1/*
  2 * common.c - C code for kernel entry and exit
  3 * Copyright (c) 2015 Andrew Lutomirski
  4 * GPL v2
  5 *
  6 * Based on asm and ptrace code by many authors.  The code here originated
  7 * in ptrace.c and signal.c.
  8 */
  9
 10#include <linux/kernel.h>
 11#include <linux/sched.h>
 12#include <linux/mm.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>
 24
 25#include <asm/desc.h>
 26#include <asm/traps.h>
 27#include <asm/vdso.h>
 28#include <asm/uaccess.h>
 29#include <asm/cpufeature.h>
 30
 31#define CREATE_TRACE_POINTS
 32#include <trace/events/syscalls.h>
 33
 34static struct thread_info *pt_regs_to_thread_info(struct pt_regs *regs)
 35{
 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);
 39}
 40
 41#ifdef CONFIG_CONTEXT_TRACKING
 42/* Called on entry from user mode with IRQs off. */
 43__visible void enter_from_user_mode(void)
 44{
 45	CT_WARN_ON(ct_state() != CONTEXT_USER);
 46	user_exit();
 47}
 48#else
 49static inline void enter_from_user_mode(void) {}
 50#endif
 51
 52static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
 53{
 54#ifdef CONFIG_X86_64
 55	if (arch == AUDIT_ARCH_X86_64) {
 56		audit_syscall_entry(regs->orig_ax, regs->di,
 57				    regs->si, regs->dx, regs->r10);
 58	} else
 59#endif
 60	{
 61		audit_syscall_entry(regs->orig_ax, regs->bx,
 62				    regs->cx, regs->dx, regs->si);
 63	}
 64}
 65
 66/*
 67 * We can return 0 to resume the syscall or anything else to go to phase
 68 * 2.  If we resume the syscall, we need to put something appropriate in
 69 * regs->orig_ax.
 70 *
 71 * NB: We don't have full pt_regs here, but regs->orig_ax and regs->ax
 72 * are fully functional.
 73 *
 74 * For phase 2's benefit, our return value is:
 75 * 0:			resume the syscall
 76 * 1:			go to phase 2; no seccomp phase 2 needed
 77 * anything else:	go to phase 2; pass return value to seccomp
 78 */
 79unsigned long syscall_trace_enter_phase1(struct pt_regs *regs, u32 arch)
 80{
 81	struct thread_info *ti = pt_regs_to_thread_info(regs);
 82	unsigned long ret = 0;
 83	u32 work;
 84
 85	if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
 86		BUG_ON(regs != task_pt_regs(current));
 87
 88	work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
 89
 90#ifdef CONFIG_SECCOMP
 91	/*
 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.
 95	 */
 96	if (work & _TIF_SECCOMP) {
 97		struct seccomp_data sd;
 98
 99		sd.arch = arch;
100		sd.nr = regs->orig_ax;
101		sd.instruction_pointer = regs->ip;
102#ifdef CONFIG_X86_64
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;
110		} else
111#endif
112		{
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;
119		}
120
121		BUILD_BUG_ON(SECCOMP_PHASE1_OK != 0);
122		BUILD_BUG_ON(SECCOMP_PHASE1_SKIP != 1);
123
124		ret = seccomp_phase1(&sd);
125		if (ret == SECCOMP_PHASE1_SKIP) {
126			regs->orig_ax = -1;
127			ret = 0;
128		} else if (ret != SECCOMP_PHASE1_OK) {
129			return ret;  /* Go directly to phase 2 */
130		}
131
132		work &= ~_TIF_SECCOMP;
133	}
134#endif
135
136	/* Do our best to finish without phase 2. */
137	if (work == 0)
138		return ret;  /* seccomp and/or nohz only (ret == 0 here) */
139
140#ifdef CONFIG_AUDITSYSCALL
141	if (work == _TIF_SYSCALL_AUDIT) {
142		/*
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.
146		 */
147		do_audit_syscall_entry(regs, arch);
148		return 0;
149	}
150#endif
151
152	return 1;  /* Something is enabled that we can't handle in phase 1 */
153}
154
155/* Returns the syscall nr to run (which should match regs->orig_ax). */
156long syscall_trace_enter_phase2(struct pt_regs *regs, u32 arch,
157				unsigned long phase1_result)
158{
159	struct thread_info *ti = pt_regs_to_thread_info(regs);
160	long ret = 0;
161	u32 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
162
163	if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
164		BUG_ON(regs != task_pt_regs(current));
165
166#ifdef CONFIG_SECCOMP
167	/*
168	 * Call seccomp_phase2 before running the other hooks so that
169	 * they can see any changes made by a seccomp tracer.
170	 */
171	if (phase1_result > 1 && seccomp_phase2(phase1_result)) {
172		/* seccomp failures shouldn't expose any additional code. */
173		return -1;
174	}
175#endif
176
177	if (unlikely(work & _TIF_SYSCALL_EMU))
178		ret = -1L;
179
180	if ((ret || test_thread_flag(TIF_SYSCALL_TRACE)) &&
181	    tracehook_report_syscall_entry(regs))
182		ret = -1L;
183
184	if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
185		trace_sys_enter(regs, regs->orig_ax);
186
187	do_audit_syscall_entry(regs, arch);
188
189	return ret ?: regs->orig_ax;
190}
191
192long syscall_trace_enter(struct pt_regs *regs)
193{
194	u32 arch = is_ia32_task() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
195	unsigned long phase1_result = syscall_trace_enter_phase1(regs, arch);
196
197	if (phase1_result == 0)
198		return regs->orig_ax;
199	else
200		return syscall_trace_enter_phase2(regs, arch, phase1_result);
201}
202
203#define EXIT_TO_USERMODE_LOOP_FLAGS				\
204	(_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE |	\
205	 _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
206
207static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
208{
209	/*
210	 * In order to return to user mode, we need to have IRQs off with
211	 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
212	 * _TIF_UPROBE, or _TIF_NEED_RESCHED set.  Several of these flags
213	 * can be set at any time on preemptable kernels if we have IRQs on,
214	 * so we need to loop.  Disabling preemption wouldn't help: doing the
215	 * work to clear some of the flags can sleep.
216	 */
217	while (true) {
218		/* We have work to do. */
219		local_irq_enable();
220
221		if (cached_flags & _TIF_NEED_RESCHED)
222			schedule();
223
224		if (cached_flags & _TIF_UPROBE)
225			uprobe_notify_resume(regs);
226
227		/* deal with pending signal delivery */
228		if (cached_flags & _TIF_SIGPENDING)
229			do_signal(regs);
230
231		if (cached_flags & _TIF_NOTIFY_RESUME) {
232			clear_thread_flag(TIF_NOTIFY_RESUME);
233			tracehook_notify_resume(regs);
234		}
235
236		if (cached_flags & _TIF_USER_RETURN_NOTIFY)
237			fire_user_return_notifiers();
238
239		/* Disable IRQs and retry */
240		local_irq_disable();
241
242		cached_flags = READ_ONCE(pt_regs_to_thread_info(regs)->flags);
243
244		if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
245			break;
246
247	}
248}
249
250/* Called with IRQs disabled. */
251__visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
252{
253	struct thread_info *ti = pt_regs_to_thread_info(regs);
254	u32 cached_flags;
255
256	if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
257		local_irq_disable();
258
259	lockdep_sys_exit();
260
261	cached_flags = READ_ONCE(ti->flags);
262
263	if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
264		exit_to_usermode_loop(regs, cached_flags);
265
266#ifdef CONFIG_COMPAT
267	/*
268	 * Compat syscalls set TS_COMPAT.  Make sure we clear it before
269	 * returning to user mode.  We need to clear it *after* signal
270	 * handling, because syscall restart has a fixup for compat
271	 * syscalls.  The fixup is exercised by the ptrace_syscall_32
272	 * selftest.
273	 */
274	ti->status &= ~TS_COMPAT;
275#endif
276
277	user_enter();
278}
279
280#define SYSCALL_EXIT_WORK_FLAGS				\
281	(_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT |	\
282	 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
283
284static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
285{
286	bool step;
287
288	audit_syscall_exit(regs);
289
290	if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
291		trace_sys_exit(regs, regs->ax);
292
293	/*
294	 * If TIF_SYSCALL_EMU is set, we only get here because of
295	 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
296	 * We already reported this syscall instruction in
297	 * syscall_trace_enter().
298	 */
299	step = unlikely(
300		(cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
301		== _TIF_SINGLESTEP);
302	if (step || cached_flags & _TIF_SYSCALL_TRACE)
303		tracehook_report_syscall_exit(regs, step);
304}
305
306/*
307 * Called with IRQs on and fully valid regs.  Returns with IRQs off in a
308 * state such that we can immediately switch to user mode.
309 */
310__visible inline void syscall_return_slowpath(struct pt_regs *regs)
311{
312	struct thread_info *ti = pt_regs_to_thread_info(regs);
313	u32 cached_flags = READ_ONCE(ti->flags);
314
315	CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
316
317	if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
318	    WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
319		local_irq_enable();
320
321	/*
322	 * First do one-time work.  If these work items are enabled, we
323	 * want to run them exactly once per syscall exit with IRQs on.
324	 */
325	if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
326		syscall_slow_exit_work(regs, cached_flags);
327
328	local_irq_disable();
329	prepare_exit_to_usermode(regs);
330}
331
332#ifdef CONFIG_X86_64
333__visible void do_syscall_64(struct pt_regs *regs)
334{
335	struct thread_info *ti = pt_regs_to_thread_info(regs);
336	unsigned long nr = regs->orig_ax;
337
338	enter_from_user_mode();
339	local_irq_enable();
340
341	if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
342		nr = syscall_trace_enter(regs);
343
344	/*
345	 * NB: Native and x32 syscalls are dispatched from the same
346	 * table.  The only functional difference is the x32 bit in
347	 * regs->orig_ax, which changes the behavior of some syscalls.
348	 */
349	if (likely((nr & __SYSCALL_MASK) < NR_syscalls)) {
350		regs->ax = sys_call_table[nr & __SYSCALL_MASK](
351			regs->di, regs->si, regs->dx,
352			regs->r10, regs->r8, regs->r9);
353	}
354
355	syscall_return_slowpath(regs);
356}
357#endif
358
359#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
360/*
361 * Does a 32-bit syscall.  Called with IRQs on in CONTEXT_KERNEL.  Does
362 * all entry and exit work and returns with IRQs off.  This function is
363 * extremely hot in workloads that use it, and it's usually called from
364 * do_fast_syscall_32, so forcibly inline it to improve performance.
365 */
366static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
367{
368	struct thread_info *ti = pt_regs_to_thread_info(regs);
369	unsigned int nr = (unsigned int)regs->orig_ax;
370
371#ifdef CONFIG_IA32_EMULATION
372	ti->status |= TS_COMPAT;
373#endif
374
375	if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
376		/*
377		 * Subtlety here: if ptrace pokes something larger than
378		 * 2^32-1 into orig_ax, this truncates it.  This may or
379		 * may not be necessary, but it matches the old asm
380		 * behavior.
381		 */
382		nr = syscall_trace_enter(regs);
383	}
384
385	if (likely(nr < IA32_NR_syscalls)) {
386		/*
387		 * It's possible that a 32-bit syscall implementation
388		 * takes a 64-bit parameter but nonetheless assumes that
389		 * the high bits are zero.  Make sure we zero-extend all
390		 * of the args.
391		 */
392		regs->ax = ia32_sys_call_table[nr](
393			(unsigned int)regs->bx, (unsigned int)regs->cx,
394			(unsigned int)regs->dx, (unsigned int)regs->si,
395			(unsigned int)regs->di, (unsigned int)regs->bp);
396	}
397
398	syscall_return_slowpath(regs);
399}
400
401/* Handles int $0x80 */
402__visible void do_int80_syscall_32(struct pt_regs *regs)
403{
404	enter_from_user_mode();
405	local_irq_enable();
406	do_syscall_32_irqs_on(regs);
407}
408
409/* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
410__visible long do_fast_syscall_32(struct pt_regs *regs)
411{
412	/*
413	 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
414	 * convention.  Adjust regs so it looks like we entered using int80.
415	 */
416
417	unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
418		vdso_image_32.sym_int80_landing_pad;
419
420	/*
421	 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
422	 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
423	 * Fix it up.
424	 */
425	regs->ip = landing_pad;
426
427	enter_from_user_mode();
428
429	local_irq_enable();
430
431	/* Fetch EBP from where the vDSO stashed it. */
432	if (
433#ifdef CONFIG_X86_64
434		/*
435		 * Micro-optimization: the pointer we're following is explicitly
436		 * 32 bits, so it can't be out of range.
437		 */
438		__get_user(*(u32 *)&regs->bp,
439			    (u32 __user __force *)(unsigned long)(u32)regs->sp)
440#else
441		get_user(*(u32 *)&regs->bp,
442			 (u32 __user __force *)(unsigned long)(u32)regs->sp)
443#endif
444		) {
445
446		/* User code screwed up. */
447		local_irq_disable();
448		regs->ax = -EFAULT;
449		prepare_exit_to_usermode(regs);
450		return 0;	/* Keep it simple: use IRET. */
451	}
452
453	/* Now this is just like a normal syscall. */
454	do_syscall_32_irqs_on(regs);
455
456#ifdef CONFIG_X86_64
457	/*
458	 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
459	 * SYSRETL is available on all 64-bit CPUs, so we don't need to
460	 * bother with SYSEXIT.
461	 *
462	 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
463	 * because the ECX fixup above will ensure that this is essentially
464	 * never the case.
465	 */
466	return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
467		regs->ip == landing_pad &&
468		(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
469#else
470	/*
471	 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
472	 *
473	 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
474	 * because the ECX fixup above will ensure that this is essentially
475	 * never the case.
476	 *
477	 * We don't allow syscalls at all from VM86 mode, but we still
478	 * need to check VM, because we might be returning from sys_vm86.
479	 */
480	return static_cpu_has(X86_FEATURE_SEP) &&
481		regs->cs == __USER_CS && regs->ss == __USER_DS &&
482		regs->ip == landing_pad &&
483		(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
484#endif
485}
486#endif