Message ID | 20231127094525.1366740-1-jolsa@kernel.org (mailing list archive) |
---|---|
State | Changes Requested |
Delegated to: | BPF |
Headers | show |
Series | [bpf] bpf, x64: Fix prog_array_map_poke_run map poke update | expand |
On Mon, Nov 27, 2023 at 10:45:25AM +0100, Jiri Olsa wrote: > Lee pointed out issue found by syscaller [0] hitting BUG in prog > array map poke update in prog_array_map_poke_run function due to > bpf_arch_text_poke error return value. > > There's race window where bpf_arch_text_poke can fail due to missing > bpf program kallsym symbols, which is accounted for with check for > -EINVAL in BUG_ON. > > The problem is that in such case we won't update the tail call jump > and cause imballance for the next tail call update check which will > fail with -EBUSY in __bpf_arch_text_poke. > > I'm hitting following race during the program load: > > CPU 0 CPU 1 > > bpf_prog_load > bpf_check > do_misc_fixups > prog_array_map_poke_track {} > > map_update_elem > bpf_fd_array_map_update_elem > prog_array_map_poke_run > > bpf_arch_text_poke returns -EINVAL > > bpf_prog_kallsyms_add > > After bpf_arch_text_poke (CPU 1) fails to update the tail call jump, > the next poke update fails on expected jump instruction check in > __bpf_arch_text_poke with -EBUSY and triggers the BUG_ON in > prog_array_map_poke_run. > > Similar race exists on the program unload. > > Fixing this by calling directly __bpf_arch_text_poke and skipping the bpf > symbol check like we do in bpf_tail_call_direct_fixup. This way the > prog_array_map_poke_run does not depend on bpf program having the kallsym > symbol in place. > > [0] https://syzkaller.appspot.com/bug?extid=97a4fe20470e9bc30810 > > Cc: Lee Jones <lee@kernel.org> > Cc: Maciej Fijalkowski <maciej.fijalkowski@intel.com> > Fixes: ebf7d1f508a7 ("bpf, x64: rework pro/epilogue and tailcall handling in JIT") > Reported-by: syzbot+97a4fe20470e9bc30810@syzkaller.appspotmail.com > Tested-by: Lee Jones <lee@kernel.org> > Signed-off-by: Jiri Olsa <jolsa@kernel.org> hum, this breaks non x86 builds.. will send new version jirka > --- > arch/x86/net/bpf_jit_comp.c | 4 ++-- > include/linux/bpf.h | 2 ++ > kernel/bpf/arraymap.c | 31 +++++++++++-------------------- > 3 files changed, 15 insertions(+), 22 deletions(-) > > diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c > index 8c10d9abc239..35c2988caf29 100644 > --- a/arch/x86/net/bpf_jit_comp.c > +++ b/arch/x86/net/bpf_jit_comp.c > @@ -391,8 +391,8 @@ static int emit_jump(u8 **pprog, void *func, void *ip) > return emit_patch(pprog, func, ip, 0xE9); > } > > -static int __bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, > - void *old_addr, void *new_addr) > +int __bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, > + void *old_addr, void *new_addr) > { > const u8 *nop_insn = x86_nops[5]; > u8 old_insn[X86_PATCH_SIZE]; > diff --git a/include/linux/bpf.h b/include/linux/bpf.h > index 6762dac3ef76..c28a8563e845 100644 > --- a/include/linux/bpf.h > +++ b/include/linux/bpf.h > @@ -3174,6 +3174,8 @@ enum bpf_text_poke_type { > > int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, > void *addr1, void *addr2); > +int __bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, > + void *old_addr, void *new_addr); > > void *bpf_arch_text_copy(void *dst, void *src, size_t len); > int bpf_arch_text_invalidate(void *dst, size_t len); > diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c > index 2058e89b5ddd..0b5afa2ec17a 100644 > --- a/kernel/bpf/arraymap.c > +++ b/kernel/bpf/arraymap.c > @@ -1044,20 +1044,11 @@ static void prog_array_map_poke_run(struct bpf_map *map, u32 key, > * activated, so tail call updates can arrive from here > * while JIT is still finishing its final fixup for > * non-activated poke entries. > - * 3) On program teardown, the program's kallsym entry gets > - * removed out of RCU callback, but we can only untrack > - * from sleepable context, therefore bpf_arch_text_poke() > - * might not see that this is in BPF text section and > - * bails out with -EINVAL. As these are unreachable since > - * RCU grace period already passed, we simply skip them. > - * 4) Also programs reaching refcount of zero while patching > + * 3) Also programs reaching refcount of zero while patching > * is in progress is okay since we're protected under > * poke_mutex and untrack the programs before the JIT > - * buffer is freed. When we're still in the middle of > - * patching and suddenly kallsyms entry of the program > - * gets evicted, we just skip the rest which is fine due > - * to point 3). > - * 5) Any other error happening below from bpf_arch_text_poke() > + * buffer is freed. > + * 4) Any error happening below from __bpf_arch_text_poke() > * is a unexpected bug. > */ > if (!READ_ONCE(poke->tailcall_target_stable)) > @@ -1073,33 +1064,33 @@ static void prog_array_map_poke_run(struct bpf_map *map, u32 key, > new_addr = new ? (u8 *)new->bpf_func + poke->adj_off : NULL; > > if (new) { > - ret = bpf_arch_text_poke(poke->tailcall_target, > + ret = __bpf_arch_text_poke(poke->tailcall_target, > BPF_MOD_JUMP, > old_addr, new_addr); > - BUG_ON(ret < 0 && ret != -EINVAL); > + BUG_ON(ret < 0); > if (!old) { > - ret = bpf_arch_text_poke(poke->tailcall_bypass, > + ret = __bpf_arch_text_poke(poke->tailcall_bypass, > BPF_MOD_JUMP, > poke->bypass_addr, > NULL); > - BUG_ON(ret < 0 && ret != -EINVAL); > + BUG_ON(ret < 0); > } > } else { > - ret = bpf_arch_text_poke(poke->tailcall_bypass, > + ret = __bpf_arch_text_poke(poke->tailcall_bypass, > BPF_MOD_JUMP, > old_bypass_addr, > poke->bypass_addr); > - BUG_ON(ret < 0 && ret != -EINVAL); > + BUG_ON(ret < 0); > /* let other CPUs finish the execution of program > * so that it will not possible to expose them > * to invalid nop, stack unwind, nop state > */ > if (!ret) > synchronize_rcu(); > - ret = bpf_arch_text_poke(poke->tailcall_target, > + ret = __bpf_arch_text_poke(poke->tailcall_target, > BPF_MOD_JUMP, > old_addr, NULL); > - BUG_ON(ret < 0 && ret != -EINVAL); > + BUG_ON(ret < 0); > } > } > } > -- > 2.43.0 >
Hi Jiri, kernel test robot noticed the following build errors: [auto build test ERROR on bpf/master] url: https://github.com/intel-lab-lkp/linux/commits/Jiri-Olsa/bpf-x64-Fix-prog_array_map_poke_run-map-poke-update/20231127-174900 base: https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf.git master patch link: https://lore.kernel.org/r/20231127094525.1366740-1-jolsa%40kernel.org patch subject: [PATCH bpf] bpf, x64: Fix prog_array_map_poke_run map poke update config: parisc-randconfig-r071-20231127 (https://download.01.org/0day-ci/archive/20231127/202311272311.WsiMBsbq-lkp@intel.com/config) compiler: hppa-linux-gcc (GCC) 13.2.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20231127/202311272311.WsiMBsbq-lkp@intel.com/reproduce) If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <lkp@intel.com> | Closes: https://lore.kernel.org/oe-kbuild-all/202311272311.WsiMBsbq-lkp@intel.com/ All errors (new ones prefixed by >>): hppa-linux-ld: kernel/bpf/arraymap.o: in function `prog_array_map_poke_run': >> (.text+0x103c): undefined reference to `__bpf_arch_text_poke' >> hppa-linux-ld: (.text+0x106c): undefined reference to `__bpf_arch_text_poke' hppa-linux-ld: (.text+0x1090): undefined reference to `__bpf_arch_text_poke' hppa-linux-ld: (.text+0x10c0): undefined reference to `__bpf_arch_text_poke'
Hi Jiri, kernel test robot noticed the following build errors: [auto build test ERROR on bpf/master] url: https://github.com/intel-lab-lkp/linux/commits/Jiri-Olsa/bpf-x64-Fix-prog_array_map_poke_run-map-poke-update/20231127-174900 base: https://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf.git master patch link: https://lore.kernel.org/r/20231127094525.1366740-1-jolsa%40kernel.org patch subject: [PATCH bpf] bpf, x64: Fix prog_array_map_poke_run map poke update config: i386-randconfig-061-20231127 (https://download.01.org/0day-ci/archive/20231127/202311272245.sevnkuSF-lkp@intel.com/config) compiler: gcc-9 (Debian 9.3.0-22) 9.3.0 reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20231127/202311272245.sevnkuSF-lkp@intel.com/reproduce) If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags | Reported-by: kernel test robot <lkp@intel.com> | Closes: https://lore.kernel.org/oe-kbuild-all/202311272245.sevnkuSF-lkp@intel.com/ All errors (new ones prefixed by >>): ld: kernel/bpf/arraymap.o: in function `prog_array_map_poke_run': >> kernel/bpf/arraymap.c:1067: undefined reference to `__bpf_arch_text_poke' >> ld: kernel/bpf/arraymap.c:1079: undefined reference to `__bpf_arch_text_poke' ld: kernel/bpf/arraymap.c:1090: undefined reference to `__bpf_arch_text_poke' ld: kernel/bpf/arraymap.c:1072: undefined reference to `__bpf_arch_text_poke' vim +1067 kernel/bpf/arraymap.c 1014 1015 static void prog_array_map_poke_run(struct bpf_map *map, u32 key, 1016 struct bpf_prog *old, 1017 struct bpf_prog *new) 1018 { 1019 u8 *old_addr, *new_addr, *old_bypass_addr; 1020 struct prog_poke_elem *elem; 1021 struct bpf_array_aux *aux; 1022 1023 aux = container_of(map, struct bpf_array, map)->aux; 1024 WARN_ON_ONCE(!mutex_is_locked(&aux->poke_mutex)); 1025 1026 list_for_each_entry(elem, &aux->poke_progs, list) { 1027 struct bpf_jit_poke_descriptor *poke; 1028 int i, ret; 1029 1030 for (i = 0; i < elem->aux->size_poke_tab; i++) { 1031 poke = &elem->aux->poke_tab[i]; 1032 1033 /* Few things to be aware of: 1034 * 1035 * 1) We can only ever access aux in this context, but 1036 * not aux->prog since it might not be stable yet and 1037 * there could be danger of use after free otherwise. 1038 * 2) Initially when we start tracking aux, the program 1039 * is not JITed yet and also does not have a kallsyms 1040 * entry. We skip these as poke->tailcall_target_stable 1041 * is not active yet. The JIT will do the final fixup 1042 * before setting it stable. The various 1043 * poke->tailcall_target_stable are successively 1044 * activated, so tail call updates can arrive from here 1045 * while JIT is still finishing its final fixup for 1046 * non-activated poke entries. 1047 * 3) Also programs reaching refcount of zero while patching 1048 * is in progress is okay since we're protected under 1049 * poke_mutex and untrack the programs before the JIT 1050 * buffer is freed. 1051 * 4) Any error happening below from __bpf_arch_text_poke() 1052 * is a unexpected bug. 1053 */ 1054 if (!READ_ONCE(poke->tailcall_target_stable)) 1055 continue; 1056 if (poke->reason != BPF_POKE_REASON_TAIL_CALL) 1057 continue; 1058 if (poke->tail_call.map != map || 1059 poke->tail_call.key != key) 1060 continue; 1061 1062 old_bypass_addr = old ? NULL : poke->bypass_addr; 1063 old_addr = old ? (u8 *)old->bpf_func + poke->adj_off : NULL; 1064 new_addr = new ? (u8 *)new->bpf_func + poke->adj_off : NULL; 1065 1066 if (new) { > 1067 ret = __bpf_arch_text_poke(poke->tailcall_target, 1068 BPF_MOD_JUMP, 1069 old_addr, new_addr); 1070 BUG_ON(ret < 0); 1071 if (!old) { 1072 ret = __bpf_arch_text_poke(poke->tailcall_bypass, 1073 BPF_MOD_JUMP, 1074 poke->bypass_addr, 1075 NULL); 1076 BUG_ON(ret < 0); 1077 } 1078 } else { > 1079 ret = __bpf_arch_text_poke(poke->tailcall_bypass, 1080 BPF_MOD_JUMP, 1081 old_bypass_addr, 1082 poke->bypass_addr); 1083 BUG_ON(ret < 0); 1084 /* let other CPUs finish the execution of program 1085 * so that it will not possible to expose them 1086 * to invalid nop, stack unwind, nop state 1087 */ 1088 if (!ret) 1089 synchronize_rcu(); 1090 ret = __bpf_arch_text_poke(poke->tailcall_target, 1091 BPF_MOD_JUMP, 1092 old_addr, NULL); 1093 BUG_ON(ret < 0); 1094 } 1095 } 1096 } 1097 } 1098
diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c index 8c10d9abc239..35c2988caf29 100644 --- a/arch/x86/net/bpf_jit_comp.c +++ b/arch/x86/net/bpf_jit_comp.c @@ -391,8 +391,8 @@ static int emit_jump(u8 **pprog, void *func, void *ip) return emit_patch(pprog, func, ip, 0xE9); } -static int __bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, - void *old_addr, void *new_addr) +int __bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, + void *old_addr, void *new_addr) { const u8 *nop_insn = x86_nops[5]; u8 old_insn[X86_PATCH_SIZE]; diff --git a/include/linux/bpf.h b/include/linux/bpf.h index 6762dac3ef76..c28a8563e845 100644 --- a/include/linux/bpf.h +++ b/include/linux/bpf.h @@ -3174,6 +3174,8 @@ enum bpf_text_poke_type { int bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, void *addr1, void *addr2); +int __bpf_arch_text_poke(void *ip, enum bpf_text_poke_type t, + void *old_addr, void *new_addr); void *bpf_arch_text_copy(void *dst, void *src, size_t len); int bpf_arch_text_invalidate(void *dst, size_t len); diff --git a/kernel/bpf/arraymap.c b/kernel/bpf/arraymap.c index 2058e89b5ddd..0b5afa2ec17a 100644 --- a/kernel/bpf/arraymap.c +++ b/kernel/bpf/arraymap.c @@ -1044,20 +1044,11 @@ static void prog_array_map_poke_run(struct bpf_map *map, u32 key, * activated, so tail call updates can arrive from here * while JIT is still finishing its final fixup for * non-activated poke entries. - * 3) On program teardown, the program's kallsym entry gets - * removed out of RCU callback, but we can only untrack - * from sleepable context, therefore bpf_arch_text_poke() - * might not see that this is in BPF text section and - * bails out with -EINVAL. As these are unreachable since - * RCU grace period already passed, we simply skip them. - * 4) Also programs reaching refcount of zero while patching + * 3) Also programs reaching refcount of zero while patching * is in progress is okay since we're protected under * poke_mutex and untrack the programs before the JIT - * buffer is freed. When we're still in the middle of - * patching and suddenly kallsyms entry of the program - * gets evicted, we just skip the rest which is fine due - * to point 3). - * 5) Any other error happening below from bpf_arch_text_poke() + * buffer is freed. + * 4) Any error happening below from __bpf_arch_text_poke() * is a unexpected bug. */ if (!READ_ONCE(poke->tailcall_target_stable)) @@ -1073,33 +1064,33 @@ static void prog_array_map_poke_run(struct bpf_map *map, u32 key, new_addr = new ? (u8 *)new->bpf_func + poke->adj_off : NULL; if (new) { - ret = bpf_arch_text_poke(poke->tailcall_target, + ret = __bpf_arch_text_poke(poke->tailcall_target, BPF_MOD_JUMP, old_addr, new_addr); - BUG_ON(ret < 0 && ret != -EINVAL); + BUG_ON(ret < 0); if (!old) { - ret = bpf_arch_text_poke(poke->tailcall_bypass, + ret = __bpf_arch_text_poke(poke->tailcall_bypass, BPF_MOD_JUMP, poke->bypass_addr, NULL); - BUG_ON(ret < 0 && ret != -EINVAL); + BUG_ON(ret < 0); } } else { - ret = bpf_arch_text_poke(poke->tailcall_bypass, + ret = __bpf_arch_text_poke(poke->tailcall_bypass, BPF_MOD_JUMP, old_bypass_addr, poke->bypass_addr); - BUG_ON(ret < 0 && ret != -EINVAL); + BUG_ON(ret < 0); /* let other CPUs finish the execution of program * so that it will not possible to expose them * to invalid nop, stack unwind, nop state */ if (!ret) synchronize_rcu(); - ret = bpf_arch_text_poke(poke->tailcall_target, + ret = __bpf_arch_text_poke(poke->tailcall_target, BPF_MOD_JUMP, old_addr, NULL); - BUG_ON(ret < 0 && ret != -EINVAL); + BUG_ON(ret < 0); } } }