bpf, fix for subprogs with tailcalls
Subsystems (8)
John Fastabend
reviewerv2 cover bpf, fix for subprogs with tailcalls
This fixes a use-after-free when using subprogs and tailcalls and adds a test case to trigger the use-after-free.
v2: original patch missed fixups needed in out label as Daniel caught
John Fastabend (2):
bpf: track subprog poke correctly, fix use-after-free
bpf: selftest to verify mixing bpf2bpf calls and tailcalls with insn
patch
arch/x86/net/bpf_jit_comp.c | 4 ++ include/linux/bpf.h | 1 + kernel/bpf/core.c | 7 ++- kernel/bpf/verifier.c | 45 ++++--------------- .../selftests/bpf/prog_tests/tailcalls.c | 36 ++++++++++----- .../selftests/bpf/progs/tailcall_bpf2bpf4.c | 21 ++++++++- 6 files changed, 65 insertions(+), 49 deletions(-)
Signature · 1 line
2.17.1
John Fastabend
reviewerPATCH v2 1/2 bpf: track subprog poke correctly, fix use-after-free
Subprograms are calling map_poke_track but on program release there is no hook to call map_poke_untrack. But on prog release the aux memory is freed even though we still have a reference to it in the element list of the map aux data.
So when we run map_poke_run() we end up accessing free'd memory. This triggers with KASAN in prog_array_map_poke_run() shown here.
[ 402.824686] ==================================================================
[ 402.824689] BUG: KASAN: use-after-free in prog_array_map_poke_run+0xc2/0x34e
[ 402.824698] Read of size 4 at addr ffff8881905a7940 by task hubble-fgs/4337 [ 402.824705] CPU: 1 PID: 4337 Comm: hubble-fgs Tainted: G I 5.12.0+ #399
[ 402.824715] Call Trace:
[ 402.824719] dump_stack+0x93/0xc2
[ 402.824727] print_address_description.constprop.0+0x1a/0x140
[ 402.824736] ? prog_array_map_poke_run+0xc2/0x34e
[ 402.824740] ? prog_array_map_poke_run+0xc2/0x34e
[ 402.824744] kasan_report.cold+0x7c/0xd8
[ 402.824752] ? prog_array_map_poke_run+0xc2/0x34e
[ 402.824757] prog_array_map_poke_run+0xc2/0x34e
[ 402.824765] bpf_fd_array_map_update_elem+0x124/0x1a0The elements concerned are walked like this,
for (i = 0; i < elem->aux->size_poke_tab; i++) {
poke = &elem->aux->poke_tab[i];So the access to size_poke_tab is the 4B read, verified by checking offsets in the KASAN dump,
[ 402.825004] The buggy address belongs to the object at ffff8881905a7800
which belongs to the cache kmalloc-1k of size 1024
[ 402.825008] The buggy address is located 320 bytes inside of
1024-byte region [ffff8881905a7800, ffff8881905a7c00)With pahol output,
struct bpf_prog_aux {
...
/* --- cacheline 5 boundary (320 bytes) --- */
u32 size_poke_tab; /* 320 4 */
...In general subprograms do not manage their own data structures. For example btf func_info, linfo, etc are just pointers to the prog structure. This allows reference counting and cleanup to be done on the main prog.
The aux->poke_tab struct however did not follow this logic. The initial fix for above use after free further embedded subprogram tracking of poke data tracking into the subprogram with proper reference counting. However, Daniel and Alexei questioned why we were treating these objects specially. I agree its unnecessary.
The fix here removes the per subprogram poke data structure alloc and map tracking and instead simply points the aux->poke_tab pointer at the main programs poke_tab. This way map tracking is done on the origin program and we do not need to manage them per subprogram. A couple small complication arise here. First on bpf_prog_free_deferred(), where we unwind the prog reference counting and kfree objects, we need to ensure that we don't try to double free the poke_tab when free'ing the subprog structures. This is easily solved by NULL'ing the poke_tab pointer.
The second detail is to ensure that per subprog jit logic only does fixups on poke_tab[] entries it owns. To do this we add a pointer in the poke structure to point at the subprog value so JITs can easily check while walking the poke_tab structure if the current entry belongs to the current program. This change is necessary per JIT. See x86/net/bpf_jit_compo.c func bpf_tail_call_direct_fixup() for the details. Only x86 is currently using the poke_tab struct so we only need to fixup the x86 JIT.
On the error path we omit cleaning up the poke->aux field because these are only ever referenced from the JIT side, but on error we will never make it to the JIT so its fine to leave them dangling. Removing these pointers would complicate the error path for no reason.
--- arch/x86/net/bpf_jit_comp.c | 4 ++++ include/linux/bpf.h | 1 + kernel/bpf/core.c | 7 +++++- kernel/bpf/verifier.c | 45 +++++++------------------------------ 4 files changed, 19 insertions(+), 38 deletions(-)
arch/x86/net/bpf_jit_comp.c +4 -0
@@ -576,6 +576,10 @@ static void bpf_tail_call_direct_fixup(struct bpf_prog *prog)for (i =0 ; i < prog->aux->size_poke_tab; i++) { poke = &prog->aux->poke_tab[i]; + +if (poke->aux && poke->aux != prog->aux) +continue ; +WARN_ON_ONCE (READ_ONCE (poke->tailcall_target_stable));if (poke->reason !=BPF_POKE_REASON_TAIL_CALL )
include/linux/bpf.h +1 -0
@@ -777,6 +777,7 @@ struct bpf_jit_poke_descriptor {void *tailcall_target;void *tailcall_bypass;void *bypass_addr; +void *aux; union { struct { struct bpf_map *map;
kernel/bpf/core.c +6 -1
@@ -2211,8 +2211,13 @@ static void bpf_prog_free_deferred(struct work_struct *work)#endif if (aux->dst_trampoline)bpf_trampoline_put (aux->dst_trampoline); -for (i =0 ; i < aux->func_cnt; i++) +for (i =0 ; i < aux->func_cnt; i++) { +/* poke_tab in subprogs are links to main prog and are +* freed above so delete link without kfree. +*/ + aux->func[i]->aux->poke_tab =NULL ;bpf_jit_free (aux->func[i]); + }if (aux->func_cnt) {kfree (aux->func);bpf_prog_unlock_free (aux->prog);
kernel/bpf/verifier.c +8 -37
67 lines not on this page · raw message
Signature · 1 line
2.17.1
John Fastabend
reviewerPATCH v2 2/2 bpf: selftest to verify mixing bpf2bpf calls and tailcalls with insn patch
In reply to John FastabendThis adds some extra noise to the tailcall_bpf2bpf4 tests that will cause verify to patch insns. This then moves around subprog start/end insn index and poke descriptor insn index to ensure that verify and JIT will continue to track these correctly.
If done correctly verifier should pass this program same as before and JIT should emit tail call logic.
--- .../selftests/bpf/prog_tests/tailcalls.c | 36 +++++++++++++------ .../selftests/bpf/progs/tailcall_bpf2bpf4.c | 21 ++++++++++- 2 files changed, 46 insertions(+), 11 deletions(-)
tools/testing/selftests/bpf/prog_tests/tailcalls.c +26 -10
@@ -715,6 +715,8 @@ static void test_tailcall_bpf2bpf_3(void)bpf_object__close (obj); } +#include "tailcall_bpf2bpf4.skel.h" +/* test_tailcall_bpf2bpf_4 checks that tailcall counter is correctly preserved * across tailcalls combined with bpf2bpf calls. for making sure that tailcall * counter behaves correctly, bpf program will go through following flow:
@@ -727,10 +729,15 @@ static void test_tailcall_bpf2bpf_3(void)* the loop begins. At the end of the test make sure that the global counter is * equal to 31, because tailcall counter includes the first two tailcalls * whereas global counter is incremented only on loop presented on flow above. +* +* The noise parameter is used to insert bpf_map_update calls into the logic +* to force verifier to patch instructions. This allows us to ensure jump +* logic remains correct with instruction movement. */ -static void test_tailcall_bpf2bpf_4 (void ) +static void test_tailcall_bpf2bpf_4 (bool noise) { -int err, map_fd, prog_fd, main_fd, data_fd, i, val; +int err, map_fd, prog_fd, main_fd, data_fd, i; +struct tailcall_bpf2bpf4__bss val;struct bpf_map *prog_array, *data_map;struct bpf_program *prog;struct bpf_object *obj;
@@ -774,11 +781,6 @@ static void test_tailcall_bpf2bpf_4(void) goto out; } - err =bpf_prog_test_run (main_fd,1 , &pkt_v4,sizeof (pkt_v4),0 , - &duration, &retval,NULL ); -CHECK (err || retval !=sizeof (pkt_v4) *3 ,"tailcall" ,"err %d errno %d retval %d\n" , - err, errno, retval); - data_map =bpf_object__find_map_by_name (obj,"tailcall.bss" );if (CHECK_FAIL (!data_map || !bpf_map__is_internal (data_map)))return ;
@@ -787,10 +789,22 @@ static void test_tailcall_bpf2bpf_4(void)if (CHECK_FAIL (map_fd <0 ))return ; + i =0 ; + val.noise = noise; + val.count =0 ; + err =bpf_map_update_elem (data_fd, &i, &val,BPF_ANY ); +if (CHECK_FAIL (err)) + goto out; + + err =bpf_prog_test_run (main_fd,1 , &pkt_v4,sizeof (pkt_v4),0 , + &duration, &retval,NULL ); +CHECK (err || retval !=sizeof (pkt_v4) *3 ,"tailcall" ,"err %d errno %d retval %d\n" , + err, errno, retval); + i =0 ; err =bpf_map_lookup_elem (data_fd, &i, &val); -CHECK (err || val !=31 ,"tailcall count" ,"err %d errno %d count %d\n" , - err, errno, val); +CHECK (err || val.count !=31 ,"tailcall count" ,"err %d errno %d count %d\n" , + err, errno, val.count); out:bpf_object__close (obj);
@@ -815,5 +829,7 @@ void test_tailcalls(void)if (test__start_subtest ("tailcall_bpf2bpf_3" ))test_tailcall_bpf2bpf_3 ();if (test__start_subtest ("tailcall_bpf2bpf_4" )) -test_tailcall_bpf2bpf_4 (); +test_tailcall_bpf2bpf_4 (false); +if (test__start_subtest ("tailcall_bpf2bpf_5" )) +test_tailcall_bpf2bpf_4 (true); }
tools/testing/selftests/bpf/progs/tailcall_bpf2bpf4.c +20 -1
@@ -2,6 +2,13 @@#include <linux/bpf.h> #include <bpf/bpf_helpers.h> +struct { + __uint(type, BPF_MAP_TYPE_ARRAY); + __uint(max_entries, 1); + __uint(key_size, sizeof(__u32)); + __uint(value_size, sizeof(__u32)); +} nop_table SEC(".maps"); +struct { __uint(type, BPF_MAP_TYPE_PROG_ARRAY); __uint(max_entries, 3);
@@ -9,11 +16,23 @@ struct {__uint (value_size,sizeof (__u32)); } jmp_table SEC(".maps"); -static volatile int count; +int count =0 ; +int noise =0 ; + +__always_inline int subprog_noise(void) +{ +__u32 key =0 ; + +bpf_printk ("hello noisy subprog %d\n" , key); +bpf_map_lookup_elem (&nop_table, &key); +return 0 ; +}__noinline int subprog_tail_2(struct __sk_buff *skb) { +if (noise) +subprog_noise ();bpf_tail_call_static (skb, &jmp_table,2 );return skb->len *3 ; }
Signature · 1 line
2.17.1
Re: [PATCH bpf v2 2/2] bpf: selftest to verify mixing bpf2bpf calls and tailcalls with insn patch
tools/testing/selftests/bpf/progs/tailcall_bpf2bpf4.c struct { · line 29
Quoted 13 lines · John Fastabend
tools/testing/selftests/bpf/progs/tailcall_bpf2bpf4.c +11
-1-static volatile int count; +int count =0 ; +int noise =0 ; + +__always_inline int subprog_noise(void) +{ +__u32 key =0 ; + +bpf_printk ("hello noisy subprog %d\n" , key); +bpf_map_lookup_elem (&nop_table, &key); +return 0 ; +}
This selftest patch I had to apply manually due to conflicts. I've also removed the noisy printk before pushing. I verified that I saw the spam before removing it. The patch 1 looks great. Thanks a lot for the fix.
John Fastabend
reviewerRe: [PATCH bpf v2 2/2] bpf: selftest to verify mixing bpf2bpf calls and tailcalls with insn patch
Quoted 17 lines · Alexei Starovoitov
tools/testing/selftests/bpf/progs/tailcall_bpf2bpf4.c struct { · line 29
Quoted 13 lines · John Fastabend
tools/testing/selftests/bpf/progs/tailcall_bpf2bpf4.c +11
-1-static volatile int count; +int count =0 ; +int noise =0 ; + +__always_inline int subprog_noise(void) +{ +__u32 key =0 ; + +bpf_printk ("hello noisy subprog %d\n" , key); +bpf_map_lookup_elem (&nop_table, &key); +return 0 ; +}This selftest patch I had to apply manually due to conflicts. I've also removed the noisy printk before pushing.
Thanks! I had the printk there to verify the code was actually being run, but forgot to remove it before submitting.
Quoted 1 line
I verified that I saw the spam before removing it.
Great.
Quoted 1 line
The patch 1 looks great. Thanks a lot for the fix.