| Message ID | 20221107092032.178235-4-yangjihong1@huawei.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | BPF |
| Headers | show |
| Series | bpf: Support kernel function call in 32-bit ARM | expand |
On Mon, Nov 7, 2022 at 1:23 AM Yang Jihong <yangjihong1@huawei.com> wrote: > > bpf_core_patch_insn modifies load's mem size from 8 bytes to 4 bytes. > As a result, the bpf check fails, we need to skip adjust mem size to fit > the verifier. > > Signed-off-by: Yang Jihong <yangjihong1@huawei.com> > --- > tools/lib/bpf/libbpf.c | 34 +++++++++++++++++++++++++++++----- > 1 file changed, 29 insertions(+), 5 deletions(-) > > diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c > index 184ce1684dcd..e1c21b631a0b 100644 > --- a/tools/lib/bpf/libbpf.c > +++ b/tools/lib/bpf/libbpf.c > @@ -5634,6 +5634,28 @@ static int bpf_core_resolve_relo(struct bpf_program *prog, > targ_res); > } > > +static bool > +bpf_core_patch_insn_skip(const struct btf *local_btf, const struct bpf_insn *insn, > + const struct bpf_core_relo_res *res) > +{ > + __u8 class; > + const struct btf_type *orig_t; > + > + class = BPF_CLASS(insn->code); > + orig_t = btf_type_by_id(local_btf, res->orig_type_id); > + > + /* > + * verifier has to see a load of a pointer as a 8-byte load, > + * CO_RE should not screws up access, bpf_core_patch_insn modifies > + * load's mem size from 8 bytes to 4 bytes in 32-bit arch, > + * so we skip adjust mem size. > + */ Nope, this is only for BPF UAPI context types like __sk_buff (right now). fentry/fexit/raw_tp_btf programs traversing kernel types and following pointers actually need this to work correctly. Don't do this. > + if (class == BPF_LDX && btf_is_ptr(orig_t)) > + return true; > + > + return false; > +} > + > static int > bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) > { > @@ -5730,11 +5752,13 @@ bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) > goto out; > } > > - err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res); > - if (err) { > - pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n", > - prog->name, i, insn_idx, err); > - goto out; > + if (!bpf_core_patch_insn_skip(obj->btf, insn, &targ_res)) { > + err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res); > + if (err) { > + pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n", > + prog->name, i, insn_idx, err); > + goto out; > + } > } > } > } > -- > 2.30.GIT >
Hello, On 2022/11/8 9:22, Andrii Nakryiko wrote: > On Mon, Nov 7, 2022 at 1:23 AM Yang Jihong <yangjihong1@huawei.com> wrote: >> >> bpf_core_patch_insn modifies load's mem size from 8 bytes to 4 bytes. >> As a result, the bpf check fails, we need to skip adjust mem size to fit >> the verifier. >> >> Signed-off-by: Yang Jihong <yangjihong1@huawei.com> >> --- >> tools/lib/bpf/libbpf.c | 34 +++++++++++++++++++++++++++++----- >> 1 file changed, 29 insertions(+), 5 deletions(-) >> >> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c >> index 184ce1684dcd..e1c21b631a0b 100644 >> --- a/tools/lib/bpf/libbpf.c >> +++ b/tools/lib/bpf/libbpf.c >> @@ -5634,6 +5634,28 @@ static int bpf_core_resolve_relo(struct bpf_program *prog, >> targ_res); >> } >> >> +static bool >> +bpf_core_patch_insn_skip(const struct btf *local_btf, const struct bpf_insn *insn, >> + const struct bpf_core_relo_res *res) >> +{ >> + __u8 class; >> + const struct btf_type *orig_t; >> + >> + class = BPF_CLASS(insn->code); >> + orig_t = btf_type_by_id(local_btf, res->orig_type_id); >> + >> + /* >> + * verifier has to see a load of a pointer as a 8-byte load, >> + * CO_RE should not screws up access, bpf_core_patch_insn modifies >> + * load's mem size from 8 bytes to 4 bytes in 32-bit arch, >> + * so we skip adjust mem size. >> + */ > > Nope, this is only for BPF UAPI context types like __sk_buff (right > now). fentry/fexit/raw_tp_btf programs traversing kernel types and > following pointers actually need this to work correctly. Don't do > this. Distinguishing BPF UAPI context from kernel type requires some work. According to current situation, the solution of patch2 is relatively simple. Thanks, Yang
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 184ce1684dcd..e1c21b631a0b 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -5634,6 +5634,28 @@ static int bpf_core_resolve_relo(struct bpf_program *prog, targ_res); } +static bool +bpf_core_patch_insn_skip(const struct btf *local_btf, const struct bpf_insn *insn, + const struct bpf_core_relo_res *res) +{ + __u8 class; + const struct btf_type *orig_t; + + class = BPF_CLASS(insn->code); + orig_t = btf_type_by_id(local_btf, res->orig_type_id); + + /* + * verifier has to see a load of a pointer as a 8-byte load, + * CO_RE should not screws up access, bpf_core_patch_insn modifies + * load's mem size from 8 bytes to 4 bytes in 32-bit arch, + * so we skip adjust mem size. + */ + if (class == BPF_LDX && btf_is_ptr(orig_t)) + return true; + + return false; +} + static int bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) { @@ -5730,11 +5752,13 @@ bpf_object__relocate_core(struct bpf_object *obj, const char *targ_btf_path) goto out; } - err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res); - if (err) { - pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n", - prog->name, i, insn_idx, err); - goto out; + if (!bpf_core_patch_insn_skip(obj->btf, insn, &targ_res)) { + err = bpf_core_patch_insn(prog->name, insn, insn_idx, rec, i, &targ_res); + if (err) { + pr_warn("prog '%s': relo #%d: failed to patch insn #%u: %d\n", + prog->name, i, insn_idx, err); + goto out; + } } } }
bpf_core_patch_insn modifies load's mem size from 8 bytes to 4 bytes. As a result, the bpf check fails, we need to skip adjust mem size to fit the verifier. Signed-off-by: Yang Jihong <yangjihong1@huawei.com> --- tools/lib/bpf/libbpf.c | 34 +++++++++++++++++++++++++++++----- 1 file changed, 29 insertions(+), 5 deletions(-)