nftnl_parse_type typ
 			      enum nftnl_parse_input input)
 {
 	int ret;
-	struct nftnl_parse_err perr;
+	struct nftnl_parse_err perr = {};
 
 	switch (type) {
 	case NFTNL_PARSE_XML:
src/rule.c +1 -1
index dada00e1cc99f..8aeefbe19a9ed 100644
@@ -700,7 +700,7 @@ static int nftnl_rule_do_parse(struct nftnl_rule *r, enum nftnl_parse_type type,
 			     enum nftnl_parse_input input)
 {
 	int ret;
-	struct nftnl_parse_err perr;
+	struct nftnl_parse_err perr = {};
 
 	switch (type) {
 	case NFTNL_PARSE_XML:
src/set.c +1 -1
index 7a41e8caeb3c0..47b6ef46a9c34 100644
@@ -781,7 +781,7 @@ static int nftnl_set_do_parse(struct nftnl_set *s, enum nftnl_parse_type type,
 			    enum nftnl_parse_input input)
 {
 	int ret;
-	struct nftnl_parse_err perr;
+	struct nftnl_parse_err perr = {};
 
 	switch (type) {
 	case NFTNL_PARSE_XML:
src/table.c +1 -1
index bb46716950b98..845fd68168797 100644
@@ -360,7 +360,7 @@ static int nftnl_table_do_parse(struct nftnl_table *t, enum nftnl_parse_type typ
 			      enum nftnl_parse_input input)
 {
 	int ret;
-	struct nftnl_parse_err perr;
+	struct nftnl_parse_err perr = {};
 
 	switch (type) {
 	case NFTNL_PARSE_XML:
Signature · 1 line
2.8.2
author reviewer
#11 lore raw

PATCH 7/7 ruleset: Initialize ctx.flags before calling nftnl_ruleset_ctx_set()

In reply to Phil Sutter

The called function otherwise accesses uninitialized data.

---
 src/ruleset.c | 2 ++
 1 file changed, 2 insertions(+)
src/ruleset.c +2 -0
index 93cf95ab61e15..a2d67cb550179 100644
@@ -551,6 +551,7 @@ static int nftnl_ruleset_json_parse(const void *json,
 
 	ctx.cb = cb;
 	ctx.format = type;
+	ctx.flags = 0;
 
 	ctx.set_list = nftnl_set_list_alloc();
 	if (ctx.set_list == NULL)
@@ -682,6 +683,7 @@ static int nftnl_ruleset_xml_parse(const void *xml, struct nftnl_parse_err *err,
 
 	ctx.cb = cb;
 	ctx.format = type;
+	ctx.flags = 0;
 
 	ctx.set_list = nftnl_set_list_alloc();
 	if (ctx.set_list == NULL)
Signature · 1 line
2.8.2
maintainer Applied
#12 lore raw
In reply to Phil Sutter
Quoted 4 lines · Phil Sutter

The following series aims at fixing a number of issues identified by Coverity tool. Due to limited familiarity with the whole code layout, I am not sure all of them are really valid, but I tried my best to verify the concerns are legitimate and worth fixing.

Applied, 1/7, 3/7, 4/7, 5/7, 7/7.

Kept back 2/7 and 6/7.

I think 2/7 you can send a new version.

Thanks Phil.

#13 lore raw
Quoted 9 lines · Pablo Neira Ayuso
Quoted 4 lines · Phil Sutter

The following series aims at fixing a number of issues identified by Coverity tool. Due to limited familiarity with the whole code layout, I am not sure all of them are really valid, but I tried my best to verify the concerns are legitimate and worth fixing.

Applied, 1/7, 3/7, 4/7, 5/7, 7/7.

Kept back 2/7 and 6/7.

Sorry, I mean I kept back 5/7.

Quoted 3 lines

I think 2/7 you can send a new version.

Thanks Phil.

author reviewer
#14 lore raw
Quoted 7 lines · Pablo Neira Ayuso
Quoted 6 lines · Pablo Neira Ayuso
Quoted 4 lines · Phil Sutter

The following series aims at fixing a number of issues identified by Coverity tool. Due to limited familiarity with the whole code layout, I am not sure all of them are really valid, but I tried my best to verify the concerns are legitimate and worth fixing.

Quoted 5 lines
Quoted 3 lines

Applied, 1/7, 3/7, 4/7, 5/7, 7/7.

Kept back 2/7 and 6/7.

Sorry, I mean I kept back 5/7.

Actually, you didn't. Patch 5/7 slipped through (d26feca2c9c19). Feel free to revert if it's useless.

Thanks, Phil

#15 lore raw
Quoted 8 lines · Phil Sutter
Quoted 7 lines · Pablo Neira Ayuso
Quoted 6 lines · Pablo Neira Ayuso
Quoted 4 lines · Phil Sutter

The following series aims at fixing a number of issues identified by Coverity tool. Due to limited familiarity with the whole code layout, I am not sure all of them are really valid, but I tried my best to verify the concerns are legitimate and worth fixing.

Quoted 8 lines
Quoted 5 lines
Quoted 3 lines

Applied, 1/7, 3/7, 4/7, 5/7, 7/7.

Kept back 2/7 and 6/7.

Sorry, I mean I kept back 5/7.

Actually, you didn't. Patch 5/7 slipped through (d26feca2c9c19). Feel free to revert if it's useless.

Done.