CVE-2026-98299: tcp: do not let tcp_rmem be set below 4096

In the Linux kernel, the following vulnerability has been resolved:

tcp: do not let tcprmem be set below 4096

We can hit a division by zero crash in tcprcvbufgrow() and tcprcvspaceadjust():

divide error: 0000 [#1] PREEMPT SMP RIP: 0010:tcprcvbufgrow+0x187/0x450 net/ipv4/tcpinput.c:939 ... grow = divu64(((u64)rcvwin << 1) (newval - oldval), oldval);

The division uses oldval = tp->rcvqspace.space as divisor. When tp->rcvqspace.space is zero, this leads to a divide-by-zero exception.

tp->rcvqspace.space is initialized in tcpinitbufferspace(): tp->rcvqspace.space = min3(tp->rcvssthresh, tp->rcvwnd, (u32)TCPINITCWND tp->advmss);

If tcprmem[1] is configured to very small values (such as 1), sk->skrcvbuf is initialized to 1. Then tcpfullspace(sk), which computes (sk->skrcvbuf scalingratio) >> 8, truncates to 0. This sets tp->windowclamp = 0, tp->rcvssthresh = 0, and tp->rcvqspace.space = 0. Later, when data arrives and DRS is invoked, tcprcvbufgrow() divides by oldval == 0.

Back in 2015, commit b1cb59cf2efe ("net: sysctlnetcore: check SNDBUF and RCVBUF for min length") ensured that net.core.rmemdefault and net.core.rmemmax cannot be set below SOCKMINRCVBUF. Similarly, SORCVBUF setsockopt enforces maxt(int, val 2, SOCKMINRCVBUF).

However, net.ipv4.tcprmem still had .extra1 = SYSCTLONE, allowing arbitrarily small values.

Because SOCKMINRCVBUF depends on sizeof(struct skbuff) and cacheline alignment, its value varies across architectures and configuration options. Using a fixed constant of 4096 ensures a predictable, architecture- independent lower bound that is safely above SOCKMINRCVBUF everywhere and matches the documented 4K default.

Fix this by setting tcprmem.extra1 to 4096 and updating the documentation.