| Message ID | 20200430121845.10388-1-pablo@netfilter.org |
|---|---|
| State | Accepted |
| Delegated to: | Pablo Neira |
| Headers | show |
| Series | [nft] rule: memleak in __do_add_setelems() | expand |
On Thu, Apr 30, 2020 at 02:18:45PM +0200, Pablo Neira Ayuso wrote: > This patch invokes interval_map_decompose() with named sets: > > ==3402== 2,352 (128 direct, 2,224 indirect) bytes in 1 blocks are definitely lost in loss record 9 of 9 > ==3402== at 0x483577F: malloc (vg_replace_malloc.c:299) > ==3402== by 0x48996A8: xmalloc (utils.c:36) > ==3402== by 0x4899778: xzalloc (utils.c:65) > ==3402== by 0x487CB46: expr_alloc (expression.c:45) > ==3402== by 0x487E2A0: mapping_expr_alloc (expression.c:1140) > ==3402== by 0x4898AA8: interval_map_decompose (segtree.c:1095) > ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1569) > ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1559) > ==3402== by 0x4877936: do_command (rule.c:2710) > ==3402== by 0x489F1CB: nft_netlink.isra.5 (libnftables.c:42) > ==3402== by 0x489FB07: nft_run_cmd_from_filename (libnftables.c:508) > ==3402== by 0x10A9AA: main (main.c:455) > > Fixes: dd44081d91ce ("segtree: Fix add and delete of element in same batch") This fixes the problem for anonymous sets, still named sets are showing a memleak.
Hi Pablo, On Thu, Apr 30, 2020 at 03:08:20PM +0200, Pablo Neira Ayuso wrote: > On Thu, Apr 30, 2020 at 02:18:45PM +0200, Pablo Neira Ayuso wrote: > > This patch invokes interval_map_decompose() with named sets: > > > > ==3402== 2,352 (128 direct, 2,224 indirect) bytes in 1 blocks are definitely lost in loss record 9 of 9 > > ==3402== at 0x483577F: malloc (vg_replace_malloc.c:299) > > ==3402== by 0x48996A8: xmalloc (utils.c:36) > > ==3402== by 0x4899778: xzalloc (utils.c:65) > > ==3402== by 0x487CB46: expr_alloc (expression.c:45) > > ==3402== by 0x487E2A0: mapping_expr_alloc (expression.c:1140) > > ==3402== by 0x4898AA8: interval_map_decompose (segtree.c:1095) > > ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1569) > > ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1559) > > ==3402== by 0x4877936: do_command (rule.c:2710) > > ==3402== by 0x489F1CB: nft_netlink.isra.5 (libnftables.c:42) > > ==3402== by 0x489FB07: nft_run_cmd_from_filename (libnftables.c:508) > > ==3402== by 0x10A9AA: main (main.c:455) > > > > Fixes: dd44081d91ce ("segtree: Fix add and delete of element in same batch") > > This fixes the problem for anonymous sets, still named sets are > showing a memleak. The change is strange: My fix (dd44081d91ce) was about anonymous sets. Since you make the added code apply to non-anonymous sets only, I would expect for my testcase to start failing again (I didn't test it, though). Are we maybe missing a free() somewhere instead? Cheers, Phil
On Thu, Apr 30, 2020 at 03:52:17PM +0200, Phil Sutter wrote: > Hi Pablo, > > On Thu, Apr 30, 2020 at 03:08:20PM +0200, Pablo Neira Ayuso wrote: > > On Thu, Apr 30, 2020 at 02:18:45PM +0200, Pablo Neira Ayuso wrote: > > > This patch invokes interval_map_decompose() with named sets: > > > > > > ==3402== 2,352 (128 direct, 2,224 indirect) bytes in 1 blocks are definitely lost in loss record 9 of 9 > > > ==3402== at 0x483577F: malloc (vg_replace_malloc.c:299) > > > ==3402== by 0x48996A8: xmalloc (utils.c:36) > > > ==3402== by 0x4899778: xzalloc (utils.c:65) > > > ==3402== by 0x487CB46: expr_alloc (expression.c:45) > > > ==3402== by 0x487E2A0: mapping_expr_alloc (expression.c:1140) > > > ==3402== by 0x4898AA8: interval_map_decompose (segtree.c:1095) > > > ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1569) > > > ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1559) > > > ==3402== by 0x4877936: do_command (rule.c:2710) > > > ==3402== by 0x489F1CB: nft_netlink.isra.5 (libnftables.c:42) > > > ==3402== by 0x489FB07: nft_run_cmd_from_filename (libnftables.c:508) > > > ==3402== by 0x10A9AA: main (main.c:455) > > > > > > Fixes: dd44081d91ce ("segtree: Fix add and delete of element in same batch") > > > > This fixes the problem for anonymous sets, still named sets are > > showing a memleak. > > The change is strange: My fix (dd44081d91ce) was about anonymous sets. It was about named sets, right? # nft 'add element t s { 22-25 }; delete element t s { 22-25 }' I think the cache update is still not needed for anonymous sets, even if this was not the right fix indeed. > Since you make the added code apply to non-anonymous sets only, I would > expect for my testcase to start failing again (I didn't test it, > though). > > Are we maybe missing a free() somewhere instead? I think I found the root cause: https://marc.info/?l=netfilter-devel&m=158825784609307&w=2
On Thu, Apr 30, 2020 at 04:47:14PM +0200, Pablo Neira Ayuso wrote: > On Thu, Apr 30, 2020 at 03:52:17PM +0200, Phil Sutter wrote: > > Hi Pablo, > > > > On Thu, Apr 30, 2020 at 03:08:20PM +0200, Pablo Neira Ayuso wrote: > > > On Thu, Apr 30, 2020 at 02:18:45PM +0200, Pablo Neira Ayuso wrote: > > > > This patch invokes interval_map_decompose() with named sets: > > > > > > > > ==3402== 2,352 (128 direct, 2,224 indirect) bytes in 1 blocks are definitely lost in loss record 9 of 9 > > > > ==3402== at 0x483577F: malloc (vg_replace_malloc.c:299) > > > > ==3402== by 0x48996A8: xmalloc (utils.c:36) > > > > ==3402== by 0x4899778: xzalloc (utils.c:65) > > > > ==3402== by 0x487CB46: expr_alloc (expression.c:45) > > > > ==3402== by 0x487E2A0: mapping_expr_alloc (expression.c:1140) > > > > ==3402== by 0x4898AA8: interval_map_decompose (segtree.c:1095) > > > > ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1569) > > > > ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1559) > > > > ==3402== by 0x4877936: do_command (rule.c:2710) > > > > ==3402== by 0x489F1CB: nft_netlink.isra.5 (libnftables.c:42) > > > > ==3402== by 0x489FB07: nft_run_cmd_from_filename (libnftables.c:508) > > > > ==3402== by 0x10A9AA: main (main.c:455) > > > > > > > > Fixes: dd44081d91ce ("segtree: Fix add and delete of element in same batch") > > > > > > This fixes the problem for anonymous sets, still named sets are > > > showing a memleak. > > > > The change is strange: My fix (dd44081d91ce) was about anonymous sets. > > It was about named sets, right? > > # nft 'add element t s { 22-25 }; delete element t s { 22-25 }' Oh, hehe. I was staring at exactly that line and concluded it's about anonymous sets. Maybe I should call it a day. %) [...] > I think I found the root cause: > > https://marc.info/?l=netfilter-devel&m=158825784609307&w=2 Good catch! Thanks, Phil
On Thu, Apr 30, 2020 at 05:41:06PM +0200, Phil Sutter wrote: > [...] > > I think I found the root cause: > > > > https://marc.info/?l=netfilter-devel&m=158825784609307&w=2 > > Good catch! I just pushed this out, thanks for reviewing.
diff --git a/src/rule.c b/src/rule.c index 633ca13639ad..9e80c0251947 100644 --- a/src/rule.c +++ b/src/rule.c @@ -1563,7 +1563,8 @@ static int __do_add_setelems(struct netlink_ctx *ctx, struct set *set, if (mnl_nft_setelem_add(ctx, set, expr, flags) < 0) return -1; - if (set->init != NULL && + if (!set_is_anonymous(set->flags) && + set->init != NULL && set->flags & NFT_SET_INTERVAL && set->desc.field_count <= 1) { interval_map_decompose(expr);
This patch invokes interval_map_decompose() with named sets: ==3402== 2,352 (128 direct, 2,224 indirect) bytes in 1 blocks are definitely lost in loss record 9 of 9 ==3402== at 0x483577F: malloc (vg_replace_malloc.c:299) ==3402== by 0x48996A8: xmalloc (utils.c:36) ==3402== by 0x4899778: xzalloc (utils.c:65) ==3402== by 0x487CB46: expr_alloc (expression.c:45) ==3402== by 0x487E2A0: mapping_expr_alloc (expression.c:1140) ==3402== by 0x4898AA8: interval_map_decompose (segtree.c:1095) ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1569) ==3402== by 0x4872BDF: __do_add_setelems (rule.c:1559) ==3402== by 0x4877936: do_command (rule.c:2710) ==3402== by 0x489F1CB: nft_netlink.isra.5 (libnftables.c:42) ==3402== by 0x489FB07: nft_run_cmd_from_filename (libnftables.c:508) ==3402== by 0x10A9AA: main (main.c:455) Fixes: dd44081d91ce ("segtree: Fix add and delete of element in same batch") Signed-off-by: Pablo Neira Ayuso <pablo@netfilter.org> --- src/rule.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)