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/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/uprobes.h> 25 #include <linux/livepatch.h> 26 #include <linux/syscalls.h> 27 28 #include <asm/desc.h> 29 #include <asm/traps.h> 30 #include <asm/vdso.h> 31 #include <linux/uaccess.h> 32 #include <asm/cpufeature.h> 33 34 #define CREATE_TRACE_POINTS 35 #include <trace/events/syscalls.h> 36 37 #ifdef CONFIG_CONTEXT_TRACKING 38 /* Called on entry from user mode with IRQs off. */ 39 __visible inline void enter_from_user_mode(void) 40 { 41 CT_WARN_ON(ct_state() != CONTEXT_USER); 42 user_exit_irqoff(); 43 } 44 #else 45 static inline void enter_from_user_mode(void) {} 46 #endif 47 48 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch) 49 { 50 #ifdef CONFIG_X86_64 51 if (arch == AUDIT_ARCH_X86_64) { 52 audit_syscall_entry(regs->orig_ax, regs->di, 53 regs->si, regs->dx, regs->r10); 54 } else 55 #endif 56 { 57 audit_syscall_entry(regs->orig_ax, regs->bx, 58 regs->cx, regs->dx, regs->si); 59 } 60 } 61 62 /* 63 * Returns the syscall nr to run (which should match regs->orig_ax) or -1 64 * to skip the syscall. 65 */ 66 static long syscall_trace_enter(struct pt_regs *regs) 67 { 68 u32 arch = in_ia32_syscall() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64; 69 70 struct thread_info *ti = current_thread_info(); 71 unsigned long ret = 0; 72 bool emulated = false; 73 u32 work; 74 75 if (IS_ENABLED(CONFIG_DEBUG_ENTRY)) 76 BUG_ON(regs != task_pt_regs(current)); 77 78 work = READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY; 79 80 if (unlikely(work & _TIF_SYSCALL_EMU)) 81 emulated = true; 82 83 if ((emulated || (work & _TIF_SYSCALL_TRACE)) && 84 tracehook_report_syscall_entry(regs)) 85 return -1L; 86 87 if (emulated) 88 return -1L; 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 preemptable 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 /* deal with pending signal delivery */ 157 if (cached_flags & _TIF_SIGPENDING) 158 do_signal(regs); 159 160 if (cached_flags & _TIF_NOTIFY_RESUME) { 161 clear_thread_flag(TIF_NOTIFY_RESUME); 162 tracehook_notify_resume(regs); 163 } 164 165 if (cached_flags & _TIF_USER_RETURN_NOTIFY) 166 fire_user_return_notifiers(); 167 168 if (cached_flags & _TIF_PATCH_PENDING) 169 klp_update_patch_state(current); 170 171 /* Disable IRQs and retry */ 172 local_irq_disable(); 173 174 cached_flags = READ_ONCE(current_thread_info()->flags); 175 176 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS)) 177 break; 178 } 179 } 180 181 /* Called with IRQs disabled. */ 182 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs) 183 { 184 struct thread_info *ti = current_thread_info(); 185 u32 cached_flags; 186 187 addr_limit_user_check(); 188 189 lockdep_assert_irqs_disabled(); 190 lockdep_sys_exit(); 191 192 cached_flags = READ_ONCE(ti->flags); 193 194 if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS)) 195 exit_to_usermode_loop(regs, cached_flags); 196 197 #ifdef CONFIG_COMPAT 198 /* 199 * Compat syscalls set TS_COMPAT. Make sure we clear it before 200 * returning to user mode. We need to clear it *after* signal 201 * handling, because syscall restart has a fixup for compat 202 * syscalls. The fixup is exercised by the ptrace_syscall_32 203 * selftest. 204 * 205 * We also need to clear TS_REGS_POKED_I386: the 32-bit tracer 206 * special case only applies after poking regs and before the 207 * very next return to user mode. 208 */ 209 current->thread.status &= ~(TS_COMPAT|TS_I386_REGS_POKED); 210 #endif 211 212 user_enter_irqoff(); 213 } 214 215 #define SYSCALL_EXIT_WORK_FLAGS \ 216 (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT | \ 217 _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT) 218 219 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags) 220 { 221 bool step; 222 223 audit_syscall_exit(regs); 224 225 if (cached_flags & _TIF_SYSCALL_TRACEPOINT) 226 trace_sys_exit(regs, regs->ax); 227 228 /* 229 * If TIF_SYSCALL_EMU is set, we only get here because of 230 * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP). 231 * We already reported this syscall instruction in 232 * syscall_trace_enter(). 233 */ 234 step = unlikely( 235 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU)) 236 == _TIF_SINGLESTEP); 237 if (step || cached_flags & _TIF_SYSCALL_TRACE) 238 tracehook_report_syscall_exit(regs, step); 239 } 240 241 /* 242 * Called with IRQs on and fully valid regs. Returns with IRQs off in a 243 * state such that we can immediately switch to user mode. 244 */ 245 __visible inline void syscall_return_slowpath(struct pt_regs *regs) 246 { 247 struct thread_info *ti = current_thread_info(); 248 u32 cached_flags = READ_ONCE(ti->flags); 249 250 CT_WARN_ON(ct_state() != CONTEXT_KERNEL); 251 252 if (IS_ENABLED(CONFIG_PROVE_LOCKING) && 253 WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax)) 254 local_irq_enable(); 255 256 /* 257 * First do one-time work. If these work items are enabled, we 258 * want to run them exactly once per syscall exit with IRQs on. 259 */ 260 if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS)) 261 syscall_slow_exit_work(regs, cached_flags); 262 263 local_irq_disable(); 264 prepare_exit_to_usermode(regs); 265 } 266 267 #ifdef CONFIG_X86_64 268 __visible void do_syscall_64(struct pt_regs *regs) 269 { 270 struct thread_info *ti = current_thread_info(); 271 unsigned long nr = regs->orig_ax; 272 273 enter_from_user_mode(); 274 local_irq_enable(); 275 276 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) 277 nr = syscall_trace_enter(regs); 278 279 /* 280 * NB: Native and x32 syscalls are dispatched from the same 281 * table. The only functional difference is the x32 bit in 282 * regs->orig_ax, which changes the behavior of some syscalls. 283 */ 284 if (likely((nr & __SYSCALL_MASK) < NR_syscalls)) { 285 regs->ax = sys_call_table[nr & __SYSCALL_MASK]( 286 regs->di, regs->si, regs->dx, 287 regs->r10, regs->r8, regs->r9); 288 } 289 290 syscall_return_slowpath(regs); 291 } 292 #endif 293 294 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION) 295 /* 296 * Does a 32-bit syscall. Called with IRQs on in CONTEXT_KERNEL. Does 297 * all entry and exit work and returns with IRQs off. This function is 298 * extremely hot in workloads that use it, and it's usually called from 299 * do_fast_syscall_32, so forcibly inline it to improve performance. 300 */ 301 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs) 302 { 303 struct thread_info *ti = current_thread_info(); 304 unsigned int nr = (unsigned int)regs->orig_ax; 305 306 #ifdef CONFIG_IA32_EMULATION 307 current->thread.status |= TS_COMPAT; 308 #endif 309 310 if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) { 311 /* 312 * Subtlety here: if ptrace pokes something larger than 313 * 2^32-1 into orig_ax, this truncates it. This may or 314 * may not be necessary, but it matches the old asm 315 * behavior. 316 */ 317 nr = syscall_trace_enter(regs); 318 } 319 320 if (likely(nr < IA32_NR_syscalls)) { 321 /* 322 * It's possible that a 32-bit syscall implementation 323 * takes a 64-bit parameter but nonetheless assumes that 324 * the high bits are zero. Make sure we zero-extend all 325 * of the args. 326 */ 327 regs->ax = ia32_sys_call_table[nr]( 328 (unsigned int)regs->bx, (unsigned int)regs->cx, 329 (unsigned int)regs->dx, (unsigned int)regs->si, 330 (unsigned int)regs->di, (unsigned int)regs->bp); 331 } 332 333 syscall_return_slowpath(regs); 334 } 335 336 /* Handles int $0x80 */ 337 __visible void do_int80_syscall_32(struct pt_regs *regs) 338 { 339 enter_from_user_mode(); 340 local_irq_enable(); 341 do_syscall_32_irqs_on(regs); 342 } 343 344 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */ 345 __visible long do_fast_syscall_32(struct pt_regs *regs) 346 { 347 /* 348 * Called using the internal vDSO SYSENTER/SYSCALL32 calling 349 * convention. Adjust regs so it looks like we entered using int80. 350 */ 351 352 unsigned long landing_pad = (unsigned long)current->mm->context.vdso + 353 vdso_image_32.sym_int80_landing_pad; 354 355 /* 356 * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward 357 * so that 'regs->ip -= 2' lands back on an int $0x80 instruction. 358 * Fix it up. 359 */ 360 regs->ip = landing_pad; 361 362 enter_from_user_mode(); 363 364 local_irq_enable(); 365 366 /* Fetch EBP from where the vDSO stashed it. */ 367 if ( 368 #ifdef CONFIG_X86_64 369 /* 370 * Micro-optimization: the pointer we're following is explicitly 371 * 32 bits, so it can't be out of range. 372 */ 373 __get_user(*(u32 *)®s->bp, 374 (u32 __user __force *)(unsigned long)(u32)regs->sp) 375 #else 376 get_user(*(u32 *)®s->bp, 377 (u32 __user __force *)(unsigned long)(u32)regs->sp) 378 #endif 379 ) { 380 381 /* User code screwed up. */ 382 local_irq_disable(); 383 regs->ax = -EFAULT; 384 prepare_exit_to_usermode(regs); 385 return 0; /* Keep it simple: use IRET. */ 386 } 387 388 /* Now this is just like a normal syscall. */ 389 do_syscall_32_irqs_on(regs); 390 391 #ifdef CONFIG_X86_64 392 /* 393 * Opportunistic SYSRETL: if possible, try to return using SYSRETL. 394 * SYSRETL is available on all 64-bit CPUs, so we don't need to 395 * bother with SYSEXIT. 396 * 397 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP, 398 * because the ECX fixup above will ensure that this is essentially 399 * never the case. 400 */ 401 return regs->cs == __USER32_CS && regs->ss == __USER_DS && 402 regs->ip == landing_pad && 403 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0; 404 #else 405 /* 406 * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT. 407 * 408 * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP, 409 * because the ECX fixup above will ensure that this is essentially 410 * never the case. 411 * 412 * We don't allow syscalls at all from VM86 mode, but we still 413 * need to check VM, because we might be returning from sys_vm86. 414 */ 415 return static_cpu_has(X86_FEATURE_SEP) && 416 regs->cs == __USER_CS && regs->ss == __USER_DS && 417 regs->ip == landing_pad && 418 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0; 419 #endif 420 } 421 #endif 422