From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001 From: DroidDeck Date: Tue, 29 Sep 2026 00:00:00 +0000 Subject: [PATCH] seccomp: trace ioctl by request, drop exit stops the kernel does not need --- src/syscall/seccomp.c | 98 ++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 97 insertions(+), 1 deletion(-) diff --git a/src/syscall/seccomp.c b/src/syscall/seccomp.c index 0b381bb..1116ab2 100644 --- a/src/syscall/seccomp.c +++ b/src/syscall/seccomp.c @@ -39,6 +39,11 @@ #include /* uint*_t, UINT*_MAX, */ #include #include +#include +#include +#include +#include +#include #include /* assert(3), */ #include "syscall/seccomp.h" @@ -166,6 +171,46 @@ static int add_trace_prlimit_stack(struct sock_fprog *program, word_t syscall, i return add_statements(program, LENGTH_TRACE_PRLIMIT_STACK, statements); } + + + + + + + + + + +static int add_trace_ioctl_requests(struct sock_fprog *program, word_t syscall, int flag) +{ + const size_t request_offset = offsetof(struct seccomp_data, args[1]); + + #define LENGTH_TRACE_IOCTL 10 + struct sock_filter statements[LENGTH_TRACE_IOCTL] = { + + + BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, syscall, 0, 9), + + + BPF_STMT(BPF_LD + BPF_W + BPF_ABS, request_offset), + + BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, TCSETSF, 5, 0), + BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, TCGETS2, 4, 0), + BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, TCSETS2, 3, 0), + BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, TCSETSW2, 2, 0), + BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, TCSETSF2, 1, 0), + BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, (uint32_t) FICLONE, 0, 1), + + BPF_STMT(BPF_RET + BPF_K, SECCOMP_RET_TRACE + flag), + BPF_STMT(BPF_RET + BPF_K, SECCOMP_RET_ALLOW) + }; + + DEBUG_FILTER("FILTER: trace if syscall == %ld and a termios2/TCSETSF/FICLONE request\n", + syscall); + + return add_statements(program, LENGTH_TRACE_IOCTL, statements); +} + static size_t trace_length(Sysnum sysnum) { switch (sysnum) { @@ -174,6 +219,8 @@ static size_t trace_length(Sysnum sysnum) return LENGTH_TRACE_SYSCALL_IF_ARG; case PR_prlimit64: return LENGTH_TRACE_PRLIMIT_STACK; + case PR_ioctl: + return LENGTH_TRACE_IOCTL; default: return LENGTH_TRACE_SYSCALL; } @@ -191,6 +238,8 @@ static int add_trace(struct sock_fprog *program, Sysnum sysnum, word_t syscall, return add_trace_syscall_if_arg(program, syscall, 0, RLIMIT_STACK, flag); case PR_prlimit64: return add_trace_prlimit_stack(program, syscall, flag); + case PR_ioctl: + return add_trace_ioctl_requests(program, syscall, flag); default: return add_trace_syscall(program, syscall, flag); } @@ -329,7 +378,7 @@ static int set_seccomp_filters(const FilteredSysnum *sysnums) struct sock_fprog program = { .len = 0, .filter = NULL }; size_t section_length; - size_t i, j, k; + size_t i, j, k, pass; int status; status = new_program_filter(&program); @@ -355,8 +404,14 @@ static int set_seccomp_filters(const FilteredSysnum *sysnums) if (status < 0) goto end; + + + for (pass = 0; pass < 2; pass++) for (j = 0; j < seccomp_archs[i].nb_abis; j++) { for (k = 0; sysnums[k].value != PR_void; k++) { + if ((sysnums[k].value == PR_ioctl) != (pass == 0)) + continue; + /* Get the architecture specific syscall number. */ syscall = detranslate_sysnum(seccomp_archs[i].abis[j], sysnums[k].value); if (syscall == SYSCALL_AVOIDER) @@ -572,12 +627,53 @@ int enable_syscall_filtering(const Tracee *tracee) return 0; } + + + + + + + + + +static void drop_unneeded_exit_stops(void) +{ + static bool done = false; + struct utsname utsname; + int major, minor; + size_t i; + + if (done) + return; + done = true; + + if (uname(&utsname) < 0 || sscanf(utsname.release, "%d.%d", &major, &minor) != 2) + return; + + for (i = 0; proot_sysnums[i].value != PR_void; i++) { + switch (proot_sysnums[i].value) { + case PR_faccessat2: + if (major > 5 || (major == 5 && minor >= 8)) + proot_sysnums[i].flags &= ~FILTER_SYSEXIT; + break; + case PR_statx: + if (major > 4 || (major == 4 && minor >= 11)) + proot_sysnums[i].flags &= ~FILTER_SYSEXIT; + break; + default: + break; + } + } +} + static int collect_filtered_sysnums(TALLOC_CTX *context, const Tracee *tracee, FilteredSysnum **filtered_sysnums) { Extension *extension; int status; + drop_unneeded_exit_stops(); + /* Add the sysnums required by PRoot to the list of filtered * sysnums. TODO: only if path translation is required. */ status = merge_filtered_sysnums(context, filtered_sysnums, proot_sysnums);