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/mm.h> 14 #include <linux/smp.h> 15 #include <linux/errno.h> 16 #include <linux/ptrace.h> 17 #include <linux/tracehook.h> 18 #include <linux/audit.h> 19 #include <linux/seccomp.h> 20 #include <linux/signal.h> 21 #include <linux/export.h> 22 #include <linux/context_tracking.h> 23 #include <linux/user-return-notifier.h> 24 #include <linux/nospec.h> 25 #include <linux/uprobes.h> 26 #include <linux/livepatch.h> 27 #include <linux/syscalls.h> 28 #include <linux/uaccess.h> 29 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 39 #define CREATE_TRACE_POINTS 40 #include <trace/events/syscalls.h> 41 42 #ifdef CONFIG_CONTEXT_TRACKING 43 /* Called on entry from user mode with IRQs off. */ 44 __visible inline void enter_from_user_mode(void) 45 { 46 CT_WARN_ON(ct_state() != CONTEXT_USER); 47 user_exit_irqoff(); 48 } 49 #else 50 static inline void enter_from_user_mode(void) {} 51 #endif 52 53 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch) 54 { 55 #ifdef CONFIG_X86_64 56 if (arch == AUDIT_ARCH_X86_64) { 57 audit_syscall_entry(regs->orig_ax, regs->di, 58 regs->si, regs->dx, regs->r10); 59 } else 60 #endif 61 { 62 audit_syscall_entry(regs->orig_ax, regs->bx, 63 regs->cx, regs->dx, regs->si); 64 } 65 } 66 67 /* 68 * Returns the syscall nr to run (which should match regs->orig_ax) or -1 69 * to skip the syscall. 70 */ 71 static long syscall_trace_enter(struct pt_regs *regs) 72 { 73 u32 arch = in_ia32_syscall() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64; 74 75 struct thread_info *ti = current_thread_info(); 76 unsigned long ret = 0; 77 u32 work; 78 79 if (IS_ENABLED(CONFIG_DEBUG_ENTRY)) 80 BUG_ON(regs != task_pt_regs(current)); 81 82 work = READ_ONCE(ti->flags); 83 84 if (work & (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_EMU)) { 85 ret = tracehook_report_syscall_entry(regs); 86 if (ret || (work & _TIF_SYSCALL_EMU)) 87 return -1L; 88 } 89 90 #ifdef CONFIG_SECCOMP 91 /* 92 * Do seccomp after ptrace, to catch any tracer changes. 93 */ 94 if (work & _TIF_SECCOMP) { 95 struct seccomp_data sd; 96 97 sd.arch = arch; 98 sd.nr = regs->orig_ax; 99 sd.instruction_pointer = regs->ip; 100 #ifdef CONFIG_X86_64 101 if (arch == AUDIT_ARCH_X86_64) { 102 sd.args[0] = regs->di; 103 sd.args[1] = regs->si; 104 sd.args[2] = regs->dx; 105 sd.args[3] = regs->r10; 106 sd.args[4] = regs->r8; 107 sd.args[5] = regs->r9; 108 } else 109 #endif 110 { 111 sd.args[0] = regs->bx; 112 sd.args[1] = regs->cx; 113 sd.args[2] = regs->dx; 114 sd.args[3] = regs->si; 115 sd.args[4] = regs->di; 116 sd.args[5] = regs->bp; 117 } 118 119 ret = __secure_computing(&sd); 120 if (ret == -1) 121 return ret; 122 } 123 #endif 124 125 if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT))) 126 trace_sys_enter(regs, regs->orig_ax); 127 128 do_audit_syscall_entry(regs, arch); 129 130 return ret ?: regs->orig_ax; 131 } 132 133 #define EXIT_TO_USERMODE_LOOP_FLAGS \ 134 (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE | \ 135 _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY | _TIF_PATCH_PENDING) 136 137 static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags) 138 { 139 /* 140 * In order to return to user mode, we need to have IRQs off with 141 * none of EXIT_TO_USERMODE_LOOP_FLAGS set. Several of these flags 142 * can be set at any time on preemptible kernels if we have IRQs on, 143 * so we need to loop. Disabling preemption wouldn't help: doing the 144 * work to clear some of the flags can sleep. 145 */ 146 while (true) { 147 /* We have work to do. */ 148 local_irq_enable(); 149 150 if (cached_flags & _TIF_NEED_RESCHED) 151 schedule(); 152 153 if (cached_flags & _TIF_UPROBE) 154 uprobe_notify_resume(regs); 155 156 if (cached_flags & _TIF_PATCH_PENDING) 157 klp_update_patch_state(current); 158 159 /* deal with pending signal delivery */ 160 if (cached_flags & _TIF_SIGPENDING) 161 do_signal(regs); 162 163 if (cached_flags & _TIF_NOTIFY_RESUME) { 164 clear_thread_flag(TIF_NOTIFY_RESUME); 165 tracehook_notify_resume(regs); 166 rseq_handle_notify_resume(NULL, regs); 167 } 168 169 if (cached_flags & _TIF_USER_RETURN_NOTIFY) 170 fire_user_return_notifiers(); 171 172 /* Disable IRQs and retry */ 173 local_irq_disable(); 174 175 cached_flags = READ_ONCE(current_thread_info()->flags); 176 177 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS)) 178 break; 179 } 180 } 181 182 /* Called with IRQs disabled. */ 183 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs) 184 { 185 struct thread_info *ti = current_thread_info(); 186 u32 cached_flags; 187 188 addr_limit_user_check(); 189 190 lockdep_assert_irqs_disabled(); 191 lockdep_sys_exit(); 192 193 cached_flags = READ_ONCE(ti->flags); 194 195 if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS)) 196 exit_to_usermode_loop(regs, cached_flags); 197 198 /* Reload ti->flags; we may have rescheduled above. */ 199 cached_flags = READ_ONCE(ti->flags); 200 201 if (unlikely(cached_flags & _TIF_IO_BITMAP)) 202 tss_update_io_bitmap(); 203 204 fpregs_assert_state_consistent(); 205 if (unlikely(cached_flags & _TIF_NEED_FPU_LOAD)) 206 switch_fpu_return(); 207 208 #ifdef CONFIG_COMPAT 209 /* 210 * Compat syscalls set TS_COMPAT. Make sure we clear it before 211 * returning to user mode. We need to clear it *after* signal 212 * handling, because syscall restart has a fixup for compat 213 * syscalls. The fixup is exercised by the ptrace_syscall_32 214 * selftest. 215 * 216 * We also need to clear TS_REGS_POKED_I386: the 32-bit tracer 217 * special case only applies after poking regs and before the 218 * very next return to user mode. 219 */ 220 ti->status &= ~(TS_COMPAT|TS_I386_REGS_POKED); 221 #endif 222 223 user_enter_irqoff(); 224 225 mds_user_clear_cpu_buffers(); 226 } 227 228 #define SYSCALL_EXIT_WORK_FLAGS \ 229 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \ 230 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT) 231 232 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags) 233 { 234 bool step; 235 236 audit_syscall_exit(regs); 237 238 if (cached_flags & _TIF_SYSCALL_TRACEPOINT) 239 trace_sys_exit(regs, regs->ax); 240 241 /* 242 * If TIF_SYSCALL_EMU is set, we only get here because of 243 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP). 244 * We already reported this syscall instruction in 245 * syscall_trace_enter(). 246 */ 247 step = unlikely( 248 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU)) 249 == _TIF_SINGLESTEP); 250 if (step || cached_flags & _TIF_SYSCALL_TRACE) 251 tracehook_report_syscall_exit(regs, step); 252 } 253 254 /* 255 * Called with IRQs on and fully valid regs. Returns with IRQs off in a 256 * state such that we can immediately switch to user mode. 257 */ 258 __visible inline void syscall_return_slowpath(struct pt_regs *regs) 259 { 260 struct thread_info *ti = current_thread_info(); 261 u32 cached_flags = READ_ONCE(ti->flags); 262 263 CT_WARN_ON(ct_state() != CONTEXT_KERNEL); 264 265 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && 266 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax)) 267 local_irq_enable(); 268 269 rseq_syscall(regs); 270 271 /* 272 * First do one-time work. If these work items are enabled, we 273 * want to run them exactly once per syscall exit with IRQs on. 274 */ 275 if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS)) 276 syscall_slow_exit_work(regs, cached_flags); 277 278 local_irq_disable(); 279 prepare_exit_to_usermode(regs); 280 } 281 282 #ifdef CONFIG_X86_64 283 __visible void do_syscall_64(unsigned long nr, struct pt_regs *regs) 284 { 285 struct thread_info *ti; 286 287 enter_from_user_mode(); 288 local_irq_enable(); 289 ti = current_thread_info(); 290 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) 291 nr = syscall_trace_enter(regs); 292 293 if (likely(nr < NR_syscalls)) { 294 nr = array_index_nospec(nr, NR_syscalls); 295 regs->ax = sys_call_table[nr](regs); 296 #ifdef CONFIG_X86_X32_ABI 297 } else if (likely((nr & __X32_SYSCALL_BIT) && 298 (nr & ~__X32_SYSCALL_BIT) < X32_NR_syscalls)) { 299 nr = array_index_nospec(nr & ~__X32_SYSCALL_BIT, 300 X32_NR_syscalls); 301 regs->ax = x32_sys_call_table[nr](regs); 302 #endif 303 } 304 305 syscall_return_slowpath(regs); 306 } 307 #endif 308 309 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) 310 /* 311 * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does 312 * all entry and exit work and returns with IRQs off. This function is 313 * extremely hot in workloads that use it, and it's usually called from 314 * do_fast_syscall_32, so forcibly inline it to improve performance. 315 */ 316 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs) 317 { 318 struct thread_info *ti = current_thread_info(); 319 unsigned int nr = (unsigned int)regs->orig_ax; 320 321 #ifdef CONFIG_IA32_EMULATION 322 ti->status |= TS_COMPAT; 323 #endif 324 325 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) { 326 /* 327 * Subtlety here: if ptrace pokes something larger than 328 * 2^32-1 into orig_ax, this truncates it. This may or 329 * may not be necessary, but it matches the old asm 330 * behavior. 331 */ 332 nr = syscall_trace_enter(regs); 333 } 334 335 if (likely(nr < IA32_NR_syscalls)) { 336 nr = array_index_nospec(nr, IA32_NR_syscalls); 337 regs->ax = ia32_sys_call_table[nr](regs); 338 } 339 340 syscall_return_slowpath(regs); 341 } 342 343 /* Handles int $0x80 */ 344 __visible void do_int80_syscall_32(struct pt_regs *regs) 345 { 346 enter_from_user_mode(); 347 local_irq_enable(); 348 do_syscall_32_irqs_on(regs); 349 } 350 351 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */ 352 __visible long do_fast_syscall_32(struct pt_regs *regs) 353 { 354 /* 355 * Called using the internal vDSO SYSENTER/SYSCALL32 calling 356 * convention. Adjust regs so it looks like we entered using int80. 357 */ 358 359 unsigned long landing_pad = (unsigned long)current->mm->context.vdso + 360 vdso_image_32.sym_int80_landing_pad; 361 362 /* 363 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward 364 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction. 365 * Fix it up. 366 */ 367 regs->ip = landing_pad; 368 369 enter_from_user_mode(); 370 371 local_irq_enable(); 372 373 /* Fetch EBP from where the vDSO stashed it. */ 374 if ( 375 #ifdef CONFIG_X86_64 376 /* 377 * Micro-optimization: the pointer we're following is explicitly 378 * 32 bits, so it can't be out of range. 379 */ 380 __get_user(*(u32 *)®s->bp, 381 (u32 __user __force *)(unsigned long)(u32)regs->sp) 382 #else 383 get_user(*(u32 *)®s->bp, 384 (u32 __user __force *)(unsigned long)(u32)regs->sp) 385 #endif 386 ) { 387 388 /* User code screwed up. */ 389 local_irq_disable(); 390 regs->ax = -EFAULT; 391 prepare_exit_to_usermode(regs); 392 return 0; /* Keep it simple: use IRET. */ 393 } 394 395 /* Now this is just like a normal syscall. */ 396 do_syscall_32_irqs_on(regs); 397 398 #ifdef CONFIG_X86_64 399 /* 400 * Opportunistic SYSRETL: if possible, try to return using SYSRETL. 401 * SYSRETL is available on all 64-bit CPUs, so we don't need to 402 * bother with SYSEXIT. 403 * 404 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP, 405 * because the ECX fixup above will ensure that this is essentially 406 * never the case. 407 */ 408 return regs->cs == __USER32_CS && regs->ss == __USER_DS && 409 regs->ip == landing_pad && 410 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0; 411 #else 412 /* 413 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT. 414 * 415 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP, 416 * because the ECX fixup above will ensure that this is essentially 417 * never the case. 418 * 419 * We don't allow syscalls at all from VM86 mode, but we still 420 * need to check VM, because we might be returning from sys_vm86. 421 */ 422 return static_cpu_has(X86_FEATURE_SEP) && 423 regs->cs == __USER_CS && regs->ss == __USER_DS && 424 regs->ip == landing_pad && 425 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0; 426 #endif 427 } 428 #endif 429 430 SYSCALL_DEFINE0(ni_syscall) 431 { 432 return -ENOSYS; 433 } 434