Skip to contentSkip to conversation

bpf: Reject attaching fexit to __noreturn functions

PATCH Superseded Yafang Shao 5 participants 19 messages Feb 11, 2025 – Feb 14, 2025

Subsystems (5)
Kselftest BPF BPF selftests BPF core +1 more subsystems

v1 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

cover bpf: Reject attaching fexit to __noreturn functions

Attaching fexit probes to functions marked with __noreturn may lead to unpredictable behavior. To avoid this, we will reject attaching probes to such functions. Currently, there is no ideal solution, so we will hardcode a check for all __noreturn functions. Since objtool already handles this, we will leverage its implementation.

Once a more robust solution is found, this workaround can be removed.

Yafang Shao (3):
objtool: Move noreturns.h to a common location
bpf: Reject attaching fexit to functions annotated with __noreturn selftests/bpf: Add selftest for attaching fexit to __noreturn
functions

 {tools/objtool => include/linux}/noreturns.h  |  0
 kernel/bpf/verifier.c                         | 10 ++++
 tools/include/linux/noreturns.h               | 52 +++++++++++++++++++
 tools/objtool/Documentation/objtool.txt       |  3 +-
 tools/objtool/check.c                         |  2 +-
 .../bpf/prog_tests/fexit_noreturns.c          | 13 +++++
 .../selftests/bpf/progs/fexit_noreturns.c     | 13 +++++
 7 files changed, 91 insertions(+), 2 deletions(-)
 rename {tools/objtool => include/linux}/noreturns.h (100%)
 create mode 100644 tools/include/linux/noreturns.h
 create mode 100644 tools/testing/selftests/bpf/prog_tests/fexit_noreturns.c
 create mode 100644 tools/testing/selftests/bpf/progs/fexit_noreturns.c
Signature · 1 line
2.43.5
#1 lore raw

PATCH 1/3 objtool: Move noreturns.h to a common location

It will used by bpf to reject attaching fexit prog to functions annotated with __noreturn.

Cc Josh Poimboeuf <[email protected]>
Cc Peter Zijlstra <[email protected]>
---
 {tools/objtool => include/linux}/noreturns.h |  0
 tools/include/linux/noreturns.h              | 52 ++++++++++++++++++++
 tools/objtool/Documentation/objtool.txt      |  3 +-
 tools/objtool/check.c                        |  2 +-
 4 files changed, 55 insertions(+), 2 deletions(-)
 rename {tools/objtool => include/linux}/noreturns.h (100%)
 create mode 100644 tools/include/linux/noreturns.h
include/linux/noreturns.h renamed from tools/objtool/noreturns.h +0 -0
similarity index 100%
rename from tools/objtool/noreturns.h
rename to include/linux/noreturns.h
tools/include/linux/noreturns.h new file +52 -0

52 lines not on this page · raw message

tools/objtool/Documentation/objtool.txt +2 -1
index 7c3ee959b63c..726db5b2b1a5 100644
@@ -326,7 +326,8 @@ the objtool maintainers.
 
    The call from foo() to bar() doesn't return, but bar() is missing the
    __noreturn annotation.  NOTE: In addition to annotating the function
-   with __noreturn, please also add it to tools/objtool/noreturns.h.
+   with __noreturn, please also add it to tools/include/linux/noreturns.h and
+   include/linux/noreturns.h.
 
 4. file.o: warning: objtool: func(): can't find starting instruction
    or
tools/objtool/check.c +1 -1
index 753dbc4f8198..2940ddc56b1a 100644
@@ -250,7 +250,7 @@ static bool __dead_end_function(struct objtool_file *file, struct symbol *func,
 
 #define NORETURN(func) __stringify(func),
 	static const char * const global_noreturns[] = {
-#include "noreturns.h"
+#include <linux/noreturns.h>
 	};
 #undef NORETURN
 
Signature · 1 line
2.43.5
maintainer
#2 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

On Tue, 2025-02-11 at 10:33 +0800, Yafang Shao wrote:

[...]

Quoted 6 lines
tools/include/linux/noreturns.h new file +0 -0
new file mode 100644
index 000000000000..b2174894f9f7

tools/include/linux/noreturns.h line 8

Quoted 8 lines
tools/include/linux/noreturns.h new file +8 -0
@@ -0,0 +1,52 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+/*
+ * This is a (sorted!) list of all known __noreturn functions in the kernel.
+ * It's needed for objtool to properly reverse-engineer the control flow graph.
+ *
+ * Yes, this is unfortunate.  A better solution is in the works.
+ */

I'm probably out of context for this discussion, sorry if I'm raising points already discussed.

The DW_AT_noreturn attribute is defined for DWARF. A simple script like [1] could be used to find all functions with this attribute known to DWARF. Using this script I see several functions present in my kernel but not present in the NORETURN list from this patch:
- abort
- devtmpfs_work_loop
- play_dead
- rcu_gp_kthread
- rcu_tasks_kthread

All these are marked as FUNC symbols when doing 'readelf --symbols vmlinux'.

'pahole' could be modified to look for DW_AT_noreturn attributes and add this information in BTF. E.g. by adding special btf_decl_tag to corresponding FUNC definitions. This won't work if kernel is compiled w/o BTF, of-course.

[1] https://gist.github.com/eddyz87/d8513a731dfe7e2be52b346aef1de353

Quoted 44 lines
tools/include/linux/noreturns.h new file +44 -0
+NORETURN(__fortify_panic)
+NORETURN(__ia32_sys_exit)
+NORETURN(__ia32_sys_exit_group)
+NORETURN(__kunit_abort)
+NORETURN(__module_put_and_kthread_exit)
+NORETURN(__stack_chk_fail)
+NORETURN(__tdx_hypercall_failed)
+NORETURN(__ubsan_handle_builtin_unreachable)
+NORETURN(__x64_sys_exit)
+NORETURN(__x64_sys_exit_group)
+NORETURN(arch_cpu_idle_dead)
+NORETURN(bch2_trans_in_restart_error)
+NORETURN(bch2_trans_restart_error)
+NORETURN(bch2_trans_unlocked_error)
+NORETURN(cpu_bringup_and_idle)
+NORETURN(cpu_startup_entry)
+NORETURN(do_exit)
+NORETURN(do_group_exit)
+NORETURN(do_task_dead)
+NORETURN(ex_handler_msr_mce)
+NORETURN(hlt_play_dead)
+NORETURN(hv_ghcb_terminate)
+NORETURN(kthread_complete_and_exit)
+NORETURN(kthread_exit)
+NORETURN(kunit_try_catch_throw)
+NORETURN(machine_real_restart)
+NORETURN(make_task_dead)
+NORETURN(mpt_halt_firmware)
+NORETURN(nmi_panic_self_stop)
+NORETURN(panic)
+NORETURN(panic_smp_self_stop)
+NORETURN(rest_init)
+NORETURN(rewind_stack_and_make_dead)
+NORETURN(rust_begin_unwind)
+NORETURN(rust_helper_BUG)
+NORETURN(sev_es_terminate)
+NORETURN(snp_abort)
+NORETURN(start_kernel)
+NORETURN(stop_this_cpu)
+NORETURN(usercopy_abort)
+NORETURN(x86_64_start_kernel)
+NORETURN(x86_64_start_reservations)
+NORETURN(xen_cpu_bringup_again)
+NORETURN(xen_start_kernel)

[...]

#3 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

Quoted 14 lines · Eduard Zingerman

On Tue, 2025-02-11 at 10:33 +0800, Yafang Shao wrote:

[...]

Quoted 9 lines
tools/include/linux/noreturns.h new file +3 -0
new file mode 100644
index 000000000000..b2174894f9f7
@@ -0,0 +1,52 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+
+/*
Quoted 8 lines

tools/include/linux/noreturns.h line 8

Quoted 5 lines
tools/include/linux/noreturns.h new file +5 -0
+ * This is a (sorted!) list of all known __noreturn functions in the kernel.
+ * It's needed for objtool to properly reverse-engineer the control flow graph.
+ *
+ * Yes, this is unfortunate.  A better solution is in the works.
+ */

I'm probably out of context for this discussion, sorry if I'm raising points already discussed.

Please refer to the discussion at
https://lore.kernel.org/bpf/CAADnVQKXgPTQsjUDB3tjZ46aPWvoEcxBCnDXro8WPtNhkGNFyg@mail.gmail.com/
for more details.

Quoted 11 lines

The DW_AT_noreturn attribute is defined for DWARF. A simple script like [1] could be used to find all functions with this attribute known to DWARF. Using this script I see several functions present in my kernel but not present in the NORETURN list from this patch:
- abort
- devtmpfs_work_loop
- play_dead
- rcu_gp_kthread
- rcu_tasks_kthread

Quoted 8 lines

All these are marked as FUNC symbols when doing 'readelf --symbols vmlinux'.

'pahole' could be modified to look for DW_AT_noreturn attributes and add this information in BTF. E.g. by adding special btf_decl_tag to corresponding FUNC definitions. This won't work if kernel is compiled w/o BTF, of-course.

[1] https://gist.github.com/eddyz87/d8513a731dfe7e2be52b346aef1de353

Thank you for your example—it’s been really helpful. Following Alexei's suggestion, we’ll address this in the next step. We need to apply the fix to the current kernel and backport it, so for now, the simplest approach is to implement the current solution.

--
Regards
Yafang

maintainer
#4 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

In reply to Eduard Zingerman
Quoted 13 lines · Eduard Zingerman

I'm probably out of context for this discussion, sorry if I'm raising points already discussed.

The DW_AT_noreturn attribute is defined for DWARF. A simple script like [1] could be used to find all functions with this attribute known to DWARF. Using this script I see several functions present in my kernel but not present in the NORETURN list from this patch:
- abort
- devtmpfs_work_loop
- play_dead
- rcu_gp_kthread
- rcu_tasks_kthread

Quoted 8 lines

All these are marked as FUNC symbols when doing 'readelf --symbols vmlinux'.

'pahole' could be modified to look for DW_AT_noreturn attributes and add this information in BTF. E.g. by adding special btf_decl_tag to corresponding FUNC definitions. This won't work if kernel is compiled w/o BTF, of-course.

[1] https://gist.github.com/eddyz87/d8513a731dfe7e2be52b346aef1de353

I also suggested this, I agree this is a much better way to go. noreturns.h is manually maintained based on objtool warnings and I'm not surprised it has missing entries.

Alexei mentioned 30k+ noreturns, but when I eliminate dups and __compiletime_assert_* it's a much smaller list:

$ ./noreturn_printer vmlinux |sort |uniq |grep -v compiletime_assert
arch_cpu_idle_dead               external idle.c
arch_cpu_idle_dead               external process.c
cpu_startup_entry                external cpu.h
cpu_startup_entry                external idle.c
do_exit                          external exit.c
do_exit                          external kernel.h
do_group_exit                    external exit.c
do_group_exit                    external task.h
do_task_dead                     external core.c
do_task_dead                     external task.h
doublefault_shim                 external doublefault_32.c
ex_handler_msr_mce               external core.c
ex_handler_msr_mce               external extable.h
__fortify_panic                  external fortify-string.h
__fortify_panic                  external string_helpers.c
i386_start_kernel                external head32.c
__ia32_sys_exit                  external syscalls_32.h
__ia32_sys_exit_group            external syscalls_32.h
kthread_complete_and_exit        external kthread.c
kthread_exit                     external kthread.c
kthread_exit                     external kthread.h
machine_real_restart             external reboot.c
make_task_dead                   external exit.c
__module_put_and_kthread_exit    external main.c
__module_put_and_kthread_exit    external module.h
nmi_panic_self_stop              external panic.c
panic                            external panic.c
panic                            external panic.h
panic_smp_self_stop              external panic.c
play_dead                                 process.c
rcu_gp_kthread                            tree.c
rcu_tasks_kthread                         tasks.h
rest_init                                 main.c
rewind_stack_and_make_dead       external dumpstack.c
start_kernel                     external main.c
start_kernel                     external start_kernel.h
stop_this_cpu                    external process.c
stop_this_cpu                    external processor.h

Also, for objtool we could use something based on your program to autogenerate noreturns.h. Only problem is, objtool doesn't currently have a dependency on CONFIG_DEBUG_INFO. Another option we've considered is compiler annotations (or compiler plugins).

Signature · 1 line
Josh
#5 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

Quoted 71 lines · Josh Poimboeuf
Quoted 13 lines · Eduard Zingerman

I'm probably out of context for this discussion, sorry if I'm raising points already discussed.

The DW_AT_noreturn attribute is defined for DWARF. A simple script like [1] could be used to find all functions with this attribute known to DWARF. Using this script I see several functions present in my kernel but not present in the NORETURN list from this patch:
- abort
- devtmpfs_work_loop
- play_dead
- rcu_gp_kthread
- rcu_tasks_kthread

Quoted 8 lines

All these are marked as FUNC symbols when doing 'readelf --symbols vmlinux'.

'pahole' could be modified to look for DW_AT_noreturn attributes and add this information in BTF. E.g. by adding special btf_decl_tag to corresponding FUNC definitions. This won't work if kernel is compiled w/o BTF, of-course.

[1] https://gist.github.com/eddyz87/d8513a731dfe7e2be52b346aef1de353

I also suggested this, I agree this is a much better way to go. noreturns.h is manually maintained based on objtool warnings and I'm not surprised it has missing entries.

Alexei mentioned 30k+ noreturns, but when I eliminate dups and __compiletime_assert_* it's a much smaller list:

$ ./noreturn_printer vmlinux |sort |uniq |grep -v compiletime_assert
arch_cpu_idle_dead               external idle.c
arch_cpu_idle_dead               external process.c
cpu_startup_entry                external cpu.h
cpu_startup_entry                external idle.c
do_exit                          external exit.c
do_exit                          external kernel.h
do_group_exit                    external exit.c
do_group_exit                    external task.h
do_task_dead                     external core.c
do_task_dead                     external task.h
doublefault_shim                 external doublefault_32.c
ex_handler_msr_mce               external core.c
ex_handler_msr_mce               external extable.h
__fortify_panic                  external fortify-string.h
__fortify_panic                  external string_helpers.c
i386_start_kernel                external head32.c
__ia32_sys_exit                  external syscalls_32.h
__ia32_sys_exit_group            external syscalls_32.h
kthread_complete_and_exit        external kthread.c
kthread_exit                     external kthread.c
kthread_exit                     external kthread.h
machine_real_restart             external reboot.c
make_task_dead                   external exit.c
__module_put_and_kthread_exit    external main.c
__module_put_and_kthread_exit    external module.h
nmi_panic_self_stop              external panic.c
panic                            external panic.c
panic                            external panic.h
panic_smp_self_stop              external panic.c
play_dead                                 process.c
rcu_gp_kthread                            tree.c
rcu_tasks_kthread                         tasks.h
rest_init                                 main.c
rewind_stack_and_make_dead       external dumpstack.c
start_kernel                     external main.c
start_kernel                     external start_kernel.h
stop_this_cpu                    external process.c
stop_this_cpu                    external processor.h
Quoted 2 lines

Also, for objtool we could use something based on your program to autogenerate noreturns.h.

Automatically generating noreturns.h is a great idea, as it would make the file accessible for use by others as well.

Quoted 2 lines

Only problem is, objtool doesn't currently
have a dependency on CONFIG_DEBUG_INFO.

Is there any reason we can't make it dependent on CONFIG_DEBUG_INFO?"

Quoted 2 lines

Another option we've considered
is compiler annotations (or compiler plugins).

Signature · 2 lines
Regards
Yafang
maintainer
#6 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

Quoted 5 lines · Yafang Shao
Quoted 2 lines · Josh Poimboeuf

Only problem is, objtool doesn't currently
have a dependency on CONFIG_DEBUG_INFO.

Is there any reason we can't make it dependent on CONFIG_DEBUG_INFO?"

Objtool is enabled by default on x86 and is pretty much required at this point. We definitely don't want to force enable CONFIG_DEBUG_INFO as that will slow the build down considerably.

Signature · 1 line
Josh
maintainer
#7 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

Quoted 9 lines · Josh Poimboeuf
Quoted 5 lines · Yafang Shao
Quoted 2 lines · Josh Poimboeuf

Only problem is, objtool doesn't currently
have a dependency on CONFIG_DEBUG_INFO.

Is there any reason we can't make it dependent on CONFIG_DEBUG_INFO?"

Objtool is enabled by default on x86 and is pretty much required at this point.

x86_64, but yeah, disabling objtool for x86_64 is going to be *really* hard.

i386 is still limping along without -- mostly because nobody cared enough to make it work.

Quoted 2 lines

We definitely don't want to force enable CONFIG_DEBUG_INFO as that will slow the build down considerably.

This, very much this. Same reason I hardly every use clang to build a kernel -- just too slow.

maintainer
#8 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

In reply to Josh Poimboeuf
Quoted 4 lines · Josh Poimboeuf

Also, for objtool we could use something based on your program to autogenerate noreturns.h. Only problem is, objtool doesn't currently have a dependency on CONFIG_DEBUG_INFO. Another option we've considered is compiler annotations (or compiler plugins).

But we don't need to re-generate the file on every build, right? We can have every DBUG_INFO build verify the file is still complete.

maintainer
#9 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

Quoted 9 lines · Peter Zijlstra
Quoted 4 lines · Josh Poimboeuf

Also, for objtool we could use something based on your program to autogenerate noreturns.h. Only problem is, objtool doesn't currently have a dependency on CONFIG_DEBUG_INFO. Another option we've considered is compiler annotations (or compiler plugins).

But we don't need to re-generate the file on every build, right? We can have every DBUG_INFO build verify the file is still complete.

Hm, yeah we could do that. It's still not ideal, as a human now has to go fix those warnings...

But it's probably better than the current situation.

Signature · 1 line
Josh
maintainer
#10 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

In reply to Yafang Shao
Quoted 9 lines · Yafang Shao

It will used by bpf to reject attaching fexit prog to functions annotated with __noreturn.

Cc Josh Poimboeuf <[email protected]>
Cc Peter Zijlstra <[email protected]>
---
 {tools/objtool => include/linux}/noreturns.h |  0
 tools/include/linux/noreturns.h              | 52 ++++++++++++++++++++

Instead of moving the file, please make a copy and add it to tools/objtool/sync-check.sh.

Signature · 1 line
Josh
#11 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

Quoted 5 lines · Josh Poimboeuf
Quoted 3 lines · Yafang Shao

It will used by bpf to reject attaching fexit prog to functions annotated with __noreturn.

Quoted 8 lines
Quoted 6 lines
Cc Josh Poimboeuf <[email protected]>
Cc Peter Zijlstra <[email protected]>
---
 {tools/objtool => include/linux}/noreturns.h |  0
 tools/include/linux/noreturns.h              | 52 ++++++++++++++++++++

Instead of moving the file, please make a copy

Are you suggesting we keep tools/objtool/noreturns.h as is and simply copy it to include/linux/noreturns.h? In other words, the two files would be tools/objtool/noreturns.h and include/linux/noreturns.h.

Quoted 2 lines

and add it to
tools/objtool/sync-check.sh.

Sure.

Signature · 2 lines
Regards
Yafang
maintainer
#12 lore raw

Re: [PATCH bpf-next 1/3] objtool: Move noreturns.h to a common location

Quoted 13 lines · Yafang Shao
Quoted 12 lines · Josh Poimboeuf
Quoted 9 lines · Yafang Shao

It will used by bpf to reject attaching fexit prog to functions annotated with __noreturn.

Cc Josh Poimboeuf <[email protected]>
Cc Peter Zijlstra <[email protected]>
---
 {tools/objtool => include/linux}/noreturns.h |  0
 tools/include/linux/noreturns.h              | 52 ++++++++++++++++++++
Quoted 5 lines
Quoted 1 line

Instead of moving the file, please make a copy

Are you suggesting we keep tools/objtool/noreturns.h as is and simply copy it to include/linux/noreturns.h? In other words, the two files would be tools/objtool/noreturns.h and include/linux/noreturns.h.

Right.

Signature · 1 line
Josh
#13 lore raw

PATCH 2/3 bpf: Reject attaching fexit to functions annotated with __noreturn

In reply to Yafang Shao

If we attach fexit to a function annotated with __noreturn, it will cause an issue that the bpf trampoline image will be left over even if the bpf link has been destroyed. Take attaching do_exit() for example. The fexit works as follows,

bpf_trampoline
+ __bpf_tramp_enter
+ percpu_ref_get(&tr->pcref);

+ call do_exit()

+ __bpf_tramp_exit
+ percpu_ref_put(&tr->pcref);

Since do_exit() never returns, the refcnt of the trampoline image is never decremented, preventing it from being freed. This can be verified with as follows,

  $ bpftool link show                                   <<<< nothing output
  $ grep "bpf_trampoline_[0-9]" /proc/kallsyms
  ffffffffc04cb000 t bpf_trampoline_6442526459    [bpf] <<<< leftover

With this change, attaching fexit probes to functions like do_exit() will be rejected.

$ ./fexit
libbpf: prog 'fexit': BPF program load failed: -EINVAL
libbpf: prog 'fexit': -- BEGIN PROG LOAD LOG --
processed 0 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
-- END PROG LOAD LOG --
libbpf: prog 'fexit': failed to load: -EINVAL
libbpf: failed to load object 'fexit_bpf'
libbpf: failed to load BPF skeleton 'fexit_bpf': -EINVAL
failed to load BPF object -22
---
 kernel/bpf/verifier.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
kernel/bpf/verifier.c +10 -0
index 9971c03adfd5..f7224fc61e0c 100644
@@ -22841,6 +22841,13 @@ BTF_ID(func, __rcu_read_unlock)
 #endif
 BTF_SET_END(btf_id_deny)
 
+/* The functions annotated with __noreturn are denied. */
+BTF_SET_START(fexit_deny)
+#define NORETURN(fn) BTF_ID(func, fn)
+#include <linux/noreturns.h>
+#undef NORETURN
+BTF_SET_END(fexit_deny)
+
 static bool can_be_sleepable(struct bpf_prog *prog)
 {
 	if (prog->type == BPF_PROG_TYPE_TRACING) {
@@ -22929,6 +22936,9 @@ static int check_attach_btf_id(struct bpf_verifier_env *env)
 	} else if (prog->type == BPF_PROG_TYPE_TRACING &&
 		   btf_id_set_contains(&btf_id_deny, btf_id)) {
 		return -EINVAL;
+	} else if (prog->expected_attach_type == BPF_TRACE_FEXIT &&
+		   btf_id_set_contains(&fexit_deny, btf_id)) {
+		return -EINVAL;
 	}
 
 	key = bpf_trampoline_compute_key(tgt_prog, prog->aux->attach_btf, btf_id);
Signature · 1 line
2.43.5
reviewer
#14 lore raw

Re: [PATCH bpf-next 2/3] bpf: Reject attaching fexit to functions annotated with __noreturn

Quoted 60 lines · Yafang Shao

If we attach fexit to a function annotated with __noreturn, it will cause an issue that the bpf trampoline image will be left over even if the bpf link has been destroyed. Take attaching do_exit() for example. The fexit works as follows,

bpf_trampoline
+ __bpf_tramp_enter
+ percpu_ref_get(&tr->pcref);

+ call do_exit()

+ __bpf_tramp_exit
+ percpu_ref_put(&tr->pcref);

Since do_exit() never returns, the refcnt of the trampoline image is never decremented, preventing it from being freed. This can be verified with as follows,

  $ bpftool link show                                   <<<< nothing output
  $ grep "bpf_trampoline_[0-9]" /proc/kallsyms
  ffffffffc04cb000 t bpf_trampoline_6442526459    [bpf] <<<< leftover

With this change, attaching fexit probes to functions like do_exit() will be rejected.

$ ./fexit
libbpf: prog 'fexit': BPF program load failed: -EINVAL
libbpf: prog 'fexit': -- BEGIN PROG LOAD LOG --
processed 0 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
-- END PROG LOAD LOG --
libbpf: prog 'fexit': failed to load: -EINVAL
libbpf: failed to load object 'fexit_bpf'
libbpf: failed to load BPF skeleton 'fexit_bpf': -EINVAL
failed to load BPF object -22
---
 kernel/bpf/verifier.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
kernel/bpf/verifier.c +7 -0
index 9971c03adfd5..f7224fc61e0c 100644
@@ -22841,6 +22841,13 @@ BTF_ID(func, __rcu_read_unlock)
 #endif
 BTF_SET_END(btf_id_deny)
 
+/* The functions annotated with __noreturn are denied. */
+BTF_SET_START(fexit_deny)
+#define NORETURN(fn) BTF_ID(func, fn)
+#include <linux/noreturns.h>
+#undef NORETURN
+BTF_SET_END(fexit_deny)
+
 static bool can_be_sleepable(struct bpf_prog *prog)
 {
        if (prog->type == BPF_PROG_TYPE_TRACING) {

kernel/bpf/verifier.c static int check_attach_btf_id(struct bpf_verifier_env *env) · line 22940

Quoted 5 lines
kernel/bpf/verifier.c +2 -0
@@ -22929,6 +22936,9 @@ static int check_attach_btf_id(struct bpf_verifier_env *env)
        } else if (prog->type == BPF_PROG_TYPE_TRACING &&
                   btf_id_set_contains(&btf_id_deny, btf_id)) {
                return -EINVAL;
+       } else if (prog->expected_attach_type == BPF_TRACE_FEXIT &&
+                  btf_id_set_contains(&fexit_deny, btf_id)) {

Please add a verifier log (with verbose()) here, so that the user knows why the program failed to attach.

Thanks,
Song

Quoted 7 lines
kernel/bpf/verifier.c +1 -1
+               return -EINVAL;
        }
 
        key = bpf_trampoline_compute_key(tgt_prog, prog->aux->attach_btf, btf_id);
--

2.43.5

#15 lore raw

Re: [PATCH bpf-next 2/3] bpf: Reject attaching fexit to functions annotated with __noreturn

Quoted 62 lines · Song Liu
Quoted 60 lines · Yafang Shao

If we attach fexit to a function annotated with __noreturn, it will cause an issue that the bpf trampoline image will be left over even if the bpf link has been destroyed. Take attaching do_exit() for example. The fexit works as follows,

bpf_trampoline
+ __bpf_tramp_enter
+ percpu_ref_get(&tr->pcref);

+ call do_exit()

+ __bpf_tramp_exit
+ percpu_ref_put(&tr->pcref);

Since do_exit() never returns, the refcnt of the trampoline image is never decremented, preventing it from being freed. This can be verified with as follows,

  $ bpftool link show                                   <<<< nothing output
  $ grep "bpf_trampoline_[0-9]" /proc/kallsyms
  ffffffffc04cb000 t bpf_trampoline_6442526459    [bpf] <<<< leftover

With this change, attaching fexit probes to functions like do_exit() will be rejected.

$ ./fexit
libbpf: prog 'fexit': BPF program load failed: -EINVAL
libbpf: prog 'fexit': -- BEGIN PROG LOAD LOG --
processed 0 insns (limit 1000000) max_states_per_insn 0 total_states 0 peak_states 0 mark_read 0
-- END PROG LOAD LOG --
libbpf: prog 'fexit': failed to load: -EINVAL
libbpf: failed to load object 'fexit_bpf'
libbpf: failed to load BPF skeleton 'fexit_bpf': -EINVAL
failed to load BPF object -22
---
 kernel/bpf/verifier.c | 10 ++++++++++
 1 file changed, 10 insertions(+)
kernel/bpf/verifier.c +7 -0
index 9971c03adfd5..f7224fc61e0c 100644
@@ -22841,6 +22841,13 @@ BTF_ID(func, __rcu_read_unlock)
 #endif
 BTF_SET_END(btf_id_deny)
 
+/* The functions annotated with __noreturn are denied. */
+BTF_SET_START(fexit_deny)
+#define NORETURN(fn) BTF_ID(func, fn)
+#include <linux/noreturns.h>
+#undef NORETURN
+BTF_SET_END(fexit_deny)
+
 static bool can_be_sleepable(struct bpf_prog *prog)
 {
        if (prog->type == BPF_PROG_TYPE_TRACING) {
@@ -22929,6 +22936,9 @@ static int check_attach_btf_id(struct bpf_verifier_env *env)
Quoted 8 lines

kernel/bpf/verifier.c static int check_attach_btf_id(struct bpf_verifier_env *env) · line 22940

Quoted 5 lines
kernel/bpf/verifier.c +2 -0
        } else if (prog->type == BPF_PROG_TYPE_TRACING &&
                   btf_id_set_contains(&btf_id_deny, btf_id)) {
                return -EINVAL;
+       } else if (prog->expected_attach_type == BPF_TRACE_FEXIT &&
+                  btf_id_set_contains(&fexit_deny, btf_id)) {

Please add a verifier log (with verbose()) here, so that the user knows why the program failed to attach.

Good suggestion. I will add it.

Signature · 2 lines
Regards
Yafang
#16 lore raw

PATCH 3/3 selftests/bpf: Add selftest for attaching fexit to __noreturn functions

In reply to Yafang Shao

The reuslt:

  $ tools/testing/selftests/bpf/test_progs --name=fexit_noreturns
  #99      fexit_noreturns:OK
  Summary: 1/0 PASSED, 0 SKIPPED, 0 FAILED
---
 .../selftests/bpf/prog_tests/fexit_noreturns.c      | 13 +++++++++++++
 tools/testing/selftests/bpf/progs/fexit_noreturns.c | 13 +++++++++++++
 2 files changed, 26 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/fexit_noreturns.c
 create mode 100644 tools/testing/selftests/bpf/progs/fexit_noreturns.c
tools/testing/selftests/bpf/prog_tests/fexit_noreturns.c new file +13 -0
new file mode 100644
index 000000000000..588362275ed7
@@ -0,0 +1,13 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <test_progs.h>
+#include "fexit_noreturns.skel.h"
+
+void test_fexit_noreturns(void)
+{
+	struct fexit_noreturns *fexit_skel;
+
+	fexit_skel = fexit_noreturns__open_and_load();
+	ASSERT_NULL(fexit_skel, "fexit_load");
+	ASSERT_EQ(errno, EINVAL, "can't load fexit_noreturns");
+}
tools/testing/selftests/bpf/progs/fexit_noreturns.c new file +13 -0
new file mode 100644
index 000000000000..003aafe2b896
@@ -0,0 +1,13 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/bpf.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+char _license[] SEC("license") = "GPL";
+
+SEC("fexit/do_exit")
+int BPF_PROG(noreturns)
+{
+	return 0;
+}
Signature · 1 line
2.43.5
reviewer
#17 lore raw

Re: [PATCH bpf-next 3/3] selftests/bpf: Add selftest for attaching fexit to __noreturn functions

Quoted 40 lines · Yafang Shao

The reuslt:

  $ tools/testing/selftests/bpf/test_progs --name=fexit_noreturns
  #99      fexit_noreturns:OK
  Summary: 1/0 PASSED, 0 SKIPPED, 0 FAILED
---
 .../selftests/bpf/prog_tests/fexit_noreturns.c      | 13 +++++++++++++
 tools/testing/selftests/bpf/progs/fexit_noreturns.c | 13 +++++++++++++
 2 files changed, 26 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/fexit_noreturns.c
 create mode 100644 tools/testing/selftests/bpf/progs/fexit_noreturns.c
tools/testing/selftests/bpf/prog_tests/fexit_noreturns.c new file +13 -0
new file mode 100644
index 000000000000..588362275ed7
@@ -0,0 +1,13 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <test_progs.h>
+#include "fexit_noreturns.skel.h"
+
+void test_fexit_noreturns(void)
+{
+       struct fexit_noreturns *fexit_skel;
+
+       fexit_skel = fexit_noreturns__open_and_load();
+       ASSERT_NULL(fexit_skel, "fexit_load");
+       ASSERT_EQ(errno, EINVAL, "can't load fexit_noreturns");
+}
tools/testing/selftests/bpf/progs/fexit_noreturns.c new file +0 -0
new file mode 100644
index 000000000000..003aafe2b896

tools/testing/selftests/bpf/progs/fexit_noreturns.c line 9

Quoted 9 lines
tools/testing/selftests/bpf/progs/fexit_noreturns.c new file +9 -0
@@ -0,0 +1,13 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/bpf.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+char _license[] SEC("license") = "GPL";
+
+SEC("fexit/do_exit")

We can add:

__failure __msg(<verifier log added in 2/3>)

here.

Quoted 7 lines
tools/testing/selftests/bpf/progs/fexit_noreturns.c new file +4 -1
+int BPF_PROG(noreturns)
+{
+       return 0;
+}
--

2.43.5

Then, test_fexit_noreturns above can simply be:

void test_fexit_noreturns(void)

{
        RUN_TESTS(fexit_noreturns);
}

Thanks,
Song

#18 lore raw

Re: [PATCH bpf-next 3/3] selftests/bpf: Add selftest for attaching fexit to __noreturn functions

Quoted 67 lines · Song Liu
Quoted 40 lines · Yafang Shao

The reuslt:

  $ tools/testing/selftests/bpf/test_progs --name=fexit_noreturns
  #99      fexit_noreturns:OK
  Summary: 1/0 PASSED, 0 SKIPPED, 0 FAILED
---
 .../selftests/bpf/prog_tests/fexit_noreturns.c      | 13 +++++++++++++
 tools/testing/selftests/bpf/progs/fexit_noreturns.c | 13 +++++++++++++
 2 files changed, 26 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/prog_tests/fexit_noreturns.c
 create mode 100644 tools/testing/selftests/bpf/progs/fexit_noreturns.c
tools/testing/selftests/bpf/prog_tests/fexit_noreturns.c new file +13 -0
new file mode 100644
index 000000000000..588362275ed7
@@ -0,0 +1,13 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <test_progs.h>
+#include "fexit_noreturns.skel.h"
+
+void test_fexit_noreturns(void)
+{
+       struct fexit_noreturns *fexit_skel;
+
+       fexit_skel = fexit_noreturns__open_and_load();
+       ASSERT_NULL(fexit_skel, "fexit_load");
+       ASSERT_EQ(errno, EINVAL, "can't load fexit_noreturns");
+}
tools/testing/selftests/bpf/progs/fexit_noreturns.c new file +0 -0
new file mode 100644
index 000000000000..003aafe2b896

tools/testing/selftests/bpf/progs/fexit_noreturns.c line 9

Quoted 9 lines
tools/testing/selftests/bpf/progs/fexit_noreturns.c new file +9 -0
@@ -0,0 +1,13 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <linux/bpf.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+
+char _license[] SEC("license") = "GPL";
+
+SEC("fexit/do_exit")

We can add:

__failure __msg(<verifier log added in 2/3>)

here.

Quoted 7 lines
tools/testing/selftests/bpf/progs/fexit_noreturns.c new file +4 -1
+int BPF_PROG(noreturns)
+{
+       return 0;
+}
--

2.43.5

Then, test_fexit_noreturns above can simply be:

Quoted 5 lines

void test_fexit_noreturns(void)

{
        RUN_TESTS(fexit_noreturns);
}

Thanks for your suggestion; it really helps simplify the code.

Signature · 2 lines
Regards
Yafang