[RFC] [PATCH] netfilter: Fix kmemleak false positive reports

From: zhe.he
Date: Wed Oct 24 2018 - 13:32:48 EST

From: He Zhe <zhe.he@xxxxxxxxxxxxx>

unreferenced object 0xffff9643edb89900 (size 256):
comm "sd-resolve", pid 220, jiffies 4295016710 (age 208.256s)
hex dump (first 32 bytes):
01 00 00 00 00 00 00 00 03 00 74 f3 ba b1 b6 b5 ..........t.....
65 3e 00 00 00 00 00 00 90 f9 a0 ed 43 96 ff ff e>..........C...
[<0000000070d5b185>] kmem_cache_alloc+0x146/0x200
[<0000000007a27faa>] __nf_conntrack_alloc.isra.13+0x4d/0x170 [nf_conntrack]
[<00000000ecc5b0ec>] init_conntrack+0x6a/0x2f0 [nf_conntrack]
[<000000003d38809f>] nf_conntrack_in+0x2c5/0x360 [nf_conntrack]
[<000000001fe154e3>] ipv4_conntrack_local+0x5d/0x70 [nf_conntrack_ipv4]
[<0000000027adadb2>] nf_hook_slow+0x48/0xd0
[<000000009893511f>] __ip_local_out+0xbd/0xf0
[<00000000d68cbd2f>] ip_local_out+0x1c/0x50
[<00000000995e2f37>] ip_send_skb+0x19/0x40
[<000000003d95f220>] udp_send_skb.isra.5+0x157/0x360
[<00000000ebc25968>] udp_sendmsg+0x9d8/0xc10
[<000000003bef56ec>] inet_sendmsg+0x3e/0xf0
[<000000008d23e405>] sock_sendmsg+0x1d/0x30
[<000000008c297097>] ___sys_sendmsg+0x108/0x2b0
[<00000000f15a806c>] __sys_sendmmsg+0xba/0x1c0
[<00000000e195d2cf>] __x64_sys_sendmmsg+0x24/0x30

In __nf_conntrack_confirm, object ct can be referenced to by the stack variable
ct and the members of ct->tuplehash. kmemleak needs at least one of them to find
the ct object during scan.

When the ct object is moved from the unconfirmed hlist to the confirmed hlist.
kmemleak cannot see ct object if things happen in the following order and thus
give the above false positive report.
1) The ct object is removed from the unconfirmed hlist.
2) kmemleak scans data/bss sections(heap scan passes without heap reference).
3) The ct object is added to confirmed hlist and the variable ct is destroyed as
the function returns.
4) kmemleak scans task stacks(stack scan passes without stack reference).

This patch marks ct object as not a leak.

Signed-off-by: He Zhe <zhe.he@xxxxxxxxxxxxx>
Cc: pablo@xxxxxxxxxxxxx
Cc: kadlec@xxxxxxxxxxxxxxxxx
Cc: fw@xxxxxxxxx
Cc: davem@xxxxxxxxxxxxx
Cc: catalin.marinas@xxxxxxx
So far this is only observed in v4.18, not in v4.19. But the case seems apply
to both.

net/netfilter/nf_conntrack_core.c | 3 +++
1 file changed, 3 insertions(+)

diff --git a/net/netfilter/nf_conntrack_core.c b/net/netfilter/nf_conntrack_core.c
index a676d5f..067365d 100644
--- a/net/netfilter/nf_conntrack_core.c
+++ b/net/netfilter/nf_conntrack_core.c
@@ -35,6 +35,7 @@
#include <linux/mm.h>
#include <linux/nsproxy.h>
#include <linux/rculist_nulls.h>
+#include <linux/kmemleak.h>

#include <net/netfilter/nf_conntrack.h>
#include <net/netfilter/nf_conntrack_l4proto.h>
@@ -1282,6 +1283,8 @@ __nf_conntrack_alloc(struct net *net,
if (ct == NULL)
goto out;

+ kmemleak_not_leak(ct);
ct->tuplehash[IP_CT_DIR_ORIGINAL].tuple = *orig;
ct->tuplehash[IP_CT_DIR_ORIGINAL].hnnode.pprev = NULL;