syscall.c 4.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160
  1. // SPDX-License-Identifier: GPL-2.0
  2. #include <linux/compiler.h>
  3. #include <linux/context_tracking.h>
  4. #include <linux/errno.h>
  5. #include <linux/nospec.h>
  6. #include <linux/ptrace.h>
  7. #include <linux/randomize_kstack.h>
  8. #include <linux/syscalls.h>
  9. #include <asm/debug-monitors.h>
  10. #include <asm/exception.h>
  11. #include <asm/fpsimd.h>
  12. #include <asm/syscall.h>
  13. #include <asm/thread_info.h>
  14. #include <asm/unistd.h>
  15. #include <asm/unistd_compat_32.h>
  16. long compat_arm_syscall(struct pt_regs *regs, int scno);
  17. long sys_ni_syscall(void);
  18. static long do_ni_syscall(struct pt_regs *regs, int scno)
  19. {
  20. if (is_compat_task()) {
  21. long ret = compat_arm_syscall(regs, scno);
  22. if (ret != -ENOSYS)
  23. return ret;
  24. }
  25. return sys_ni_syscall();
  26. }
  27. static long __invoke_syscall(struct pt_regs *regs, syscall_fn_t syscall_fn)
  28. {
  29. return syscall_fn(regs);
  30. }
  31. static void invoke_syscall(struct pt_regs *regs, unsigned int scno,
  32. unsigned int sc_nr,
  33. const syscall_fn_t syscall_table[])
  34. {
  35. long ret;
  36. add_random_kstack_offset();
  37. if (scno < sc_nr) {
  38. syscall_fn_t syscall_fn;
  39. syscall_fn = syscall_table[array_index_nospec(scno, sc_nr)];
  40. ret = __invoke_syscall(regs, syscall_fn);
  41. } else {
  42. ret = do_ni_syscall(regs, scno);
  43. }
  44. syscall_set_return_value(current, regs, 0, ret);
  45. /*
  46. * This value will get limited by KSTACK_OFFSET_MAX(), which is 10
  47. * bits. The actual entropy will be further reduced by the compiler
  48. * when applying stack alignment constraints: the AAPCS mandates a
  49. * 16-byte aligned SP at function boundaries, which will remove the
  50. * 4 low bits from any entropy chosen here.
  51. *
  52. * The resulting 6 bits of entropy is seen in SP[9:4].
  53. */
  54. choose_random_kstack_offset(get_random_u16());
  55. }
  56. static inline bool has_syscall_work(unsigned long flags)
  57. {
  58. return unlikely(flags & _TIF_SYSCALL_WORK);
  59. }
  60. static void el0_svc_common(struct pt_regs *regs, int scno, int sc_nr,
  61. const syscall_fn_t syscall_table[])
  62. {
  63. unsigned long flags = read_thread_flags();
  64. regs->orig_x0 = regs->regs[0];
  65. regs->syscallno = scno;
  66. /*
  67. * BTI note:
  68. * The architecture does not guarantee that SPSR.BTYPE is zero
  69. * on taking an SVC, so we could return to userspace with a
  70. * non-zero BTYPE after the syscall.
  71. *
  72. * This shouldn't matter except when userspace is explicitly
  73. * doing something stupid, such as setting PROT_BTI on a page
  74. * that lacks conforming BTI/PACIxSP instructions, falling
  75. * through from one executable page to another with differing
  76. * PROT_BTI, or messing with BTYPE via ptrace: in such cases,
  77. * userspace should not be surprised if a SIGILL occurs on
  78. * syscall return.
  79. *
  80. * So, don't touch regs->pstate & PSR_BTYPE_MASK here.
  81. * (Similarly for HVC and SMC elsewhere.)
  82. */
  83. if (flags & _TIF_MTE_ASYNC_FAULT) {
  84. /*
  85. * Process the asynchronous tag check fault before the actual
  86. * syscall. do_notify_resume() will send a signal to userspace
  87. * before the syscall is restarted.
  88. */
  89. syscall_set_return_value(current, regs, -ERESTARTNOINTR, 0);
  90. return;
  91. }
  92. if (has_syscall_work(flags)) {
  93. /*
  94. * The de-facto standard way to skip a system call using ptrace
  95. * is to set the system call to -1 (NO_SYSCALL) and set x0 to a
  96. * suitable error code for consumption by userspace. However,
  97. * this cannot be distinguished from a user-issued syscall(-1)
  98. * and so we must set x0 to -ENOSYS here in case the tracer doesn't
  99. * issue the skip and we fall into trace_exit with x0 preserved.
  100. *
  101. * This is slightly odd because it also means that if a tracer
  102. * sets the system call number to -1 but does not initialise x0,
  103. * then x0 will be preserved for all system calls apart from a
  104. * user-issued syscall(-1). However, requesting a skip and not
  105. * setting the return value is unlikely to do anything sensible
  106. * anyway.
  107. */
  108. if (scno == NO_SYSCALL)
  109. syscall_set_return_value(current, regs, -ENOSYS, 0);
  110. scno = syscall_trace_enter(regs);
  111. if (scno == NO_SYSCALL)
  112. goto trace_exit;
  113. }
  114. invoke_syscall(regs, scno, sc_nr, syscall_table);
  115. /*
  116. * The tracing status may have changed under our feet, so we have to
  117. * check again. However, if we were tracing entry, then we always trace
  118. * exit regardless, as the old entry assembly did.
  119. */
  120. if (!has_syscall_work(flags) && !IS_ENABLED(CONFIG_DEBUG_RSEQ)) {
  121. flags = read_thread_flags();
  122. if (!has_syscall_work(flags) && !(flags & _TIF_SINGLESTEP))
  123. return;
  124. }
  125. trace_exit:
  126. syscall_trace_exit(regs);
  127. }
  128. void do_el0_svc(struct pt_regs *regs)
  129. {
  130. el0_svc_common(regs, regs->regs[8], __NR_syscalls, sys_call_table);
  131. }
  132. #ifdef CONFIG_COMPAT
  133. void do_el0_svc_compat(struct pt_regs *regs)
  134. {
  135. el0_svc_common(regs, regs->regs[7], __NR_compat32_syscalls,
  136. compat_sys_call_table);
  137. }
  138. #endif