arch/x86/entry/common.c

   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/nospec.h>
  25 #include <linux/uprobes.h>
  26 #include <linux/livepatch.h>
  27 #include <linux/syscalls.h>
  28
  29 #include <asm/desc.h>
  30 #include <asm/traps.h>
  31 #include <asm/vdso.h>
  32 #include <linux/uaccess.h>
  33 #include <asm/cpufeature.h>
  34 #include <asm/nospec-branch.h>
  35
  36 #define CREATE_TRACE_POINTS
  37 #include <trace/events/syscalls.h>
  38
  39 #ifdef CONFIG_CONTEXT_TRACKING
  40 /* Called on entry from user mode with IRQs off. */
  41 __visible inline void enter_from_user_mode(void)
  42 {
  43         CT_WARN_ON(ct_state() != CONTEXT_USER);
  44         user_exit_irqoff();
  45 }
  46 #else
  47 static inline void enter_from_user_mode(void) {}
  48 #endif
  49
  50 static void do_audit_syscall_entry(struct pt_regs *regs, u32 arch)
  51 {
  52 #ifdef CONFIG_X86_64
  53         if (arch == AUDIT_ARCH_X86_64) {
  54                 audit_syscall_entry(regs->orig_ax, regs->di,
  55                                     regs->si, regs->dx, regs->r10);
  56         } else
  57 #endif
  58         {
  59                 audit_syscall_entry(regs->orig_ax, regs->bx,
  60                                     regs->cx, regs->dx, regs->si);
  61         }
  62 }
  63
  64 /*
  65  * Returns the syscall nr to run (which should match regs->orig_ax) or -1
  66  * to skip the syscall.
  67  */
  68 static long syscall_trace_enter(struct pt_regs *regs)
  69 {
  70         u32 arch = in_ia32_syscall() ? AUDIT_ARCH_I386 : AUDIT_ARCH_X86_64;
  71
  72         struct thread_info *ti = current_thread_info();
  73         unsigned long ret = 0;
  74         bool emulated = false;
  75         u32 work;
  76
  77         if (IS_ENABLED(CONFIG_DEBUG_ENTRY))
  78                 BUG_ON(regs != task_pt_regs(current));
  79
  80         work = ACCESS_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY;
  81
  82         if (unlikely(work & _TIF_SYSCALL_EMU))
  83                 emulated = true;
  84
  85         if ((emulated || (work & _TIF_SYSCALL_TRACE)) &&
  86             tracehook_report_syscall_entry(regs))
  87                 return -1L;
  88
  89         if (emulated)
  90                 return -1L;
  91
  92 #ifdef CONFIG_SECCOMP
  93         /*
  94          * Do seccomp after ptrace, to catch any tracer changes.
  95          */
  96         if (work & _TIF_SECCOMP) {
  97                 struct seccomp_data sd;
  98
  99                 sd.arch = arch;
 100                 sd.nr = regs->orig_ax;
 101                 sd.instruction_pointer = regs->ip;
 102 #ifdef CONFIG_X86_64
 103                 if (arch == AUDIT_ARCH_X86_64) {
 104                         sd.args[0] = regs->di;
 105                         sd.args[1] = regs->si;
 106                         sd.args[2] = regs->dx;
 107                         sd.args[3] = regs->r10;
 108                         sd.args[4] = regs->r8;
 109                         sd.args[5] = regs->r9;
 110                 } else
 111 #endif
 112                 {
 113                         sd.args[0] = regs->bx;
 114                         sd.args[1] = regs->cx;
 115                         sd.args[2] = regs->dx;
 116                         sd.args[3] = regs->si;
 117                         sd.args[4] = regs->di;
 118                         sd.args[5] = regs->bp;
 119                 }
 120
 121                 ret = __secure_computing(&sd);
 122                 if (ret == -1)
 123                         return ret;
 124         }
 125 #endif
 126
 127         if (unlikely(test_thread_flag(TIF_SYSCALL_TRACEPOINT)))
 128                 trace_sys_enter(regs, regs->orig_ax);
 129
 130         do_audit_syscall_entry(regs, arch);
 131
 132         return ret ?: regs->orig_ax;
 133 }
 134
 135 #define EXIT_TO_USERMODE_LOOP_FLAGS                             \
 136         (_TIF_SIGPENDING | _TIF_NOTIFY_RESUME | _TIF_UPROBE |   \
 137          _TIF_NEED_RESCHED | _TIF_USER_RETURN_NOTIFY | _TIF_PATCH_PENDING)
 138
 139 static void exit_to_usermode_loop(struct pt_regs *regs, u32 cached_flags)
 140 {
 141         /*
 142          * In order to return to user mode, we need to have IRQs off with
 143          * none of EXIT_TO_USERMODE_LOOP_FLAGS set.  Several of these flags
 144          * can be set at any time on preemptable kernels if we have IRQs on,
 145          * so we need to loop.  Disabling preemption wouldn't help: doing the
 146          * work to clear some of the flags can sleep.
 147          */
 148         while (true) {
 149                 /* We have work to do. */
 150                 local_irq_enable();
 151
 152                 if (cached_flags & _TIF_NEED_RESCHED)
 153                         schedule();
 154
 155                 if (cached_flags & _TIF_UPROBE)
 156                         uprobe_notify_resume(regs);
 157
 158                 /* deal with pending signal delivery */
 159                 if (cached_flags & _TIF_SIGPENDING)
 160                         do_signal(regs);
 161
 162                 if (cached_flags & _TIF_NOTIFY_RESUME) {
 163                         clear_thread_flag(TIF_NOTIFY_RESUME);
 164                         tracehook_notify_resume(regs);
 165                 }
 166
 167                 if (cached_flags & _TIF_USER_RETURN_NOTIFY)
 168                         fire_user_return_notifiers();
 169
 170                 if (cached_flags & _TIF_PATCH_PENDING)
 171                         klp_update_patch_state(current);
 172
 173                 /* Disable IRQs and retry */
 174                 local_irq_disable();
 175
 176                 cached_flags = READ_ONCE(current_thread_info()->flags);
 177
 178                 if (!(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
 179                         break;
 180         }
 181 }
 182
 183 /* Called with IRQs disabled. */
 184 __visible inline void prepare_exit_to_usermode(struct pt_regs *regs)
 185 {
 186         struct thread_info *ti = current_thread_info();
 187         u32 cached_flags;
 188
 189         addr_limit_user_check();
 190
 191         if (IS_ENABLED(CONFIG_PROVE_LOCKING) && WARN_ON(!irqs_disabled()))
 192                 local_irq_disable();
 193
 194         lockdep_sys_exit();
 195
 196         cached_flags = READ_ONCE(ti->flags);
 197
 198         if (unlikely(cached_flags & EXIT_TO_USERMODE_LOOP_FLAGS))
 199                 exit_to_usermode_loop(regs, cached_flags);
 200
 201 #ifdef CONFIG_COMPAT
 202         /*
 203          * Compat syscalls set TS_COMPAT.  Make sure we clear it before
 204          * returning to user mode.  We need to clear it *after* signal
 205          * handling, because syscall restart has a fixup for compat
 206          * syscalls.  The fixup is exercised by the ptrace_syscall_32
 207          * selftest.
 208          *
 209          * We also need to clear TS_REGS_POKED_I386: the 32-bit tracer
 210          * special case only applies after poking regs and before the
 211          * very next return to user mode.
 212          */
 213         ti->status &= ~(TS_COMPAT|TS_I386_REGS_POKED);
 214 #endif
 215
 216         user_enter_irqoff();
 217
 218         mds_user_clear_cpu_buffers();
 219 }
 220
 221 #define SYSCALL_EXIT_WORK_FLAGS                         \
 222         (_TIF_SYSCALL_TRACE | _TIF_SYSCALL_AUDIT |      \
 223          _TIF_SINGLESTEP | _TIF_SYSCALL_TRACEPOINT)
 224
 225 static void syscall_slow_exit_work(struct pt_regs *regs, u32 cached_flags)
 226 {
 227         bool step;
 228
 229         audit_syscall_exit(regs);
 230
 231         if (cached_flags & _TIF_SYSCALL_TRACEPOINT)
 232                 trace_sys_exit(regs, regs->ax);
 233
 234         /*
 235          * If TIF_SYSCALL_EMU is set, we only get here because of
 236          * TIF_SINGLESTEP (i.e. this is PTRACE_SYSEMU_SINGLESTEP).
 237          * We already reported this syscall instruction in
 238          * syscall_trace_enter().
 239          */
 240         step = unlikely(
 241                 (cached_flags & (_TIF_SINGLESTEP | _TIF_SYSCALL_EMU))
 242                 == _TIF_SINGLESTEP);
 243         if (step || cached_flags & _TIF_SYSCALL_TRACE)
 244                 tracehook_report_syscall_exit(regs, step);
 245 }
 246
 247 /*
 248  * Called with IRQs on and fully valid regs.  Returns with IRQs off in a
 249  * state such that we can immediately switch to user mode.
 250  */
 251 __visible inline void syscall_return_slowpath(struct pt_regs *regs)
 252 {
 253         struct thread_info *ti = current_thread_info();
 254         u32 cached_flags = READ_ONCE(ti->flags);
 255
 256         CT_WARN_ON(ct_state() != CONTEXT_KERNEL);
 257
 258         if (IS_ENABLED(CONFIG_PROVE_LOCKING) &&
 259             WARN(irqs_disabled(), "syscall %ld left IRQs disabled", regs->orig_ax))
 260                 local_irq_enable();
 261
 262         /*
 263          * First do one-time work.  If these work items are enabled, we
 264          * want to run them exactly once per syscall exit with IRQs on.
 265          */
 266         if (unlikely(cached_flags & SYSCALL_EXIT_WORK_FLAGS))
 267                 syscall_slow_exit_work(regs, cached_flags);
 268
 269         local_irq_disable();
 270         prepare_exit_to_usermode(regs);
 271 }
 272
 273 #ifdef CONFIG_X86_64
 274 __visible void do_syscall_64(struct pt_regs *regs)
 275 {
 276         struct thread_info *ti = current_thread_info();
 277         unsigned long nr = regs->orig_ax;
 278
 279         enter_from_user_mode();
 280         local_irq_enable();
 281
 282         if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY)
 283                 nr = syscall_trace_enter(regs);
 284
 285         /*
 286          * NB: Native and x32 syscalls are dispatched from the same
 287          * table.  The only functional difference is the x32 bit in
 288          * regs->orig_ax, which changes the behavior of some syscalls.
 289          */
 290         if (likely((nr & __SYSCALL_MASK) < NR_syscalls)) {
 291                 nr = array_index_nospec(nr & __SYSCALL_MASK, NR_syscalls);
 292                 regs->ax = sys_call_table[nr](
 293                         regs->di, regs->si, regs->dx,
 294                         regs->r10, regs->r8, regs->r9);
 295         }
 296
 297         syscall_return_slowpath(regs);
 298 }
 299 #endif
 300
 301 #if defined(CONFIG_X86_32) || defined(CONFIG_IA32_EMULATION)
 302 /*
 303  * Does a 32-bit syscall.  Called with IRQs on in CONTEXT_KERNEL.  Does
 304  * all entry and exit work and returns with IRQs off.  This function is
 305  * extremely hot in workloads that use it, and it's usually called from
 306  * do_fast_syscall_32, so forcibly inline it to improve performance.
 307  */
 308 static __always_inline void do_syscall_32_irqs_on(struct pt_regs *regs)
 309 {
 310         struct thread_info *ti = current_thread_info();
 311         unsigned int nr = (unsigned int)regs->orig_ax;
 312
 313 #ifdef CONFIG_IA32_EMULATION
 314         ti->status |= TS_COMPAT;
 315 #endif
 316
 317         if (READ_ONCE(ti->flags) & _TIF_WORK_SYSCALL_ENTRY) {
 318                 /*
 319                  * Subtlety here: if ptrace pokes something larger than
 320                  * 2^32-1 into orig_ax, this truncates it.  This may or
 321                  * may not be necessary, but it matches the old asm
 322                  * behavior.
 323                  */
 324                 nr = syscall_trace_enter(regs);
 325         }
 326
 327         if (likely(nr < IA32_NR_syscalls)) {
 328                 nr = array_index_nospec(nr, IA32_NR_syscalls);
 329                 /*
 330                  * It's possible that a 32-bit syscall implementation
 331                  * takes a 64-bit parameter but nonetheless assumes that
 332                  * the high bits are zero.  Make sure we zero-extend all
 333                  * of the args.
 334                  */
 335                 regs->ax = ia32_sys_call_table[nr](
 336                         (unsigned int)regs->bx, (unsigned int)regs->cx,
 337                         (unsigned int)regs->dx, (unsigned int)regs->si,
 338                         (unsigned int)regs->di, (unsigned int)regs->bp);
 339         }
 340
 341         syscall_return_slowpath(regs);
 342 }
 343
 344 /* Handles int $0x80 */
 345 __visible void do_int80_syscall_32(struct pt_regs *regs)
 346 {
 347         enter_from_user_mode();
 348         local_irq_enable();
 349         do_syscall_32_irqs_on(regs);
 350 }
 351
 352 /* Returns 0 to return using IRET or 1 to return using SYSEXIT/SYSRETL. */
 353 __visible long do_fast_syscall_32(struct pt_regs *regs)
 354 {
 355         /*
 356          * Called using the internal vDSO SYSENTER/SYSCALL32 calling
 357          * convention.  Adjust regs so it looks like we entered using int80.
 358          */
 359
 360         unsigned long landing_pad = (unsigned long)current->mm->context.vdso +
 361                 vdso_image_32.sym_int80_landing_pad;
 362
 363         /*
 364          * SYSENTER loses EIP, and even SYSCALL32 needs us to skip forward
 365          * so that 'regs->ip -= 2' lands back on an int $0x80 instruction.
 366          * Fix it up.
 367          */
 368         regs->ip = landing_pad;
 369
 370         enter_from_user_mode();
 371
 372         local_irq_enable();
 373
 374         /* Fetch EBP from where the vDSO stashed it. */
 375         if (
 376 #ifdef CONFIG_X86_64
 377                 /*
 378                  * Micro-optimization: the pointer we're following is explicitly
 379                  * 32 bits, so it can't be out of range.
 380                  */
 381                 __get_user(*(u32 *)&regs->bp,
 382                             (u32 __user __force *)(unsigned long)(u32)regs->sp)
 383 #else
 384                 get_user(*(u32 *)&regs->bp,
 385                          (u32 __user __force *)(unsigned long)(u32)regs->sp)
 386 #endif
 387                 ) {
 388
 389                 /* User code screwed up. */
 390                 local_irq_disable();
 391                 regs->ax = -EFAULT;
 392                 prepare_exit_to_usermode(regs);
 393                 return 0;       /* Keep it simple: use IRET. */
 394         }
 395
 396         /* Now this is just like a normal syscall. */
 397         do_syscall_32_irqs_on(regs);
 398
 399 #ifdef CONFIG_X86_64
 400         /*
 401          * Opportunistic SYSRETL: if possible, try to return using SYSRETL.
 402          * SYSRETL is available on all 64-bit CPUs, so we don't need to
 403          * bother with SYSEXIT.
 404          *
 405          * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
 406          * because the ECX fixup above will ensure that this is essentially
 407          * never the case.
 408          */
 409         return regs->cs == __USER32_CS && regs->ss == __USER_DS &&
 410                 regs->ip == landing_pad &&
 411                 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF)) == 0;
 412 #else
 413         /*
 414          * Opportunistic SYSEXIT: if possible, try to return using SYSEXIT.
 415          *
 416          * Unlike 64-bit opportunistic SYSRET, we can't check that CX == IP,
 417          * because the ECX fixup above will ensure that this is essentially
 418          * never the case.
 419          *
 420          * We don't allow syscalls at all from VM86 mode, but we still
 421          * need to check VM, because we might be returning from sys_vm86.
 422          */
 423         return static_cpu_has(X86_FEATURE_SEP) &&
 424                 regs->cs == __USER_CS && regs->ss == __USER_DS &&
 425                 regs->ip == landing_pad &&
 426                 (regs->flags & (X86_EFLAGS_RF | X86_EFLAGS_TF | X86_EFLAGS_VM)) == 0;
 427 #endif
 428 }
 429 #endif