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>
29 #include <asm/cpufeature.h>
31 #define CREATE_TRACE_POINTS
32 #include <trace/events/syscalls.h>
34 static struct thread_info *pt_regs_to_thread_info(struct pt_regs *regs)
36 unsigned long top_of_stack =
37 (unsigned long)(regs + 1) + TOP_OF_KERNEL_STACK_PADDING;
38 return (struct thread_info *)(top_of_stack - THREAD_SIZE);
41 #ifdef CONFIG_CONTEXT_TRACKING
42 /* Called on entry from user mode with IRQs off. */
43 __visible void enter_from_user_mode(void)
45 CT_WARN_ON(ct_state() != CONTEXT_USER);
50 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
53 if (arch == AUDIT_ARCH_X86_64) {
54 audit_syscall_entry(regs->orig_ax, regs->di,
55 regs->si, regs->dx, regs->r10);
59 audit_syscall_entry(regs->orig_ax, regs->bx,
60 regs->cx, regs->dx, regs->si);
65 * We can return 0 to resume the syscall or anything else to go to phase
66 * 2. If we resume the syscall, we need to put something appropriate in
69 * NB: We don't have full pt_regs here, but regs->orig_ax and regs->ax
70 * are fully functional.
72 * For phase 2's benefit, our return value is:
73 * 0: resume the syscall
74 * 1: go to phase 2; no seccomp phase 2 needed
75 * anything else: go to phase 2; pass return value to seccomp
77 unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
79 struct thread_info *ti = pt_regs_to_thread_info(regs);
80 unsigned long ret = 0;
83 if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
84 BUG_ON(regs != task_pt_regs(current));
86 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
88 #ifdef CONFIG_CONTEXT_TRACKING
90 * If TIF_NOHZ is set, we are required to call user_exit() before
91 * doing anything that could touch RCU.
93 if (work & _TIF_NOHZ) {
94 enter_from_user_mode();
101 * Do seccomp first -- it should minimize exposure of other
102 * code, and keeping seccomp fast is probably more valuable
103 * than the rest of this.
105 if (work & _TIF_SECCOMP) {
106 struct seccomp_data sd;
109 sd.nr = regs->orig_ax;
110 sd.instruction_pointer = regs->ip;
112 if (arch == AUDIT_ARCH_X86_64) {
113 sd.args[0] = regs->di;
114 sd.args[1] = regs->si;
115 sd.args[2] = regs->dx;
116 sd.args[3] = regs->r10;
117 sd.args[4] = regs->r8;
118 sd.args[5] = regs->r9;
122 sd.args[0] = regs->bx;
123 sd.args[1] = regs->cx;
124 sd.args[2] = regs->dx;
125 sd.args[3] = regs->si;
126 sd.args[4] = regs->di;
127 sd.args[5] = regs->bp;
130 BUILD_BUG_ON(SECCOMP_PHASE1_OK != 0);
131 BUILD_BUG_ON(SECCOMP_PHASE1_SKIP != 1);
133 ret = seccomp_phase1(&sd);
134 if (ret == SECCOMP_PHASE1_SKIP) {
137 } else if (ret != SECCOMP_PHASE1_OK) {
138 return ret; /* Go directly to phase 2 */
141 work &= ~_TIF_SECCOMP;
145 /* Do our best to finish without phase 2. */
147 return ret; /* seccomp and/or nohz only (ret == 0 here) */
149 #ifdef CONFIG_AUDITSYSCALL
150 if (work == _TIF_SYSCALL_AUDIT) {
152 * If there is no more work to be done except auditing,
153 * then audit in phase 1. Phase 2 always audits, so, if
154 * we audit here, then we can't go on to phase 2.
156 do_audit_syscall_entry(regs, arch);
161 return 1; /* Something is enabled that we can't handle in phase 1 */
164 /* Returns the syscall nr to run (which should match regs->orig_ax). */
165 long syscall_trace_enter_phase2(struct pt_regs *regs, u32 arch,
166 unsigned long phase1_result)
168 struct thread_info *ti = pt_regs_to_thread_info(regs);
170 u32 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
172 if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
173 BUG_ON(regs != task_pt_regs(current));
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 #define EXIT_TO_USERMODE_LOOP_FLAGS \
213 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
214 _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
216 static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
219 * In order to return to user mode, we need to have IRQs off with
220 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
221 * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
222 * can be set at any time on preemptable kernels if we have IRQs on,
223 * so we need to loop. Disabling preemption wouldn't help: doing the
224 * work to clear some of the flags can sleep.
227 /* We have work to do. */
230 if (cached_flags & _TIF_NEED_RESCHED)
233 if (cached_flags & _TIF_UPROBE)
234 uprobe_notify_resume(regs);
236 /* deal with pending signal delivery */
237 if (cached_flags & _TIF_SIGPENDING)
240 if (cached_flags & _TIF_NOTIFY_RESUME) {
241 clear_thread_flag(TIF_NOTIFY_RESUME);
242 tracehook_notify_resume(regs);
245 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
246 fire_user_return_notifiers();
248 /* Disable IRQs and retry */
251 cached_flags = READ_ONCE(pt_regs_to_thread_info(regs)->flags);
253 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
259 /* Called with IRQs disabled. */
260 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
262 struct thread_info *ti = pt_regs_to_thread_info(regs);
265 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
270 cached_flags = READ_ONCE(ti->flags);
272 if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
273 exit_to_usermode_loop(regs, cached_flags);
277 * Compat syscalls set TS_COMPAT. Make sure we clear it before
278 * returning to user mode. We need to clear it *after* signal
279 * handling, because syscall restart has a fixup for compat
280 * syscalls. The fixup is exercised by the ptrace_syscall_32
283 ti->status &= ~TS_COMPAT;
289 #define SYSCALL_EXIT_WORK_FLAGS \
290 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
291 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
293 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
297 audit_syscall_exit(regs);
299 if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
300 trace_sys_exit(regs, regs->ax);
303 * If TIF_SYSCALL_EMU is set, we only get here because of
304 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
305 * We already reported this syscall instruction in
306 * syscall_trace_enter().
309 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
311 if (step || cached_flags & _TIF_SYSCALL_TRACE)
312 tracehook_report_syscall_exit(regs, step);
316 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
317 * state such that we can immediately switch to user mode.
319 __visible inline void syscall_return_slowpath(struct pt_regs *regs)
321 struct thread_info *ti = pt_regs_to_thread_info(regs);
322 u32 cached_flags = READ_ONCE(ti->flags);
324 CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
326 if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
327 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
331 * First do one-time work. If these work items are enabled, we
332 * want to run them exactly once per syscall exit with IRQs on.
334 if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
335 syscall_slow_exit_work(regs, cached_flags);
338 prepare_exit_to_usermode(regs);
342 __visible void do_syscall_64(struct pt_regs *regs)
344 struct thread_info *ti = pt_regs_to_thread_info(regs);
345 unsigned long nr = regs->orig_ax;
349 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
350 nr = syscall_trace_enter(regs);
353 * NB: Native and x32 syscalls are dispatched from the same
354 * table. The only functional difference is the x32 bit in
355 * regs->orig_ax, which changes the behavior of some syscalls.
357 if (likely((nr & __SYSCALL_MASK) < NR_syscalls)) {
358 regs->ax = sys_call_table[nr & __SYSCALL_MASK](
359 regs->di, regs->si, regs->dx,
360 regs->r10, regs->r8, regs->r9);
363 syscall_return_slowpath(regs);
367 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
369 * Does a 32-bit syscall. Called with IRQs on and does all entry and
370 * exit work and returns with IRQs off. This function is extremely hot
371 * in workloads that use it, and it's usually called from
372 * do_fast_syscall_32, so forcibly inline it to improve performance.
375 /* 32-bit kernels use a trap gate for INT80, and the asm code calls here. */
378 /* 64-bit kernels use do_syscall_32_irqs_off() instead. */
381 __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
383 struct thread_info *ti = pt_regs_to_thread_info(regs);
384 unsigned int nr = (unsigned int)regs->orig_ax;
386 #ifdef CONFIG_IA32_EMULATION
387 ti->status |= TS_COMPAT;
390 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
392 * Subtlety here: if ptrace pokes something larger than
393 * 2^32-1 into orig_ax, this truncates it. This may or
394 * may not be necessary, but it matches the old asm
397 nr = syscall_trace_enter(regs);
400 if (likely(nr < IA32_NR_syscalls)) {
402 * It's possible that a 32-bit syscall implementation
403 * takes a 64-bit parameter but nonetheless assumes that
404 * the high bits are zero. Make sure we zero-extend all
407 regs->ax = ia32_sys_call_table[nr](
408 (unsigned int)regs->bx, (unsigned int)regs->cx,
409 (unsigned int)regs->dx, (unsigned int)regs->si,
410 (unsigned int)regs->di, (unsigned int)regs->bp);
413 syscall_return_slowpath(regs);
417 /* Handles INT80 on 64-bit kernels */
418 __visible void do_syscall_32_irqs_off(struct pt_regs *regs)
421 do_syscall_32_irqs_on(regs);
425 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
426 __visible long do_fast_syscall_32(struct pt_regs *regs)
429 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
430 * convention. Adjust regs so it looks like we entered using int80.
433 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
434 vdso_image_32.sym_int80_landing_pad;
437 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
438 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
441 regs->ip = landing_pad;
444 * Fetch EBP from where the vDSO stashed it.
446 * WARNING: We are in CONTEXT_USER and RCU isn't paying attention!
452 * Micro-optimization: the pointer we're following is explicitly
453 * 32 bits, so it can't be out of range.
455 __get_user(*(u32 *)®s->bp,
456 (u32 __user __force *)(unsigned long)(u32)regs->sp)
458 get_user(*(u32 *)®s->bp,
459 (u32 __user __force *)(unsigned long)(u32)regs->sp)
463 /* User code screwed up. */
466 #ifdef CONFIG_CONTEXT_TRACKING
467 enter_from_user_mode();
469 prepare_exit_to_usermode(regs);
470 return 0; /* Keep it simple: use IRET. */
473 /* Now this is just like a normal syscall. */
474 do_syscall_32_irqs_on(regs);
478 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
479 * SYSRETL is available on all 64-bit CPUs, so we don't need to
480 * bother with SYSEXIT.
482 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
483 * because the ECX fixup above will ensure that this is essentially
486 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
487 regs->ip == landing_pad &&
488 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
491 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
493 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
494 * because the ECX fixup above will ensure that this is essentially
497 * We don't allow syscalls at all from VM86 mode, but we still
498 * need to check VM, because we might be returning from sys_vm86.
500 return static_cpu_has(X86_FEATURE_SEP) &&
501 regs->cs == __USER_CS && regs->ss == __USER_DS &&
502 regs->ip == landing_pad &&
503 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;