Similar to previous patch, support target-less definitions like
SEC("fentry"), SEC("freplace"), etc. For such BTF-backed program types
it is expected that user will specify BTF target programmatically at
runtime using bpf_program__set_attach_target() *before* load phase. If
not, libbpf will report this as an error.
Aslo use SEC_ATTACH_BTF flag instead of explicitly listing a set of
types that are expected to require attach_btf_id. This was an accidental
omission during custom SEC() support refactoring.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20220428185349.3799599-3-andrii@kernel.org
In a lot of cases the target of kprobe/kretprobe, tracepoint, raw
tracepoint, etc BPF program might not be known at the compilation time
and will be discovered at runtime. This was always a supported case by
libbpf, with APIs like bpf_program__attach_{kprobe,tracepoint,etc}()
accepting full target definition, regardless of what was defined in
SEC() definition in BPF source code.
Unfortunately, up till now libbpf still enforced users to specify at
least something for the fake target, e.g., SEC("kprobe/whatever"), which
is cumbersome and somewhat misleading.
This patch allows target-less SEC() definitions for basic tracing BPF
program types:
- kprobe/kretprobe;
- multi-kprobe/multi-kretprobe;
- tracepoints;
- raw tracepoints.
Such target-less SEC() definitions are meant to specify declaratively
proper BPF program type only. Attachment of them will have to be handled
programmatically using correct APIs. As such, skeleton's auto-attachment
of such BPF programs is skipped and generic bpf_program__attach() will
fail, if attempted, due to the lack of enough target information.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Song Liu <songliubraving@fb.com>
Link: https://lore.kernel.org/bpf/20220428185349.3799599-2-andrii@kernel.org
Teach libbpf to post-process BPF verifier log on BPF program load
failure and detect known error patterns to provide user with more
context.
Currently there is one such common situation: an "unguarded" failed BPF
CO-RE relocation. While failing CO-RE relocation is expected, it is
expected to be property guarded in BPF code such that BPF verifier
always eliminates BPF instructions corresponding to such failed CO-RE
relos as dead code. In cases when user failed to take such precautions,
BPF verifier provides the best log it can:
123: (85) call unknown#195896080
invalid func unknown#195896080
Such incomprehensible log error is due to libbpf "poisoning" BPF
instruction that corresponds to failed CO-RE relocation by replacing it
with invalid `call 0xbad2310` instruction (195896080 == 0xbad2310 reads
"bad relo" if you squint hard enough).
Luckily, libbpf has all the necessary information to look up CO-RE
relocation that failed and provide more human-readable description of
what's going on:
5: <invalid CO-RE relocation>
failed to resolve CO-RE relocation <byte_off> [6] struct task_struct___bad.fake_field_subprog (0:2 @ offset 8)
This hopefully makes it much easier to understand what's wrong with
user's BPF program without googling magic constants.
This BPF verifier log fixup is setup to be extensible and is going to be
used for at least one other upcoming feature of libbpf in follow up patches.
Libbpf is parsing lines of BPF verifier log starting from the very end.
Currently it processes up to 10 lines of code looking for familiar
patterns. This avoids wasting lots of CPU processing huge verifier logs
(especially for log_level=2 verbosity level). Actual verification error
should normally be found in last few lines, so this should work
reliably.
If libbpf needs to expand log beyond available log_buf_size, it
truncates the end of the verifier log. Given verifier log normally ends
with something like:
processed 2 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
... truncating this on program load error isn't too bad (end user can
always increase log size, if it needs to get complete log).
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220426004511.2691730-10-andrii@kernel.org
Refactor how CO-RE relocation is formatted. Now it dumps human-readable
representation, currently used by libbpf in either debug or error
message output during CO-RE relocation resolution process, into provided
buffer. This approach allows for better reuse of this functionality
outside of CO-RE relocation resolution, which we'll use in next patch
for providing better error message for BPF verifier rejecting BPF
program due to unguarded failed CO-RE relocation.
It also gets rid of annoying "stitching" of libbpf_print() calls, which
was the only place where we did this.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220426004511.2691730-8-andrii@kernel.org
Previously, libbpf recorded CO-RE relocations with insns_idx resolved
according to finalized subprog locations (which are appended at the end
of entry BPF program) to simplify the job of light skeleton generator.
This is necessary because once subprogs' instructions are appended to
main entry BPF program all the subprog instruction indices are shifted
and that shift is different for each entry (main) BPF program, so it's
generally impossible to map final absolute insn_idx of the finalized BPF
program to their original locations inside subprograms.
This information is now going to be used not only during light skeleton
generation, but also to map absolute instruction index to subprog's
instruction and its corresponding CO-RE relocation. So start recording
these relocations always, not just when obj->gen_loader is set.
This information is going to be freed at the end of bpf_object__load()
step, as before (but this can change in the future if there will be
a need for this information post load step).
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220426004511.2691730-7-andrii@kernel.org
Instead of using ELF section names as a joining key between .BTF.ext and
corresponding BPF programs, pre-build .BTF.ext section number to ELF
section index mapping during bpf_object__open() and use it later for
matching .BTF.ext information (func/line info or CO-RE relocations) to
their respective BPF programs and subprograms.
This simplifies corresponding joining logic and let's libbpf do
manipulations with BPF program's ELF sections like dropping leading '?'
character for non-autoloaded programs. Original joining logic in
bpf_object__relocate_core() (see relevant comment that's now removed)
was never elegant, so it's a good improvement regardless. But it also
avoids unnecessary internal assumptions about preserving original ELF
section name as BPF program's section name (which was broken when
SEC("?abc") support was added).
Fixes: a3820c481112 ("libbpf: Support opting out from autoloading BPF programs declaratively")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220426004511.2691730-5-andrii@kernel.org
Fix the bug in bpf_object__relocate_core() which can lead to finding
invalid matching BPF program when processing CO-RE relocation. IF
matching program is not found, last encountered program will be assumed
to be correct program and thus error detection won't detect the problem.
Fixes: 9c82a63cf370 ("libbpf: Fix CO-RE relocs against .text section")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220426004511.2691730-4-andrii@kernel.org
libbpf pretends it knows actual limit of BPF program instructions based
on UAPI headers it compiled with. There is neither any guarantee that
UAPI headers match host kernel, nor BPF verifier actually uses
BPF_MAXINSNS constant anymore. Just drop unhelpful "guess", BPF verifier
will emit actual reason for failure in its logs anyways.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220426004511.2691730-3-andrii@kernel.org
Extending the code in previous commits, introduce referenced kptr
support, which needs to be tagged using 'kptr_ref' tag instead. Unlike
unreferenced kptr, referenced kptr have a lot more restrictions. In
addition to the type matching, only a newly introduced bpf_kptr_xchg
helper is allowed to modify the map value at that offset. This transfers
the referenced pointer being stored into the map, releasing the
references state for the program, and returning the old value and
creating new reference state for the returned pointer.
Similar to unreferenced pointer case, return value for this case will
also be PTR_TO_BTF_ID_OR_NULL. The reference for the returned pointer
must either be eventually released by calling the corresponding release
function, otherwise it must be transferred into another map.
It is also allowed to call bpf_kptr_xchg with a NULL pointer, to clear
the value, and obtain the old value if any.
BPF_LDX, BPF_STX, and BPF_ST cannot access referenced kptr. A future
commit will permit using BPF_LDX for such pointers, but attempt at
making it safe, since the lifetime of object won't be guaranteed.
There are valid reasons to enforce the restriction of permitting only
bpf_kptr_xchg to operate on referenced kptr. The pointer value must be
consistent in face of concurrent modification, and any prior values
contained in the map must also be released before a new one is moved
into the map. To ensure proper transfer of this ownership, bpf_kptr_xchg
returns the old value, which the verifier would require the user to
either free or move into another map, and releases the reference held
for the pointer being moved in.
In the future, direct BPF_XCHG instruction may also be permitted to work
like bpf_kptr_xchg helper.
Note that process_kptr_func doesn't have to call
check_helper_mem_access, since we already disallow rdonly/wronly flags
for map, which is what check_map_access_type checks, and we already
ensure the PTR_TO_MAP_VALUE refers to kptr by obtaining its off_desc,
so check_map_access is also not required.
Signed-off-by: Kumar Kartikeya Dwivedi <memxor@gmail.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220424214901.2743946-4-memxor@gmail.com
Teach bpf_link_create() to fallback to bpf_raw_tracepoint_open() on
older kernels for programs that are attachable through
BPF_RAW_TRACEPOINT_OPEN. This makes bpf_link_create() more unified and
convenient interface for creating bpf_link-based attachments.
With this approach end users can just use bpf_link_create() for
tp_btf/fentry/fexit/fmod_ret/lsm program attachments without needing to
care about kernel support, as libbpf will handle this transparently. On
the other hand, as newer features (like BPF cookie) are added to
LINK_CREATE interface, they will be readily usable though the same
bpf_link_create() API without any major refactoring from user's
standpoint.
bpf_program__attach_btf_id() is now using bpf_link_create() internally
as well and will take advantaged of this unified interface when BPF
cookie is added for fentry/fexit.
Doing proactive feature detection of LINK_CREATE support for
fentry/tp_btf/etc is quite involved. It requires parsing vmlinux BTF,
determining some stable and guaranteed to be in all kernels versions
target BTF type (either raw tracepoint or fentry target function),
actually attaching this program and thus potentially affecting the
performance of the host kernel briefly, etc. So instead we are taking
much simpler "lazy" approach of falling back to
bpf_raw_tracepoint_open() call only if initial LINK_CREATE command
fails. For modern kernels this will mean zero added overhead, while
older kernels will incur minimal overhead with a single fast-failing
LINK_CREATE call.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Reviewed-by: Kui-Feng Lee <kuifeng@fb.com>
Link: https://lore.kernel.org/bpf/20220421033945.3602803-3-andrii@kernel.org
Add riscv-specific USDT argument specification parsing logic.
riscv USDT argument format is shown below:
- Memory dereference case:
"size@off(reg)", e.g. "-8@-88(s0)"
- Constant value case:
"size@val", e.g. "4@5"
- Register read case:
"size@reg", e.g. "-8@a1"
s8 will be marked as poison while it's a reg of riscv, we need
to alias it in advance. Both RV32 and RV64 have been tested.
Signed-off-by: Pu Lehui <pulehui@huawei.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220419145238.482134-3-pulehui@huawei.com
Establish SEC("?abc") naming convention (i.e., adding question mark in
front of otherwise normal section name) that allows to set corresponding
program's autoload property to false. This is effectively just
a declarative way to do bpf_program__set_autoload(prog, false).
Having a way to do this declaratively in BPF code itself is useful and
convenient for various scenarios. E.g., for testing, when BPF object
consists of multiple independent BPF programs that each needs to be
tested separately. Opting out all of them by default and then setting
autoload to true for just one of them at a time simplifies testing code
(see next patch for few conversions in BPF selftests taking advantage of
this new feature).
Another real-world use case is in libbpf-tools for cases when different
BPF programs have to be picked depending on particulars of the host
kernel due to various incompatible changes (like kernel function renames
or signature change, or to pick kprobe vs fentry depending on
corresponding kernel support for the latter). Marking all the different
BPF program candidates as non-autoloaded declaratively makes this more
obvious in BPF source code and allows simpler code in user-space code.
When BPF program marked as SEC("?abc") it is otherwise treated just like
SEC("abc") and bpf_program__section_name() reported will be "abc".
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220419002452.632125-1-andrii@kernel.org
Parsing of USDT arguments is architecture-specific. On aarch64 it is
relatively easy since registers used are x[0-31] and sp. Format is
slightly different compared to x86_64. Possible forms are:
- "size@[reg[,offset]]" for dereferences, e.g. "-8@[sp,76]" and "-4@[sp]";
- "size@reg" for register values, e.g. "-4@x0";
- "size@value" for raw values, e.g. "-8@1".
Signed-off-by: Alan Maguire <alan.maguire@oracle.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/1649690496-1902-2-git-send-email-alan.maguire@oracle.com
Background:
Libbpf automatically replaces calls to BPF bpf_probe_read_{kernel,user}
[_str]() helpers with bpf_probe_read[_str](), if libbpf detects that
kernel doesn't support new APIs. Specifically, libbpf invokes the
probe_kern_probe_read_kernel function to load a small eBPF program into
the kernel in which bpf_probe_read_kernel API is invoked and lets the
kernel checks whether the new API is valid. If the loading fails, libbpf
considers the new API invalid and replaces it with the old API.
static int probe_kern_probe_read_kernel(void)
{
struct bpf_insn insns[] = {
BPF_MOV64_REG(BPF_REG_1, BPF_REG_10), /* r1 = r10 (fp) */
BPF_ALU64_IMM(BPF_ADD, BPF_REG_1, -8), /* r1 += -8 */
BPF_MOV64_IMM(BPF_REG_2, 8), /* r2 = 8 */
BPF_MOV64_IMM(BPF_REG_3, 0), /* r3 = 0 */
BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, BPF_FUNC_probe_read_kernel),
BPF_EXIT_INSN(),
};
int fd, insn_cnt = ARRAY_SIZE(insns);
fd = bpf_prog_load(BPF_PROG_TYPE_KPROBE, NULL,
"GPL", insns, insn_cnt, NULL);
return probe_fd(fd);
}
Bug:
On older kernel versions [0], the kernel checks whether the version
number provided in the bpf syscall, matches the LINUX_VERSION_CODE.
If not matched, the bpf syscall fails. eBPF However, the
probe_kern_probe_read_kernel code does not set the kernel version
number provided to the bpf syscall, which causes the loading process
alwasys fails for old versions. It means that libbpf will replace the
new API with the old one even the kernel supports the new one.
Solution:
After a discussion in [1], the solution is using BPF_PROG_TYPE_TRACEPOINT
program type instead of BPF_PROG_TYPE_KPROBE because kernel does not
enfoce version check for tracepoint programs. I test the patch in old
kernels (4.18 and 4.19) and it works well.
[0] https://elixir.bootlin.com/linux/v4.19/source/kernel/bpf/syscall.c#L1360
[1] Closes: https://github.com/libbpf/libbpf/issues/473
Signed-off-by: Runqing Yang <rainkin1993@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220409144928.27499-1-rainkin1993@gmail.com
Enable the following options in Kconfig for x86-64 and s390x:
CONFIG_NETFILTER_SYNPROXY=y
CONFIG_NETFILTER_XT_TARGET_CT=y
CONFIG_NETFILTER_XT_MATCH_STATE=y
CONFIG_IP_NF_FILTER=y
CONFIG_IP_NF_TARGET_SYNPROXY=y
CONFIG_IP_NF_RAW=y
These options are needed to run the selftests for the new BPF SYN cookie
helpers.
Signed-off-by: Maxim Mikityanskiy <maximmi@nvidia.com>
Add `ethtool` as a dependency to the rootfs image.
Tested by running and building the rootfs images with both
`sudo ./mkrootfs_arch.sh`
and
`sudo ./mkrootfs_debian.sh`
and running in qemu with:
```
wget https://libbpf-ci.s3-us-west-1.amazonaws.com/x86_64/vmlinuz-5.5.0
rootfs_img=rootfs.img kernel_bzimage=vmlinuz-5.5.0
mkdir rootfs
touch rootfs.img
truncate -s 2G rootfs.img
sudo mount -o loop rootfs.img rootfs
cat ~/Downloads/libbpf-vmtest-rootfs-2022.04.25.tar.zst | sudo tar -C rootfs -I zstd -xvf -
sudo install -m 755 -o root -g root /dev/stdin rootfs/etc/rcS.d/S50-startup <<'EOF'
ethtool -h
cat /etc/issue
EOF
qemu-system-x86_64 -nodefaults -display none -serial mon:stdio -enable-kvm -m 4G -drive file="${rootfs_img}",format=raw,index=1,media=disk,if=virtio,cache=none -kernel "${kernel_bzimage}" -append "root=/dev/vda rw console=ttyS0,115200"
```
The last block printed ethtool's help, confirming the presence of
ethtool in the rootfs.
`libbpf-vmtest-rootfs-2022.04.25.tar.zst` was generated and uploaded to S3. INDEX in libbpf/ci needs to be changed to make the CI pick it up.
Use __weak __hidden for bpf_usdt_xxx() APIs instead of much more
confusing `static inline __noinline`. This was previously impossible due
to libbpf erroring out on CO-RE relocations pointing to eliminated weak
subprogs. Now that previous patch fixed this issue, switch back to
__weak __hidden as it's a more direct way of specifying the desired
behavior.
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220408181425.2287230-3-andrii@kernel.org
During BPF static linking, all the ELF relocations and .BTF.ext
information (including CO-RE relocations) are preserved for __weak
subprograms that were logically overriden by either previous weak
subprogram instance or by corresponding "strong" (non-weak) subprogram.
This is just how native user-space linkers work, nothing new.
But libbpf is over-zealous when processing CO-RE relocation to error out
when CO-RE relocation belonging to such eliminated weak subprogram is
encountered. Instead of erroring out on this expected situation, log
debug-level message and skip the relocation.
Fixes: db2b8b06423c ("libbpf: Support CO-RE relocations for multi-prog sections")
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220408181425.2287230-2-andrii@kernel.org
During BTF fix up for global variables, global variable can be global
weak and will have STB_WEAK binding in ELF. Support such global
variables in addition to non-weak ones.
This is not the problem when using BPF static linking, as BPF static
linker "fixes up" BTF during generation so that libbpf doesn't have to
do it anymore during bpf_object__open(), which led to this not being
noticed for a while, along with a pretty rare (currently) use of __weak
variables and maps.
Reported-by: Hengqi Chen <hengqi.chen@gmail.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Link: https://lore.kernel.org/bpf/20220407230446.3980075-2-andrii@kernel.org
The logic is superficially similar to that of x86, but the small
differences (no need for register table and dynamic allocation of
register names, no $ sign before constants) make maintaining a common
implementation too burdensome. Therefore simply add a s390x-specific
version of parse_usdt_arg().
Note that while bcc supports index registers, this patch does not. This
should not be a problem in most cases, since s390 uses a default value
"nor" for STAP_SDT_ARG_CONSTRAINT.
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220407214411.257260-4-iii@linux.ibm.com
BPF_USDT_ARG_REG_DEREF handling always reads 8 bytes, regardless of
the actual argument size. On little-endian the relevant argument bits
end up in the lower bits of val, and later on the code that handles
all the argument types expects them to be there.
On big-endian they end up in the upper bits of val, breaking that
expectation. Fix by right-shifting val on big-endian.
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220407214411.257260-3-iii@linux.ibm.com
Fix several typos and references to non-existing headers.
Also use __BYTE_ORDER__ instead of __BYTE_ORDER for consistency with
the rest of the bpf code - see commit 45f2bebc8079 ("libbpf: Fix
endianness detection in BPF_CORE_READ_BITFIELD_PROBED()") for
rationale).
Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Link: https://lore.kernel.org/bpf/20220407214411.257260-2-iii@linux.ibm.com
As reported by Naresh:
perf build errors on i386 [1] on Linux next-20220407 [2]
usdt.c:1181:5: error: "__x86_64__" is not defined, evaluates to 0
[-Werror=undef]
1181 | #if __x86_64__
| ^~~~~~~~~~
usdt.c:1196:5: error: "__x86_64__" is not defined, evaluates to 0
[-Werror=undef]
1196 | #if __x86_64__
| ^~~~~~~~~~
cc1: all warnings being treated as errors
Use #ifdef instead of #if to avoid this.
Fixes: 4c59e584d158 ("libbpf: Add x86-specific USDT arg spec parsing logic")
Reported-by: Naresh Kamboju <naresh.kamboju@linaro.org>
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20220407203842.3019904-1-andrii@kernel.org
