CVE-2026-98588

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

bpf, sockmap: Reject max entries > INT MAX in sock map alloc

sock map alloc() only rejects max entries == 0 and otherwise allows any u32 value. sock map free() then walks the sks[] array with a signed int iterator:

int i;
for (i = 0; i < stab->map.max entries; i++)
	struct sock **psk = &stab->sks[i];

When a SOCKMAP is created with max entries = 0xffffffff (UINT MAX), the allocation of 32 GiB can succeed on large-memory hosts. During free the counter reaches 0x80000000, wraps to INT MIN, is sign-extended by movslq and turned into a ~16 GiB negative offset from stab->sks, pointing far below the allocation.

The faulting access is an xchg() write in sock map free(). Without KASAN, the same out-of-bounds write can fault on an unmapped vmalloc page or corrupt an unrelated allocation if that vmalloc address is populated. On a KASAN kernel with CONFIG KASAN VMALLOC=y, the shadow check for that address hits an unmapped shadow page and oopses first:

BUG: unable to handle page fault for address: fffff521b59c5a00 RIP: 0010:kasan check range+0x107/0x190 Call Trace: sock map free+0x93/0x190 map create+0x68d/0xb30 sys bpf+0x21e/0x2e70

Vmcore confirmed stab->map.max entries == 0xffffffff, stab->sks == 0xffffc911ace2d000, and the faulting address sks + (s64)INT MIN * 8 exactly at 0xffffc90dace2d000. The same buggy path is reached on the normal close()/bpf map free deferred() path whenever such a map is destroyed.

sock map alloc() used to bound its allocation size through bpf map charge init(), but the bound was dropped when rlimit-based memory accounting was removed. Reject max entries > INT MAX at creation time so the signed iterator in sock map free() never sees a value that would overflow.

Triggered by syzkaller and reproduced on both a 6.6-based KASAN kernel and the upstream v7.3-rc2 kernel.