Difference between net.core.rmem_max and net.ipv4.tcp_rmem

Question

What's the difference between net.core.rmem_max and the third value of net.ipv4.tcp_rmem? Which has the higher priority for tcp connections?

For below two examples, what's the max buffer for tcp connections?

Case 1:
sysctl -w net.core.rmem_max=7388608
sysctl -w net.ipv4.tcp_rmem='4096 87380 8388608'

Case 2:
sysctl -w net.core.rmem_max=8388608
sysctl -w net.ipv4.tcp_rmem='4096 87380 7388608'

Solution

Core is the overall max receive buffer, while tcp relates to just that protocol.

As for the priority-question: It seems that the tcp-setting will take precendence over the common max setting, which is a bit confusing. Setting max has no effect on the current tcp setting (just tested on CentOS 5).

A more correct description would have been: default_max - but that was propably too longish.

Comments

[Nils]

Priority related to tcp?

[bydsky]

@Nils Yes, for tcp connections.

[nh2]

Your explanation makes sense, but this conflicts with what man tcp says about tcp_rmem's max value: the maximum size of the receive buffer used by each TCP socket. This value does not override the global net.core.rmem_max - see also stackoverflow.com/questions/31546835/…. Is man tcp wrong?

[Nils]

@nh2 That would not be the first time where a man page is wrong.

[Wildcard]

How exactly did you test it?

[Jordan Pilat]

@Nils, simply reading the values won't tell you if one overrides another -- you have to actually try to get a TCP buffer which exceeds the net.core.[wmem/rmem]_max buffer in order to test out such overriding.

[nh2]

I've reported the apparent man page bug here: bugzilla.kernel.org/show_bug.cgi?id=209327