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