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 static struct thread_info *pt_regs_to_thread_info(struct pt_regs *regs)
35 unsigned long top_of_stack =
36 (unsigned long)(regs + 1) + TOP_OF_KERNEL_STACK_PADDING;
37 return (struct thread_info *)(top_of_stack - THREAD_SIZE);
40 #ifdef CONFIG_CONTEXT_TRACKING
41 /* Called on entry from user mode with IRQs off. */
42 __visible void enter_from_user_mode(void)
44 CT_WARN_ON(ct_state() != CONTEXT_USER);
49 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
52 if (arch == AUDIT_ARCH_X86_64) {
53 audit_syscall_entry(regs->orig_ax, regs->di,
54 regs->si, regs->dx, regs->r10);
58 audit_syscall_entry(regs->orig_ax, regs->bx,
59 regs->cx, regs->dx, regs->si);
64 * We can return 0 to resume the syscall or anything else to go to phase
65 * 2. If we resume the syscall, we need to put something appropriate in
68 * NB: We don't have full pt_regs here, but regs->orig_ax and regs->ax
69 * are fully functional.
71 * For phase 2's benefit, our return value is:
72 * 0: resume the syscall
73 * 1: go to phase 2; no seccomp phase 2 needed
74 * anything else: go to phase 2; pass return value to seccomp
76 unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
78 struct thread_info *ti = pt_regs_to_thread_info(regs);
79 unsigned long ret = 0;
82 if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
83 BUG_ON(regs != task_pt_regs(current));
85 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
87 #ifdef CONFIG_CONTEXT_TRACKING
89 * If TIF_NOHZ is set, we are required to call user_exit() before
90 * doing anything that could touch RCU.
92 if (work & _TIF_NOHZ) {
93 enter_from_user_mode();
100 * Do seccomp first -- it should minimize exposure of other
101 * code, and keeping seccomp fast is probably more valuable
102 * than the rest of this.
104 if (work & _TIF_SECCOMP) {
105 struct seccomp_data sd;
108 sd.nr = regs->orig_ax;
109 sd.instruction_pointer = regs->ip;
111 if (arch == AUDIT_ARCH_X86_64) {
112 sd.args[0] = regs->di;
113 sd.args[1] = regs->si;
114 sd.args[2] = regs->dx;
115 sd.args[3] = regs->r10;
116 sd.args[4] = regs->r8;
117 sd.args[5] = regs->r9;
121 sd.args[0] = regs->bx;
122 sd.args[1] = regs->cx;
123 sd.args[2] = regs->dx;
124 sd.args[3] = regs->si;
125 sd.args[4] = regs->di;
126 sd.args[5] = regs->bp;
129 BUILD_BUG_ON(SECCOMP_PHASE1_OK != 0);
130 BUILD_BUG_ON(SECCOMP_PHASE1_SKIP != 1);
132 ret = seccomp_phase1(&sd);
133 if (ret == SECCOMP_PHASE1_SKIP) {
136 } else if (ret != SECCOMP_PHASE1_OK) {
137 return ret; /* Go directly to phase 2 */
140 work &= ~_TIF_SECCOMP;
144 /* Do our best to finish without phase 2. */
146 return ret; /* seccomp and/or nohz only (ret == 0 here) */
148 #ifdef CONFIG_AUDITSYSCALL
149 if (work == _TIF_SYSCALL_AUDIT) {
151 * If there is no more work to be done except auditing,
152 * then audit in phase 1. Phase 2 always audits, so, if
153 * we audit here, then we can't go on to phase 2.
155 do_audit_syscall_entry(regs, arch);
160 return 1; /* Something is enabled that we can't handle in phase 1 */
163 /* Returns the syscall nr to run (which should match regs->orig_ax). */
164 long syscall_trace_enter_phase2(struct pt_regs *regs, u32 arch,
165 unsigned long phase1_result)
167 struct thread_info *ti = pt_regs_to_thread_info(regs);
169 u32 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
171 if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
172 BUG_ON(regs != task_pt_regs(current));
175 * If we stepped into a sysenter/syscall insn, it trapped in
176 * kernel mode; do_debug() cleared TF and set TIF_SINGLESTEP.
177 * If user-mode had set TF itself, then it's still clear from
178 * do_debug() and we need to set it again to restore the user
179 * state. If we entered on the slow path, TF was already set.
181 if (work & _TIF_SINGLESTEP)
182 regs->flags |= X86_EFLAGS_TF;
184 #ifdef CONFIG_SECCOMP
186 * Call seccomp_phase2 before running the other hooks so that
187 * they can see any changes made by a seccomp tracer.
189 if (phase1_result > 1 && seccomp_phase2(phase1_result)) {
190 /* seccomp failures shouldn't expose any additional code. */
195 if (unlikely(work & _TIF_SYSCALL_EMU))
198 if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
199 tracehook_report_syscall_entry(regs))
202 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
203 trace_sys_enter(regs, regs->orig_ax);
205 do_audit_syscall_entry(regs, arch);
207 return ret ?: regs->orig_ax;
210 long syscall_trace_enter(struct pt_regs *regs)
212 u32 arch = is_ia32_task() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
213 unsigned long phase1_result = syscall_trace_enter_phase1(regs, arch);
215 if (phase1_result == 0)
216 return regs->orig_ax;
218 return syscall_trace_enter_phase2(regs, arch, phase1_result);
221 #define EXIT_TO_USERMODE_LOOP_FLAGS \
222 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
223 _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
225 static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
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.
236 /* We have work to do. */
239 if (cached_flags & _TIF_NEED_RESCHED)
242 if (cached_flags & _TIF_UPROBE)
243 uprobe_notify_resume(regs);
245 /* deal with pending signal delivery */
246 if (cached_flags & _TIF_SIGPENDING)
249 if (cached_flags & _TIF_NOTIFY_RESUME) {
250 clear_thread_flag(TIF_NOTIFY_RESUME);
251 tracehook_notify_resume(regs);
254 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
255 fire_user_return_notifiers();
257 /* Disable IRQs and retry */
260 cached_flags = READ_ONCE(pt_regs_to_thread_info(regs)->flags);
262 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
268 /* Called with IRQs disabled. */
269 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
271 struct thread_info *ti = pt_regs_to_thread_info(regs);
274 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
279 cached_flags = READ_ONCE(ti->flags);
281 if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
282 exit_to_usermode_loop(regs, cached_flags);
286 * Compat syscalls set TS_COMPAT. Make sure we clear it before
287 * returning to user mode. We need to clear it *after* signal
288 * handling, because syscall restart has a fixup for compat
289 * syscalls. The fixup is exercised by the ptrace_syscall_32
292 ti->status &= ~TS_COMPAT;
298 #define SYSCALL_EXIT_WORK_FLAGS \
299 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
300 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
302 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
306 audit_syscall_exit(regs);
308 if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
309 trace_sys_exit(regs, regs->ax);
312 * If TIF_SYSCALL_EMU is set, we only get here because of
313 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
314 * We already reported this syscall instruction in
315 * syscall_trace_enter().
318 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
320 if (step || cached_flags & _TIF_SYSCALL_TRACE)
321 tracehook_report_syscall_exit(regs, step);
325 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
326 * state such that we can immediately switch to user mode.
328 __visible inline void syscall_return_slowpath(struct pt_regs *regs)
330 struct thread_info *ti = pt_regs_to_thread_info(regs);
331 u32 cached_flags = READ_ONCE(ti->flags);
333 CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
335 if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
336 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
340 * First do one-time work. If these work items are enabled, we
341 * want to run them exactly once per syscall exit with IRQs on.
343 if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
344 syscall_slow_exit_work(regs, cached_flags);
347 prepare_exit_to_usermode(regs);
350 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
352 * Does a 32-bit syscall. Called with IRQs on and does all entry and
353 * exit work and returns with IRQs off. This function is extremely hot
354 * in workloads that use it, and it's usually called from
355 * do_fast_syscall_32, so forcibly inline it to improve performance.
358 /* 32-bit kernels use a trap gate for INT80, and the asm code calls here. */
361 /* 64-bit kernels use do_syscall_32_irqs_off() instead. */
364 __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
366 struct thread_info *ti = pt_regs_to_thread_info(regs);
367 unsigned int nr = (unsigned int)regs->orig_ax;
369 #ifdef CONFIG_IA32_EMULATION
370 ti->status |= TS_COMPAT;
373 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
375 * Subtlety here: if ptrace pokes something larger than
376 * 2^32-1 into orig_ax, this truncates it. This may or
377 * may not be necessary, but it matches the old asm
380 nr = syscall_trace_enter(regs);
383 if (likely(nr < IA32_NR_syscalls)) {
385 * It's possible that a 32-bit syscall implementation
386 * takes a 64-bit parameter but nonetheless assumes that
387 * the high bits are zero. Make sure we zero-extend all
390 regs->ax = ia32_sys_call_table[nr](
391 (unsigned int)regs->bx, (unsigned int)regs->cx,
392 (unsigned int)regs->dx, (unsigned int)regs->si,
393 (unsigned int)regs->di, (unsigned int)regs->bp);
396 syscall_return_slowpath(regs);
400 /* Handles INT80 on 64-bit kernels */
401 __visible void do_syscall_32_irqs_off(struct pt_regs *regs)
404 do_syscall_32_irqs_on(regs);
408 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
409 __visible long do_fast_syscall_32(struct pt_regs *regs)
412 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
413 * convention. Adjust regs so it looks like we entered using int80.
416 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
417 vdso_image_32.sym_int80_landing_pad;
420 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
421 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
424 regs->ip = landing_pad;
427 * Fetch EBP from where the vDSO stashed it.
429 * WARNING: We are in CONTEXT_USER and RCU isn't paying attention!
435 * Micro-optimization: the pointer we're following is explicitly
436 * 32 bits, so it can't be out of range.
438 __get_user(*(u32 *)®s->bp,
439 (u32 __user __force *)(unsigned long)(u32)regs->sp)
441 get_user(*(u32 *)®s->bp,
442 (u32 __user __force *)(unsigned long)(u32)regs->sp)
446 /* User code screwed up. */
449 #ifdef CONFIG_CONTEXT_TRACKING
450 enter_from_user_mode();
452 prepare_exit_to_usermode(regs);
453 return 0; /* Keep it simple: use IRET. */
456 /* Now this is just like a normal syscall. */
457 do_syscall_32_irqs_on(regs);
461 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
462 * SYSRETL is available on all 64-bit CPUs, so we don't need to
463 * bother with SYSEXIT.
465 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
466 * because the ECX fixup above will ensure that this is essentially
469 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
470 regs->ip == landing_pad &&
471 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
474 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
476 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
477 * because the ECX fixup above will ensure that this is essentially
480 * We don't allow syscalls at all from VM86 mode, but we still
481 * need to check VM, because we might be returning from sys_vm86.
483 return static_cpu_has(X86_FEATURE_SEP) &&
484 regs->cs == __USER_CS && regs->ss == __USER_DS &&
485 regs->ip == landing_pad &&
486 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;