Message ID | 20201205025140.443115-1-andrii@kernel.org (mailing list archive) |
---|---|
State | Superseded |
Delegated to: | BPF |
Headers | show |
Series | [bpf-next] libbpf: support module BTF for BPF_TYPE_ID_TARGET CO-RE relocation | expand |
Context | Check | Description |
---|---|---|
netdev/cover_letter | success | Link |
netdev/fixes_present | success | Link |
netdev/patch_count | success | Link |
netdev/tree_selection | success | Clearly marked for bpf-next |
netdev/subject_prefix | success | Link |
netdev/source_inline | success | Was 0 now: 0 |
netdev/verify_signedoff | success | Link |
netdev/module_param | success | Was 0 now: 0 |
netdev/build_32bit | success | Errors and warnings before: 0 this patch: 0 |
netdev/kdoc | success | Errors and warnings before: 0 this patch: 0 |
netdev/verify_fixes | success | Link |
netdev/checkpatch | success | total: 0 errors, 0 warnings, 0 checks, 50 lines checked |
netdev/build_allmodconfig_warn | success | Errors and warnings before: 0 this patch: 0 |
netdev/header_inline | success | Link |
netdev/stable | success | Stable not CCed |
On Fri, 4 Dec 2020, Andrii Nakryiko wrote: > When Clang emits ldimm64 instruction for BPF_TYPE_ID_TARGET CO-RE relocation, > put module BTF FD, containing target type, into upper 32 bits of imm64. > > Because this FD is internal to libbpf, it's very cumbersome to test this in > selftests. Manual testing was performed with debug log messages sprinkled > across selftests and libbpf, confirming expected values are substituted. > Better testing will be performed as part of the work adding module BTF types > support to bpf_snprintf_btf() helpers. > > Cc: Alan Maguire <alan.maguire@oracle.com> > Signed-off-by: Andrii Nakryiko <andrii@kernel.org> Thanks so much for doing this Andrii! When I tested, I ran into a problem; it turns out when a module struct such as "veth_stats" is used, it's classified as BTF_KIND_FWD, and as a result when we iterate over the modules and look in the veth module, "struct veth_stats" does not match since its module kind (BTF_KIND_STRUCT) does not match the candidate kind (BTF_KIND_FWD). I'm kind of out of my depth here, but the below patch (on top of your patch) worked. However without it - when we find 0 candidate matches - as well as not substituting the module object id/type id - we hit a segfault: Program terminated with signal 11, Segmentation fault. #0 0x0000000000480bf9 in bpf_core_calc_relo (prog=0x4d6ba40, relo=0x4d70e7c, relo_idx=0, local_spec=0x7ffe2cf17b00, targ_spec=0x0, res=0x7ffe2cf17ae0) at libbpf.c:4408 4408 switch (kind) { Missing separate debuginfos, use: debuginfo-install elfutils-libelf-0.172-2.el7.x86_64 glibc-2.17-196.el7.x86_64 libattr-2.4.46-13.el7.x86_64 libcap-2.22-9.el7.x86_64 libgcc-4.8.5-36.0.1.el7_6.2.x86_64 zlib-1.2.7-18.el7.x86_64 (gdb) bt #0 0x0000000000480bf9 in bpf_core_calc_relo (prog=0x4d6ba40, relo=0x4d70e7c, relo_idx=0, local_spec=0x7ffe2cf17b00, targ_spec=0x0, res=0x7ffe2cf17ae0) at libbpf.c:4408 The dereferences of targ_spec in bpf_core_recalc_relo() seem to be the cause; that function is called with a NULL targ_spec when 0 candidates are found, so it's possible we'd need to guard those accesses for cases where a bogus type was passed in and no candidates were found. If the below looks good would it make sense to roll it into your patch or will I add it to my v3 patch series? Thanks again for your help with this! Alan >From 08040730dbff6c5d7636927777ac85a71c10827f Mon Sep 17 00:00:00 2001 From: Alan Maguire <alan.maguire@oracle.com> Date: Sun, 6 Dec 2020 01:10:28 +0100 Subject: [PATCH] libbpf: handle fwd kinds when checking candidate relocations for modules when a struct belonging to a module is being assessed, it will be designated a fwd kind (BTF_KIND_FWD); when matching candidate types constraints on exact type matching need to be relaxed to ensure that such structures are found successfully. Introduce kinds_match() function to handle this comparison. Signed-off-by: Alan Maguire <alan.maguire@oracle.com> --- tools/lib/bpf/libbpf.c | 24 +++++++++++++++++++++--- 1 file changed, 21 insertions(+), 3 deletions(-) diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 539956f..00fdb30 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -4673,6 +4673,24 @@ static void bpf_core_free_cands(struct core_cand_list *cands) free(cands); } +/* module-specific structs will have relo kind set to fwd, so as + * well as handling exact matches, struct has to match fwd kind. + */ +static bool kinds_match(const struct btf_type *type1, + const struct btf_type *type2) +{ + __u8 kind1 = btf_kind(type1); + __u8 kind2 = btf_kind(type2); + + if (kind1 == kind2) + return true; + if (kind1 == BTF_KIND_STRUCT && kind2 == BTF_KIND_FWD) + return true; + if (kind1 == BTF_KIND_FWD && kind2 == BTF_KIND_STRUCT) + return true; + return false; +} + static int bpf_core_add_cands(struct core_cand *local_cand, size_t local_essent_len, const struct btf *targ_btf, @@ -4689,7 +4707,7 @@ static int bpf_core_add_cands(struct core_cand *local_cand, n = btf__get_nr_types(targ_btf); for (i = targ_start_id; i <= n; i++) { t = btf__type_by_id(targ_btf, i); - if (btf_kind(t) != btf_kind(local_cand->t)) + if (!kinds_match(t, local_cand->t)) continue; targ_name = btf__name_by_offset(targ_btf, t->name_off); @@ -5057,7 +5075,7 @@ static int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id /* caller made sure that names match (ignoring flavor suffix) */ local_type = btf__type_by_id(local_btf, local_id); targ_type = btf__type_by_id(targ_btf, targ_id); - if (btf_kind(local_type) != btf_kind(targ_type)) + if (!kinds_match(local_type, targ_type)) return 0; recur: @@ -5070,7 +5088,7 @@ static int bpf_core_types_are_compat(const struct btf *local_btf, __u32 local_id if (!local_type || !targ_type) return -EINVAL; - if (btf_kind(local_type) != btf_kind(targ_type)) + if (!kinds_match(local_type, targ_type)) return 0; switch (btf_kind(local_type)) {
On Fri, 4 Dec 2020, Andrii Nakryiko wrote: > When Clang emits ldimm64 instruction for BPF_TYPE_ID_TARGET CO-RE relocation, > put module BTF FD, containing target type, into upper 32 bits of imm64. > > Because this FD is internal to libbpf, it's very cumbersome to test this in > selftests. Manual testing was performed with debug log messages sprinkled > across selftests and libbpf, confirming expected values are substituted. > Better testing will be performed as part of the work adding module BTF types > support to bpf_snprintf_btf() helpers. > > Cc: Alan Maguire <alan.maguire@oracle.com> > Signed-off-by: Andrii Nakryiko <andrii@kernel.org> > --- > tools/lib/bpf/libbpf.c | 19 ++++++++++++++++--- > 1 file changed, 16 insertions(+), 3 deletions(-) > > diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c > index 9be88a90a4aa..539956f7920a 100644 > --- a/tools/lib/bpf/libbpf.c > +++ b/tools/lib/bpf/libbpf.c > @@ -4795,6 +4795,7 @@ static int load_module_btfs(struct bpf_object *obj) > > mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; > > + btf__set_fd(btf, fd); > mod_btf->btf = btf; > mod_btf->id = id; > mod_btf->fd = fd; > @@ -5445,6 +5446,10 @@ struct bpf_core_relo_res > __u32 orig_type_id; > __u32 new_sz; > __u32 new_type_id; > + /* FD of the module BTF containing the target candidate, or 0 for > + * vmlinux BTF > + */ > + int btf_obj_fd; > }; > > /* Calculate original and target relocation values, given local and target > @@ -5469,6 +5474,7 @@ static int bpf_core_calc_relo(const struct bpf_program *prog, > res->fail_memsz_adjust = false; > res->orig_sz = res->new_sz = 0; > res->orig_type_id = res->new_type_id = 0; > + res->btf_obj_fd = 0; > > if (core_relo_is_field_based(relo->kind)) { > err = bpf_core_calc_field_relo(prog, relo, local_spec, > @@ -5519,6 +5525,9 @@ static int bpf_core_calc_relo(const struct bpf_program *prog, > } else if (core_relo_is_type_based(relo->kind)) { > err = bpf_core_calc_type_relo(relo, local_spec, &res->orig_val); > err = err ?: bpf_core_calc_type_relo(relo, targ_spec, &res->new_val); > + if (!err && relo->kind == BPF_TYPE_ID_TARGET && > + targ_spec->btf != prog->obj->btf_vmlinux) > + res->btf_obj_fd = btf__fd(targ_spec->btf); Sorry about this Andrii, but I'm a bit stuck here. I'm struggling to get tests working where the obj fd is used to designate the module BTF. Unless I'm missing something there are a few problems: - the fd association is removed by libbpf when the BPF program has loaded; the module fds are closed and the module BTF is discarded. However even if that isn't done (and as you mentioned, we could hold onto BTF that is in use, and I commented out the code that does that to test) - there's another problem: - I can't see a way to use the object fd value we set here later in BPF program context; btf_get_by_fd() returns -EBADF as the fd is associated with the module BTF in the test's process context, not necessarily in the context that the BPF program is running. Would it be possible in this case to use object id? Or is there another way to handle the fd->module BTF association that we need to make in BPF program context that I'm missing? - A more long-term issue; if we use fds to specify module BTFs and write the object fd into the program, we can pin the BPF program such that it outlives fds that refer to its associated BTF. So unless we pinned the BTF too, any code that assumed the BTF fd-> module mapping was valid would start to break once the user-space side went away and the pinned program persisted. Maybe there are solutions to these problems that I'm missing of course, but for now I'm not sure how to get things working. Thanks again for your help with this! Alan
On Mon, Dec 07, 2020 at 04:38:16PM +0000, Alan Maguire wrote: > On Fri, 4 Dec 2020, Andrii Nakryiko wrote: > > > When Clang emits ldimm64 instruction for BPF_TYPE_ID_TARGET CO-RE relocation, > > put module BTF FD, containing target type, into upper 32 bits of imm64. > > > > Because this FD is internal to libbpf, it's very cumbersome to test this in > > selftests. Manual testing was performed with debug log messages sprinkled > > across selftests and libbpf, confirming expected values are substituted. > > Better testing will be performed as part of the work adding module BTF types > > support to bpf_snprintf_btf() helpers. > > > > Cc: Alan Maguire <alan.maguire@oracle.com> > > Signed-off-by: Andrii Nakryiko <andrii@kernel.org> > > --- > > tools/lib/bpf/libbpf.c | 19 ++++++++++++++++--- > > 1 file changed, 16 insertions(+), 3 deletions(-) > > > > diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c > > index 9be88a90a4aa..539956f7920a 100644 > > --- a/tools/lib/bpf/libbpf.c > > +++ b/tools/lib/bpf/libbpf.c > > @@ -4795,6 +4795,7 @@ static int load_module_btfs(struct bpf_object *obj) > > > > mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; > > > > + btf__set_fd(btf, fd); > > mod_btf->btf = btf; > > mod_btf->id = id; > > mod_btf->fd = fd; > > @@ -5445,6 +5446,10 @@ struct bpf_core_relo_res > > __u32 orig_type_id; > > __u32 new_sz; > > __u32 new_type_id; > > + /* FD of the module BTF containing the target candidate, or 0 for > > + * vmlinux BTF > > + */ > > + int btf_obj_fd; > > }; > > > > /* Calculate original and target relocation values, given local and target > > @@ -5469,6 +5474,7 @@ static int bpf_core_calc_relo(const struct bpf_program *prog, > > res->fail_memsz_adjust = false; > > res->orig_sz = res->new_sz = 0; > > res->orig_type_id = res->new_type_id = 0; > > + res->btf_obj_fd = 0; > > > > if (core_relo_is_field_based(relo->kind)) { > > err = bpf_core_calc_field_relo(prog, relo, local_spec, > > @@ -5519,6 +5525,9 @@ static int bpf_core_calc_relo(const struct bpf_program *prog, > > } else if (core_relo_is_type_based(relo->kind)) { > > err = bpf_core_calc_type_relo(relo, local_spec, &res->orig_val); > > err = err ?: bpf_core_calc_type_relo(relo, targ_spec, &res->new_val); > > + if (!err && relo->kind == BPF_TYPE_ID_TARGET && > > + targ_spec->btf != prog->obj->btf_vmlinux) > > + res->btf_obj_fd = btf__fd(targ_spec->btf); > > Sorry about this Andrii, but I'm a bit stuck here. > > I'm struggling to get tests working where the obj fd is used to designate > the module BTF. Unless I'm missing something there are a few problems: > > - the fd association is removed by libbpf when the BPF program has loaded; > the module fds are closed and the module BTF is discarded. However even if > that isn't done (and as you mentioned, we could hold onto BTF that is in > use, and I commented out the code that does that to test) - there's > another problem: > - I can't see a way to use the object fd value we set here later in BPF > program context; btf_get_by_fd() returns -EBADF as the fd is associated > with the module BTF in the test's process context, not necessarily in > the context that the BPF program is running. Would it be possible in this > case to use object id? Or is there another way to handle the fd->module > BTF association that we need to make in BPF program context that I'm > missing? > - A more long-term issue; if we use fds to specify module BTFs and write > the object fd into the program, we can pin the BPF program such that it > outlives fds that refer to its associated BTF. So unless we pinned the > BTF too, any code that assumed the BTF fd-> module mapping was valid would > start to break once the user-space side went away and the pinned program > persisted. All of the above are not issues. They are features of FD based approach. When the program refers to btf via fd the verifier needs to increment btf's refcnt so it won't go away while the prog is running. For module's BTF it means that the module can be unloaded, but its BTF may stay around if there is a prog that needs to access it. I think the missing piece in the above is that btf_get_by_fd() should be done at load time instead of program run-time. Everything FD based needs to behave similar to map_fds where ld_imm64 insn contains map_fd that gets converted to map_ptr by the verifier at load time. In this case single ld_imm64 with 32-bit FD + 32-bit btf_id is not enough. So either libbpf or the verifier need to insert additional instruction. I'm not sure yet how to extend 'struct btf_ptr' cleanly, so it looks good from C side. In the other patch I saw: struct btf_ptr { void *ptr; __u32 type_id; - __u32 flags; /* BTF ptr flags; unused at present. */ + __u32 obj_id; /* BTF object; vmlinux if 0 */ }; The removal of flags cannot be done, since it will break progs. Probably something like this: struct btf_ptr { void *ptr; __u32 type_id; __u32 flags; __u64 btf_obj_fd; /* this is 32-bit FD for libbpf which will become pointer after load */ }; would be the most convenient from the bpf prog side. The ld_imm64 init of btf_obj_fd will be replaced with absolute btf pointer by the verifier. So when bpf_snprintf_btf() is called the prog will pass the kernel internal pointer of struct btf to the helper. No extra run-time checks needed. bpf_snprintf_btf() would print that type_id within given struct btf object. libbpf would need to deal with two relos. One to store btf_id from bpf_core_type_id_kernel() into type_id. And another to find module's BTF and store its FD into btf_obj_fd with ld_imm64. I'm still thinking to how to frame that cleanly from C side. Other ideas?
On Sat, Dec 5, 2020 at 4:38 PM Alan Maguire <alan.maguire@oracle.com> wrote: > > On Fri, 4 Dec 2020, Andrii Nakryiko wrote: > > > When Clang emits ldimm64 instruction for BPF_TYPE_ID_TARGET CO-RE relocation, > > put module BTF FD, containing target type, into upper 32 bits of imm64. > > > > Because this FD is internal to libbpf, it's very cumbersome to test this in > > selftests. Manual testing was performed with debug log messages sprinkled > > across selftests and libbpf, confirming expected values are substituted. > > Better testing will be performed as part of the work adding module BTF types > > support to bpf_snprintf_btf() helpers. > > > > Cc: Alan Maguire <alan.maguire@oracle.com> > > Signed-off-by: Andrii Nakryiko <andrii@kernel.org> > > Thanks so much for doing this Andrii! When I tested, I ran into a problem; > it turns out when a module struct such as "veth_stats" is used, it's > classified as BTF_KIND_FWD, and as a result when we iterate over > the modules and look in the veth module, "struct veth_stats" does not > match since its module kind (BTF_KIND_STRUCT) does not match the candidate > kind (BTF_KIND_FWD). I'm kind of out of my depth here, but the below > patch (on top of your patch) worked. I'm not quite clear on the situation. BTF_KIND_FWD is for the local type or the remote type? Maybe a small example would help, before we go straight to assuming FWD can be always resolved into a concrete STRUCT/UNION. > However without it - when we find > 0 candidate matches - as well as not substituting the module object > id/type id - we hit a segfault: Yep, I missed the null check in: targ_spec->btf != prog->obj->btf_vmlinux I'll fix that. > > Program terminated with signal 11, Segmentation fault. > #0 0x0000000000480bf9 in bpf_core_calc_relo (prog=0x4d6ba40, > relo=0x4d70e7c, > relo_idx=0, local_spec=0x7ffe2cf17b00, targ_spec=0x0, > res=0x7ffe2cf17ae0) > at libbpf.c:4408 > 4408 switch (kind) { > Missing separate debuginfos, use: debuginfo-install > elfutils-libelf-0.172-2.el7.x86_64 glibc-2.17-196.el7.x86_64 > libattr-2.4.46-13.el7.x86_64 libcap-2.22-9.el7.x86_64 > libgcc-4.8.5-36.0.1.el7_6.2.x86_64 zlib-1.2.7-18.el7.x86_64 > (gdb) bt > #0 0x0000000000480bf9 in bpf_core_calc_relo (prog=0x4d6ba40, > relo=0x4d70e7c, > relo_idx=0, local_spec=0x7ffe2cf17b00, targ_spec=0x0, > res=0x7ffe2cf17ae0) > at libbpf.c:4408 > > > The dereferences of targ_spec in bpf_core_recalc_relo() seem > to be the cause; that function is called with a NULL targ_spec > when 0 candidates are found, so it's possible we'd need to > guard those accesses for cases where a bogus type was passed > in and no candidates were found. If the below looks good would > it make sense to roll it into your patch or will I add it to my > v3 patch series? > > Thanks again for your help with this! > > Alan > > From 08040730dbff6c5d7636927777ac85a71c10827f Mon Sep 17 00:00:00 2001 > From: Alan Maguire <alan.maguire@oracle.com> > Date: Sun, 6 Dec 2020 01:10:28 +0100 > Subject: [PATCH] libbpf: handle fwd kinds when checking candidate relocations > for modules > > when a struct belonging to a module is being assessed, it will be > designated a fwd kind (BTF_KIND_FWD); when matching candidate > types constraints on exact type matching need to be relaxed to > ensure that such structures are found successfully. Introduce > kinds_match() function to handle this comparison. > > Signed-off-by: Alan Maguire <alan.maguire@oracle.com> > --- > tools/lib/bpf/libbpf.c | 24 +++++++++++++++++++++--- > 1 file changed, 21 insertions(+), 3 deletions(-) > [...]
On Mon, Dec 7, 2020 at 7:12 PM Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote: > > On Mon, Dec 07, 2020 at 04:38:16PM +0000, Alan Maguire wrote: > > On Fri, 4 Dec 2020, Andrii Nakryiko wrote: > > > > > When Clang emits ldimm64 instruction for BPF_TYPE_ID_TARGET CO-RE relocation, > > > put module BTF FD, containing target type, into upper 32 bits of imm64. > > > > > > Because this FD is internal to libbpf, it's very cumbersome to test this in > > > selftests. Manual testing was performed with debug log messages sprinkled > > > across selftests and libbpf, confirming expected values are substituted. > > > Better testing will be performed as part of the work adding module BTF types > > > support to bpf_snprintf_btf() helpers. > > > > > > Cc: Alan Maguire <alan.maguire@oracle.com> > > > Signed-off-by: Andrii Nakryiko <andrii@kernel.org> > > > --- > > > tools/lib/bpf/libbpf.c | 19 ++++++++++++++++--- > > > 1 file changed, 16 insertions(+), 3 deletions(-) > > > > > > diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c > > > index 9be88a90a4aa..539956f7920a 100644 > > > --- a/tools/lib/bpf/libbpf.c > > > +++ b/tools/lib/bpf/libbpf.c > > > @@ -4795,6 +4795,7 @@ static int load_module_btfs(struct bpf_object *obj) > > > > > > mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; > > > > > > + btf__set_fd(btf, fd); > > > mod_btf->btf = btf; > > > mod_btf->id = id; > > > mod_btf->fd = fd; > > > @@ -5445,6 +5446,10 @@ struct bpf_core_relo_res > > > __u32 orig_type_id; > > > __u32 new_sz; > > > __u32 new_type_id; > > > + /* FD of the module BTF containing the target candidate, or 0 for > > > + * vmlinux BTF > > > + */ > > > + int btf_obj_fd; > > > }; > > > > > > /* Calculate original and target relocation values, given local and target > > > @@ -5469,6 +5474,7 @@ static int bpf_core_calc_relo(const struct bpf_program *prog, > > > res->fail_memsz_adjust = false; > > > res->orig_sz = res->new_sz = 0; > > > res->orig_type_id = res->new_type_id = 0; > > > + res->btf_obj_fd = 0; > > > > > > if (core_relo_is_field_based(relo->kind)) { > > > err = bpf_core_calc_field_relo(prog, relo, local_spec, > > > @@ -5519,6 +5525,9 @@ static int bpf_core_calc_relo(const struct bpf_program *prog, > > > } else if (core_relo_is_type_based(relo->kind)) { > > > err = bpf_core_calc_type_relo(relo, local_spec, &res->orig_val); > > > err = err ?: bpf_core_calc_type_relo(relo, targ_spec, &res->new_val); > > > + if (!err && relo->kind == BPF_TYPE_ID_TARGET && > > > + targ_spec->btf != prog->obj->btf_vmlinux) > > > + res->btf_obj_fd = btf__fd(targ_spec->btf); > > > > Sorry about this Andrii, but I'm a bit stuck here. > > > > I'm struggling to get tests working where the obj fd is used to designate > > the module BTF. Unless I'm missing something there are a few problems: > > > > - the fd association is removed by libbpf when the BPF program has loaded; > > the module fds are closed and the module BTF is discarded. However even if > > that isn't done (and as you mentioned, we could hold onto BTF that is in > > use, and I commented out the code that does that to test) - there's > > another problem: > > - I can't see a way to use the object fd value we set here later in BPF > > program context; btf_get_by_fd() returns -EBADF as the fd is associated > > with the module BTF in the test's process context, not necessarily in > > the context that the BPF program is running. Would it be possible in this > > case to use object id? Or is there another way to handle the fd->module > > BTF association that we need to make in BPF program context that I'm > > missing? > > - A more long-term issue; if we use fds to specify module BTFs and write > > the object fd into the program, we can pin the BPF program such that it > > outlives fds that refer to its associated BTF. So unless we pinned the > > BTF too, any code that assumed the BTF fd-> module mapping was valid would > > start to break once the user-space side went away and the pinned program > > persisted. > > All of the above are not issues. They are features of FD based approach. > When the program refers to btf via fd the verifier needs to increment btf's refcnt > so it won't go away while the prog is running. For module's BTF it means > that the module can be unloaded, but its BTF may stay around if there is a prog > that needs to access it. > I think the missing piece in the above is that btf_get_by_fd() should be > done at load time instead of program run-time. > Everything FD based needs to behave similar to map_fds where ld_imm64 insn > contains map_fd that gets converted to map_ptr by the verifier at load time. Right. I was going to extend verifier to do the same for all used BTF objects as part of ksym support for module BTFs. So totally agree. Just didn't need it so far. > In this case single ld_imm64 with 32-bit FD + 32-bit btf_id is not enough. > So either libbpf or the verifier need to insert additional instruction. So this part I haven't investigated in detail yet. But, if we are just talking about keeping struct btf * pointer + BTF type id (u32) in a single ldimm64, we actually have enough space by using both off + imm fields in both parts of ldimm64 instruction. Gives exactly 8 + 4 bytes. But I don't know if the problem you are referring to is in the JIT part. Also, for the ldimm64 instruction generated by __builtin_btf_type_id(), btf fd + btf type id are always the same, regardless of code path, so we can easily use bpf_insn_aux_data to keep any extra data there, no? > I'm not sure yet how to extend 'struct btf_ptr' cleanly, so it looks good > from C side. > In the other patch I saw: > struct btf_ptr { > void *ptr; > __u32 type_id; > - __u32 flags; /* BTF ptr flags; unused at present. */ > + __u32 obj_id; /* BTF object; vmlinux if 0 */ > }; > The removal of flags cannot be done, since it will break progs. This was something that I suggested to avoid extra logic based on the size of btf_ptr. Not super critical. The idea was that flags so far were always enforced to be zero, which make it backwards compatible and we can now re-use it instead for module BTF fd. If we need flags later, then we can extend it. But as I said, it's not a critical part of the design, so I won't fight that :) > Probably something like this: > struct btf_ptr { > void *ptr; > __u32 type_id; > __u32 flags; > __u64 btf_obj_fd; /* this is 32-bit FD for libbpf which will become pointer after load */ > }; > would be the most convenient from the bpf prog side. The ld_imm64 init of > btf_obj_fd will be replaced with absolute btf pointer by the verifier. So when > bpf_snprintf_btf() is called the prog will pass the kernel internal pointer > of struct btf to the helper. No extra run-time checks needed. > bpf_snprintf_btf() would print that type_id within given struct btf object. > libbpf would need to deal with two relos. One to store btf_id from > bpf_core_type_id_kernel() into type_id. And another to find module's BTF and > store its FD into btf_obj_fd with ld_imm64. I'm still thinking to how to frame So the latter we can do as yet another type of type-based CO-RE relocation, if needed. But if we do that, we should probably revert current __builtin_type_id(TYPE_ID_REMOTE) to just emit 32-bit register assignment (no ldimm64). As for how the verifier would translate such FD into struct btf * pointer. We have something similar today with ldimm64 with BPF_PSEUDO_BTF_ID, which resolves into kernel variables. Let's think if we can re-use that, or we can just add another BPF_PSEUDO_xxx "flavor"? > that cleanly from C side. > Other ideas? I'll need to think a bit more about this. But some thoughts I've got so far.
On Mon, 7 Dec 2020, Andrii Nakryiko wrote: > On Sat, Dec 5, 2020 at 4:38 PM Alan Maguire <alan.maguire@oracle.com> wrote: > > Thanks so much for doing this Andrii! When I tested, I ran into a problem; > > it turns out when a module struct such as "veth_stats" is used, it's > > classified as BTF_KIND_FWD, and as a result when we iterate over > > the modules and look in the veth module, "struct veth_stats" does not > > match since its module kind (BTF_KIND_STRUCT) does not match the candidate > > kind (BTF_KIND_FWD). I'm kind of out of my depth here, but the below > > patch (on top of your patch) worked. > > I'm not quite clear on the situation. BTF_KIND_FWD is for the local > type or the remote type? Maybe a small example would help, before we > go straight to assuming FWD can be always resolved into a concrete > STRUCT/UNION. > The local type was BTF_KIND_FWD, and the target type was BTF_KIND_STRUCT IIRC; I'll try and get some libbpf debug output for you showing the relocation info. If it helps, I think the situation was this; I was referencing __builtin_btf_type_id(struct veth_stats), and hadn't included a BTF-generated veth header, so I'm guessing libbpf classified it as a fwd declaration. My patch was a bit too general I suspect in that it assumed that either target or local could be BTF_KIND_FWD and should match BTF_KIND_STRUCT in local/target, wheres I _think_ the local only should permit BTF_KIND_FWD. Does that make sense? > > > However without it - when we find > > 0 candidate matches - as well as not substituting the module object > > id/type id - we hit a segfault: > > Yep, I missed the null check in: > > targ_spec->btf != prog->obj->btf_vmlinux > > I'll fix that. > Thanks! I think the core_reloc selftests trigger the segfault also if you need a test case to verify. Alan
On Mon, 7 Dec 2020, Andrii Nakryiko wrote: > On Mon, Dec 7, 2020 at 7:12 PM Alexei Starovoitov > <alexei.starovoitov@gmail.com> wrote: > > > > On Mon, Dec 07, 2020 at 04:38:16PM +0000, Alan Maguire wrote: > > > Sorry about this Andrii, but I'm a bit stuck here. > > > > > > I'm struggling to get tests working where the obj fd is used to designate > > > the module BTF. Unless I'm missing something there are a few problems: > > > > > > - the fd association is removed by libbpf when the BPF program has loaded; > > > the module fds are closed and the module BTF is discarded. However even if > > > that isn't done (and as you mentioned, we could hold onto BTF that is in > > > use, and I commented out the code that does that to test) - there's > > > another problem: > > > - I can't see a way to use the object fd value we set here later in BPF > > > program context; btf_get_by_fd() returns -EBADF as the fd is associated > > > with the module BTF in the test's process context, not necessarily in > > > the context that the BPF program is running. Would it be possible in this > > > case to use object id? Or is there another way to handle the fd->module > > > BTF association that we need to make in BPF program context that I'm > > > missing? > > > - A more long-term issue; if we use fds to specify module BTFs and write > > > the object fd into the program, we can pin the BPF program such that it > > > outlives fds that refer to its associated BTF. So unless we pinned the > > > BTF too, any code that assumed the BTF fd-> module mapping was valid would > > > start to break once the user-space side went away and the pinned program > > > persisted. > > > > All of the above are not issues. They are features of FD based approach. > > When the program refers to btf via fd the verifier needs to increment btf's refcnt > > so it won't go away while the prog is running. For module's BTF it means > > that the module can be unloaded, but its BTF may stay around if there is a prog > > that needs to access it. > > I think the missing piece in the above is that btf_get_by_fd() should be > > done at load time instead of program run-time. > > Everything FD based needs to behave similar to map_fds where ld_imm64 insn > > contains map_fd that gets converted to map_ptr by the verifier at load time. > > Right. I was going to extend verifier to do the same for all used BTF > objects as part of ksym support for module BTFs. So totally agree. > Just didn't need it so far. > Does this approach prevent more complex run-time specification of BTF object fd though? For example, I've been working on a simple tracer focused on kernel debugging; it uses a BPF map entry for each kernel function that is traced. User-space populates the map entry with BTF type ids for the function arguments/return value, and when the BPF program runs it uses the instruction pointer to look up the map entry for that function, and uses bpf_snprintf_btf() to write the string representations of the function arguments/return values. I'll send out an RFC soon, but longer-term I was hoping to extend it to support module-specific types. Would a dynamic case like that - where the BTF module fd is looked up in a map entry during program execution (rather than derived via __btf_builtin_type_id()) work too? Thanks! Alan
On Tue, Dec 08, 2020 at 10:13:35PM +0000, Alan Maguire wrote: > On Mon, 7 Dec 2020, Andrii Nakryiko wrote: > > > On Mon, Dec 7, 2020 at 7:12 PM Alexei Starovoitov > > <alexei.starovoitov@gmail.com> wrote: > > > > > > On Mon, Dec 07, 2020 at 04:38:16PM +0000, Alan Maguire wrote: > > > > Sorry about this Andrii, but I'm a bit stuck here. > > > > > > > > I'm struggling to get tests working where the obj fd is used to designate > > > > the module BTF. Unless I'm missing something there are a few problems: > > > > > > > > - the fd association is removed by libbpf when the BPF program has loaded; > > > > the module fds are closed and the module BTF is discarded. However even if > > > > that isn't done (and as you mentioned, we could hold onto BTF that is in > > > > use, and I commented out the code that does that to test) - there's > > > > another problem: > > > > - I can't see a way to use the object fd value we set here later in BPF > > > > program context; btf_get_by_fd() returns -EBADF as the fd is associated > > > > with the module BTF in the test's process context, not necessarily in > > > > the context that the BPF program is running. Would it be possible in this > > > > case to use object id? Or is there another way to handle the fd->module > > > > BTF association that we need to make in BPF program context that I'm > > > > missing? > > > > - A more long-term issue; if we use fds to specify module BTFs and write > > > > the object fd into the program, we can pin the BPF program such that it > > > > outlives fds that refer to its associated BTF. So unless we pinned the > > > > BTF too, any code that assumed the BTF fd-> module mapping was valid would > > > > start to break once the user-space side went away and the pinned program > > > > persisted. > > > > > > All of the above are not issues. They are features of FD based approach. > > > When the program refers to btf via fd the verifier needs to increment btf's refcnt > > > so it won't go away while the prog is running. For module's BTF it means > > > that the module can be unloaded, but its BTF may stay around if there is a prog > > > that needs to access it. > > > I think the missing piece in the above is that btf_get_by_fd() should be > > > done at load time instead of program run-time. > > > Everything FD based needs to behave similar to map_fds where ld_imm64 insn > > > contains map_fd that gets converted to map_ptr by the verifier at load time. > > > > Right. I was going to extend verifier to do the same for all used BTF > > objects as part of ksym support for module BTFs. So totally agree. > > Just didn't need it so far. > > > > Does this approach prevent more complex run-time specification of BTF > object fd though? For example, I've been working on a simple tracer > focused on kernel debugging; it uses a BPF map entry for each kernel > function that is traced. User-space populates the map entry with BTF type > ids for the function arguments/return value, and when the BPF program > runs it uses the instruction pointer to look up the map entry for that > function, and uses bpf_snprintf_btf() to write the string representations > of the function arguments/return values. I'll send out an RFC soon, > but longer-term I was hoping to extend it to support module-specific > types. Would a dynamic case like that - where the BTF module fd is looked > up in a map entry during program execution (rather than derived via > __btf_builtin_type_id()) work too? Thanks! fd has to be resolved in the process context. bpf prog can read fd number from the map, but that number is meaningless. Say we allow using btf_obj_id+btf_id, how user space will know these two numbers? Some new libbpf api that searches for it? An extension to libbpf_find_vmlinux_btf_id() ? I was hoping that this api will stay semi-internal. But say it's extended. The user space will store a pair of numbers into a map and what program are going to do with it? If it's printing struct veth_stats contents it should have attached to a corresponding function in the veth module via fentry or something. The prog has hard coded logic in C with specific pointer to print. The prog has its type right there. Why would the prog take a pointer from one place, but it's type_id from the map? That's not realistic. Where it would potentially make sense is what I think you're descring where single kprobe style prog attached to many places and args of those places are stored in a map and the prog selects them with map_lookup with key=PT_REGS_IP ? And passes pointers into bpf_snprintf_btf() from PT_REGS_PARM1() ? I see why that is useful, but it's so racy. By the time the map is populated those btf_obj_id+btf_id could be invalid. I think instead of doing this in user space the program needs an access to vmlinux+mods BTFs. Sort-of like proposed bpf helper to return ksym based on IP there could be a helper to figure out btf_id+btf_obj_POINTER based on IP. Then there will no need for external map to populate. Would that solve your use case?
On Tue, 8 Dec 2020, Alexei Starovoitov wrote: > On Tue, Dec 08, 2020 at 10:13:35PM +0000, Alan Maguire wrote: > > > > Does this approach prevent more complex run-time specification of BTF > > object fd though? For example, I've been working on a simple tracer > > focused on kernel debugging; it uses a BPF map entry for each kernel > > function that is traced. User-space populates the map entry with BTF type > > ids for the function arguments/return value, and when the BPF program > > runs it uses the instruction pointer to look up the map entry for that > > function, and uses bpf_snprintf_btf() to write the string representations > > of the function arguments/return values. I'll send out an RFC soon, > > but longer-term I was hoping to extend it to support module-specific > > types. Would a dynamic case like that - where the BTF module fd is looked > > up in a map entry during program execution (rather than derived via > > __btf_builtin_type_id()) work too? Thanks! > > fd has to be resolved in the process context. bpf prog can read fd > number from the map, but that number is meaningless. > Say we allow using btf_obj_id+btf_id, how user space will know these > two numbers? Some new libbpf api that searches for it? > An extension to libbpf_find_vmlinux_btf_id() ? I was hoping that this api > will stay semi-internal. But say it's extended. > The user space will store a pair of numbers into a map and > what program are going to do with it? > If it's printing struct veth_stats contents it should have attached to > a corresponding function in the veth module via fentry or something. > The prog has hard coded logic in C with specific pointer to print. > The prog has its type right there. Why would the prog take a pointer > from one place, but it's type_id from the map? That's not realistic. > Where it would potentially make sense is what I think you're descring > where single kprobe style prog attached to many places and args of > those places are stored in a map and the prog selects them with > map_lookup with key=PT_REGS_IP ? Right, that's exactly it. A pair of generic tracing BPF programs are used, and they attach to kprobe/kretprobes, and when they run they use the arguments plus the map details about BTF ids of those arguments to run bpf_snprintf_btf(), and send perf events to userspace containing the results. > And passes pointers into bpf_snprintf_btf() from PT_REGS_PARM1() ? Exactly. > I see why that is useful, but it's so racy. By the time the map > is populated those btf_obj_id+btf_id could be invalid. > I think instead of doing this in user space the program needs an access > to vmlinux+mods BTFs. Sort-of like proposed bpf helper to return ksym > based on IP there could be a helper to figure out btf_id+btf_obj_POINTER > based on IP. Then there will no need for external map to populate. > Would that solve your use case? That would be fantastic! We could do that from the context passed into a kprobe program as the IP in struct pt_regs points at the function. kretprobes seems a bit trickier as in that case the IP in struct pt_regs is actually set to kretprobe_trampoline rather than the function we're returning from due to how kretprobes work; maybe there's another way to get it in that case though.. Alan
On Wed, Dec 09, 2020 at 11:21:43PM +0000, Alan Maguire wrote: > Right, that's exactly it. A pair of generic tracing BPF programs are > used, and they attach to kprobe/kretprobes, and when they run they > use the arguments plus the map details about BTF ids of those > arguments to run bpf_snprintf_btf(), and send perf events to > userspace containing the results. ... > That would be fantastic! We could do that from the context passed > into a kprobe program as the IP in struct pt_regs points at the > function. kretprobes seems a bit trickier as in that case the IP in > struct pt_regs is actually set to kretprobe_trampoline rather than > the function we're returning from due to how kretprobes work; maybe > there's another way to get it in that case though.. Yeah. kprobe's IP doesn't match kretprobe's IP which makes such tracing use cases more complicated. Also kretprobe is quite slow. See prog_tests/test_overhead and selftests/bpf/bench. I think the key realization is that the user space knows all IPs it will attach to. It has to know all IPs otherwise hashmap{key=ip, value=btf_data} is not possible. Obvious, right ? What it means that we can use this key observation to build better interfaces at all layers. kprobes are slow to setup one by one. It's also slow to execute. fentry/fexit is slow to setup, but fast to execute. Jiri proposed a batching api for fentry, but it doesn't quite make sense from api perspective since user space has to give different bpf prog for every fentry. bpf trampoline is unique for every target fentry kernel function. The batched attach would make sense for kprobe because one prog can be attached everywhere. But kprobe is slow. This thought process justifies an addition of a new program type where one program can attach to multiple fentry. Since fentry ctx is no longer fixed the verifier won't be able to track btf_id-s of arguments, but btf based pointer walking is fast and powerful, so if btf is passed into the program there could be a helper that does dynamic cast from long to PTR_TO_BTF_ID. Since such new fentry prog will have btf in the context and there will be no need for user space to populate hashmap and mess with IPs. And the best part that batched attach will not only be desired, but mandatory part of the api. So I'm proposing to extend BPF_PROG_LOAD cmd with an array of pairs (attach_obj_fd, attach_btf_id). The fentry prog in .c file might even have a regex in attach pattern: SEC("fentry/sys_*") int BPF_PROG(test, struct btf *btf_obj, __u32 btf_id, __u64 arg1, __u64 arg2, ...__u64 arg6) { struct btf_ptr ptr1 = { .ptr = arg1, .type_id = bpf_core_type_id_kernel(struct foo), .btf_obj = btf_obj, }, ptr2 = { .ptr = arg2, .type_id = bpf_core_type_id_kernel(struct bar), .btf_obj = btf_obj, }; bpf_snprintf_btf(,, &ptr1, sizeof(ptr1), ); bpf_snprintf_btf(,, &ptr1, sizeof(ptr2), ); } libbpf will process the attach regex and find all matching functions in the kernel and in the kernel modules. Then it will pass this list of (fd,btf_id) pairs to the kernel. The kernel will find IP addresses and BTFs of all functions. It will generate single bpf trampoline to handle all the functions. Either one trampoline or multiple trampolines is an implementation detail. It could be one trampoline that does lookup based on IP to find btf_obj, btf_id to pass into the program or multiple trampolines that share most of the code with N unique trampoline prefixes with hardcoded btf_obj, btf_id. The argument save/restore code can be the same for all fentries. The same way we can support single fexit prog attaching to multiple kernel functions. And even single fmod_ret prog attaching to multiple. The batching part will make attaching to thousands of functions efficient. We can use batched text_poke_bp, etc. As far as dynamic btf casting helper we could do something like this: SEC("fentry/sys_*") int BPF_PROG(test, struct btf *btf_obj, __u32 btf_id, __u64 arg1, __u64 arg2, ...__u64 arg6) { struct sk_buff *skb; struct task_struct *task; skb = bpf_dynamic_cast(btf_obj, btf_id, 1, arg1, bpf_core_type_id_kernel(skb)); task = bpf_dynamic_cast(btf_obj, btf_id, 2, arg2, bpf_core_type_id_kernel(task)); skb->len + task->status; } The dynamic part of the helper will walk btf of func_proto that was pointed to by 'btf_id' argument. It will find Nth argument and if argument's btf_id matches the last u32 passed into bpf_dynamic_cast() it will return ptr_to_btf_id. The verifier needs 5th u32 arg to know const value of btf_id during verification. The execution time of this casting helper will be pretty fast. Thoughts?
On Tue, 15 Dec 2020, Alexei Starovoitov wrote: > On Wed, Dec 09, 2020 at 11:21:43PM +0000, Alan Maguire wrote: > > Right, that's exactly it. A pair of generic tracing BPF programs are > > used, and they attach to kprobe/kretprobes, and when they run they > > use the arguments plus the map details about BTF ids of those arguments to > > run bpf_snprintf_btf(), and send perf events to userspace containing the > > results. > ... > > That would be fantastic! We could do that from the context passed into a > > kprobe program as the IP in struct pt_regs points at the function. > > kretprobes seems a bit trickier as in that case the IP in struct pt_regs is > > actually set to kretprobe_trampoline rather than the function we're > > returning from due to how kretprobes work; maybe there's another way to get > > it in that case though.. > > Yeah. kprobe's IP doesn't match kretprobe's IP which makes such tracing > use cases more complicated. Also kretprobe is quite slow. See > prog_tests/test_overhead and selftests/bpf/bench. > I think the key realization is that the user space knows all IPs > it will attach to. It has to know all IPs otherwise > hashmap{key=ip, value=btf_data} is not possible. > Obvious, right ? What it means that we can use this key observation > to build better interfaces at all layers. kprobes are slow to > setup one by one. It's also slow to execute. fentry/fexit is slow > to setup, but fast to execute. Jiri proposed a batching api for > fentry, but it doesn't quite make sense from api perspective > since user space has to give different bpf prog for every fentry. > bpf trampoline is unique for every target fentry kernel function. > The batched attach would make sense for kprobe because one prog > can be attached everywhere. But kprobe is slow. > This thought process justifies an addition of a new program > type where one program can attach to multiple fentry. > Since fentry ctx is no longer fixed the verifier won't be able to > track btf_id-s of arguments, but btf based pointer walking is fast > and powerful, so if btf is passed into the program there could > be a helper that does dynamic cast from long to PTR_TO_BTF_ID. > Since such new fentry prog will have btf in the context and > there will be no need for user space to populate hashmap and > mess with IPs. And the best part that batched attach will not > only be desired, but mandatory part of the api. > So I'm proposing to extend BPF_PROG_LOAD cmd with an array of > pairs (attach_obj_fd, attach_btf_id). > The fentry prog in .c file might even have a regex in attach pattern: > SEC("fentry/sys_*") > int BPF_PROG(test, struct btf *btf_obj, __u32 btf_id, __u64 arg1, > __u64 arg2, ...__u64 arg6) > { > struct btf_ptr ptr1 = { > .ptr = arg1, > .type_id = bpf_core_type_id_kernel(struct foo), > .btf_obj = btf_obj, > }, > ptr2 = { > .ptr = arg2, > .type_id = bpf_core_type_id_kernel(struct bar), > .btf_obj = btf_obj, > }; > bpf_snprintf_btf(,, &ptr1, sizeof(ptr1), ); > bpf_snprintf_btf(,, &ptr1, sizeof(ptr2), ); > } > > libbpf will process the attach regex and find all matching functions in > the kernel and in the kernel modules. Then it will pass this list of > (fd,btf_id) pairs to the kernel. The kernel will find IP addresses and > BTFs of all functions. It will generate single bpf trampoline to handle > all the functions. Either one trampoline or multiple trampolines is an > implementation detail. It could be one trampoline that does lookup based > on IP to find btf_obj, btf_id to pass into the program or multiple > trampolines that share most of the code with N unique trampoline > prefixes with hardcoded btf_obj, btf_id. The argument save/restore code > can be the same for all fentries. The same way we can support single > fexit prog attaching to multiple kernel functions. And even single > fmod_ret prog attaching to multiple. The batching part will make > attaching to thousands of functions efficient. We can use batched > text_poke_bp, etc. > > As far as dynamic btf casting helper we could do something like this: > SEC("fentry/sys_*") > int BPF_PROG(test, struct btf *btf_obj, __u32 btf_id, __u64 arg1, __u64 > arg2, ...__u64 arg6) > { > struct sk_buff *skb; > struct task_struct *task; > > skb = bpf_dynamic_cast(btf_obj, btf_id, 1, arg1, > bpf_core_type_id_kernel(skb)); > task = bpf_dynamic_cast(btf_obj, btf_id, 2, arg2, > bpf_core_type_id_kernel(task)); > skb->len + task->status; > } > The dynamic part of the helper will walk btf of func_proto that was > pointed to by 'btf_id' argument. It will find Nth argument and > if argument's btf_id matches the last u32 passed into bpf_dynamic_cast() > it will return ptr_to_btf_id. The verifier needs 5th u32 arg to know > const value of btf_id during verification. > The execution time of this casting helper will be pretty fast. > Thoughts? > > From a bpf programmer's perspective, the above sounds fantastic and opens up a bunch of new possibilities. For example, a program that attaches to a bunch of networking functions at once and uses dynamic casts to find the skb argument could help trace packet flow through the stack without having to match exact function signatures. From a mechanics perspective I wonder if we could take a similar approach to the cgroup storage, and use the bpf prog array structure to store a struct btf * and any other site-specific metadata at attach time? Then when the array runs we could set a per-cpu variable such that helpers could pick up that info if needed. Having the function argument BTF ids gets us nearly the whole way there from a generic tracer perspective - I can now attach my generic tracing program to an arbitrary function via fentry/fexit and get the BTF ids of the arguments or return value, and even better can do it with wildcarding. There is an additional use case though - at least for the ksnoop program I'm working on at least - and it's where we access members and need their type ids too. The ksnoop program (which I wanted to send out last week but due to a system move I'm temporarily not able to access the code, sorry about that, hoping it'll be back online early next week) operates in two modes; - default mode where we trace function arguments for kprobe and return value for kretprobe; that's covered by the above; and - a mode where the user specifies what they want. For example running $ ksnoop "ip_send_skb" ...is an example of default mode, this will trace entry/return and print arguments and return values, while $ ksnoop "ip_send_skb(skb)" ...will trace the skb argument only, and $ ksnoop "ip_send_skb(skb->sk)" ...will trace the skb->sk value. The user-space side of the program matches the function/arg name and looks up the referenced type, setting it in the function's map. For field references such as skb->sk, it also records offset and whether that offset is a pointer (as is the case for skb->sk) - in such cases we need to read the offset value via bpf_probe_read() and use it in bpf_snprintf_btf() along with the referenced type. Only a single simple reference like the above is supported currently, but multiple levels of reference could be made to work too. This latter case would be awkward to support programmatically in BPF program context I think, though I'm sure it could be done. To turn back to our earlier conversation, your concern as I understand it was that pre-specifying module BTF type ids in a map is racy, and I'd like to dig into that a bit more if you don't mind because I think some form of user-space-specified BTF ids may be the easiest approach for more flexible generic tracing that covers more than function arguments. I assume the race you're concerned about is caused by the module unloading after the BTF ids have been set in the map? And then the module reappears with different BTF/type id mappings? Perhaps a helper for retrieving the struct btf * which was set at attach time would be enough? So for example something like struct btf_ptr ptr; ptr.type_id = /* retrieve from map */ ptr.obj_id = bpf_get_btf(THIS_MODULE); ...where we don't actually specify a type but a libbpf-specified fd is used to stash the associated "struct btf *" for the module in the prog array at attach. Are there still race conditions we need to worry about in a scenario like this? Thanks! Alan
On Wed, Dec 16, 2020 at 8:18 AM Alan Maguire <alan.maguire@oracle.com> wrote: > > On Tue, 15 Dec 2020, Alexei Starovoitov wrote: > > > On Wed, Dec 09, 2020 at 11:21:43PM +0000, Alan Maguire wrote: > > > Right, that's exactly it. A pair of generic tracing BPF programs are > > > used, and they attach to kprobe/kretprobes, and when they run they > > > use the arguments plus the map details about BTF ids of those arguments to > > > run bpf_snprintf_btf(), and send perf events to userspace containing the > > > results. > > ... > > > That would be fantastic! We could do that from the context passed into a > > > kprobe program as the IP in struct pt_regs points at the function. > > > kretprobes seems a bit trickier as in that case the IP in struct pt_regs is > > > actually set to kretprobe_trampoline rather than the function we're > > > returning from due to how kretprobes work; maybe there's another way to get > > > it in that case though.. > > > > Yeah. kprobe's IP doesn't match kretprobe's IP which makes such tracing > > use cases more complicated. Also kretprobe is quite slow. See > > prog_tests/test_overhead and selftests/bpf/bench. > > I think the key realization is that the user space knows all IPs > > it will attach to. It has to know all IPs otherwise > > hashmap{key=ip, value=btf_data} is not possible. > > Obvious, right ? What it means that we can use this key observation > > to build better interfaces at all layers. kprobes are slow to > > setup one by one. It's also slow to execute. fentry/fexit is slow > > to setup, but fast to execute. Jiri proposed a batching api for > > fentry, but it doesn't quite make sense from api perspective > > since user space has to give different bpf prog for every fentry. > > bpf trampoline is unique for every target fentry kernel function. > > The batched attach would make sense for kprobe because one prog > > can be attached everywhere. But kprobe is slow. > > This thought process justifies an addition of a new program > > type where one program can attach to multiple fentry. > > Since fentry ctx is no longer fixed the verifier won't be able to > > track btf_id-s of arguments, but btf based pointer walking is fast > > and powerful, so if btf is passed into the program there could > > be a helper that does dynamic cast from long to PTR_TO_BTF_ID. > > Since such new fentry prog will have btf in the context and > > there will be no need for user space to populate hashmap and > > mess with IPs. And the best part that batched attach will not > > only be desired, but mandatory part of the api. > > So I'm proposing to extend BPF_PROG_LOAD cmd with an array of > > pairs (attach_obj_fd, attach_btf_id). Beyond knowing its attach btf_id, such BPF programs probably need to store some additional application-specific data per each attached function, so if we can extend this to also accept an arbitrary long, in addition to fd+id, that would be great. It can be used as an array index for some extra configuration, or to correlate fentry with fexit, etc, etc. No need for an arbitrary amount of data, one long is enough to then use other means to get to arbitrary program data. > > The fentry prog in .c file might even have a regex in attach pattern: > > SEC("fentry/sys_*") > > int BPF_PROG(test, struct btf *btf_obj, __u32 btf_id, __u64 arg1, > > __u64 arg2, ...__u64 arg6) So I like the overall idea, but I'm going to nitpick on some details :) This still looks, feels and behaves like fentry/fexit program, so hopefully it won't be really an entirely new type of BPF program (at least from user-space point of view). Also, not a big fan of having struct btf * and separate btf_id. Maybe we can just abstract it into an opaque (or not opaque) struct representing fully-qualified BTF type. Further, if instead of passing it as an input argument, we can just have an BPF helper to return this, it would be even better and could be actually generalized to other BTF-powered BPF programs. So something like: struct btf_type_id *id = bpf_get_btf_type_id(ctx); ... .type_id = id->type_id; .btf_obj = id->btf; As Alan proposed below, this btf and type id could be stored at the attachment point and just fetched through program context. > > { > > struct btf_ptr ptr1 = { > > .ptr = arg1, > > .type_id = bpf_core_type_id_kernel(struct foo), > > .btf_obj = btf_obj, > > }, > > ptr2 = { > > .ptr = arg2, > > .type_id = bpf_core_type_id_kernel(struct bar), > > .btf_obj = btf_obj, > > }; > > bpf_snprintf_btf(,, &ptr1, sizeof(ptr1), ); > > bpf_snprintf_btf(,, &ptr1, sizeof(ptr2), ); > > } > > > > libbpf will process the attach regex and find all matching functions in > > the kernel and in the kernel modules. Then it will pass this list of > > (fd,btf_id) pairs to the kernel. The kernel will find IP addresses and > > BTFs of all functions. It will generate single bpf trampoline to handle > > all the functions. Either one trampoline or multiple trampolines is an > > implementation detail. It could be one trampoline that does lookup based > > on IP to find btf_obj, btf_id to pass into the program or multiple > > trampolines that share most of the code with N unique trampoline > > prefixes with hardcoded btf_obj, btf_id. The argument save/restore code > > can be the same for all fentries. The same way we can support single > > fexit prog attaching to multiple kernel functions. And even single > > fmod_ret prog attaching to multiple. The batching part will make > > attaching to thousands of functions efficient. We can use batched > > text_poke_bp, etc. Yep, agree. And extra information that we'll have to store per each attachment more than offsets the extra overhead that we'd pay for thousands of BPF program copies we'd need otherwise to achieve the same functionality, if attached through existing fentry/fexit APIs. > > > > As far as dynamic btf casting helper we could do something like this: > > SEC("fentry/sys_*") > > int BPF_PROG(test, struct btf *btf_obj, __u32 btf_id, __u64 arg1, __u64 > > arg2, ...__u64 arg6) > > { > > struct sk_buff *skb; > > struct task_struct *task; > > > > skb = bpf_dynamic_cast(btf_obj, btf_id, 1, arg1, > > bpf_core_type_id_kernel(skb)); > > task = bpf_dynamic_cast(btf_obj, btf_id, 2, arg2, > > bpf_core_type_id_kernel(task)); > > skb->len + task->status; > > } > > The dynamic part of the helper will walk btf of func_proto that was > > pointed to by 'btf_id' argument. It will find Nth argument and > > if argument's btf_id matches the last u32 passed into bpf_dynamic_cast() > > it will return ptr_to_btf_id. The verifier needs 5th u32 arg to know > > const value of btf_id during verification. > > The execution time of this casting helper will be pretty fast. > > Thoughts? So I haven't dug through the code to see if it's doable, but this API seems unnecessarily complicated and error-prone. Plus it's not clear how verifier can validate that btf_obj and btf_id correctly describe arg1 and it does match "1" argument position. I.e., if I mess up and for skb specify bpf_dynamic_cast(btf_obj, 123123, 2, arg3, bpf_core_type_id_kernel(skb)), will verifier stop me? Note that I specified an arbitrary integer for the second argument. Or at least, would we get a NULL in runtime? Or we'll just re-interpret arg3 (incorrectly) as a sk_buff? I might be missing something here, of course. But this seems more "verifiable" and nicer to use, even though it won't substituting an arbitrary btf_id and btf_obj (but that's sort of a goal, I think): skb = bpf_get_btf_arg(ctx, 1, bpf_core_type_id_kernel(skb)); So btf_obj and btf_id of attach point will be taken from the program's context, as well as btf info of an argument #1 (and additionally verified that argument number makes sense at all). Note also that bpf_core_type_id_kernel(skb) ideally would encode type ID and BTF object FD of a module, if the destination type is defined in a module. If we do want explicitly providing btf_obj+btf_id, then I think having it in a separate struct that verifier can track as a PTR_TO_BTF_ID and verify it's valid would make this API better (see above). > > > > > > From a bpf programmer's perspective, the above sounds fantastic and opens > up a bunch of new possibilities. For example, a program that attaches to > a bunch of networking functions at once and uses dynamic casts to find the > skb argument could help trace packet flow through the stack without having > to match exact function signatures. From a mechanics perspective I > wonder if we could take a similar approach to the cgroup storage, and use > the bpf prog array structure to store a struct btf * and any other > site-specific metadata at attach time? Then when the array runs we could > set a per-cpu variable such that helpers could pick up that info if > needed. I don't think per-cpu variable will work for sleepable BPF programs. I think storing it in whatever BPF context struct is a more reliable way. But I agree that remembering struct btf * and attach btf ID at the attachment point makes a lot of stuff simpler. > > Having the function argument BTF ids gets us nearly the whole way there > from a generic tracer perspective - I can now attach my generic tracing > program to an arbitrary function via fentry/fexit and get the BTF ids > of the arguments or return value, and even better can do it with wildcarding. Yeah, it's a nice interface and libbpf can support that. I think in more advanced cases (e.g., matching fentry with fexit), libbpf should still provide APIs to specify extra user-provided long for each attachment point, so that it's possible to correlate "related" BPF programs. > There is an additional use case though - at least for the ksnoop program > I'm working on at least - and it's where we access members and need their > type ids too. The ksnoop program (which I wanted to send out last week but due > to a system move I'm temporarily not able to access the code, sorry about that, > hoping it'll be back online early next week) operates in two modes; > > - default mode where we trace function arguments for kprobe and return value > for kretprobe; that's covered by the above; and > - a mode where the user specifies what they want. For example running > > $ ksnoop "ip_send_skb" > > ...is an example of default mode, this will trace entry/return and print > arguments and return values, while > > $ ksnoop "ip_send_skb(skb)" > > ...will trace the skb argument only, and > > $ ksnoop "ip_send_skb(skb->sk)" > > ...will trace the skb->sk value. The user-space side of the program > matches the function/arg name and looks up the referenced type, setting it > in the function's map. For field references such as skb->sk, it also > records offset and whether that offset is a pointer (as is the case for > skb->sk) - in such cases we need to read the offset value via bpf_probe_read() > and use it in bpf_snprintf_btf() along with the referenced type. Only a > single simple reference like the above is supported currently, but > multiple levels of reference could be made to work too. I think this is normally done with Clang compiling the code at runtime (like trace.py), right? I assume you are trying to find a way to do that without code generation at runtime, am I right? I honestly don't yet see how this can be done easily... > > This latter case would be awkward to support programmatically in BPF > program context I think, though I'm sure it could be done. To turn back > to our earlier conversation, your concern as I understand it was that > pre-specifying module BTF type ids in a map is racy, and I'd like to dig > into that a bit more if you don't mind because I think some form of > user-space-specified BTF ids may be the easiest approach for more flexible > generic tracing that covers more than function arguments. > > I assume the race you're concerned about is caused by the module unloading > after the BTF ids have been set in the map? And then the module reappears > with different BTF/type id mappings? Perhaps a helper for retrieving > the struct btf * which was set at attach time would be enough? > > So for example something like > > > struct btf_ptr ptr; > > ptr.type_id = /* retrieve from map */ > ptr.obj_id = bpf_get_btf(THIS_MODULE); On re-reading this this seems like something I was proposing above with bpf_get_btf_type_id(), right? THIS_MODULE part is confusing, though. > > ...where we don't actually specify a type but a libbpf-specified fd > is used to stash the associated "struct btf *" for the module in the > prog array at attach. Are there still race conditions we need to worry > about in a scenario like this? Thanks! > > Alan
On Wed, Dec 16, 2020 at 02:27:23PM -0800, Andrii Nakryiko wrote: > > But this seems more "verifiable" and nicer to use, even though it > won't substituting an arbitrary btf_id and btf_obj (but that's sort of > a goal, I think): > > skb = bpf_get_btf_arg(ctx, 1, bpf_core_type_id_kernel(skb)); yep. makes sense to me. Assuming that ctx has both: - BTF of the func and the helper will follow to arg's BTF at run-time to check that it matches 3rd arg btf_id. - and the actual arg values as well. So that helper will return them. > > - default mode where we trace function arguments for kprobe and return value > > for kretprobe; that's covered by the above; and > > - a mode where the user specifies what they want. For example running > > > > $ ksnoop "ip_send_skb" > > > > ...is an example of default mode, this will trace entry/return and print > > arguments and return values, while > > > > $ ksnoop "ip_send_skb(skb)" > > > > ...will trace the skb argument only, and > > > > $ ksnoop "ip_send_skb(skb->sk)" > > > > ...will trace the skb->sk value. The user-space side of the program > > matches the function/arg name and looks up the referenced type, setting it > > in the function's map. For field references such as skb->sk, it also > > records offset and whether that offset is a pointer (as is the case for > > skb->sk) - in such cases we need to read the offset value via bpf_probe_read() > > and use it in bpf_snprintf_btf() along with the referenced type. Only a > > single simple reference like the above is supported currently, but > > multiple levels of reference could be made to work too. Alan, I'm not sure why the last example is so different form the first two. I think ksnoop tool will generate the program on the fly, right? So it can generate normal LDX insn with CO-RE relocation (instead of bpf_probe_read) to access skb->sk. It can also add relo for that LDX to point to struct sk_buff's btf_id defined inside prog's BTF. The 'sk' offset inside bpf program and inside BTF can be anything: 0, 4, ... libbpf relocation logic will find the right offset in kernel's sk_buff. If ksnoop doesn't have an ability to parse vmlinux.h file or kernel's BTF it can 'cheat'. If the cmdline looks like: $ ksnoop "ip_send_skb(skb->sk)" It can generate BTF: struct sk_buff { struct sock *sk; }; If cmdline looks like: $ ksnoop "ip_send_skb(skb->sock)" It can generate BTF: struct sk_buff { struct sock *sock; }; Obviously there is no 'sock' field inside kernel's struct sk_buff, but tool doesn't need to care. It can let libbpf do the checking and match fields properly. > > into that a bit more if you don't mind because I think some form of > > user-space-specified BTF ids may be the easiest approach for more flexible > > generic tracing that covers more than function arguments. I think you're trying to figure out kernel's btf_ids in ksnoop tool. I suggest to leave that job to libbpf. Generate local BTFs in ksnoop with CO-RE relocs and let libbpf handle insn patching. No FDs to worry about from ksnoop side either.
On Wed, 16 Dec 2020, Alexei Starovoitov wrote: > > > $ ksnoop "ip_send_skb(skb->sk)" > > > > > > ...will trace the skb->sk value. The user-space side of the program > > > matches the function/arg name and looks up the referenced type, setting it > > > in the function's map. For field references such as skb->sk, it also > > > records offset and whether that offset is a pointer (as is the case for > > > skb->sk) - in such cases we need to read the offset value via bpf_probe_read() > > > and use it in bpf_snprintf_btf() along with the referenced type. Only a > > > single simple reference like the above is supported currently, but > > > multiple levels of reference could be made to work too. > > Alan, > > I'm not sure why the last example is so different form the first two. > I think ksnoop tool will generate the program on the fly, right? Nope, the BPF program is hard-coded; it adapts to different functions through use of the map entries describing function signatures and their BTF ids, and other associated tracing info. The aim is to provide a generic tracing tool which displays kernel function arguments but doesn't require LLVM/clang on the target, just a kernel built with BTF and libbpf. Sorry this wasn't clearer in my explanation; I'm working on rewriting the code and will send it out ASAP. > So it can generate normal LDX insn with CO-RE relocation (instead of bpf_probe_read) > to access skb->sk. It can also add relo for that LDX to point to > struct sk_buff's btf_id defined inside prog's BTF. > The 'sk' offset inside bpf program and inside BTF can be anything: 0, 4, ... > libbpf relocation logic will find the right offset in kernel's sk_buff. > If ksnoop doesn't have an ability to parse vmlinux.h file or kernel's BTF > it can 'cheat'. > If the cmdline looks like: > $ ksnoop "ip_send_skb(skb->sk)" > It can generate BTF: > struct sk_buff { > struct sock *sk; > }; > > If cmdline looks like: > $ ksnoop "ip_send_skb(skb->sock)" > It can generate BTF: > struct sk_buff { > struct sock *sock; > }; > Obviously there is no 'sock' field inside kernel's struct sk_buff, but tool > doesn't need to care. It can let libbpf do the checking and match > fields properly. > > > > into that a bit more if you don't mind because I think some form of > > > user-space-specified BTF ids may be the easiest approach for more flexible > > > generic tracing that covers more than function arguments. > > I think you're trying to figure out kernel's btf_ids in ksnoop tool. Yep. > I suggest to leave that job to libbpf. Generate local BTFs in ksnoop > with CO-RE relocs and let libbpf handle insn patching. > No FDs to worry about from ksnoop side either. > The current approach doesn't rely on instruction patching outside of limited CORE use around struct pt_regs fields (args, IP, etc) which shouldn't require LLVM/clang availability on the target system. I'll try and get it ready for RFC submission by the weekend so you can see more details of the approach. Thanks! Alan
On Thu, Dec 17, 2020 at 05:46:42PM +0000, Alan Maguire wrote: > > > On Wed, 16 Dec 2020, Alexei Starovoitov wrote: > > > > > $ ksnoop "ip_send_skb(skb->sk)" > > > > > > > > ...will trace the skb->sk value. The user-space side of the program > > > > matches the function/arg name and looks up the referenced type, setting it > > > > in the function's map. For field references such as skb->sk, it also > > > > records offset and whether that offset is a pointer (as is the case for > > > > skb->sk) - in such cases we need to read the offset value via bpf_probe_read() > > > > and use it in bpf_snprintf_btf() along with the referenced type. Only a > > > > single simple reference like the above is supported currently, but > > > > multiple levels of reference could be made to work too. > > > > Alan, > > > > I'm not sure why the last example is so different form the first two. > > I think ksnoop tool will generate the program on the fly, right? > > Nope, the BPF program is hard-coded; it adapts to different functions > through use of the map entries describing function signatures and their > BTF ids, and other associated tracing info. The aim is to provide a > generic tracing tool which displays kernel function arguments but > doesn't require LLVM/clang on the target, just a kernel built with BTF > and libbpf. Sorry this wasn't clearer in my explanation; I'm working > on rewriting the code and will send it out ASAP. > > > So it can generate normal LDX insn with CO-RE relocation (instead of bpf_probe_read) > > to access skb->sk. It can also add relo for that LDX to point to > > struct sk_buff's btf_id defined inside prog's BTF. > > The 'sk' offset inside bpf program and inside BTF can be anything: 0, 4, ... > > libbpf relocation logic will find the right offset in kernel's sk_buff. > > If ksnoop doesn't have an ability to parse vmlinux.h file or kernel's BTF > > it can 'cheat'. > > If the cmdline looks like: > > $ ksnoop "ip_send_skb(skb->sk)" > > It can generate BTF: > > struct sk_buff { > > struct sock *sk; > > }; > > > > If cmdline looks like: > > $ ksnoop "ip_send_skb(skb->sock)" > > It can generate BTF: > > struct sk_buff { > > struct sock *sock; > > }; > > Obviously there is no 'sock' field inside kernel's struct sk_buff, but tool > > doesn't need to care. It can let libbpf do the checking and match > > fields properly. > > > > > > into that a bit more if you don't mind because I think some form of > > > > user-space-specified BTF ids may be the easiest approach for more flexible > > > > generic tracing that covers more than function arguments. > > > > I think you're trying to figure out kernel's btf_ids in ksnoop tool. > > Yep. > > > I suggest to leave that job to libbpf. Generate local BTFs in ksnoop > > with CO-RE relocs and let libbpf handle insn patching. > > No FDs to worry about from ksnoop side either. > > > > The current approach doesn't rely on instruction patching outside > of limited CORE use around struct pt_regs fields (args, IP, etc) > which shouldn't require LLVM/clang availability on the target system. I'm not suggesting to use clang. Everything I proposed above is for ksnoop to do. Not for the clang.
diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c index 9be88a90a4aa..539956f7920a 100644 --- a/tools/lib/bpf/libbpf.c +++ b/tools/lib/bpf/libbpf.c @@ -4795,6 +4795,7 @@ static int load_module_btfs(struct bpf_object *obj) mod_btf = &obj->btf_modules[obj->btf_module_cnt++]; + btf__set_fd(btf, fd); mod_btf->btf = btf; mod_btf->id = id; mod_btf->fd = fd; @@ -5445,6 +5446,10 @@ struct bpf_core_relo_res __u32 orig_type_id; __u32 new_sz; __u32 new_type_id; + /* FD of the module BTF containing the target candidate, or 0 for + * vmlinux BTF + */ + int btf_obj_fd; }; /* Calculate original and target relocation values, given local and target @@ -5469,6 +5474,7 @@ static int bpf_core_calc_relo(const struct bpf_program *prog, res->fail_memsz_adjust = false; res->orig_sz = res->new_sz = 0; res->orig_type_id = res->new_type_id = 0; + res->btf_obj_fd = 0; if (core_relo_is_field_based(relo->kind)) { err = bpf_core_calc_field_relo(prog, relo, local_spec, @@ -5519,6 +5525,9 @@ static int bpf_core_calc_relo(const struct bpf_program *prog, } else if (core_relo_is_type_based(relo->kind)) { err = bpf_core_calc_type_relo(relo, local_spec, &res->orig_val); err = err ?: bpf_core_calc_type_relo(relo, targ_spec, &res->new_val); + if (!err && relo->kind == BPF_TYPE_ID_TARGET && + targ_spec->btf != prog->obj->btf_vmlinux) + res->btf_obj_fd = btf__fd(targ_spec->btf); } else if (core_relo_is_enumval_based(relo->kind)) { err = bpf_core_calc_enumval_relo(relo, local_spec, &res->orig_val); err = err ?: bpf_core_calc_enumval_relo(relo, targ_spec, &res->new_val); @@ -5725,10 +5734,14 @@ static int bpf_core_patch_insn(struct bpf_program *prog, } insn[0].imm = new_val; - insn[1].imm = 0; /* currently only 32-bit values are supported */ - pr_debug("prog '%s': relo #%d: patched insn #%d (LDIMM64) imm64 %llu -> %u\n", + /* btf_obj_fd is zero for all relos but BPF_TYPE_ID_TARGET + * with target type in the kernel module BTF + */ + insn[1].imm = res->btf_obj_fd; + pr_debug("prog '%s': relo #%d: patched insn #%d (LDIMM64) imm64 %llu -> %llu\n", prog->name, relo_idx, insn_idx, - (unsigned long long)imm, new_val); + (unsigned long long)imm, + ((unsigned long long)res->btf_obj_fd << 32) | new_val); break; } default:
When Clang emits ldimm64 instruction for BPF_TYPE_ID_TARGET CO-RE relocation, put module BTF FD, containing target type, into upper 32 bits of imm64. Because this FD is internal to libbpf, it's very cumbersome to test this in selftests. Manual testing was performed with debug log messages sprinkled across selftests and libbpf, confirming expected values are substituted. Better testing will be performed as part of the work adding module BTF types support to bpf_snprintf_btf() helpers. Cc: Alan Maguire <alan.maguire@oracle.com> Signed-off-by: Andrii Nakryiko <andrii@kernel.org> --- tools/lib/bpf/libbpf.c | 19 ++++++++++++++++--- 1 file changed, 16 insertions(+), 3 deletions(-)