Re: [PATCH v2] sched: report if filter is too large to dump

[David Miller]

From: Roman Kapl 
Date: Wed, 21 Feb 2018 09:49:31 +0100

> Should I send a v3 or something, David?

No, it's too late to fix it, sorry.

Re: [PATCH v2] sched: report if filter is too large to dump

[Roman Kapl]

On 02/21/2018 09:42 AM, Phil Sutter wrote:

Hi Roman,

On Wed, Feb 21, 2018 at 09:38:52AM +0100, Roman Kapl wrote:

On 02/21/2018 08:45 AM, Phil Sutter wrote:

On Mon, Feb 19, 2018 at 09:32:51PM +0100, Roman Kapl wrote:

Note: The error case can happen pretty easily if you create a filter
with 32 actions and have 4kb pages. Also recent versions of iproute try
to be clever with their buffer allocation size, which in turn leads to

I'm curious, what does it lead to? :)

Thanks, Phil

tc will dump all filters up to the big one, then it will stop silently.
So it will seem as if you have less filters.

The new behavior is the same, but tc will at leas print the EMSGSIZE
error. It does not handle it in any other way.

I got that, yes. Though your commit message stops mid-sentence and I
wondered what the dynamic buffer allocation "in turn leads to".

Thanks, Phil


Aaah... It leads to "smaller SKBs in kernel". I've lost that line in v2.

Should I send a v3 or something, David?

Sorry, Roman Kapl

Re: [PATCH v2] sched: report if filter is too large to dump

[Phil Sutter]

Hi Roman,

On Wed, Feb 21, 2018 at 09:38:52AM +0100, Roman Kapl wrote:
> On 02/21/2018 08:45 AM, Phil Sutter wrote:
> > On Mon, Feb 19, 2018 at 09:32:51PM +0100, Roman Kapl wrote:
> >> So far, if the filter was too large to fit in the allocated skb, the
> >> kernel did not return any error and stopped dumping. Modify the dumper
> >> so that it returns -EMSGSIZE when a filter fails to dump and it is the
> >> first filter in the skb. If we are not first, we will get a next chance
> >> with more room.
> >>
> >> I understand this is pretty near to being an API change, but the
> >> original design (silent truncation) can be considered a bug.
> >>
> >> Note: The error case can happen pretty easily if you create a filter
> >> with 32 actions and have 4kb pages. Also recent versions of iproute try
> >> to be clever with their buffer allocation size, which in turn leads to
> > I'm curious, what does it lead to? :)
> >
> > Thanks, Phil
> 
> tc will dump all filters up to the big one, then it will stop silently. 
> So it will seem as if you have less filters.
> 
> The new behavior is the same, but tc will at leas print the EMSGSIZE 
> error. It does not handle it in any other way.

I got that, yes. Though your commit message stops mid-sentence and I
wondered what the dynamic buffer allocation "in turn leads to".

Thanks, Phil

Re: [PATCH v2] sched: report if filter is too large to dump

[Roman Kapl]

On 02/21/2018 08:45 AM, Phil Sutter wrote:

Hi Roman,

On Mon, Feb 19, 2018 at 09:32:51PM +0100, Roman Kapl wrote:

So far, if the filter was too large to fit in the allocated skb, the
kernel did not return any error and stopped dumping. Modify the dumper
so that it returns -EMSGSIZE when a filter fails to dump and it is the
first filter in the skb. If we are not first, we will get a next chance
with more room.

I understand this is pretty near to being an API change, but the
original design (silent truncation) can be considered a bug.

Note: The error case can happen pretty easily if you create a filter
with 32 actions and have 4kb pages. Also recent versions of iproute try
to be clever with their buffer allocation size, which in turn leads to

I'm curious, what does it lead to? :)

Thanks, Phil


tc will dump all filters up to the big one, then it will stop silently. 
So it will seem as if you have less filters.


The new behavior is the same, but tc will at leas print the EMSGSIZE 
error. It does not handle it in any other way.

Re: [PATCH v2] sched: report if filter is too large to dump

[Phil Sutter]

Hi Roman,

On Mon, Feb 19, 2018 at 09:32:51PM +0100, Roman Kapl wrote:
> So far, if the filter was too large to fit in the allocated skb, the
> kernel did not return any error and stopped dumping. Modify the dumper
> so that it returns -EMSGSIZE when a filter fails to dump and it is the
> first filter in the skb. If we are not first, we will get a next chance
> with more room.
> 
> I understand this is pretty near to being an API change, but the
> original design (silent truncation) can be considered a bug.
> 
> Note: The error case can happen pretty easily if you create a filter
> with 32 actions and have 4kb pages. Also recent versions of iproute try
> to be clever with their buffer allocation size, which in turn leads to

I'm curious, what does it lead to? :)

Thanks, Phil

Re: [PATCH v2] sched: report if filter is too large to dump

[David Miller]

From: Roman Kapl 
Date: Mon, 19 Feb 2018 21:32:51 +0100

> So far, if the filter was too large to fit in the allocated skb, the
> kernel did not return any error and stopped dumping. Modify the dumper
> so that it returns -EMSGSIZE when a filter fails to dump and it is the
> first filter in the skb. If we are not first, we will get a next chance
> with more room.
> 
> I understand this is pretty near to being an API change, but the
> original design (silent truncation) can be considered a bug.
> 
> Note: The error case can happen pretty easily if you create a filter
> with 32 actions and have 4kb pages. Also recent versions of iproute try
> to be clever with their buffer allocation size, which in turn leads to
> 
> Signed-off-by: Roman Kapl 

Applied and queued up for -stable, thanks Roman.

Re: [PATCH v2] sched: report if filter is too large to dump

[Cong Wang]

On Mon, Feb 19, 2018 at 12:32 PM, Roman Kapl  wrote:
> So far, if the filter was too large to fit in the allocated skb, the
> kernel did not return any error and stopped dumping. Modify the dumper
> so that it returns -EMSGSIZE when a filter fails to dump and it is the
> first filter in the skb. If we are not first, we will get a next chance
> with more room.
>
> I understand this is pretty near to being an API change, but the
> original design (silent truncation) can be considered a bug.
>
> Note: The error case can happen pretty easily if you create a filter
> with 32 actions and have 4kb pages. Also recent versions of iproute try
> to be clever with their buffer allocation size, which in turn leads to
>
> Signed-off-by: Roman Kapl 

Acked-by: Cong Wang 

Re: [PATCH v2] sched: report if filter is too large to dump

[Jiri Pirko]

Mon, Feb 19, 2018 at 09:32:51PM CET, c...@rkapl.cz wrote:
>So far, if the filter was too large to fit in the allocated skb, the
>kernel did not return any error and stopped dumping. Modify the dumper
>so that it returns -EMSGSIZE when a filter fails to dump and it is the
>first filter in the skb. If we are not first, we will get a next chance
>with more room.
>
>I understand this is pretty near to being an API change, but the
>original design (silent truncation) can be considered a bug.
>
>Note: The error case can happen pretty easily if you create a filter
>with 32 actions and have 4kb pages. Also recent versions of iproute try
>to be clever with their buffer allocation size, which in turn leads to
>
>Signed-off-by: Roman Kapl 

Acked-by: Jiri Pirko 

[PATCH v2] sched: report if filter is too large to dump

[Roman Kapl]

So far, if the filter was too large to fit in the allocated skb, the
kernel did not return any error and stopped dumping. Modify the dumper
so that it returns -EMSGSIZE when a filter fails to dump and it is the
first filter in the skb. If we are not first, we will get a next chance
with more room.

I understand this is pretty near to being an API change, but the
original design (silent truncation) can be considered a bug.

Note: The error case can happen pretty easily if you create a filter
with 32 actions and have 4kb pages. Also recent versions of iproute try
to be clever with their buffer allocation size, which in turn leads to

Signed-off-by: Roman Kapl 
---
v1 -> v2: add the "progress" comment, fixed error name in commit message.

I've looked at other dumpers in rtnetnlink, there are various ways to handle
that. For example rtnl_stats_dump has:
WARN_ON((err == -EMSGSIZE) && (skb->len == 0));
rtnl_dump_ifinfo has the same logic I am proposing:
if (err < 0) 
if (skb->len) goto out;
goto out_err;
Other functions handle the error in the "wrong" way (= what we currently do).
Although it might be OK if there is no array in what they return. I have not
tested the behavior, since the only way I found was to have a device with many
VFs.

 net/sched/cls_api.c | 7 ++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/net/sched/cls_api.c b/net/sched/cls_api.c
index f21610c5da1a..92e9308bb920 100644
--- a/net/sched/cls_api.c
+++ b/net/sched/cls_api.c
@@ -1399,13 +1399,18 @@ static int tc_dump_tfilter(struct sk_buff *skb, struct 
netlink_callback *cb)
nla_get_u32(tca[TCA_CHAIN]) != chain->index)
continue;
if (!tcf_chain_dump(chain, q, parent, skb, cb,
-   index_start, ))
+   index_start, )) {
+   err = -EMSGSIZE;
break;
+   }
}
 
cb->args[0] = index;
 
 out:
+   /* If we did no progress, the error (EMSGSIZE) is real */
+   if (skb->len == 0 && err)
+   return err;
return skb->len;
 }
 
-- 
2.16.1