CVE-2022-49840

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> bpf, test_run: Fix alignment problem in bpf_prog_test_run_skb()<br /> <br /> We got a syzkaller problem because of aarch64 alignment fault<br /> if KFENCE enabled. When the size from user bpf program is an odd<br /> number, like 399, 407, etc, it will cause the struct skb_shared_info&amp;#39;s<br /> unaligned access. As seen below:<br /> <br /> BUG: KFENCE: use-after-free read in __skb_clone+0x23c/0x2a0 net/core/skbuff.c:1032<br /> <br /> Use-after-free read at 0xffff6254fffac077 (in kfence-#213):<br /> __lse_atomic_add arch/arm64/include/asm/atomic_lse.h:26 [inline]<br /> arch_atomic_add arch/arm64/include/asm/atomic.h:28 [inline]<br /> arch_atomic_inc include/linux/atomic-arch-fallback.h:270 [inline]<br /> atomic_inc include/asm-generic/atomic-instrumented.h:241 [inline]<br /> __skb_clone+0x23c/0x2a0 net/core/skbuff.c:1032<br /> skb_clone+0xf4/0x214 net/core/skbuff.c:1481<br /> ____bpf_clone_redirect net/core/filter.c:2433 [inline]<br /> bpf_clone_redirect+0x78/0x1c0 net/core/filter.c:2420<br /> bpf_prog_d3839dd9068ceb51+0x80/0x330<br /> bpf_dispatcher_nop_func include/linux/bpf.h:728 [inline]<br /> bpf_test_run+0x3c0/0x6c0 net/bpf/test_run.c:53<br /> bpf_prog_test_run_skb+0x638/0xa7c net/bpf/test_run.c:594<br /> bpf_prog_test_run kernel/bpf/syscall.c:3148 [inline]<br /> __do_sys_bpf kernel/bpf/syscall.c:4441 [inline]<br /> __se_sys_bpf+0xad0/0x1634 kernel/bpf/syscall.c:4381<br /> <br /> kfence-#213: 0xffff6254fffac000-0xffff6254fffac196, size=407, cache=kmalloc-512<br /> <br /> allocated by task 15074 on cpu 0 at 1342.585390s:<br /> kmalloc include/linux/slab.h:568 [inline]<br /> kzalloc include/linux/slab.h:675 [inline]<br /> bpf_test_init.isra.0+0xac/0x290 net/bpf/test_run.c:191<br /> bpf_prog_test_run_skb+0x11c/0xa7c net/bpf/test_run.c:512<br /> bpf_prog_test_run kernel/bpf/syscall.c:3148 [inline]<br /> __do_sys_bpf kernel/bpf/syscall.c:4441 [inline]<br /> __se_sys_bpf+0xad0/0x1634 kernel/bpf/syscall.c:4381<br /> __arm64_sys_bpf+0x50/0x60 kernel/bpf/syscall.c:4381<br /> <br /> To fix the problem, we adjust @size so that (@size + @hearoom) is a<br /> multiple of SMP_CACHE_BYTES. So we make sure the struct skb_shared_info<br /> is aligned to a cache line.