Martin mentioned that the verifier cannot assume arguments from
LSM hook sk_alloc_security being trusted since after the hook
is called, the sk ref_count is set to 1. This will overwrite
the ref_count changed by the bpf program and may cause ref_count
underflow later on.
I then further checked some other hooks. For example,
for bpf_lsm_file_alloc() hook in fs/file_table.c,
        f->f_cred = get_cred(cred);
        error = security_file_alloc(f);
        if (unlikely(error)) {
                file_free_rcu(&f->f_rcuhead);
                return ERR_PTR(error);
        }
        atomic_long_set(&f->f_count, 1);
The input parameter 'f' to security_file_alloc() cannot be trusted
as well.
Specifically, I investiaged bpf_map/bpf_prog/file/sk/task alloc/free
lsm hooks. Except bpf_map_alloc and task_alloc, arguments for all other
hooks should not be considered as trusted. This may not be a complete
list, but it covers common usage for sk and task.
Fixes: 3f00c5239344 ("bpf: Allow trusted pointers to be passed to KF_TRUSTED_ARGS kfuncs")
Signed-off-by: Yonghong Song <yhs@fb.com>
Link: https://lore.kernel.org/r/20221203204954.2043348-1-yhs@fb.com
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
                        const struct bpf_prog *prog);
 
 bool bpf_lsm_is_sleepable_hook(u32 btf_id);
+bool bpf_lsm_is_trusted(const struct bpf_prog *prog);
 
 static inline struct bpf_storage_blob *bpf_inode(
        const struct inode *inode)
        return false;
 }
 
+static inline bool bpf_lsm_is_trusted(const struct bpf_prog *prog)
+{
+       return false;
+}
+
 static inline int bpf_lsm_verify_prog(struct bpf_verifier_log *vlog,
                                      const struct bpf_prog *prog)
 {
 
 BTF_ID(func, bpf_lsm_userns_create)
 BTF_SET_END(sleepable_lsm_hooks)
 
+BTF_SET_START(untrusted_lsm_hooks)
+BTF_ID(func, bpf_lsm_bpf_map_free_security)
+BTF_ID(func, bpf_lsm_bpf_prog_alloc_security)
+BTF_ID(func, bpf_lsm_bpf_prog_free_security)
+BTF_ID(func, bpf_lsm_file_alloc_security)
+BTF_ID(func, bpf_lsm_file_free_security)
+BTF_ID(func, bpf_lsm_sk_alloc_security)
+BTF_ID(func, bpf_lsm_sk_free_security)
+BTF_ID(func, bpf_lsm_task_free)
+BTF_SET_END(untrusted_lsm_hooks)
+
 bool bpf_lsm_is_sleepable_hook(u32 btf_id)
 {
        return btf_id_set_contains(&sleepable_lsm_hooks, btf_id);
 }
 
+bool bpf_lsm_is_trusted(const struct bpf_prog *prog)
+{
+       return !btf_id_set_contains(&untrusted_lsm_hooks, prog->aux->attach_btf_id);
+}
+
 const struct bpf_prog_ops lsm_prog_ops = {
 };
 
 
 #include <linux/bpf_verifier.h>
 #include <linux/btf.h>
 #include <linux/btf_ids.h>
+#include <linux/bpf_lsm.h>
 #include <linux/skmsg.h>
 #include <linux/perf_event.h>
 #include <linux/bsearch.h>
        case BPF_PROG_TYPE_TRACING:
                return atype == BPF_TRACE_RAW_TP || atype == BPF_TRACE_ITER;
        case BPF_PROG_TYPE_LSM:
+               return bpf_lsm_is_trusted(prog);
        case BPF_PROG_TYPE_STRUCT_OPS:
                return true;
        default:
 
        {"task_kfunc_release_null", "arg#0 is ptr_or_null_ expected ptr_ or socket"},
        {"task_kfunc_release_unacquired", "release kernel function bpf_task_release expects"},
        {"task_kfunc_from_pid_no_null_check", "arg#0 is ptr_or_null_ expected ptr_ or socket"},
+       {"task_kfunc_from_lsm_task_free", "reg type unsupported for arg#0 function"},
 };
 
 static void verify_fail(const char *prog_name, const char *expected_err_msg)
 
 
        return 0;
 }
+
+SEC("lsm/task_free")
+int BPF_PROG(task_kfunc_from_lsm_task_free, struct task_struct *task)
+{
+       struct task_struct *acquired;
+
+       /* the argument of lsm task_free hook is untrusted. */
+       acquired = bpf_task_acquire(task);
+       bpf_task_release(acquired);
+       return 0;
+}