Skip to contentSkip to conversation

net: Consolidate metrics handling for ipv4 and ipv6

PATCH David Miller 3 participants 5 messages Oct 5, 2018 – Oct 6, 2018 1 message not in the archive

Subsystems (3)
lore
Share & tools
lore

David Miller

author maintainer
#1 lore raw
In reply to a message not in the archive

From: David Ahern <[email protected]>
Date: Thu, 4 Oct 2018 20:07:50 -0700

Quoted 10 lines

From: David Ahern <[email protected]>

As part of the IPv6 fib info refactoring, the intent was to make metrics handling for ipv6 identical to ipv4. One oversight in ip6_dst_destroy led to confusion and a couple of incomplete attempts at finding and fixing the resulting memory leak which was ultimately resolved by ce7ea4af0838 ("ipv6: fix memory leak on dst->_metrics").

Refactor metrics hanlding make the code really identical for v4 and v6, and add a few test cases.

Looks nice, series applied, thanks David.

#2 lore raw
Quoted 11 lines · David Miller

From: David Ahern <[email protected]>
Date: Thu, 4 Oct 2018 20:07:50 -0700

Quoted 8 lines

From: David Ahern <[email protected]>

As part of the IPv6 fib info refactoring, the intent was to make metrics handling for ipv6 identical to ipv4. One oversight in ip6_dst_destroy led to confusion and a couple of incomplete attempts at finding and fixing the resulting memory leak which was ultimately resolved by ce7ea4af0838 ("ipv6: fix memory leak on dst->_metrics").

Quoted 5 lines
Quoted 2 lines

Refactor metrics hanlding make the code really identical for v4 and v6, and add a few test cases.

Looks nice, series applied, thanks David.

Does not look well tested and reviewed to me.

#3 lore raw
Quoted 14 lines · Eric Dumazet
Quoted 11 lines · David Miller

From: David Ahern <[email protected]>
Date: Thu, 4 Oct 2018 20:07:50 -0700

Quoted 8 lines

From: David Ahern <[email protected]>

As part of the IPv6 fib info refactoring, the intent was to make metrics handling for ipv6 identical to ipv4. One oversight in ip6_dst_destroy led to confusion and a couple of incomplete attempts at finding and fixing the resulting memory leak which was ultimately resolved by ce7ea4af0838 ("ipv6: fix memory leak on dst->_metrics").

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

Refactor metrics hanlding make the code really identical for v4 and v6, and add a few test cases.

Looks nice, series applied, thanks David.

Does not look well tested and reviewed to me.

For some reason I have not received the patch series in my inbox, I only got your "series applied" message.

Commit 767a2217533fed6 ("net: common metrics init helper for FIB entries") is not correct because we need to better deal with error paths.

I will submit this more formally when I can reach my workstation in a few minutes :

net/ipv6/route.c +2 -0
index 6c1d817151cae45421dc976c5ea082b4115650be..74d97addf1af20dda0c2b6a2018e88696f9f7d5a 100644
@@ -2976,6 +2976,8 @@ static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
        rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len);
        if (IS_ERR(rt->fib6_metrics)) {
                err = PTR_ERR(rt->fib6_metrics);
+               /* Do not leave garbage there. */
+               rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
                goto out;
        }
 
#4 lore raw
Quoted 11 lines · Eric Dumazet

Commit 767a2217533fed6 ("net: common metrics init helper for FIB entries") is not correct because we need to better deal with error paths.

I will submit this more formally when I can reach my workstation in a few minutes :

net/ipv6/route.c +0 -0
index 6c1d817151cae45421dc976c5ea082b4115650be..74d97addf1af20dda0c2b6a2018e88696f9f7d5a 100644

net/ipv6/route.c static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg, · line 2983

Quoted 8 lines
net/ipv6/route.c +2 -0
@@ -2976,6 +2976,8 @@ static struct fib6_info *ip6_route_info_create(struct fib6_config *cfg,
        rt->fib6_metrics = ip_fib_metrics_init(net, cfg->fc_mx, cfg->fc_mx_len);
        if (IS_ERR(rt->fib6_metrics)) {
                err = PTR_ERR(rt->fib6_metrics);
+               /* Do not leave garbage there. */
+               rt->fib6_metrics = (struct dst_metrics *)&dst_default_metrics;
                goto out;
        }
 

Yes, I was focused on the memory leaks and cleanup and forgot to purposely fail the alloc for the metrics. The above is one way to fix it. Thanks for spotting it.

author maintainer
#5 lore raw
In reply to Eric Dumazet

From: Eric Dumazet <[email protected]>
Date: Fri, 5 Oct 2018 06:08:57 -0700

Quoted 2 lines

For some reason I have not received the patch series in my inbox, I only got your "series applied" message.

I see what happened.

I did something stupid on vger.kernel.org yesterday which ran a partition out of disk space, and some postings got lost as a result.

Sorry about that.

Quoted 4 lines

Commit 767a2217533fed6 ("net: common metrics init helper for FIB entries") is not correct because we need to better deal with error paths.

I will submit this more formally when I can reach my workstation in a few minutes :

Thanks Eric.