CVE-2021-47300

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> bpf: Fix tail_call_reachable rejection for interpreter when jit failed<br /> <br /> During testing of f263a81451c1 ("bpf: Track subprog poke descriptors correctly<br /> and fix use-after-free") under various failure conditions, for example, when<br /> jit_subprogs() fails and tries to clean up the program to be run under the<br /> interpreter, we ran into the following freeze:<br /> <br /> [...]<br /> #127/8 tailcall_bpf2bpf_3:FAIL<br /> [...]<br /> [ 92.041251] BUG: KASAN: slab-out-of-bounds in ___bpf_prog_run+0x1b9d/0x2e20<br /> [ 92.042408] Read of size 8 at addr ffff88800da67f68 by task test_progs/682<br /> [ 92.043707]<br /> [ 92.044030] CPU: 1 PID: 682 Comm: test_progs Tainted: G O 5.13.0-53301-ge6c08cb33a30-dirty #87<br /> [ 92.045542] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.13.0-1ubuntu1 04/01/2014<br /> [ 92.046785] Call Trace:<br /> [ 92.047171] ? __bpf_prog_run_args64+0xc0/0xc0<br /> [ 92.047773] ? __bpf_prog_run_args32+0x8b/0xb0<br /> [ 92.048389] ? __bpf_prog_run_args64+0xc0/0xc0<br /> [ 92.049019] ? ktime_get+0x117/0x130<br /> [...] // few hundred [similar] lines more<br /> [ 92.659025] ? ktime_get+0x117/0x130<br /> [ 92.659845] ? __bpf_prog_run_args64+0xc0/0xc0<br /> [ 92.660738] ? __bpf_prog_run_args32+0x8b/0xb0<br /> [ 92.661528] ? __bpf_prog_run_args64+0xc0/0xc0<br /> [ 92.662378] ? print_usage_bug+0x50/0x50<br /> [ 92.663221] ? print_usage_bug+0x50/0x50<br /> [ 92.664077] ? bpf_ksym_find+0x9c/0xe0<br /> [ 92.664887] ? ktime_get+0x117/0x130<br /> [ 92.665624] ? kernel_text_address+0xf5/0x100<br /> [ 92.666529] ? __kernel_text_address+0xe/0x30<br /> [ 92.667725] ? unwind_get_return_address+0x2f/0x50<br /> [ 92.668854] ? ___bpf_prog_run+0x15d4/0x2e20<br /> [ 92.670185] ? ktime_get+0x117/0x130<br /> [ 92.671130] ? __bpf_prog_run_args64+0xc0/0xc0<br /> [ 92.672020] ? __bpf_prog_run_args32+0x8b/0xb0<br /> [ 92.672860] ? __bpf_prog_run_args64+0xc0/0xc0<br /> [ 92.675159] ? ktime_get+0x117/0x130<br /> [ 92.677074] ? lock_is_held_type+0xd5/0x130<br /> [ 92.678662] ? ___bpf_prog_run+0x15d4/0x2e20<br /> [ 92.680046] ? ktime_get+0x117/0x130<br /> [ 92.681285] ? __bpf_prog_run32+0x6b/0x90<br /> [ 92.682601] ? __bpf_prog_run64+0x90/0x90<br /> [ 92.683636] ? lock_downgrade+0x370/0x370<br /> [ 92.684647] ? mark_held_locks+0x44/0x90<br /> [ 92.685652] ? ktime_get+0x117/0x130<br /> [ 92.686752] ? lockdep_hardirqs_on+0x79/0x100<br /> [ 92.688004] ? ktime_get+0x117/0x130<br /> [ 92.688573] ? __cant_migrate+0x2b/0x80<br /> [ 92.689192] ? bpf_test_run+0x2f4/0x510<br /> [ 92.689869] ? bpf_test_timer_continue+0x1c0/0x1c0<br /> [ 92.690856] ? rcu_read_lock_bh_held+0x90/0x90<br /> [ 92.691506] ? __kasan_slab_alloc+0x61/0x80<br /> [ 92.692128] ? eth_type_trans+0x128/0x240<br /> [ 92.692737] ? __build_skb+0x46/0x50<br /> [ 92.693252] ? bpf_prog_test_run_skb+0x65e/0xc50<br /> [ 92.693954] ? bpf_prog_test_run_raw_tp+0x2d0/0x2d0<br /> [ 92.694639] ? __fget_light+0xa1/0x100<br /> [ 92.695162] ? bpf_prog_inc+0x23/0x30<br /> [ 92.695685] ? __sys_bpf+0xb40/0x2c80<br /> [ 92.696324] ? bpf_link_get_from_fd+0x90/0x90<br /> [ 92.697150] ? mark_held_locks+0x24/0x90<br /> [ 92.698007] ? lockdep_hardirqs_on_prepare+0x124/0x220<br /> [ 92.699045] ? finish_task_switch+0xe6/0x370<br /> [ 92.700072] ? lockdep_hardirqs_on+0x79/0x100<br /> [ 92.701233] ? finish_task_switch+0x11d/0x370<br /> [ 92.702264] ? __switch_to+0x2c0/0x740<br /> [ 92.703148] ? mark_held_locks+0x24/0x90<br /> [ 92.704155] ? __x64_sys_bpf+0x45/0x50<br /> [ 92.705146] ? do_syscall_64+0x35/0x80<br /> [ 92.706953] ? entry_SYSCALL_64_after_hwframe+0x44/0xae<br /> [...]<br /> <br /> Turns out that the program rejection from e411901c0b77 ("bpf: allow for tailcalls<br /> in BPF subprograms for x64 JIT") is buggy since env-&gt;prog-&gt;aux-&gt;tail_call_reachable<br /> is never true. Commit ebf7d1f508a7 ("bpf, x64: rework pro/epilogue and tailcall<br /> handling in JIT") added a tracker into check_max_stack_depth() which propagates<br /> the tail_call_reachable condition throughout the subprograms. This info is then<br /> assigned to the subprogram&amp;#39;s <br /> ---truncated---