2 * common.c - C code for kernel entry and exit
3 * Copyright (c) 2015 Andrew Lutomirski
6 * Based on asm and ptrace code by many authors. The code here originated
7 * in ptrace.c and signal.c.
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
13 #include <linux/smp.h>
14 #include <linux/errno.h>
15 #include <linux/ptrace.h>
16 #include <linux/tracehook.h>
17 #include <linux/audit.h>
18 #include <linux/seccomp.h>
19 #include <linux/signal.h>
20 #include <linux/export.h>
21 #include <linux/context_tracking.h>
22 #include <linux/user-return-notifier.h>
23 #include <linux/uprobes.h>
26 #include <asm/traps.h>
28 #include <asm/uaccess.h>
30 #define CREATE_TRACE_POINTS
31 #include <trace/events/syscalls.h>
33 #ifdef CONFIG_CONTEXT_TRACKING
34 /* Called on entry from user mode with IRQs off. */
35 __visible void enter_from_user_mode(void)
37 CT_WARN_ON(ct_state() != CONTEXT_USER);
42 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
45 if (arch == AUDIT_ARCH_X86_64) {
46 audit_syscall_entry(regs->orig_ax, regs->di,
47 regs->si, regs->dx, regs->r10);
51 audit_syscall_entry(regs->orig_ax, regs->bx,
52 regs->cx, regs->dx, regs->si);
57 * We can return 0 to resume the syscall or anything else to go to phase
58 * 2. If we resume the syscall, we need to put something appropriate in
61 * NB: We don't have full pt_regs here, but regs->orig_ax and regs->ax
62 * are fully functional.
64 * For phase 2's benefit, our return value is:
65 * 0: resume the syscall
66 * 1: go to phase 2; no seccomp phase 2 needed
67 * anything else: go to phase 2; pass return value to seccomp
69 unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
71 unsigned long ret = 0;
74 BUG_ON(regs != task_pt_regs(current));
76 work = ACCESS_ONCE(current_thread_info()->flags) &
77 _TIF_WORK_SYSCALL_ENTRY;
79 #ifdef CONFIG_CONTEXT_TRACKING
81 * If TIF_NOHZ is set, we are required to call user_exit() before
82 * doing anything that could touch RCU.
84 if (work & _TIF_NOHZ) {
85 enter_from_user_mode();
92 * Do seccomp first -- it should minimize exposure of other
93 * code, and keeping seccomp fast is probably more valuable
94 * than the rest of this.
96 if (work & _TIF_SECCOMP) {
97 struct seccomp_data sd;
100 sd.nr = regs->orig_ax;
101 sd.instruction_pointer = regs->ip;
103 if (arch == AUDIT_ARCH_X86_64) {
104 sd.args[0] = regs->di;
105 sd.args[1] = regs->si;
106 sd.args[2] = regs->dx;
107 sd.args[3] = regs->r10;
108 sd.args[4] = regs->r8;
109 sd.args[5] = regs->r9;
113 sd.args[0] = regs->bx;
114 sd.args[1] = regs->cx;
115 sd.args[2] = regs->dx;
116 sd.args[3] = regs->si;
117 sd.args[4] = regs->di;
118 sd.args[5] = regs->bp;
121 BUILD_BUG_ON(SECCOMP_PHASE1_OK != 0);
122 BUILD_BUG_ON(SECCOMP_PHASE1_SKIP != 1);
124 ret = seccomp_phase1(&sd);
125 if (ret == SECCOMP_PHASE1_SKIP) {
128 } else if (ret != SECCOMP_PHASE1_OK) {
129 return ret; /* Go directly to phase 2 */
132 work &= ~_TIF_SECCOMP;
136 /* Do our best to finish without phase 2. */
138 return ret; /* seccomp and/or nohz only (ret == 0 here) */
140 #ifdef CONFIG_AUDITSYSCALL
141 if (work == _TIF_SYSCALL_AUDIT) {
143 * If there is no more work to be done except auditing,
144 * then audit in phase 1. Phase 2 always audits, so, if
145 * we audit here, then we can't go on to phase 2.
147 do_audit_syscall_entry(regs, arch);
152 return 1; /* Something is enabled that we can't handle in phase 1 */
155 /* Returns the syscall nr to run (which should match regs->orig_ax). */
156 long syscall_trace_enter_phase2(struct pt_regs *regs, u32 arch,
157 unsigned long phase1_result)
160 u32 work = ACCESS_ONCE(current_thread_info()->flags) &
161 _TIF_WORK_SYSCALL_ENTRY;
163 BUG_ON(regs != task_pt_regs(current));
166 * If we stepped into a sysenter/syscall insn, it trapped in
167 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
168 * If user-mode had set TF itself, then it's still clear from
169 * do_debug() and we need to set it again to restore the user
170 * state. If we entered on the slow path, TF was already set.
172 if (work & _TIF_SINGLESTEP)
173 regs->flags |= X86_EFLAGS_TF;
175 #ifdef CONFIG_SECCOMP
177 * Call seccomp_phase2 before running the other hooks so that
178 * they can see any changes made by a seccomp tracer.
180 if (phase1_result > 1 && seccomp_phase2(phase1_result)) {
181 /* seccomp failures shouldn't expose any additional code. */
186 if (unlikely(work & _TIF_SYSCALL_EMU))
189 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
190 tracehook_report_syscall_entry(regs))
193 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
194 trace_sys_enter(regs, regs->orig_ax);
196 do_audit_syscall_entry(regs, arch);
198 return ret ?: regs->orig_ax;
201 long syscall_trace_enter(struct pt_regs *regs)
203 u32 arch = is_ia32_task() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
204 unsigned long phase1_result = syscall_trace_enter_phase1(regs, arch);
206 if (phase1_result == 0)
207 return regs->orig_ax;
209 return syscall_trace_enter_phase2(regs, arch, phase1_result);
212 static struct thread_info *pt_regs_to_thread_info(struct pt_regs *regs)
214 unsigned long top_of_stack =
215 (unsigned long)(regs + 1) + TOP_OF_KERNEL_STACK_PADDING;
216 return (struct thread_info *)(top_of_stack - THREAD_SIZE);
219 /* Called with IRQs disabled. */
220 __visible void prepare_exit_to_usermode(struct pt_regs *regs)
222 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
228 * In order to return to user mode, we need to have IRQs off with
229 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
230 * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
231 * can be set at any time on preemptable kernels if we have IRQs on,
232 * so we need to loop. Disabling preemption wouldn't help: doing the
233 * work to clear some of the flags can sleep.
237 READ_ONCE(pt_regs_to_thread_info(regs)->flags);
239 if (!(cached_flags & (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME |
240 _TIF_UPROBE | _TIF_NEED_RESCHED |
241 _TIF_USER_RETURN_NOTIFY)))
244 /* We have work to do. */
247 if (cached_flags & _TIF_NEED_RESCHED)
250 if (cached_flags & _TIF_UPROBE)
251 uprobe_notify_resume(regs);
253 /* deal with pending signal delivery */
254 if (cached_flags & _TIF_SIGPENDING)
257 if (cached_flags & _TIF_NOTIFY_RESUME) {
258 clear_thread_flag(TIF_NOTIFY_RESUME);
259 tracehook_notify_resume(regs);
262 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
263 fire_user_return_notifiers();
265 /* Disable IRQs and retry */
273 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
274 * state such that we can immediately switch to user mode.
276 __visible void syscall_return_slowpath(struct pt_regs *regs)
278 struct thread_info *ti = pt_regs_to_thread_info(regs);
279 u32 cached_flags = READ_ONCE(ti->flags);
282 CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
284 if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
285 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
289 * First do one-time work. If these work items are enabled, we
290 * want to run them exactly once per syscall exit with IRQs on.
292 if (cached_flags & (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT |
293 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)) {
294 audit_syscall_exit(regs);
296 if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
297 trace_sys_exit(regs, regs->ax);
300 * If TIF_SYSCALL_EMU is set, we only get here because of
301 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
302 * We already reported this syscall instruction in
303 * syscall_trace_enter().
306 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
308 if (step || cached_flags & _TIF_SYSCALL_TRACE)
309 tracehook_report_syscall_exit(regs, step);
314 * Compat syscalls set TS_COMPAT. Make sure we clear it before
315 * returning to user mode.
317 ti->status &= ~TS_COMPAT;
321 prepare_exit_to_usermode(regs);
324 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
326 * Does a 32-bit syscall. Called with IRQs on and does all entry and
327 * exit work and returns with IRQs off. This function is extremely hot
328 * in workloads that use it, and it's usually called from
329 * do_fast_syscall_32, so forcibly inline it to improve performance.
331 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
333 struct thread_info *ti = pt_regs_to_thread_info(regs);
334 unsigned int nr = (unsigned int)regs->orig_ax;
336 #ifdef CONFIG_IA32_EMULATION
337 ti->status |= TS_COMPAT;
340 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
342 * Subtlety here: if ptrace pokes something larger than
343 * 2^32-1 into orig_ax, this truncates it. This may or
344 * may not be necessary, but it matches the old asm
347 nr = syscall_trace_enter(regs);
350 if (likely(nr < IA32_NR_syscalls)) {
352 * It's possible that a 32-bit syscall implementation
353 * takes a 64-bit parameter but nonetheless assumes that
354 * the high bits are zero. Make sure we zero-extend all
357 regs->ax = ia32_sys_call_table[nr](
358 (unsigned int)regs->bx, (unsigned int)regs->cx,
359 (unsigned int)regs->dx, (unsigned int)regs->si,
360 (unsigned int)regs->di, (unsigned int)regs->bp);
363 syscall_return_slowpath(regs);
366 /* Handles int $0x80 */
367 __visible void do_int80_syscall_32(struct pt_regs *regs)
370 do_syscall_32_irqs_on(regs);
373 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
374 __visible long do_fast_syscall_32(struct pt_regs *regs)
377 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
378 * convention. Adjust regs so it looks like we entered using int80.
381 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
382 vdso_image_32.sym_int80_landing_pad;
385 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
386 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
389 regs->ip = landing_pad;
392 * Fetch ECX from where the vDSO stashed it.
394 * WARNING: We are in CONTEXT_USER and RCU isn't paying attention!
400 * Micro-optimization: the pointer we're following is explicitly
401 * 32 bits, so it can't be out of range.
403 __get_user(*(u32 *)®s->cx,
404 (u32 __user __force *)(unsigned long)(u32)regs->sp)
406 get_user(*(u32 *)®s->cx,
407 (u32 __user __force *)(unsigned long)(u32)regs->sp)
411 /* User code screwed up. */
414 #ifdef CONFIG_CONTEXT_TRACKING
415 enter_from_user_mode();
417 prepare_exit_to_usermode(regs);
418 return 0; /* Keep it simple: use IRET. */
421 /* Now this is just like a normal syscall. */
422 do_syscall_32_irqs_on(regs);
426 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
427 * SYSRETL is available on all 64-bit CPUs, so we don't need to
428 * bother with SYSEXIT.
430 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
431 * because the ECX fixup above will ensure that this is essentially
434 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
435 regs->ip == landing_pad &&
436 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
439 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
441 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
442 * because the ECX fixup above will ensure that this is essentially
445 * We don't allow syscalls at all from VM86 mode, but we still
446 * need to check VM, because we might be returning from sys_vm86.
448 return static_cpu_has(X86_FEATURE_SEP) &&
449 regs->cs == __USER_CS && regs->ss == __USER_DS &&
450 regs->ip == landing_pad &&
451 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;