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/nospec.h>
24 #include <linux/uprobes.h>
25
26 #include <asm/desc.h>
27 #include <asm/traps.h>
28 #include <asm/vdso.h>
29 #include <asm/uaccess.h>
30 #include <asm/cpufeature.h>
31
32 #define CREATE_TRACE_POINTS
33 #include <trace/events/syscalls.h>
34
35 #ifdef CONFIG_CONTEXT_TRACKING
36 /* Called on entry from user mode with IRQs off. */
enter_from_user_mode(void)37 __visible inline void enter_from_user_mode(void)
38 {
39 CT_WARN_ON(ct_state() != CONTEXT_USER);
40 user_exit_irqoff();
41 }
42 #else
enter_from_user_mode(void)43 static inline void enter_from_user_mode(void) {}
44 #endif
45
do_audit_syscall_entry(struct pt_regs * regs,u32 arch)46 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
47 {
48 #ifdef CONFIG_X86_64
49 if (arch == AUDIT_ARCH_X86_64) {
50 audit_syscall_entry(regs->orig_ax, regs->di,
51 regs->si, regs->dx, regs->r10);
52 } else
53 #endif
54 {
55 audit_syscall_entry(regs->orig_ax, regs->bx,
56 regs->cx, regs->dx, regs->si);
57 }
58 }
59
60 /*
61 * Returns the syscall nr to run (which should match regs->orig_ax) or -1
62 * to skip the syscall.
63 */
syscall_trace_enter(struct pt_regs * regs)64 static long syscall_trace_enter(struct pt_regs *regs)
65 {
66 u32 arch = in_ia32_syscall() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
67
68 struct thread_info *ti = current_thread_info();
69 unsigned long ret = 0;
70 bool emulated = false;
71 u32 work;
72
73 if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
74 BUG_ON(regs != task_pt_regs(current));
75
76 work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
77
78 if (unlikely(work & _TIF_SYSCALL_EMU))
79 emulated = true;
80
81 if ((emulated || (work & _TIF_SYSCALL_TRACE)) &&
82 tracehook_report_syscall_entry(regs))
83 return -1L;
84
85 if (emulated)
86 return -1L;
87
88 #ifdef CONFIG_SECCOMP
89 /*
90 * Do seccomp after ptrace, to catch any tracer changes.
91 */
92 if (work & _TIF_SECCOMP) {
93 struct seccomp_data sd;
94
95 sd.arch = arch;
96 sd.nr = regs->orig_ax;
97 sd.instruction_pointer = regs->ip;
98 #ifdef CONFIG_X86_64
99 if (arch == AUDIT_ARCH_X86_64) {
100 sd.args[0] = regs->di;
101 sd.args[1] = regs->si;
102 sd.args[2] = regs->dx;
103 sd.args[3] = regs->r10;
104 sd.args[4] = regs->r8;
105 sd.args[5] = regs->r9;
106 } else
107 #endif
108 {
109 sd.args[0] = regs->bx;
110 sd.args[1] = regs->cx;
111 sd.args[2] = regs->dx;
112 sd.args[3] = regs->si;
113 sd.args[4] = regs->di;
114 sd.args[5] = regs->bp;
115 }
116
117 ret = __secure_computing(&sd);
118 if (ret == -1)
119 return ret;
120 }
121 #endif
122
123 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
124 trace_sys_enter(regs, regs->orig_ax);
125
126 do_audit_syscall_entry(regs, arch);
127
128 return ret ?: regs->orig_ax;
129 }
130
131 #define EXIT_TO_USERMODE_LOOP_FLAGS \
132 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
133 _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY)
134
exit_to_usermode_loop(struct pt_regs * regs,u32 cached_flags)135 static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
136 {
137 /*
138 * In order to return to user mode, we need to have IRQs off with
139 * none of _TIF_SIGPENDING, _TIF_NOTIFY_RESUME, _TIF_USER_RETURN_NOTIFY,
140 * _TIF_UPROBE, or _TIF_NEED_RESCHED set. Several of these flags
141 * can be set at any time on preemptable kernels if we have IRQs on,
142 * so we need to loop. Disabling preemption wouldn't help: doing the
143 * work to clear some of the flags can sleep.
144 */
145 while (true) {
146 /* We have work to do. */
147 local_irq_enable();
148
149 if (cached_flags & _TIF_NEED_RESCHED)
150 schedule();
151
152 if (cached_flags & _TIF_UPROBE)
153 uprobe_notify_resume(regs);
154
155 /* deal with pending signal delivery */
156 if (cached_flags & _TIF_SIGPENDING)
157 do_signal(regs);
158
159 if (cached_flags & _TIF_NOTIFY_RESUME) {
160 clear_thread_flag(TIF_NOTIFY_RESUME);
161 tracehook_notify_resume(regs);
162 }
163
164 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
165 fire_user_return_notifiers();
166
167 /* Disable IRQs and retry */
168 local_irq_disable();
169
170 cached_flags = READ_ONCE(current_thread_info()->flags);
171
172 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
173 break;
174 }
175 }
176
177 /* Called with IRQs disabled. */
prepare_exit_to_usermode(struct pt_regs * regs)178 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
179 {
180 struct thread_info *ti = current_thread_info();
181 u32 cached_flags;
182
183 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
184 local_irq_disable();
185
186 lockdep_sys_exit();
187
188 cached_flags = READ_ONCE(ti->flags);
189
190 if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
191 exit_to_usermode_loop(regs, cached_flags);
192
193 #ifdef CONFIG_COMPAT
194 /*
195 * Compat syscalls set TS_COMPAT. Make sure we clear it before
196 * returning to user mode. We need to clear it *after* signal
197 * handling, because syscall restart has a fixup for compat
198 * syscalls. The fixup is exercised by the ptrace_syscall_32
199 * selftest.
200 *
201 * We also need to clear TS_REGS_POKED_I386: the 32-bit tracer
202 * special case only applies after poking regs and before the
203 * very next return to user mode.
204 */
205 ti->status &= ~(TS_COMPAT|TS_I386_REGS_POKED);
206 #endif
207
208 user_enter_irqoff();
209 }
210
211 #define SYSCALL_EXIT_WORK_FLAGS \
212 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \
213 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
214
syscall_slow_exit_work(struct pt_regs * regs,u32 cached_flags)215 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
216 {
217 bool step;
218
219 audit_syscall_exit(regs);
220
221 if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
222 trace_sys_exit(regs, regs->ax);
223
224 /*
225 * If TIF_SYSCALL_EMU is set, we only get here because of
226 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
227 * We already reported this syscall instruction in
228 * syscall_trace_enter().
229 */
230 step = unlikely(
231 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
232 == _TIF_SINGLESTEP);
233 if (step || cached_flags & _TIF_SYSCALL_TRACE)
234 tracehook_report_syscall_exit(regs, step);
235 }
236
237 /*
238 * Called with IRQs on and fully valid regs. Returns with IRQs off in a
239 * state such that we can immediately switch to user mode.
240 */
syscall_return_slowpath(struct pt_regs * regs)241 __visible inline void syscall_return_slowpath(struct pt_regs *regs)
242 {
243 struct thread_info *ti = current_thread_info();
244 u32 cached_flags = READ_ONCE(ti->flags);
245
246 CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
247
248 if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
249 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
250 local_irq_enable();
251
252 /*
253 * First do one-time work. If these work items are enabled, we
254 * want to run them exactly once per syscall exit with IRQs on.
255 */
256 if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
257 syscall_slow_exit_work(regs, cached_flags);
258
259 local_irq_disable();
260 prepare_exit_to_usermode(regs);
261 }
262
263 #ifdef CONFIG_X86_64
do_syscall_64(struct pt_regs * regs)264 __visible void do_syscall_64(struct pt_regs *regs)
265 {
266 struct thread_info *ti = current_thread_info();
267 unsigned long nr = regs->orig_ax;
268
269 enter_from_user_mode();
270 local_irq_enable();
271
272 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
273 nr = syscall_trace_enter(regs);
274
275 /*
276 * NB: Native and x32 syscalls are dispatched from the same
277 * table. The only functional difference is the x32 bit in
278 * regs->orig_ax, which changes the behavior of some syscalls.
279 */
280 if (likely((nr & __SYSCALL_MASK) < NR_syscalls)) {
281 nr = array_index_nospec(nr & __SYSCALL_MASK, NR_syscalls);
282 regs->ax = sys_call_table[nr](
283 regs->di, regs->si, regs->dx,
284 regs->r10, regs->r8, regs->r9);
285 }
286
287 syscall_return_slowpath(regs);
288 }
289 #endif
290
291 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
292 /*
293 * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does
294 * all entry and exit work and returns with IRQs off. This function is
295 * extremely hot in workloads that use it, and it's usually called from
296 * do_fast_syscall_32, so forcibly inline it to improve performance.
297 */
do_syscall_32_irqs_on(struct pt_regs * regs)298 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
299 {
300 struct thread_info *ti = current_thread_info();
301 unsigned int nr = (unsigned int)regs->orig_ax;
302
303 #ifdef CONFIG_IA32_EMULATION
304 ti->status |= TS_COMPAT;
305 #endif
306
307 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
308 /*
309 * Subtlety here: if ptrace pokes something larger than
310 * 2^32-1 into orig_ax, this truncates it. This may or
311 * may not be necessary, but it matches the old asm
312 * behavior.
313 */
314 nr = syscall_trace_enter(regs);
315 }
316
317 if (likely(nr < IA32_NR_syscalls)) {
318 nr = array_index_nospec(nr, IA32_NR_syscalls);
319 /*
320 * It's possible that a 32-bit syscall implementation
321 * takes a 64-bit parameter but nonetheless assumes that
322 * the high bits are zero. Make sure we zero-extend all
323 * of the args.
324 */
325 regs->ax = ia32_sys_call_table[nr](
326 (unsigned int)regs->bx, (unsigned int)regs->cx,
327 (unsigned int)regs->dx, (unsigned int)regs->si,
328 (unsigned int)regs->di, (unsigned int)regs->bp);
329 }
330
331 syscall_return_slowpath(regs);
332 }
333
334 /* Handles int $0x80 */
do_int80_syscall_32(struct pt_regs * regs)335 __visible void do_int80_syscall_32(struct pt_regs *regs)
336 {
337 enter_from_user_mode();
338 local_irq_enable();
339 do_syscall_32_irqs_on(regs);
340 }
341
342 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
do_fast_syscall_32(struct pt_regs * regs)343 __visible long do_fast_syscall_32(struct pt_regs *regs)
344 {
345 /*
346 * Called using the internal vDSO SYSENTER/SYSCALL32 calling
347 * convention. Adjust regs so it looks like we entered using int80.
348 */
349
350 unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
351 vdso_image_32.sym_int80_landing_pad;
352
353 /*
354 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
355 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
356 * Fix it up.
357 */
358 regs->ip = landing_pad;
359
360 enter_from_user_mode();
361
362 local_irq_enable();
363
364 /* Fetch EBP from where the vDSO stashed it. */
365 if (
366 #ifdef CONFIG_X86_64
367 /*
368 * Micro-optimization: the pointer we're following is explicitly
369 * 32 bits, so it can't be out of range.
370 */
371 __get_user(*(u32 *)®s->bp,
372 (u32 __user __force *)(unsigned long)(u32)regs->sp)
373 #else
374 get_user(*(u32 *)®s->bp,
375 (u32 __user __force *)(unsigned long)(u32)regs->sp)
376 #endif
377 ) {
378
379 /* User code screwed up. */
380 local_irq_disable();
381 regs->ax = -EFAULT;
382 prepare_exit_to_usermode(regs);
383 return 0; /* Keep it simple: use IRET. */
384 }
385
386 /* Now this is just like a normal syscall. */
387 do_syscall_32_irqs_on(regs);
388
389 #ifdef CONFIG_X86_64
390 /*
391 * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
392 * SYSRETL is available on all 64-bit CPUs, so we don't need to
393 * bother with SYSEXIT.
394 *
395 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
396 * because the ECX fixup above will ensure that this is essentially
397 * never the case.
398 */
399 return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
400 regs->ip == landing_pad &&
401 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
402 #else
403 /*
404 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
405 *
406 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
407 * because the ECX fixup above will ensure that this is essentially
408 * never the case.
409 *
410 * We don't allow syscalls at all from VM86 mode, but we still
411 * need to check VM, because we might be returning from sys_vm86.
412 */
413 return static_cpu_has(X86_FEATURE_SEP) &&
414 regs->cs == __USER_CS && regs->ss == __USER_DS &&
415 regs->ip == landing_pad &&
416 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
417 #endif
418 }
419 #endif
420