From a8abb0c3dc1e28454851a00f8b7333d9695d566c Mon Sep 17 00:00:00 2001 From: Kumar Kartikeya Dwivedi Date: Wed, 9 Feb 2022 12:33:23 +0530 Subject: [PATCH 1/2] bpf: Fix crash due to incorrect copy_map_value When both bpf_spin_lock and bpf_timer are present in a BPF map value, copy_map_value needs to skirt both objects when copying a value into and out of the map. However, the current code does not set both s_off and t_off in copy_map_value, which leads to a crash when e.g. bpf_spin_lock is placed in map value with bpf_timer, as bpf_map_update_elem call will be able to overwrite the other timer object. When the issue is not fixed, an overwriting can produce the following splat: [root@(none) bpf]# ./test_progs -t timer_crash [ 15.930339] bpf_testmod: loading out-of-tree module taints kernel. [ 16.037849] ================================================================== [ 16.038458] BUG: KASAN: user-memory-access in __pv_queued_spin_lock_slowpath+0x32b/0x520 [ 16.038944] Write of size 8 at addr 0000000000043ec0 by task test_progs/325 [ 16.039399] [ 16.039514] CPU: 0 PID: 325 Comm: test_progs Tainted: G OE 5.16.0+ #278 [ 16.039983] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS ArchLinux 1.15.0-1 04/01/2014 [ 16.040485] Call Trace: [ 16.040645] [ 16.040805] dump_stack_lvl+0x59/0x73 [ 16.041069] ? __pv_queued_spin_lock_slowpath+0x32b/0x520 [ 16.041427] kasan_report.cold+0x116/0x11b [ 16.041673] ? __pv_queued_spin_lock_slowpath+0x32b/0x520 [ 16.042040] __pv_queued_spin_lock_slowpath+0x32b/0x520 [ 16.042328] ? memcpy+0x39/0x60 [ 16.042552] ? pv_hash+0xd0/0xd0 [ 16.042785] ? lockdep_hardirqs_off+0x95/0xd0 [ 16.043079] __bpf_spin_lock_irqsave+0xdf/0xf0 [ 16.043366] ? bpf_get_current_comm+0x50/0x50 [ 16.043608] ? jhash+0x11a/0x270 [ 16.043848] bpf_timer_cancel+0x34/0xe0 [ 16.044119] bpf_prog_c4ea1c0f7449940d_sys_enter+0x7c/0x81 [ 16.044500] bpf_trampoline_6442477838_0+0x36/0x1000 [ 16.044836] __x64_sys_nanosleep+0x5/0x140 [ 16.045119] do_syscall_64+0x59/0x80 [ 16.045377] ? lock_is_held_type+0xe4/0x140 [ 16.045670] ? irqentry_exit_to_user_mode+0xa/0x40 [ 16.046001] ? mark_held_locks+0x24/0x90 [ 16.046287] ? asm_exc_page_fault+0x1e/0x30 [ 16.046569] ? asm_exc_page_fault+0x8/0x30 [ 16.046851] ? lockdep_hardirqs_on+0x7e/0x100 [ 16.047137] entry_SYSCALL_64_after_hwframe+0x44/0xae [ 16.047405] RIP: 0033:0x7f9e4831718d [ 16.047602] Code: b4 0c 00 0f 05 eb a9 66 0f 1f 44 00 00 f3 0f 1e fa 48 89 f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d b3 6c 0c 00 f7 d8 64 89 01 48 [ 16.048764] RSP: 002b:00007fff488086b8 EFLAGS: 00000206 ORIG_RAX: 0000000000000023 [ 16.049275] RAX: ffffffffffffffda RBX: 00007f9e48683740 RCX: 00007f9e4831718d [ 16.049747] RDX: 0000000000000000 RSI: 0000000000000000 RDI: 00007fff488086d0 [ 16.050225] RBP: 00007fff488086f0 R08: 00007fff488085d7 R09: 00007f9e4cb594a0 [ 16.050648] R10: 0000000000000000 R11: 0000000000000206 R12: 00007f9e484cde30 [ 16.051124] R13: 0000000000000000 R14: 0000000000000000 R15: 0000000000000000 [ 16.051608] [ 16.051762] ================================================================== Fixes: 68134668c17f ("bpf: Add map side support for bpf timers.") Signed-off-by: Kumar Kartikeya Dwivedi Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/20220209070324.1093182-2-memxor@gmail.com --- include/linux/bpf.h | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/include/linux/bpf.h b/include/linux/bpf.h index fa517ae604ad..31a83449808b 100644 --- a/include/linux/bpf.h +++ b/include/linux/bpf.h @@ -224,7 +224,8 @@ static inline void copy_map_value(struct bpf_map *map, void *dst, void *src) if (unlikely(map_value_has_spin_lock(map))) { s_off = map->spin_lock_off; s_sz = sizeof(struct bpf_spin_lock); - } else if (unlikely(map_value_has_timer(map))) { + } + if (unlikely(map_value_has_timer(map))) { t_off = map->timer_off; t_sz = sizeof(struct bpf_timer); } From a7e75016a0753c24d6c995bc02501ae35368e333 Mon Sep 17 00:00:00 2001 From: Kumar Kartikeya Dwivedi Date: Wed, 9 Feb 2022 12:33:24 +0530 Subject: [PATCH 2/2] selftests/bpf: Add test for bpf_timer overwriting crash Add a test that validates that timer value is not overwritten when doing a copy_map_value call in the kernel. Without the prior fix, this test triggers a crash. Signed-off-by: Kumar Kartikeya Dwivedi Signed-off-by: Alexei Starovoitov Link: https://lore.kernel.org/bpf/20220209070324.1093182-3-memxor@gmail.com --- .../selftests/bpf/prog_tests/timer_crash.c | 32 +++++++++++ .../testing/selftests/bpf/progs/timer_crash.c | 54 +++++++++++++++++++ 2 files changed, 86 insertions(+) create mode 100644 tools/testing/selftests/bpf/prog_tests/timer_crash.c create mode 100644 tools/testing/selftests/bpf/progs/timer_crash.c diff --git a/tools/testing/selftests/bpf/prog_tests/timer_crash.c b/tools/testing/selftests/bpf/prog_tests/timer_crash.c new file mode 100644 index 000000000000..f74b82305da8 --- /dev/null +++ b/tools/testing/selftests/bpf/prog_tests/timer_crash.c @@ -0,0 +1,32 @@ +// SPDX-License-Identifier: GPL-2.0 +#include +#include "timer_crash.skel.h" + +enum { + MODE_ARRAY, + MODE_HASH, +}; + +static void test_timer_crash_mode(int mode) +{ + struct timer_crash *skel; + + skel = timer_crash__open_and_load(); + if (!ASSERT_OK_PTR(skel, "timer_crash__open_and_load")) + return; + skel->bss->pid = getpid(); + skel->bss->crash_map = mode; + if (!ASSERT_OK(timer_crash__attach(skel), "timer_crash__attach")) + goto end; + usleep(1); +end: + timer_crash__destroy(skel); +} + +void test_timer_crash(void) +{ + if (test__start_subtest("array")) + test_timer_crash_mode(MODE_ARRAY); + if (test__start_subtest("hash")) + test_timer_crash_mode(MODE_HASH); +} diff --git a/tools/testing/selftests/bpf/progs/timer_crash.c b/tools/testing/selftests/bpf/progs/timer_crash.c new file mode 100644 index 000000000000..f8f7944e70da --- /dev/null +++ b/tools/testing/selftests/bpf/progs/timer_crash.c @@ -0,0 +1,54 @@ +// SPDX-License-Identifier: GPL-2.0 + +#include +#include +#include + +struct map_elem { + struct bpf_timer timer; + struct bpf_spin_lock lock; +}; + +struct { + __uint(type, BPF_MAP_TYPE_ARRAY); + __uint(max_entries, 1); + __type(key, int); + __type(value, struct map_elem); +} amap SEC(".maps"); + +struct { + __uint(type, BPF_MAP_TYPE_HASH); + __uint(max_entries, 1); + __type(key, int); + __type(value, struct map_elem); +} hmap SEC(".maps"); + +int pid = 0; +int crash_map = 0; /* 0 for amap, 1 for hmap */ + +SEC("fentry/do_nanosleep") +int sys_enter(void *ctx) +{ + struct map_elem *e, value = {}; + void *map = crash_map ? (void *)&hmap : (void *)&amap; + + if (bpf_get_current_task_btf()->tgid != pid) + return 0; + + *(void **)&value = (void *)0xdeadcaf3; + + bpf_map_update_elem(map, &(int){0}, &value, 0); + /* For array map, doing bpf_map_update_elem will do a + * check_and_free_timer_in_array, which will trigger the crash if timer + * pointer was overwritten, for hmap we need to use bpf_timer_cancel. + */ + if (crash_map == 1) { + e = bpf_map_lookup_elem(map, &(int){0}); + if (!e) + return 0; + bpf_timer_cancel(&e->timer); + } + return 0; +} + +char _license[] SEC("license") = "GPL";