bpf: Add bpf_get_task_cmdline kfunc
Subsystems (3)
PATCH 1/2 bpf: Add bpf_get_task_cmdline kfunc
Add the bpf_get_task_cmdline kfunc. One use case is as follows: In production environments, there are often short-lived script tasks executed, and sometimes these tasks may cause stability issues. It is desirable to detect these script tasks via eBPF. The common approach is to check the process name, but it can be difficult to distinguish specific tasks in some cases. Take the shell as an example: some tasks are started via bash xxx.sh – their process name is bash, but the script name of the task can be obtained through the cmdline. Additionally, myabe this is helpful for security auditing purposes.
--- kernel/bpf/helpers.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)
kernel/bpf/helpers.c +22 -0
@@ -2685,6 +2685,27 @@ __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)return p; } +/* +* bpf_get_task_cmdline - Get the cmdline to a buffer +* +* @task: The task whose cmdline to get. +* @buffer: The buffer to save cmdline info. +* @len: The length of the buffer. +* +* Return: the size of the cmdline field copied. Note that the copy does +* not guarantee an ending NULL byte. A negative error code on failure. +*/ +__bpf_kfunc int bpf_get_task_cmdline(struct task_struct *task, char *buffer, size_t len) +{ +int ret; + + ret =get_cmdline (task, buffer, len); +if (ret <0 ) +memset (buffer,0 , len); + +return ret; +} +/** * bpf_task_from_vpid - Find a struct task_struct from its vpid by looking it up * in the pid namespace of the current task. If a task is returned, it must
@@ -4421,6 +4442,7 @@ BTF_ID_FLAGS(func, bpf_task_get_cgroup1, KF_ACQUIRE | KF_RCU | KF_RET_NULL) BTF_ID_FLAGS(func, bpf_task_from_pid, KF_ACQUIRE | KF_RET_NULL) BTF_ID_FLAGS(func, bpf_task_from_vpid, KF_ACQUIRE | KF_RET_NULL) BTF_ID_FLAGS(func, bpf_throw) +BTF_ID_FLAGS(func, bpf_get_task_cmdline, KF_SLEEPABLE | KF_TRUSTED_ARGS)#ifdef CONFIG_BPF_EVENTS BTF_ID_FLAGS(func, bpf_send_signal_task, KF_TRUSTED_ARGS)#endif
Signature · 1 line
2.48.1
PATCH 2/2 selftests/bpf: Add bpf_get_task_cmdline test case
Create a task, call bpf_get_task_cmdline to retrieve the cmdline, and check if it succeeds.
--- tools/testing/selftests/bpf/prog_tests/task_kfunc.c | 11 +++++++++++ .../selftests/bpf/progs/task_kfunc_success.c | 13 +++++++++++++ 2 files changed, 24 insertions(+)
tools/testing/selftests/bpf/prog_tests/task_kfunc.c +11 -0
@@ -156,6 +156,10 @@ static const char * const vpid_success_tests[] = { "test_task_from_vpid_invalid", }; +static const char *const cmdline_success_tests[] = { +"test_get_task_cmdline" , +}; +void test_task_kfunc (void ) {int i;
@@ -174,5 +178,12 @@ void test_task_kfunc(void)run_vpid_success_test (vpid_success_tests[i]); } +for (i =0 ; i <ARRAY_SIZE (cmdline_success_tests); i++) { +if (!test__start_subtest (cmdline_success_tests[i])) +continue ; + +run_success_test (cmdline_success_tests[i]); + } +RUN_TESTS (task_kfunc_failure); }
tools/testing/selftests/bpf/progs/task_kfunc_success.c +13 -0
@@ -367,6 +367,19 @@ int BPF_PROG(task_kfunc_acquire_trusted_walked, struct task_struct *task, u64 clreturn 0 ; } +SEC("lsm.s/task_alloc") +int BPF_PROG (test_get_task_cmdline ,struct task_struct *task) +{ +char buf[64 ]; +int ret; + + ret =bpf_get_task_cmdline (task, buf,sizeof (buf)); +if (ret <0 ) + err =1 ; + +return 0 ; +} + SEC("syscall")int test_task_from_vpid_current (const void *ctx) {
Signature · 1 line
2.48.1
Quoted 10 lines · Tao Chen
Add the
bpf_get_task_cmdlinekfunc. One use case is as follows: In production environments, there are often short-lived script tasks executed, and sometimes these tasks may cause stability issues. It is desirable to detect these script tasks via eBPF. The common approach is to check the process name, but it can be difficult to distinguish specific tasks in some cases. Take the shell as an example: some tasks are started via bashxxx.sh– their process name is bash, but the script name of the task can be obtained through the cmdline. Additionally, myabe this is helpful for security auditing purposes.
maybe
Quoted 15 lines
Signed-off-by Tao Chen <[email protected]>--- kernel/bpf/helpers.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)kernel/bpf/helpers.c +1
-0@@ -2685,6 +2685,27 @@ __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)return p; } +/*
kernel/bpf/helpers.c __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid) · line 2698
Quoted 10 lines
kernel/bpf/helpers.c +10
-0+* bpf_get_task_cmdline - Get the cmdline to a buffer +* +* @task: The task whose cmdline to get. +* @buffer: The buffer to save cmdline info. +* @len: The length of the buffer. +* +* Return: the size of the cmdline field copied. Note that the copy does +* not guarantee an ending NULL byte. A negative error code on failure. +*/ +__bpf_kfunc int bpf_get_task_cmdline(struct task_struct *task, char *buffer, size_t len)
'size_t len' doesn't make the verifier track the size of the buffer. while 'char *buffer' tells the verifier to check that _one_ byte is available. So this is buggy.
In general the kfunc seems useful, but selftest in patch 2 is just bad
+ ret = bpf_get_task_cmdline(task, buf, sizeof(buf));
+ if (ret < 0)
+ err = 1;
+
+ return 0;
+}
it's not testing much.
Also you must explain the true motivation for the kfunc. "maybe helpful for security" is too vague.
Do you have a proprietary bpf-lsm that needs it? What is the exact use case?
pw-bot: cr
Quoted 45 lines · Alexei Starovoitov
Quoted 10 lines · Tao Chen
Add the
bpf_get_task_cmdlinekfunc. One use case is as follows: In production environments, there are often short-lived script tasks executed, and sometimes these tasks may cause stability issues. It is desirable to detect these script tasks via eBPF. The common approach is to check the process name, but it can be difficult to distinguish specific tasks in some cases. Take the shell as an example: some tasks are started via bashxxx.sh– their process name is bash, but the script name of the task can be obtained through the cmdline. Additionally, myabe this is helpful for security auditing purposes.maybe
Quoted 15 lines
Signed-off-by Tao Chen <[email protected]>--- kernel/bpf/helpers.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)kernel/bpf/helpers.c +1
-0@@ -2685,6 +2685,27 @@ __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)return p; } +/* kernel/bpf/helpers.c __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid) · line 2698
Quoted 10 lines
kernel/bpf/helpers.c +10
-0+* bpf_get_task_cmdline - Get the cmdline to a buffer +* +* @task: The task whose cmdline to get. +* @buffer: The buffer to save cmdline info. +* @len: The length of the buffer. +* +* Return: the size of the cmdline field copied. Note that the copy does +* not guarantee an ending NULL byte. A negative error code on failure. +*/ +__bpf_kfunc int bpf_get_task_cmdline(struct task_struct *task, char *buffer, size_t len)'
size_tlen' doesn't make the verifier track the size of the buffer. while 'char *buffer' tells the verifier to check that _one_ byte is available. So this is buggy.
Quoted 2 lines
In general the kfunc seems useful, but selftest in patch 2 is just bad
Besides that mm->arg_lock spinlock (which I don't think matters all that much for BPF programs), is there anything special in get_cmdline() that BPF program cannot just implemented? Ultimately, it's just copying mm->arg_start and mm->env_start zero-separated strings, no? We have bpf_copy_from_user_task_str() and also dynptr-based equivalent of it for even more variable-length flexibility. That should be all one needs, no?
Quoted 15 lines
+ ret =
bpf_get_task_cmdline(task, buf, sizeof(buf));
+ if (ret < 0)kernel/bpf/helpers.c +4
-0+ err =1 ; + +return 0 ; +}it's not testing much.
Also you must explain the true motivation for the kfunc. "maybe helpful for security" is too vague.
Do you have a proprietary bpf-lsm that needs it? What is the exact use case?pw-bot: cr
Quoted 50 lines · Andrii Nakryiko
Quoted 45 lines · Alexei Starovoitov
Quoted 10 lines · Tao Chen
Add the
bpf_get_task_cmdlinekfunc. One use case is as follows: In production environments, there are often short-lived script tasks executed, and sometimes these tasks may cause stability issues. It is desirable to detect these script tasks via eBPF. The common approach is to check the process name, but it can be difficult to distinguish specific tasks in some cases. Take the shell as an example: some tasks are started via bashxxx.sh– their process name is bash, but the script name of the task can be obtained through the cmdline. Additionally, myabe this is helpful for security auditing purposes.maybe
Quoted 15 lines
Signed-off-by Tao Chen <[email protected]>--- kernel/bpf/helpers.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)kernel/bpf/helpers.c +1
-0@@ -2685,6 +2685,27 @@ __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid)return p; } +/* kernel/bpf/helpers.c __bpf_kfunc struct task_struct *bpf_task_from_pid(s32 pid) · line 2698
Quoted 10 lines
kernel/bpf/helpers.c +10
-0+* bpf_get_task_cmdline - Get the cmdline to a buffer +* +* @task: The task whose cmdline to get. +* @buffer: The buffer to save cmdline info. +* @len: The length of the buffer. +* +* Return: the size of the cmdline field copied. Note that the copy does +* not guarantee an ending NULL byte. A negative error code on failure. +*/ +__bpf_kfunc int bpf_get_task_cmdline(struct task_struct *task, char *buffer, size_t len)'
size_tlen' doesn't make the verifier track the size of the buffer. while 'char *buffer' tells the verifier to check that _one_ byte is available. So this is buggy.Quoted 2 lines
In general the kfunc seems useful, but selftest in patch 2 is just bad
Quoted 8 lines
Besides that
mm->arg_lockspinlock (which I don't think matters all that much for BPF programs), is there anything special inget_cmdline()that BPF program cannot just implemented? Ultimately, it's just copyingmm->arg_startandmm->env_startzero-separated strings, no? We havebpf_copy_from_user_task_str()and also dynptr-based equivalent of it for even more variable-length flexibility. That should be all one needs, no?
From a quick look at how both are implemented, it seems that way. Hold off on this patch for now. I will move forward if we find something new.
Quoted 15 lines
Quoted 15 lines
+ ret =
bpf_get_task_cmdline(task, buf, sizeof(buf));
+ if (ret < 0)kernel/bpf/helpers.c +4
-0+ err =1 ; + +return 0 ; +}it's not testing much.
Also you must explain the true motivation for the kfunc. "maybe helpful for security" is too vague.
Do you have a proprietary bpf-lsm that needs it? What is the exact use case?pw-bot: cr
Signature · 2 lines
Best Regards Tao Chen