Linux Audio

Check our new training course

Linux kernel drivers training

Mar 31-Apr 9, 2025, special US time zones
Register
Loading...
v4.6
 
  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
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
v6.9.4
  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * common.c - C code for kernel entry and exit
  4 * Copyright (c) 2015 Andrew Lutomirski
 
  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/sched/task_stack.h>
 13#include <linux/entry-common.h>
 14#include <linux/mm.h>
 15#include <linux/smp.h>
 16#include <linux/errno.h>
 17#include <linux/ptrace.h>
 
 
 
 
 18#include <linux/export.h>
 19#include <linux/nospec.h>
 20#include <linux/syscalls.h>
 21#include <linux/uaccess.h>
 22#include <linux/init.h>
 23
 24#ifdef CONFIG_XEN_PV
 25#include <xen/xen-ops.h>
 26#include <xen/events.h>
 27#endif
 28
 29#include <asm/apic.h>
 30#include <asm/desc.h>
 31#include <asm/traps.h>
 32#include <asm/vdso.h>
 
 33#include <asm/cpufeature.h>
 34#include <asm/fpu/api.h>
 35#include <asm/nospec-branch.h>
 36#include <asm/io_bitmap.h>
 37#include <asm/syscall.h>
 38#include <asm/irq_stack.h>
 39
 40#ifdef CONFIG_X86_64
 
 
 
 
 
 
 
 
 41
 42static __always_inline bool do_syscall_x64(struct pt_regs *regs, int nr)
 
 
 43{
 44	/*
 45	 * Convert negative numbers to very high and thus out of range
 46	 * numbers for comparisons.
 47	 */
 48	unsigned int unr = nr;
 
 49
 50	if (likely(unr < NR_syscalls)) {
 51		unr = array_index_nospec(unr, NR_syscalls);
 52		regs->ax = x64_sys_call(regs, unr);
 53		return true;
 
 
 
 
 
 
 
 54	}
 55	return false;
 56}
 57
 58static __always_inline bool do_syscall_x32(struct pt_regs *regs, int nr)
 
 
 
 
 
 
 
 
 
 
 
 
 
 59{
 
 
 
 
 
 
 
 
 
 
 60	/*
 61	 * Adjust the starting offset of the table, and convert numbers
 62	 * < __X32_SYSCALL_BIT to very high and thus out of range
 63	 * numbers for comparisons.
 64	 */
 65	unsigned int xnr = nr - __X32_SYSCALL_BIT;
 
 66
 67	if (IS_ENABLED(CONFIG_X86_X32_ABI) && likely(xnr < X32_NR_syscalls)) {
 68		xnr = array_index_nospec(xnr, X32_NR_syscalls);
 69		regs->ax = x32_sys_call(regs, xnr);
 70		return true;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 71	}
 72	return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 73}
 74
 75/* Returns true to return using SYSRET, or false to use IRET */
 76__visible noinstr bool do_syscall_64(struct pt_regs *regs, int nr)
 
 77{
 78	add_random_kstack_offset();
 79	nr = syscall_enter_from_user_mode(regs, nr);
 80
 81	instrumentation_begin();
 82
 83	if (!do_syscall_x64(regs, nr) && !do_syscall_x32(regs, nr) && nr != -1) {
 84		/* Invalid system call, but still a system call. */
 85		regs->ax = __x64_sys_ni_syscall(regs);
 86	}
 87
 88	instrumentation_end();
 89	syscall_exit_to_user_mode(regs);
 90
 
 91	/*
 92	 * Check that the register state is valid for using SYSRET to exit
 93	 * to userspace.  Otherwise use the slower but fully capable IRET
 94	 * exit path.
 95	 */
 
 
 
 
 
 96
 97	/* XEN PV guests always use the IRET path */
 98	if (cpu_feature_enabled(X86_FEATURE_XENPV))
 99		return false;
100
101	/* SYSRET requires RCX == RIP and R11 == EFLAGS */
102	if (unlikely(regs->cx != regs->ip || regs->r11 != regs->flags))
103		return false;
104
105	/* CS and SS must match the values set in MSR_STAR */
106	if (unlikely(regs->cs != __USER_CS || regs->ss != __USER_DS))
107		return false;
108
109	/*
110	 * On Intel CPUs, SYSRET with non-canonical RCX/RIP will #GP
111	 * in kernel space.  This essentially lets the user take over
112	 * the kernel, since userspace controls RSP.
113	 *
114	 * TASK_SIZE_MAX covers all user-accessible addresses other than
115	 * the deprecated vsyscall page.
116	 */
117	if (unlikely(regs->ip >= TASK_SIZE_MAX))
118		return false;
119
120	/*
121	 * SYSRET cannot restore RF.  It can restore TF, but unlike IRET,
122	 * restoring TF results in a trap from userspace immediately after
123	 * SYSRET.
124	 */
125	if (unlikely(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)))
126		return false;
127
128	/* Use SYSRET to exit to userspace */
129	return true;
130}
131#endif
132
133#if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
134static __always_inline int syscall_32_enter(struct pt_regs *regs)
135{
136	if (IS_ENABLED(CONFIG_IA32_EMULATION))
137		current_thread_info()->status |= TS_COMPAT;
138
139	return (int)regs->orig_ax;
 
 
 
140}
141
142#ifdef CONFIG_IA32_EMULATION
143bool __ia32_enabled __ro_after_init = !IS_ENABLED(CONFIG_IA32_EMULATION_DEFAULT_DISABLED);
 
144
145static int ia32_emulation_override_cmdline(char *arg)
146{
147	return kstrtobool(arg, &__ia32_enabled);
148}
149early_param("ia32_emulation", ia32_emulation_override_cmdline);
150#endif
151
152/*
153 * Invoke a 32-bit syscall.  Called with IRQs on in CONTEXT_KERNEL.
154 */
155static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs, int nr)
156{
157	/*
158	 * Convert negative numbers to very high and thus out of range
159	 * numbers for comparisons.
 
 
 
 
160	 */
161	unsigned int unr = nr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
162
163	if (likely(unr < IA32_NR_syscalls)) {
164		unr = array_index_nospec(unr, IA32_NR_syscalls);
165		regs->ax = ia32_sys_call(regs, unr);
166	} else if (nr != -1) {
167		regs->ax = __ia32_sys_ni_syscall(regs);
168	}
169}
170
171#ifdef CONFIG_IA32_EMULATION
172static __always_inline bool int80_is_external(void)
173{
174	const unsigned int offs = (0x80 / 32) * 0x10;
175	const u32 bit = BIT(0x80 % 32);
176
177	/* The local APIC on XENPV guests is fake */
178	if (cpu_feature_enabled(X86_FEATURE_XENPV))
179		return false;
180
181	/*
182	 * If vector 0x80 is set in the APIC ISR then this is an external
183	 * interrupt. Either from broken hardware or injected by a VMM.
184	 *
185	 * Note: In guest mode this is only valid for secure guests where
186	 * the secure module fully controls the vAPIC exposed to the guest.
187	 */
188	return apic_read(APIC_ISR + offs) & bit;
189}
190
191/**
192 * do_int80_emulation - 32-bit legacy syscall C entry from asm
193 *
194 * This entry point can be used by 32-bit and 64-bit programs to perform
195 * 32-bit system calls.  Instances of INT $0x80 can be found inline in
196 * various programs and libraries.  It is also used by the vDSO's
197 * __kernel_vsyscall fallback for hardware that doesn't support a faster
198 * entry method.  Restarted 32-bit system calls also fall back to INT
199 * $0x80 regardless of what instruction was originally used to do the
200 * system call.
201 *
202 * This is considered a slow path.  It is not used by most libc
203 * implementations on modern hardware except during process startup.
204 *
205 * The arguments for the INT $0x80 based syscall are on stack in the
206 * pt_regs structure:
207 *   eax:				system call number
208 *   ebx, ecx, edx, esi, edi, ebp:	arg1 - arg 6
209 */
210__visible noinstr void do_int80_emulation(struct pt_regs *regs)
211{
212	int nr;
213
214	/* Kernel does not use INT $0x80! */
215	if (unlikely(!user_mode(regs))) {
216		irqentry_enter(regs);
217		instrumentation_begin();
218		panic("Unexpected external interrupt 0x80\n");
219	}
220
 
221	/*
222	 * Establish kernel context for instrumentation, including for
223	 * int80_is_external() below which calls into the APIC driver.
224	 * Identical for soft and external interrupts.
 
 
225	 */
226	enter_from_user_mode(regs);
 
227
228	instrumentation_begin();
229	add_random_kstack_offset();
230
231	/* Validate that this is a soft interrupt to the extent possible */
232	if (unlikely(int80_is_external()))
233		panic("Unexpected external interrupt 0x80\n");
234
235	/*
236	 * The low level idtentry code pushed -1 into regs::orig_ax
237	 * and regs::ax contains the syscall number.
238	 *
239	 * User tracing code (ptrace or signal handlers) might assume
240	 * that the regs::orig_ax contains a 32-bit number on invoking
241	 * a 32-bit syscall.
242	 *
243	 * Establish the syscall convention by saving the 32bit truncated
244	 * syscall number in regs::orig_ax and by invalidating regs::ax.
245	 */
246	regs->orig_ax = regs->ax & GENMASK(31, 0);
247	regs->ax = -ENOSYS;
248
249	nr = syscall_32_enter(regs);
250
251	local_irq_enable();
252	nr = syscall_enter_from_user_mode_work(regs, nr);
253	do_syscall_32_irqs_on(regs, nr);
254
255	instrumentation_end();
256	syscall_exit_to_user_mode(regs);
 
 
 
 
 
 
 
 
 
257}
258
259#ifdef CONFIG_X86_FRED
260/*
261 * A FRED-specific INT80 handler is warranted for the follwing reasons:
262 *
263 * 1) As INT instructions and hardware interrupts are separate event
264 *    types, FRED does not preclude the use of vector 0x80 for external
265 *    interrupts. As a result, the FRED setup code does not reserve
266 *    vector 0x80 and calling int80_is_external() is not merely
267 *    suboptimal but actively incorrect: it could cause a system call
268 *    to be incorrectly ignored.
269 *
270 * 2) It is called only for handling vector 0x80 of event type
271 *    EVENT_TYPE_SWINT and will never be called to handle any external
272 *    interrupt (event type EVENT_TYPE_EXTINT).
273 *
274 * 3) FRED has separate entry flows depending on if the event came from
275 *    user space or kernel space, and because the kernel does not use
276 *    INT insns, the FRED kernel entry handler fred_entry_from_kernel()
277 *    falls through to fred_bad_type() if the event type is
278 *    EVENT_TYPE_SWINT, i.e., INT insns. So if the kernel is handling
279 *    an INT insn, it can only be from a user level.
280 *
281 * 4) int80_emulation() does a CLEAR_BRANCH_HISTORY. While FRED will
282 *    likely take a different approach if it is ever needed: it
283 *    probably belongs in either fred_intx()/ fred_other() or
284 *    asm_fred_entrypoint_user(), depending on if this ought to be done
285 *    for all entries from userspace or only system
286 *    calls.
287 *
288 * 5) INT $0x80 is the fast path for 32-bit system calls under FRED.
289 */
290DEFINE_FREDENTRY_RAW(int80_emulation)
291{
292	int nr;
 
293
294	enter_from_user_mode(regs);
295
296	instrumentation_begin();
297	add_random_kstack_offset();
 
298
299	/*
300	 * FRED pushed 0 into regs::orig_ax and regs::ax contains the
301	 * syscall number.
302	 *
303	 * User tracing code (ptrace or signal handlers) might assume
304	 * that the regs::orig_ax contains a 32-bit number on invoking
305	 * a 32-bit syscall.
306	 *
307	 * Establish the syscall convention by saving the 32bit truncated
308	 * syscall number in regs::orig_ax and by invalidating regs::ax.
309	 */
310	regs->orig_ax = regs->ax & GENMASK(31, 0);
311	regs->ax = -ENOSYS;
 
 
 
 
312
313	nr = syscall_32_enter(regs);
 
 
 
 
314
 
315	local_irq_enable();
316	nr = syscall_enter_from_user_mode_work(regs, nr);
317	do_syscall_32_irqs_on(regs, nr);
318
319	instrumentation_end();
320	syscall_exit_to_user_mode(regs);
321}
322#endif
323#else /* CONFIG_IA32_EMULATION */
324
325/* Handles int $0x80 on a 32bit kernel */
326__visible noinstr void do_int80_syscall_32(struct pt_regs *regs)
327{
328	int nr = syscall_32_enter(regs);
329
330	add_random_kstack_offset();
331	/*
332	 * Subtlety here: if ptrace pokes something larger than 2^31-1 into
333	 * orig_ax, the int return value truncates it. This matches
334	 * the semantics of syscall_get_nr().
335	 */
336	nr = syscall_enter_from_user_mode(regs, nr);
337	instrumentation_begin();
 
 
 
338
339	do_syscall_32_irqs_on(regs, nr);
340
341	instrumentation_end();
342	syscall_exit_to_user_mode(regs);
343}
344#endif /* !CONFIG_IA32_EMULATION */
345
346static noinstr bool __do_fast_syscall_32(struct pt_regs *regs)
 
 
 
 
 
 
 
347{
348	int nr = syscall_32_enter(regs);
349	int res;
350
351	add_random_kstack_offset();
352	/*
353	 * This cannot use syscall_enter_from_user_mode() as it has to
354	 * fetch EBP before invoking any of the syscall entry work
355	 * functions.
356	 */
357	syscall_enter_from_user_mode_prepare(regs);
358
359	instrumentation_begin();
360	/* Fetch EBP from where the vDSO stashed it. */
361	if (IS_ENABLED(CONFIG_X86_64)) {
362		/*
363		 * Micro-optimization: the pointer we're following is
364		 * explicitly 32 bits, so it can't be out of range.
 
 
365		 */
366		res = __get_user(*(u32 *)&regs->bp,
367			 (u32 __user __force *)(unsigned long)(u32)regs->sp);
368	} else {
369		res = get_user(*(u32 *)&regs->bp,
370		       (u32 __user __force *)(unsigned long)(u32)regs->sp);
371	}
372
373	if (res) {
374		/* User code screwed up. */
375		regs->ax = -EFAULT;
376
377		local_irq_disable();
378		instrumentation_end();
379		irqentry_exit_to_user_mode(regs);
380		return false;
 
 
 
381	}
382
383	nr = syscall_enter_from_user_mode_work(regs, nr);
 
384
385	/* Now this is just like a normal syscall. */
386	do_syscall_32_irqs_on(regs, nr);
387
388	instrumentation_end();
389	syscall_exit_to_user_mode(regs);
390	return true;
391}
392
393/* Returns true to return using SYSEXIT/SYSRETL, or false to use IRET */
394__visible noinstr bool do_fast_syscall_32(struct pt_regs *regs)
395{
396	/*
397	 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
398	 * convention.  Adjust regs so it looks like we entered using int80.
399	 */
 
400	unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
401					vdso_image_32.sym_int80_landing_pad;
402
403	/*
404	 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
405	 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
406	 * Fix it up.
407	 */
408	regs->ip = landing_pad;
409
410	/* Invoke the syscall. If it failed, keep it simple: use IRET. */
411	if (!__do_fast_syscall_32(regs))
412		return false;
413
414	/*
415	 * Check that the register state is valid for using SYSRETL/SYSEXIT
416	 * to exit to userspace.  Otherwise use the slower but fully capable
417	 * IRET exit path.
418	 */
419
420	/* XEN PV guests always use the IRET path */
421	if (cpu_feature_enabled(X86_FEATURE_XENPV))
422		return false;
423
424	/* EIP must point to the VDSO landing pad */
425	if (unlikely(regs->ip != landing_pad))
426		return false;
427
428	/* CS and SS must match the values set in MSR_STAR */
429	if (unlikely(regs->cs != __USER32_CS || regs->ss != __USER_DS))
430		return false;
431
432	/* If the TF, RF, or VM flags are set, use IRET */
433	if (unlikely(regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)))
434		return false;
435
436	/* Use SYSRETL/SYSEXIT to exit to userspace */
437	return true;
438}
439
440/* Returns true to return using SYSEXIT/SYSRETL, or false to use IRET */
441__visible noinstr bool do_SYSENTER_32(struct pt_regs *regs)
442{
443	/* SYSENTER loses RSP, but the vDSO saved it in RBP. */
444	regs->sp = regs->bp;
445
446	/* SYSENTER clobbers EFLAGS.IF.  Assume it was set in usermode. */
447	regs->flags |= X86_EFLAGS_IF;
448
449	return do_fast_syscall_32(regs);
450}
451#endif
 
452
453SYSCALL_DEFINE0(ni_syscall)
454{
455	return -ENOSYS;
456}
 
 
457
458#ifdef CONFIG_XEN_PV
459#ifndef CONFIG_PREEMPTION
460/*
461 * Some hypercalls issued by the toolstack can take many 10s of
462 * seconds. Allow tasks running hypercalls via the privcmd driver to
463 * be voluntarily preempted even if full kernel preemption is
464 * disabled.
465 *
466 * Such preemptible hypercalls are bracketed by
467 * xen_preemptible_hcall_begin() and xen_preemptible_hcall_end()
468 * calls.
469 */
470DEFINE_PER_CPU(bool, xen_in_preemptible_hcall);
471EXPORT_SYMBOL_GPL(xen_in_preemptible_hcall);
472
473/*
474 * In case of scheduling the flag must be cleared and restored after
475 * returning from schedule as the task might move to a different CPU.
476 */
477static __always_inline bool get_and_clear_inhcall(void)
478{
479	bool inhcall = __this_cpu_read(xen_in_preemptible_hcall);
480
481	__this_cpu_write(xen_in_preemptible_hcall, false);
482	return inhcall;
483}
484
485static __always_inline void restore_inhcall(bool inhcall)
486{
487	__this_cpu_write(xen_in_preemptible_hcall, inhcall);
488}
489#else
490static __always_inline bool get_and_clear_inhcall(void) { return false; }
491static __always_inline void restore_inhcall(bool inhcall) { }
 
 
 
 
 
 
 
 
 
 
 
 
492#endif
493
494static void __xen_pv_evtchn_do_upcall(struct pt_regs *regs)
495{
496	struct pt_regs *old_regs = set_irq_regs(regs);
497
498	inc_irq_stat(irq_hv_callback_count);
499
500	xen_evtchn_do_upcall();
501
502	set_irq_regs(old_regs);
503}
504
505__visible noinstr void xen_pv_evtchn_do_upcall(struct pt_regs *regs)
506{
507	irqentry_state_t state = irqentry_enter(regs);
508	bool inhcall;
509
510	instrumentation_begin();
511	run_sysvec_on_irqstack_cond(__xen_pv_evtchn_do_upcall, regs);
512
513	inhcall = get_and_clear_inhcall();
514	if (inhcall && !WARN_ON_ONCE(state.exit_rcu)) {
515		irqentry_exit_cond_resched();
516		instrumentation_end();
517		restore_inhcall(inhcall);
518	} else {
519		instrumentation_end();
520		irqentry_exit(regs, state);
521	}
522}
523#endif /* CONFIG_XEN_PV */