1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_ENTRYCOMMON_H
3 #define __LINUX_ENTRYCOMMON_H
4
5 #include <linux/static_call_types.h>
6 #include <linux/ptrace.h>
7 #include <linux/syscalls.h>
8 #include <linux/seccomp.h>
9 #include <linux/sched.h>
10 #include <linux/context_tracking.h>
11 #include <linux/livepatch.h>
12 #include <linux/resume_user_mode.h>
13 #include <linux/tick.h>
14 #include <linux/kmsan.h>
15
16 #include <asm/entry-common.h>
17 #include <asm/syscall.h>
18
19 /*
20 * Define dummy _TIF work flags if not defined by the architecture or for
21 * disabled functionality.
22 */
23 #ifndef _TIF_PATCH_PENDING
24 # define _TIF_PATCH_PENDING (0)
25 #endif
26
27 #ifndef _TIF_UPROBE
28 # define _TIF_UPROBE (0)
29 #endif
30
31 /*
32 * SYSCALL_WORK flags handled in syscall_enter_from_user_mode()
33 */
34 #ifndef ARCH_SYSCALL_WORK_ENTER
35 # define ARCH_SYSCALL_WORK_ENTER (0)
36 #endif
37
38 /*
39 * SYSCALL_WORK flags handled in syscall_exit_to_user_mode()
40 */
41 #ifndef ARCH_SYSCALL_WORK_EXIT
42 # define ARCH_SYSCALL_WORK_EXIT (0)
43 #endif
44
45 #define SYSCALL_WORK_ENTER (SYSCALL_WORK_SECCOMP | \
46 SYSCALL_WORK_SYSCALL_TRACEPOINT | \
47 SYSCALL_WORK_SYSCALL_TRACE | \
48 SYSCALL_WORK_SYSCALL_EMU | \
49 SYSCALL_WORK_SYSCALL_AUDIT | \
50 SYSCALL_WORK_SYSCALL_USER_DISPATCH | \
51 ARCH_SYSCALL_WORK_ENTER)
52 #define SYSCALL_WORK_EXIT (SYSCALL_WORK_SYSCALL_TRACEPOINT | \
53 SYSCALL_WORK_SYSCALL_TRACE | \
54 SYSCALL_WORK_SYSCALL_AUDIT | \
55 SYSCALL_WORK_SYSCALL_USER_DISPATCH | \
56 SYSCALL_WORK_SYSCALL_EXIT_TRAP | \
57 ARCH_SYSCALL_WORK_EXIT)
58
59 /*
60 * TIF flags handled in exit_to_user_mode_loop()
61 */
62 #ifndef ARCH_EXIT_TO_USER_MODE_WORK
63 # define ARCH_EXIT_TO_USER_MODE_WORK (0)
64 #endif
65
66 #define EXIT_TO_USER_MODE_WORK \
67 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \
68 _TIF_NEED_RESCHED | _TIF_NEED_RESCHED_LAZY | \
69 _TIF_PATCH_PENDING | _TIF_NOTIFY_SIGNAL | \
70 ARCH_EXIT_TO_USER_MODE_WORK)
71
72 /**
73 * arch_enter_from_user_mode - Architecture specific sanity check for user mode regs
74 * @regs: Pointer to currents pt_regs
75 *
76 * Defaults to an empty implementation. Can be replaced by architecture
77 * specific code.
78 *
79 * Invoked from syscall_enter_from_user_mode() in the non-instrumentable
80 * section. Use __always_inline so the compiler cannot push it out of line
81 * and make it instrumentable.
82 */
83 static __always_inline void arch_enter_from_user_mode(struct pt_regs *regs);
84
85 #ifndef arch_enter_from_user_mode
arch_enter_from_user_mode(struct pt_regs * regs)86 static __always_inline void arch_enter_from_user_mode(struct pt_regs *regs) {}
87 #endif
88
89 /**
90 * enter_from_user_mode - Establish state when coming from user mode
91 *
92 * Syscall/interrupt entry disables interrupts, but user mode is traced as
93 * interrupts enabled. Also with NO_HZ_FULL RCU might be idle.
94 *
95 * 1) Tell lockdep that interrupts are disabled
96 * 2) Invoke context tracking if enabled to reactivate RCU
97 * 3) Trace interrupts off state
98 *
99 * Invoked from architecture specific syscall entry code with interrupts
100 * disabled. The calling code has to be non-instrumentable. When the
101 * function returns all state is correct and interrupts are still
102 * disabled. The subsequent functions can be instrumented.
103 *
104 * This is invoked when there is architecture specific functionality to be
105 * done between establishing state and enabling interrupts. The caller must
106 * enable interrupts before invoking syscall_enter_from_user_mode_work().
107 */
enter_from_user_mode(struct pt_regs * regs)108 static __always_inline void enter_from_user_mode(struct pt_regs *regs)
109 {
110 arch_enter_from_user_mode(regs);
111 lockdep_hardirqs_off(CALLER_ADDR0);
112
113 CT_WARN_ON(__ct_state() != CT_STATE_USER);
114 user_exit_irqoff();
115
116 instrumentation_begin();
117 kmsan_unpoison_entry_regs(regs);
118 trace_hardirqs_off_finish();
119 instrumentation_end();
120 }
121
122 /**
123 * syscall_enter_from_user_mode_prepare - Establish state and enable interrupts
124 * @regs: Pointer to currents pt_regs
125 *
126 * Invoked from architecture specific syscall entry code with interrupts
127 * disabled. The calling code has to be non-instrumentable. When the
128 * function returns all state is correct, interrupts are enabled and the
129 * subsequent functions can be instrumented.
130 *
131 * This handles lockdep, RCU (context tracking) and tracing state, i.e.
132 * the functionality provided by enter_from_user_mode().
133 *
134 * This is invoked when there is extra architecture specific functionality
135 * to be done between establishing state and handling user mode entry work.
136 */
137 void syscall_enter_from_user_mode_prepare(struct pt_regs *regs);
138
139 long syscall_trace_enter(struct pt_regs *regs, long syscall,
140 unsigned long work);
141
142 /**
143 * syscall_enter_from_user_mode_work - Check and handle work before invoking
144 * a syscall
145 * @regs: Pointer to currents pt_regs
146 * @syscall: The syscall number
147 *
148 * Invoked from architecture specific syscall entry code with interrupts
149 * enabled after invoking syscall_enter_from_user_mode_prepare() and extra
150 * architecture specific work.
151 *
152 * Returns: The original or a modified syscall number
153 *
154 * If the returned syscall number is -1 then the syscall should be
155 * skipped. In this case the caller may invoke syscall_set_error() or
156 * syscall_set_return_value() first. If neither of those are called and -1
157 * is returned, then the syscall will fail with ENOSYS.
158 *
159 * It handles the following work items:
160 *
161 * 1) syscall_work flag dependent invocations of
162 * ptrace_report_syscall_entry(), __secure_computing(), trace_sys_enter()
163 * 2) Invocation of audit_syscall_entry()
164 */
syscall_enter_from_user_mode_work(struct pt_regs * regs,long syscall)165 static __always_inline long syscall_enter_from_user_mode_work(struct pt_regs *regs, long syscall)
166 {
167 unsigned long work = READ_ONCE(current_thread_info()->syscall_work);
168
169 if (work & SYSCALL_WORK_ENTER)
170 syscall = syscall_trace_enter(regs, syscall, work);
171
172 return syscall;
173 }
174
175 /**
176 * syscall_enter_from_user_mode - Establish state and check and handle work
177 * before invoking a syscall
178 * @regs: Pointer to currents pt_regs
179 * @syscall: The syscall number
180 *
181 * Invoked from architecture specific syscall entry code with interrupts
182 * disabled. The calling code has to be non-instrumentable. When the
183 * function returns all state is correct, interrupts are enabled and the
184 * subsequent functions can be instrumented.
185 *
186 * This is combination of syscall_enter_from_user_mode_prepare() and
187 * syscall_enter_from_user_mode_work().
188 *
189 * Returns: The original or a modified syscall number. See
190 * syscall_enter_from_user_mode_work() for further explanation.
191 */
syscall_enter_from_user_mode(struct pt_regs * regs,long syscall)192 static __always_inline long syscall_enter_from_user_mode(struct pt_regs *regs, long syscall)
193 {
194 long ret;
195
196 enter_from_user_mode(regs);
197
198 instrumentation_begin();
199 local_irq_enable();
200 ret = syscall_enter_from_user_mode_work(regs, syscall);
201 instrumentation_end();
202
203 return ret;
204 }
205
206 /**
207 * local_irq_enable_exit_to_user - Exit to user variant of local_irq_enable()
208 * @ti_work: Cached TIF flags gathered with interrupts disabled
209 *
210 * Defaults to local_irq_enable(). Can be supplied by architecture specific
211 * code.
212 */
213 static inline void local_irq_enable_exit_to_user(unsigned long ti_work);
214
215 #ifndef local_irq_enable_exit_to_user
local_irq_enable_exit_to_user(unsigned long ti_work)216 static inline void local_irq_enable_exit_to_user(unsigned long ti_work)
217 {
218 local_irq_enable();
219 }
220 #endif
221
222 /**
223 * local_irq_disable_exit_to_user - Exit to user variant of local_irq_disable()
224 *
225 * Defaults to local_irq_disable(). Can be supplied by architecture specific
226 * code.
227 */
228 static inline void local_irq_disable_exit_to_user(void);
229
230 #ifndef local_irq_disable_exit_to_user
local_irq_disable_exit_to_user(void)231 static inline void local_irq_disable_exit_to_user(void)
232 {
233 local_irq_disable();
234 }
235 #endif
236
237 /**
238 * arch_exit_to_user_mode_work - Architecture specific TIF work for exit
239 * to user mode.
240 * @regs: Pointer to currents pt_regs
241 * @ti_work: Cached TIF flags gathered with interrupts disabled
242 *
243 * Invoked from exit_to_user_mode_loop() with interrupt enabled
244 *
245 * Defaults to NOOP. Can be supplied by architecture specific code.
246 */
247 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
248 unsigned long ti_work);
249
250 #ifndef arch_exit_to_user_mode_work
arch_exit_to_user_mode_work(struct pt_regs * regs,unsigned long ti_work)251 static inline void arch_exit_to_user_mode_work(struct pt_regs *regs,
252 unsigned long ti_work)
253 {
254 }
255 #endif
256
257 /**
258 * arch_exit_to_user_mode_prepare - Architecture specific preparation for
259 * exit to user mode.
260 * @regs: Pointer to currents pt_regs
261 * @ti_work: Cached TIF flags gathered with interrupts disabled
262 *
263 * Invoked from exit_to_user_mode_prepare() with interrupt disabled as the last
264 * function before return. Defaults to NOOP.
265 */
266 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
267 unsigned long ti_work);
268
269 #ifndef arch_exit_to_user_mode_prepare
arch_exit_to_user_mode_prepare(struct pt_regs * regs,unsigned long ti_work)270 static inline void arch_exit_to_user_mode_prepare(struct pt_regs *regs,
271 unsigned long ti_work)
272 {
273 }
274 #endif
275
276 /**
277 * arch_exit_to_user_mode - Architecture specific final work before
278 * exit to user mode.
279 *
280 * Invoked from exit_to_user_mode() with interrupt disabled as the last
281 * function before return. Defaults to NOOP.
282 *
283 * This needs to be __always_inline because it is non-instrumentable code
284 * invoked after context tracking switched to user mode.
285 *
286 * An architecture implementation must not do anything complex, no locking
287 * etc. The main purpose is for speculation mitigations.
288 */
289 static __always_inline void arch_exit_to_user_mode(void);
290
291 #ifndef arch_exit_to_user_mode
arch_exit_to_user_mode(void)292 static __always_inline void arch_exit_to_user_mode(void) { }
293 #endif
294
295 /**
296 * arch_do_signal_or_restart - Architecture specific signal delivery function
297 * @regs: Pointer to currents pt_regs
298 *
299 * Invoked from exit_to_user_mode_loop().
300 */
301 void arch_do_signal_or_restart(struct pt_regs *regs);
302
303 /**
304 * exit_to_user_mode_loop - do any pending work before leaving to user space
305 */
306 unsigned long exit_to_user_mode_loop(struct pt_regs *regs,
307 unsigned long ti_work);
308
309 /**
310 * exit_to_user_mode_prepare - call exit_to_user_mode_loop() if required
311 * @regs: Pointer to pt_regs on entry stack
312 *
313 * 1) check that interrupts are disabled
314 * 2) call tick_nohz_user_enter_prepare()
315 * 3) call exit_to_user_mode_loop() if any flags from
316 * EXIT_TO_USER_MODE_WORK are set
317 * 4) check that interrupts are still disabled
318 */
exit_to_user_mode_prepare(struct pt_regs * regs)319 static __always_inline void exit_to_user_mode_prepare(struct pt_regs *regs)
320 {
321 unsigned long ti_work;
322
323 lockdep_assert_irqs_disabled();
324
325 /* Flush pending rcuog wakeup before the last need_resched() check */
326 tick_nohz_user_enter_prepare();
327
328 ti_work = read_thread_flags();
329 if (unlikely(ti_work & EXIT_TO_USER_MODE_WORK))
330 ti_work = exit_to_user_mode_loop(regs, ti_work);
331
332 arch_exit_to_user_mode_prepare(regs, ti_work);
333
334 /* Ensure that kernel state is sane for a return to userspace */
335 kmap_assert_nomap();
336 lockdep_assert_irqs_disabled();
337 lockdep_sys_exit();
338 }
339
340 /**
341 * exit_to_user_mode - Fixup state when exiting to user mode
342 *
343 * Syscall/interrupt exit enables interrupts, but the kernel state is
344 * interrupts disabled when this is invoked. Also tell RCU about it.
345 *
346 * 1) Trace interrupts on state
347 * 2) Invoke context tracking if enabled to adjust RCU state
348 * 3) Invoke architecture specific last minute exit code, e.g. speculation
349 * mitigations, etc.: arch_exit_to_user_mode()
350 * 4) Tell lockdep that interrupts are enabled
351 *
352 * Invoked from architecture specific code when syscall_exit_to_user_mode()
353 * is not suitable as the last step before returning to userspace. Must be
354 * invoked with interrupts disabled and the caller must be
355 * non-instrumentable.
356 * The caller has to invoke syscall_exit_to_user_mode_work() before this.
357 */
exit_to_user_mode(void)358 static __always_inline void exit_to_user_mode(void)
359 {
360 instrumentation_begin();
361 trace_hardirqs_on_prepare();
362 lockdep_hardirqs_on_prepare();
363 instrumentation_end();
364
365 user_enter_irqoff();
366 arch_exit_to_user_mode();
367 lockdep_hardirqs_on(CALLER_ADDR0);
368 }
369
370 /**
371 * syscall_exit_work - Handle work before returning to user mode
372 * @regs: Pointer to current pt_regs
373 * @work: Current thread syscall work
374 *
375 * Do one-time syscall specific work.
376 */
377 void syscall_exit_work(struct pt_regs *regs, unsigned long work);
378
379 /**
380 * syscall_exit_to_user_mode_work - Handle work before returning to user mode
381 * @regs: Pointer to currents pt_regs
382 *
383 * Same as step 1 and 2 of syscall_exit_to_user_mode() but without calling
384 * exit_to_user_mode() to perform the final transition to user mode.
385 *
386 * Calling convention is the same as for syscall_exit_to_user_mode() and it
387 * returns with all work handled and interrupts disabled. The caller must
388 * invoke exit_to_user_mode() before actually switching to user mode to
389 * make the final state transitions. Interrupts must stay disabled between
390 * return from this function and the invocation of exit_to_user_mode().
391 */
syscall_exit_to_user_mode_work(struct pt_regs * regs)392 static __always_inline void syscall_exit_to_user_mode_work(struct pt_regs *regs)
393 {
394 unsigned long work = READ_ONCE(current_thread_info()->syscall_work);
395 unsigned long nr = syscall_get_nr(current, regs);
396
397 CT_WARN_ON(ct_state() != CT_STATE_KERNEL);
398
399 if (IS_ENABLED(CONFIG_PROVE_LOCKING)) {
400 if (WARN(irqs_disabled(), "syscall %lu left IRQs disabled", nr))
401 local_irq_enable();
402 }
403
404 rseq_syscall(regs);
405
406 /*
407 * Do one-time syscall specific work. If these work items are
408 * enabled, we want to run them exactly once per syscall exit with
409 * interrupts enabled.
410 */
411 if (unlikely(work & SYSCALL_WORK_EXIT))
412 syscall_exit_work(regs, work);
413 local_irq_disable_exit_to_user();
414 exit_to_user_mode_prepare(regs);
415 }
416
417 /**
418 * syscall_exit_to_user_mode - Handle work before returning to user mode
419 * @regs: Pointer to currents pt_regs
420 *
421 * Invoked with interrupts enabled and fully valid regs. Returns with all
422 * work handled, interrupts disabled such that the caller can immediately
423 * switch to user mode. Called from architecture specific syscall and ret
424 * from fork code.
425 *
426 * The call order is:
427 * 1) One-time syscall exit work:
428 * - rseq syscall exit
429 * - audit
430 * - syscall tracing
431 * - ptrace (single stepping)
432 *
433 * 2) Preparatory work
434 * - Exit to user mode loop (common TIF handling). Invokes
435 * arch_exit_to_user_mode_work() for architecture specific TIF work
436 * - Architecture specific one time work arch_exit_to_user_mode_prepare()
437 * - Address limit and lockdep checks
438 *
439 * 3) Final transition (lockdep, tracing, context tracking, RCU), i.e. the
440 * functionality in exit_to_user_mode().
441 *
442 * This is a combination of syscall_exit_to_user_mode_work() (1,2) and
443 * exit_to_user_mode(). This function is preferred unless there is a
444 * compelling architectural reason to use the separate functions.
445 */
syscall_exit_to_user_mode(struct pt_regs * regs)446 static __always_inline void syscall_exit_to_user_mode(struct pt_regs *regs)
447 {
448 instrumentation_begin();
449 syscall_exit_to_user_mode_work(regs);
450 instrumentation_end();
451 exit_to_user_mode();
452 }
453
454 /**
455 * irqentry_enter_from_user_mode - Establish state before invoking the irq handler
456 * @regs: Pointer to currents pt_regs
457 *
458 * Invoked from architecture specific entry code with interrupts disabled.
459 * Can only be called when the interrupt entry came from user mode. The
460 * calling code must be non-instrumentable. When the function returns all
461 * state is correct and the subsequent functions can be instrumented.
462 *
463 * The function establishes state (lockdep, RCU (context tracking), tracing)
464 */
465 void irqentry_enter_from_user_mode(struct pt_regs *regs);
466
467 /**
468 * irqentry_exit_to_user_mode - Interrupt exit work
469 * @regs: Pointer to current's pt_regs
470 *
471 * Invoked with interrupts disabled and fully valid regs. Returns with all
472 * work handled, interrupts disabled such that the caller can immediately
473 * switch to user mode. Called from architecture specific interrupt
474 * handling code.
475 *
476 * The call order is #2 and #3 as described in syscall_exit_to_user_mode().
477 * Interrupt exit is not invoking #1 which is the syscall specific one time
478 * work.
479 */
480 void irqentry_exit_to_user_mode(struct pt_regs *regs);
481
482 #ifndef irqentry_state
483 /**
484 * struct irqentry_state - Opaque object for exception state storage
485 * @exit_rcu: Used exclusively in the irqentry_*() calls; signals whether the
486 * exit path has to invoke ct_irq_exit().
487 * @lockdep: Used exclusively in the irqentry_nmi_*() calls; ensures that
488 * lockdep state is restored correctly on exit from nmi.
489 *
490 * This opaque object is filled in by the irqentry_*_enter() functions and
491 * must be passed back into the corresponding irqentry_*_exit() functions
492 * when the exception is complete.
493 *
494 * Callers of irqentry_*_[enter|exit]() must consider this structure opaque
495 * and all members private. Descriptions of the members are provided to aid in
496 * the maintenance of the irqentry_*() functions.
497 */
498 typedef struct irqentry_state {
499 union {
500 bool exit_rcu;
501 bool lockdep;
502 };
503 } irqentry_state_t;
504 #endif
505
506 /**
507 * irqentry_enter - Handle state tracking on ordinary interrupt entries
508 * @regs: Pointer to pt_regs of interrupted context
509 *
510 * Invokes:
511 * - lockdep irqflag state tracking as low level ASM entry disabled
512 * interrupts.
513 *
514 * - Context tracking if the exception hit user mode.
515 *
516 * - The hardirq tracer to keep the state consistent as low level ASM
517 * entry disabled interrupts.
518 *
519 * As a precondition, this requires that the entry came from user mode,
520 * idle, or a kernel context in which RCU is watching.
521 *
522 * For kernel mode entries RCU handling is done conditional. If RCU is
523 * watching then the only RCU requirement is to check whether the tick has
524 * to be restarted. If RCU is not watching then ct_irq_enter() has to be
525 * invoked on entry and ct_irq_exit() on exit.
526 *
527 * Avoiding the ct_irq_enter/exit() calls is an optimization but also
528 * solves the problem of kernel mode pagefaults which can schedule, which
529 * is not possible after invoking ct_irq_enter() without undoing it.
530 *
531 * For user mode entries irqentry_enter_from_user_mode() is invoked to
532 * establish the proper context for NOHZ_FULL. Otherwise scheduling on exit
533 * would not be possible.
534 *
535 * Returns: An opaque object that must be passed to idtentry_exit()
536 */
537 irqentry_state_t noinstr irqentry_enter(struct pt_regs *regs);
538
539 /**
540 * irqentry_exit_cond_resched - Conditionally reschedule on return from interrupt
541 *
542 * Conditional reschedule with additional sanity checks.
543 */
544 void raw_irqentry_exit_cond_resched(void);
545 #ifdef CONFIG_PREEMPT_DYNAMIC
546 #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
547 #define irqentry_exit_cond_resched_dynamic_enabled raw_irqentry_exit_cond_resched
548 #define irqentry_exit_cond_resched_dynamic_disabled NULL
549 DECLARE_STATIC_CALL(irqentry_exit_cond_resched, raw_irqentry_exit_cond_resched);
550 #define irqentry_exit_cond_resched() static_call(irqentry_exit_cond_resched)()
551 #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY)
552 DECLARE_STATIC_KEY_TRUE(sk_dynamic_irqentry_exit_cond_resched);
553 void dynamic_irqentry_exit_cond_resched(void);
554 #define irqentry_exit_cond_resched() dynamic_irqentry_exit_cond_resched()
555 #endif
556 #else /* CONFIG_PREEMPT_DYNAMIC */
557 #define irqentry_exit_cond_resched() raw_irqentry_exit_cond_resched()
558 #endif /* CONFIG_PREEMPT_DYNAMIC */
559
560 /**
561 * irqentry_exit - Handle return from exception that used irqentry_enter()
562 * @regs: Pointer to pt_regs (exception entry regs)
563 * @state: Return value from matching call to irqentry_enter()
564 *
565 * Depending on the return target (kernel/user) this runs the necessary
566 * preemption and work checks if possible and required and returns to
567 * the caller with interrupts disabled and no further work pending.
568 *
569 * This is the last action before returning to the low level ASM code which
570 * just needs to return to the appropriate context.
571 *
572 * Counterpart to irqentry_enter().
573 */
574 void noinstr irqentry_exit(struct pt_regs *regs, irqentry_state_t state);
575
576 /**
577 * irqentry_nmi_enter - Handle NMI entry
578 * @regs: Pointer to currents pt_regs
579 *
580 * Similar to irqentry_enter() but taking care of the NMI constraints.
581 */
582 irqentry_state_t noinstr irqentry_nmi_enter(struct pt_regs *regs);
583
584 /**
585 * irqentry_nmi_exit - Handle return from NMI handling
586 * @regs: Pointer to pt_regs (NMI entry regs)
587 * @irq_state: Return value from matching call to irqentry_nmi_enter()
588 *
589 * Last action before returning to the low level assembly code.
590 *
591 * Counterpart to irqentry_nmi_enter().
592 */
593 void noinstr irqentry_nmi_exit(struct pt_regs *regs, irqentry_state_t irq_state);
594
595 #endif
596