Message ID | 20200630171240.2523722-1-yhs@fb.com |
---|---|
State | Accepted |
Delegated to: | BPF Maintainers |
Headers | show |
Series | bpf: fix an incorrect branch elimination by verifier | expand |
Yonghong Song wrote: > Wenbo reported an issue in [1] where a checking of null > pointer is evaluated as always false. In this particular > case, the program type is tp_btf and the pointer to > compare is a PTR_TO_BTF_ID. > > The current verifier considers PTR_TO_BTF_ID always > reprents a non-null pointer, hence all PTR_TO_BTF_ID compares > to 0 will be evaluated as always not-equal, which resulted > in the branch elimination. > > For example, > struct bpf_fentry_test_t { > struct bpf_fentry_test_t *a; > }; > int BPF_PROG(test7, struct bpf_fentry_test_t *arg) > { > if (arg == 0) > test7_result = 1; > return 0; > } > int BPF_PROG(test8, struct bpf_fentry_test_t *arg) > { > if (arg->a == 0) > test8_result = 1; > return 0; > } > > In above bpf programs, both branch arg == 0 and arg->a == 0 > are removed. This may not be what developer expected. > > The bug is introduced by Commit cac616db39c2 ("bpf: Verifier > track null pointer branch_taken with JNE and JEQ"), > where PTR_TO_BTF_ID is considered to be non-null when evaluting > pointer vs. scalar comparison. This may be added > considering we have PTR_TO_BTF_ID_OR_NULL in the verifier > as well. > > PTR_TO_BTF_ID_OR_NULL is added to explicitly requires > a non-NULL testing in selective cases. The current generic > pointer tracing framework in verifier always > assigns PTR_TO_BTF_ID so users does not need to > check NULL pointer at every pointer level like a->b->c->d. Thanks for fixing this. But, don't we really need to check for null? I'm trying to understand how we can avoid the check. If b is NULL above we will have a problem no? Also, we probably shouldn't name the type PTR_TO_BTF_ID if it can be NULL. How about renaming it in bpf-next then although it will be code churn... Or just fix the comments? Probably bpf-next content though. wdyt? In my opinion the comments and type names are really misleading as it stands. diff --git a/include/linux/bpf.h b/include/linux/bpf.h index 3d2ade703a35..18051440f886 100644 --- a/include/linux/bpf.h +++ b/include/linux/bpf.h @@ -337,7 +337,7 @@ enum bpf_reg_type { PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */ PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */ PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */ - PTR_TO_BTF_ID, /* reg points to kernel struct */ + PTR_TO_BTF_ID, /* reg points to kernel struct or NULL */ PTR_TO_BTF_ID_OR_NULL, /* reg points to kernel struct or NULL */ PTR_TO_MEM, /* reg points to valid memory region */ PTR_TO_MEM_OR_NULL, /* reg points to valid memory region or NULL */ diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 7de98906ddf4..7412f9d2f0b5 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -500,7 +500,7 @@ static const char * const reg_type_str[] = { [PTR_TO_TCP_SOCK_OR_NULL] = "tcp_sock_or_null", [PTR_TO_TP_BUFFER] = "tp_buffer", [PTR_TO_XDP_SOCK] = "xdp_sock", - [PTR_TO_BTF_ID] = "ptr_", + [PTR_TO_BTF_ID] = "ptr_or_null_", [PTR_TO_BTF_ID_OR_NULL] = "ptr_or_null_", [PTR_TO_MEM] = "mem", [PTR_TO_MEM_OR_NULL] = "mem_or_null", > > We may not want to assign every PTR_TO_BTF_ID as > PTR_TO_BTF_ID_OR_NULL as this will require a null test > before pointer dereference which may cause inconvenience > for developers. But we could avoid branch elimination > to preserve original code intention. > > This patch simply removed PTR_TO_BTD_ID from reg_type_not_null() > in verifier, which prevented the above branches from being eliminated. > > [1]: https://lore.kernel.org/bpf/79dbb7c0-449d-83eb-5f4f-7af0cc269168@fb.com/T/ > > Fixes: cac616db39c2 ("bpf: Verifier track null pointer branch_taken with JNE and JEQ") > Cc: Andrii Nakryiko <andriin@fb.com> > Cc: John Fastabend <john.fastabend@gmail.com> > Cc: Wenbo Zhang <ethercflow@gmail.com> > Signed-off-by: Yonghong Song <yhs@fb.com> > --- > kernel/bpf/verifier.c | 3 +-- > 1 file changed, 1 insertion(+), 2 deletions(-) > > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c > index 8911d0576399..94cead5a43e5 100644 > --- a/kernel/bpf/verifier.c > +++ b/kernel/bpf/verifier.c > @@ -399,8 +399,7 @@ static bool reg_type_not_null(enum bpf_reg_type type) > return type == PTR_TO_SOCKET || > type == PTR_TO_TCP_SOCK || > type == PTR_TO_MAP_VALUE || > - type == PTR_TO_SOCK_COMMON || > - type == PTR_TO_BTF_ID; > + type == PTR_TO_SOCK_COMMON; > } > > static bool reg_type_may_be_null(enum bpf_reg_type type) > -- > 2.24.1 >
On 6/30/20 10:51 AM, John Fastabend wrote: > Yonghong Song wrote: >> Wenbo reported an issue in [1] where a checking of null >> pointer is evaluated as always false. In this particular >> case, the program type is tp_btf and the pointer to >> compare is a PTR_TO_BTF_ID. >> >> The current verifier considers PTR_TO_BTF_ID always >> reprents a non-null pointer, hence all PTR_TO_BTF_ID compares >> to 0 will be evaluated as always not-equal, which resulted >> in the branch elimination. >> >> For example, >> struct bpf_fentry_test_t { >> struct bpf_fentry_test_t *a; >> }; >> int BPF_PROG(test7, struct bpf_fentry_test_t *arg) >> { >> if (arg == 0) >> test7_result = 1; >> return 0; >> } >> int BPF_PROG(test8, struct bpf_fentry_test_t *arg) >> { >> if (arg->a == 0) >> test8_result = 1; >> return 0; >> } >> >> In above bpf programs, both branch arg == 0 and arg->a == 0 >> are removed. This may not be what developer expected. >> >> The bug is introduced by Commit cac616db39c2 ("bpf: Verifier >> track null pointer branch_taken with JNE and JEQ"), >> where PTR_TO_BTF_ID is considered to be non-null when evaluting >> pointer vs. scalar comparison. This may be added >> considering we have PTR_TO_BTF_ID_OR_NULL in the verifier >> as well. >> >> PTR_TO_BTF_ID_OR_NULL is added to explicitly requires >> a non-NULL testing in selective cases. The current generic >> pointer tracing framework in verifier always >> assigns PTR_TO_BTF_ID so users does not need to >> check NULL pointer at every pointer level like a->b->c->d. > > Thanks for fixing this. > > But, don't we really need to check for null? I'm trying to > understand how we can avoid the check. If b is NULL above > we will have a problem no? It depends with particular data structure. If users are sure once pointer 'a' is valid and a->b, a->b->c, a->b->c are all valid pointers, user may just write a->b->c->d. this happens to some bcc scripts. So non-null pointer is checked. But if user thinks a->b->c is null, he may write type *p = a->b->c; if (p) p->d; Or user just takes advantage of kernel bpf guarded exception handling and do a->b->c->d even if a->b->c could be null. if the result is 0, it means a->b->c is null or major fault, otherwise it is not 0. > > Also, we probably shouldn't name the type PTR_TO_BTF_ID if > it can be NULL. How about renaming it in bpf-next then although > it will be code churn... Or just fix the comments? Probably > bpf-next content though. wdyt? In my opinion the comments and > type names are really misleading as it stands. So PTR_TO_BTF_ID actually means it may be null but not checking is enforced and pointer tracing is always allowed. PTR_TO_BTF_ID_OR_NULL means it may be null and checking against NULL is needed to allow further pointer tracing. To avoid code churn, we can add these comments in bpf-next. > > diff --git a/include/linux/bpf.h b/include/linux/bpf.h > index 3d2ade703a35..18051440f886 100644 > --- a/include/linux/bpf.h > +++ b/include/linux/bpf.h > @@ -337,7 +337,7 @@ enum bpf_reg_type { > PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */ > PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */ > PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */ > - PTR_TO_BTF_ID, /* reg points to kernel struct */ > + PTR_TO_BTF_ID, /* reg points to kernel struct or NULL */ > PTR_TO_BTF_ID_OR_NULL, /* reg points to kernel struct or NULL */ > PTR_TO_MEM, /* reg points to valid memory region */ > PTR_TO_MEM_OR_NULL, /* reg points to valid memory region or NULL */ > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c > index 7de98906ddf4..7412f9d2f0b5 100644 > --- a/kernel/bpf/verifier.c > +++ b/kernel/bpf/verifier.c > @@ -500,7 +500,7 @@ static const char * const reg_type_str[] = { > [PTR_TO_TCP_SOCK_OR_NULL] = "tcp_sock_or_null", > [PTR_TO_TP_BUFFER] = "tp_buffer", > [PTR_TO_XDP_SOCK] = "xdp_sock", > - [PTR_TO_BTF_ID] = "ptr_", > + [PTR_TO_BTF_ID] = "ptr_or_null_", > [PTR_TO_BTF_ID_OR_NULL] = "ptr_or_null_", > [PTR_TO_MEM] = "mem", > [PTR_TO_MEM_OR_NULL] = "mem_or_null", > >> >> We may not want to assign every PTR_TO_BTF_ID as >> PTR_TO_BTF_ID_OR_NULL as this will require a null test >> before pointer dereference which may cause inconvenience >> for developers. But we could avoid branch elimination >> to preserve original code intention. >> >> This patch simply removed PTR_TO_BTD_ID from reg_type_not_null() >> in verifier, which prevented the above branches from being eliminated. >> >> [1]: https://lore.kernel.org/bpf/79dbb7c0-449d-83eb-5f4f-7af0cc269168@fb.com/T/ >> >> Fixes: cac616db39c2 ("bpf: Verifier track null pointer branch_taken with JNE and JEQ") >> Cc: Andrii Nakryiko <andriin@fb.com> >> Cc: John Fastabend <john.fastabend@gmail.com> >> Cc: Wenbo Zhang <ethercflow@gmail.com> >> Signed-off-by: Yonghong Song <yhs@fb.com> >> --- >> kernel/bpf/verifier.c | 3 +-- >> 1 file changed, 1 insertion(+), 2 deletions(-) >> >> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c >> index 8911d0576399..94cead5a43e5 100644 >> --- a/kernel/bpf/verifier.c >> +++ b/kernel/bpf/verifier.c >> @@ -399,8 +399,7 @@ static bool reg_type_not_null(enum bpf_reg_type type) >> return type == PTR_TO_SOCKET || >> type == PTR_TO_TCP_SOCK || >> type == PTR_TO_MAP_VALUE || >> - type == PTR_TO_SOCK_COMMON || >> - type == PTR_TO_BTF_ID; >> + type == PTR_TO_SOCK_COMMON; >> } >> >> static bool reg_type_may_be_null(enum bpf_reg_type type) >> -- >> 2.24.1 >>
Yonghong Song wrote: > > > On 6/30/20 10:51 AM, John Fastabend wrote: > > Yonghong Song wrote: > >> Wenbo reported an issue in [1] where a checking of null > >> pointer is evaluated as always false. In this particular > >> case, the program type is tp_btf and the pointer to > >> compare is a PTR_TO_BTF_ID. > >> > >> The current verifier considers PTR_TO_BTF_ID always > >> reprents a non-null pointer, hence all PTR_TO_BTF_ID compares > >> to 0 will be evaluated as always not-equal, which resulted > >> in the branch elimination. > >> > >> For example, > >> struct bpf_fentry_test_t { > >> struct bpf_fentry_test_t *a; > >> }; > >> int BPF_PROG(test7, struct bpf_fentry_test_t *arg) > >> { > >> if (arg == 0) > >> test7_result = 1; > >> return 0; > >> } > >> int BPF_PROG(test8, struct bpf_fentry_test_t *arg) > >> { > >> if (arg->a == 0) > >> test8_result = 1; > >> return 0; > >> } > >> > >> In above bpf programs, both branch arg == 0 and arg->a == 0 > >> are removed. This may not be what developer expected. > >> > >> The bug is introduced by Commit cac616db39c2 ("bpf: Verifier > >> track null pointer branch_taken with JNE and JEQ"), > >> where PTR_TO_BTF_ID is considered to be non-null when evaluting > >> pointer vs. scalar comparison. This may be added > >> considering we have PTR_TO_BTF_ID_OR_NULL in the verifier > >> as well. > >> > >> PTR_TO_BTF_ID_OR_NULL is added to explicitly requires > >> a non-NULL testing in selective cases. The current generic > >> pointer tracing framework in verifier always > >> assigns PTR_TO_BTF_ID so users does not need to > >> check NULL pointer at every pointer level like a->b->c->d. > > > > Thanks for fixing this. > > > > But, don't we really need to check for null? I'm trying to > > understand how we can avoid the check. If b is NULL above > > we will have a problem no? > > It depends with particular data structure. > If users are sure once pointer 'a' is valid and a->b, a->b->c, a->b->c > are all valid pointers, user may just write a->b->c->d. this happens > to some bcc scripts. So non-null pointer is checked. > > But if user thinks a->b->c is null, he may write > type *p = a->b->c; > if (p) > p->d; > > Or user just takes advantage of kernel bpf guarded exception handling > and do a->b->c->d even if a->b->c could be null. > if the result is 0, it means a->b->c is null or major fault, > otherwise it is not 0. OK. > > > > > Also, we probably shouldn't name the type PTR_TO_BTF_ID if > > it can be NULL. How about renaming it in bpf-next then although > > it will be code churn... Or just fix the comments? Probably > > bpf-next content though. wdyt? In my opinion the comments and > > type names are really misleading as it stands. > > So PTR_TO_BTF_ID actually means it may be null but not checking > is enforced and pointer tracing is always allowed. > PTR_TO_BTF_ID_OR_NULL means it may be null and checking against > NULL is needed to allow further pointer tracing. > > To avoid code churn, we can add these comments in bpf-next. Agreed code churn would be not worth changing type but I'll send some patches for the comment changes. > > > > > diff --git a/include/linux/bpf.h b/include/linux/bpf.h > > index 3d2ade703a35..18051440f886 100644 > > --- a/include/linux/bpf.h > > +++ b/include/linux/bpf.h > > @@ -337,7 +337,7 @@ enum bpf_reg_type { > > PTR_TO_TCP_SOCK_OR_NULL, /* reg points to struct tcp_sock or NULL */ > > PTR_TO_TP_BUFFER, /* reg points to a writable raw tp's buffer */ > > PTR_TO_XDP_SOCK, /* reg points to struct xdp_sock */ > > - PTR_TO_BTF_ID, /* reg points to kernel struct */ > > + PTR_TO_BTF_ID, /* reg points to kernel struct or NULL */ > > PTR_TO_BTF_ID_OR_NULL, /* reg points to kernel struct or NULL */ > > PTR_TO_MEM, /* reg points to valid memory region */ > > PTR_TO_MEM_OR_NULL, /* reg points to valid memory region or NULL */ > > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c > > index 7de98906ddf4..7412f9d2f0b5 100644 > > --- a/kernel/bpf/verifier.c > > +++ b/kernel/bpf/verifier.c > > @@ -500,7 +500,7 @@ static const char * const reg_type_str[] = { > > [PTR_TO_TCP_SOCK_OR_NULL] = "tcp_sock_or_null", > > [PTR_TO_TP_BUFFER] = "tp_buffer", > > [PTR_TO_XDP_SOCK] = "xdp_sock", > > - [PTR_TO_BTF_ID] = "ptr_", > > + [PTR_TO_BTF_ID] = "ptr_or_null_", > > [PTR_TO_BTF_ID_OR_NULL] = "ptr_or_null_", > > [PTR_TO_MEM] = "mem", > > [PTR_TO_MEM_OR_NULL] = "mem_or_null", > > > >> > >> We may not want to assign every PTR_TO_BTF_ID as > >> PTR_TO_BTF_ID_OR_NULL as this will require a null test > >> before pointer dereference which may cause inconvenience > >> for developers. But we could avoid branch elimination > >> to preserve original code intention. > >> > >> This patch simply removed PTR_TO_BTD_ID from reg_type_not_null() > >> in verifier, which prevented the above branches from being eliminated. > >> > >> [1]: https://lore.kernel.org/bpf/79dbb7c0-449d-83eb-5f4f-7af0cc269168@fb.com/T/ > >> > >> Fixes: cac616db39c2 ("bpf: Verifier track null pointer branch_taken with JNE and JEQ") I added the BTF_ID in v2 of those patches :/ too bad. Thanks again for catching it. > >> Cc: Andrii Nakryiko <andriin@fb.com> > >> Cc: John Fastabend <john.fastabend@gmail.com> > >> Cc: Wenbo Zhang <ethercflow@gmail.com> > >> Signed-off-by: Yonghong Song <yhs@fb.com> > >> --- Acked-by: John Fastabend <john.fastabend@gmail.com> > >> kernel/bpf/verifier.c | 3 +-- > >> 1 file changed, 1 insertion(+), 2 deletions(-) > >> > >> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c > >> index 8911d0576399..94cead5a43e5 100644 > >> --- a/kernel/bpf/verifier.c > >> +++ b/kernel/bpf/verifier.c > >> @@ -399,8 +399,7 @@ static bool reg_type_not_null(enum bpf_reg_type type) > >> return type == PTR_TO_SOCKET || > >> type == PTR_TO_TCP_SOCK || > >> type == PTR_TO_MAP_VALUE || > >> - type == PTR_TO_SOCK_COMMON || > >> - type == PTR_TO_BTF_ID; > >> + type == PTR_TO_SOCK_COMMON; > >> } > >> > >> static bool reg_type_may_be_null(enum bpf_reg_type type) > >> -- > >> 2.24.1 > >>
On Tue, Jun 30, 2020 at 11:36 AM John Fastabend <john.fastabend@gmail.com> wrote: > > > > > > Also, we probably shouldn't name the type PTR_TO_BTF_ID if > > > it can be NULL. How about renaming it in bpf-next then although > > > it will be code churn... Or just fix the comments? Probably > > > bpf-next content though. wdyt? In my opinion the comments and > > > type names are really misleading as it stands. > > > > So PTR_TO_BTF_ID actually means it may be null but not checking > > is enforced and pointer tracing is always allowed. > > PTR_TO_BTF_ID_OR_NULL means it may be null and checking against > > NULL is needed to allow further pointer tracing. > > > > To avoid code churn, we can add these comments in bpf-next. > > Agreed code churn would be not worth changing type but I'll send > some patches for the comment changes. +1 I think for bpf tree the minimal fix is better. So I've applied this set. A follow up to bpf-next after bpf->net->linus->net-next->bpf-next would be really good. We'll make sure that all trees converge soon.
On Tue, Jun 30, 2020 at 11:46 AM Yonghong Song <yhs@fb.com> wrote: > > Wenbo reported an issue in [1] where a checking of null > pointer is evaluated as always false. In this particular > case, the program type is tp_btf and the pointer to > compare is a PTR_TO_BTF_ID. > > The current verifier considers PTR_TO_BTF_ID always > reprents a non-null pointer, hence all PTR_TO_BTF_ID compares > to 0 will be evaluated as always not-equal, which resulted > in the branch elimination. > > For example, > struct bpf_fentry_test_t { > struct bpf_fentry_test_t *a; > }; > int BPF_PROG(test7, struct bpf_fentry_test_t *arg) > { > if (arg == 0) > test7_result = 1; > return 0; > } > int BPF_PROG(test8, struct bpf_fentry_test_t *arg) > { > if (arg->a == 0) > test8_result = 1; > return 0; > } > > In above bpf programs, both branch arg == 0 and arg->a == 0 > are removed. This may not be what developer expected. > > The bug is introduced by Commit cac616db39c2 ("bpf: Verifier > track null pointer branch_taken with JNE and JEQ"), > where PTR_TO_BTF_ID is considered to be non-null when evaluting > pointer vs. scalar comparison. This may be added > considering we have PTR_TO_BTF_ID_OR_NULL in the verifier > as well. > > PTR_TO_BTF_ID_OR_NULL is added to explicitly requires > a non-NULL testing in selective cases. The current generic > pointer tracing framework in verifier always > assigns PTR_TO_BTF_ID so users does not need to > check NULL pointer at every pointer level like a->b->c->d. > > We may not want to assign every PTR_TO_BTF_ID as > PTR_TO_BTF_ID_OR_NULL as this will require a null test > before pointer dereference which may cause inconvenience > for developers. But we could avoid branch elimination > to preserve original code intention. > > This patch simply removed PTR_TO_BTD_ID from reg_type_not_null() > in verifier, which prevented the above branches from being eliminated. > > [1]: https://lore.kernel.org/bpf/79dbb7c0-449d-83eb-5f4f-7af0cc269168@fb.com/T/ > > Fixes: cac616db39c2 ("bpf: Verifier track null pointer branch_taken with JNE and JEQ") > Cc: Andrii Nakryiko <andriin@fb.com> > Cc: John Fastabend <john.fastabend@gmail.com> > Cc: Wenbo Zhang <ethercflow@gmail.com> > Signed-off-by: Yonghong Song <yhs@fb.com> > --- You missed Reported-by: tag, please add. Otherwise lgtm, thanks for fixing! Acked-by: Andrii Nakryiko <andriin@fb.com> > kernel/bpf/verifier.c | 3 +-- > 1 file changed, 1 insertion(+), 2 deletions(-) > > diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c > index 8911d0576399..94cead5a43e5 100644 > --- a/kernel/bpf/verifier.c > +++ b/kernel/bpf/verifier.c > @@ -399,8 +399,7 @@ static bool reg_type_not_null(enum bpf_reg_type type) > return type == PTR_TO_SOCKET || > type == PTR_TO_TCP_SOCK || > type == PTR_TO_MAP_VALUE || > - type == PTR_TO_SOCK_COMMON || > - type == PTR_TO_BTF_ID; > + type == PTR_TO_SOCK_COMMON; > } > > static bool reg_type_may_be_null(enum bpf_reg_type type) > -- > 2.24.1 >
On Tue, Jun 30, 2020 at 12:09 PM John Fastabend <john.fastabend@gmail.com> wrote: > > Yonghong Song wrote: > > Wenbo reported an issue in [1] where a checking of null > > pointer is evaluated as always false. In this particular > > case, the program type is tp_btf and the pointer to > > compare is a PTR_TO_BTF_ID. > > > > The current verifier considers PTR_TO_BTF_ID always > > reprents a non-null pointer, hence all PTR_TO_BTF_ID compares > > to 0 will be evaluated as always not-equal, which resulted > > in the branch elimination. > > > > For example, > > struct bpf_fentry_test_t { > > struct bpf_fentry_test_t *a; > > }; > > int BPF_PROG(test7, struct bpf_fentry_test_t *arg) > > { > > if (arg == 0) > > test7_result = 1; > > return 0; > > } > > int BPF_PROG(test8, struct bpf_fentry_test_t *arg) > > { > > if (arg->a == 0) > > test8_result = 1; > > return 0; > > } > > > > In above bpf programs, both branch arg == 0 and arg->a == 0 > > are removed. This may not be what developer expected. > > > > The bug is introduced by Commit cac616db39c2 ("bpf: Verifier > > track null pointer branch_taken with JNE and JEQ"), > > where PTR_TO_BTF_ID is considered to be non-null when evaluting > > pointer vs. scalar comparison. This may be added > > considering we have PTR_TO_BTF_ID_OR_NULL in the verifier > > as well. > > > > PTR_TO_BTF_ID_OR_NULL is added to explicitly requires > > a non-NULL testing in selective cases. The current generic > > pointer tracing framework in verifier always > > assigns PTR_TO_BTF_ID so users does not need to > > check NULL pointer at every pointer level like a->b->c->d. > > Thanks for fixing this. > > But, don't we really need to check for null? I'm trying to > understand how we can avoid the check. If b is NULL above > we will have a problem no? BPF JIT installs an exception handler for each direct memory access, so that if it turns out to be NULL, it will be caught and 0 will be assigned to the target register and BPF program will continue execution. In that sense, it simulates bpf_probe_read() behavior and significantly improves usability. > > Also, we probably shouldn't name the type PTR_TO_BTF_ID if > it can be NULL. How about renaming it in bpf-next then although > it will be code churn... Or just fix the comments? Probably > bpf-next content though. wdyt? In my opinion the comments and > type names are really misleading as it stands. > [...]
On 6/30/20 9:18 PM, Andrii Nakryiko wrote: > On Tue, Jun 30, 2020 at 11:46 AM Yonghong Song <yhs@fb.com> wrote: >> >> Wenbo reported an issue in [1] where a checking of null >> pointer is evaluated as always false. In this particular >> case, the program type is tp_btf and the pointer to >> compare is a PTR_TO_BTF_ID. >> >> The current verifier considers PTR_TO_BTF_ID always >> reprents a non-null pointer, hence all PTR_TO_BTF_ID compares >> to 0 will be evaluated as always not-equal, which resulted >> in the branch elimination. >> >> For example, >> struct bpf_fentry_test_t { >> struct bpf_fentry_test_t *a; >> }; >> int BPF_PROG(test7, struct bpf_fentry_test_t *arg) >> { >> if (arg == 0) >> test7_result = 1; >> return 0; >> } >> int BPF_PROG(test8, struct bpf_fentry_test_t *arg) >> { >> if (arg->a == 0) >> test8_result = 1; >> return 0; >> } >> >> In above bpf programs, both branch arg == 0 and arg->a == 0 >> are removed. This may not be what developer expected. >> >> The bug is introduced by Commit cac616db39c2 ("bpf: Verifier >> track null pointer branch_taken with JNE and JEQ"), >> where PTR_TO_BTF_ID is considered to be non-null when evaluting >> pointer vs. scalar comparison. This may be added >> considering we have PTR_TO_BTF_ID_OR_NULL in the verifier >> as well. >> >> PTR_TO_BTF_ID_OR_NULL is added to explicitly requires >> a non-NULL testing in selective cases. The current generic >> pointer tracing framework in verifier always >> assigns PTR_TO_BTF_ID so users does not need to >> check NULL pointer at every pointer level like a->b->c->d. >> >> We may not want to assign every PTR_TO_BTF_ID as >> PTR_TO_BTF_ID_OR_NULL as this will require a null test >> before pointer dereference which may cause inconvenience >> for developers. But we could avoid branch elimination >> to preserve original code intention. >> >> This patch simply removed PTR_TO_BTD_ID from reg_type_not_null() >> in verifier, which prevented the above branches from being eliminated. >> >> [1]: https://lore.kernel.org/bpf/79dbb7c0-449d-83eb-5f4f-7af0cc269168@fb.com/T/ >> >> Fixes: cac616db39c2 ("bpf: Verifier track null pointer branch_taken with JNE and JEQ") >> Cc: Andrii Nakryiko <andriin@fb.com> >> Cc: John Fastabend <john.fastabend@gmail.com> >> Cc: Wenbo Zhang <ethercflow@gmail.com> >> Signed-off-by: Yonghong Song <yhs@fb.com> >> --- > > You missed Reported-by: tag, please add. Agree, fixed up manually (and also pulled in your ACKs, Andrii). Thanks, Daniel
diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c index 8911d0576399..94cead5a43e5 100644 --- a/kernel/bpf/verifier.c +++ b/kernel/bpf/verifier.c @@ -399,8 +399,7 @@ static bool reg_type_not_null(enum bpf_reg_type type) return type == PTR_TO_SOCKET || type == PTR_TO_TCP_SOCK || type == PTR_TO_MAP_VALUE || - type == PTR_TO_SOCK_COMMON || - type == PTR_TO_BTF_ID; + type == PTR_TO_SOCK_COMMON; } static bool reg_type_may_be_null(enum bpf_reg_type type)
Wenbo reported an issue in [1] where a checking of null pointer is evaluated as always false. In this particular case, the program type is tp_btf and the pointer to compare is a PTR_TO_BTF_ID. The current verifier considers PTR_TO_BTF_ID always reprents a non-null pointer, hence all PTR_TO_BTF_ID compares to 0 will be evaluated as always not-equal, which resulted in the branch elimination. For example, struct bpf_fentry_test_t { struct bpf_fentry_test_t *a; }; int BPF_PROG(test7, struct bpf_fentry_test_t *arg) { if (arg == 0) test7_result = 1; return 0; } int BPF_PROG(test8, struct bpf_fentry_test_t *arg) { if (arg->a == 0) test8_result = 1; return 0; } In above bpf programs, both branch arg == 0 and arg->a == 0 are removed. This may not be what developer expected. The bug is introduced by Commit cac616db39c2 ("bpf: Verifier track null pointer branch_taken with JNE and JEQ"), where PTR_TO_BTF_ID is considered to be non-null when evaluting pointer vs. scalar comparison. This may be added considering we have PTR_TO_BTF_ID_OR_NULL in the verifier as well. PTR_TO_BTF_ID_OR_NULL is added to explicitly requires a non-NULL testing in selective cases. The current generic pointer tracing framework in verifier always assigns PTR_TO_BTF_ID so users does not need to check NULL pointer at every pointer level like a->b->c->d. We may not want to assign every PTR_TO_BTF_ID as PTR_TO_BTF_ID_OR_NULL as this will require a null test before pointer dereference which may cause inconvenience for developers. But we could avoid branch elimination to preserve original code intention. This patch simply removed PTR_TO_BTD_ID from reg_type_not_null() in verifier, which prevented the above branches from being eliminated. [1]: https://lore.kernel.org/bpf/79dbb7c0-449d-83eb-5f4f-7af0cc269168@fb.com/T/ Fixes: cac616db39c2 ("bpf: Verifier track null pointer branch_taken with JNE and JEQ") Cc: Andrii Nakryiko <andriin@fb.com> Cc: John Fastabend <john.fastabend@gmail.com> Cc: Wenbo Zhang <ethercflow@gmail.com> Signed-off-by: Yonghong Song <yhs@fb.com> --- kernel/bpf/verifier.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)