개념 설명 전체 · v6.6 / arch/arm64/kernel/syscall.c

    1 // SPDX-License-Identifier: GPL-2.0
    2 
    3 #include <linux/compiler.h>
    4 #include <linux/context_tracking.h>
    5 #include <linux/errno.h>
    6 #include <linux/nospec.h>
    7 #include <linux/ptrace.h>
    8 #include <linux/randomize_kstack.h>
    9 #include <linux/syscalls.h>
   10 
   11 #include <asm/debug-monitors.h>
   12 #include <asm/exception.h>
   13 #include <asm/fpsimd.h>
   14 #include <asm/syscall.h>
   15 #include <asm/thread_info.h>
   16 #include <asm/unistd.h>
   17 
   18 long compat_arm_syscall(struct pt_regs *regs, int scno);
   19 long sys_ni_syscall(void);
   20 
   21 static long do_ni_syscall(struct pt_regs *regs, int scno)
   22 {
   23 #ifdef CONFIG_COMPAT
   24 	long ret;
   25 	if (is_compat_task()) {
   26 		ret = compat_arm_syscall(regs, scno);
   27 		if (ret != -ENOSYS)
   28 			return ret;
   29 	}
   30 #endif
   31 
   32 	return sys_ni_syscall();
   33 }
   34 
   35 static long __invoke_syscall(struct pt_regs *regs, syscall_fn_t syscall_fn)
   36 {
   37 	return syscall_fn(regs);
   38 }
   39 
   40 static void invoke_syscall(struct pt_regs *regs, unsigned int scno,
   41 			   unsigned int sc_nr,
   42 			   const syscall_fn_t syscall_table[])
   43 {
   44 	long ret;
   45 
   46 	add_random_kstack_offset();
   47 
   48 	if (scno < sc_nr) {
   49 		syscall_fn_t syscall_fn;
   50 		syscall_fn = syscall_table[array_index_nospec(scno, sc_nr)];
   51 		ret = __invoke_syscall(regs, syscall_fn);
   52 	} else {
   53 		ret = do_ni_syscall(regs, scno);
   54 	}
   55 
   56 	syscall_set_return_value(current, regs, 0, ret);
   57 
   58 	/*
   59 	 * Ultimately, this value will get limited by KSTACK_OFFSET_MAX(),
   60 	 * but not enough for arm64 stack utilization comfort. To keep
   61 	 * reasonable stack head room, reduce the maximum offset to 9 bits.
   62 	 *
   63 	 * The actual entropy will be further reduced by the compiler when
   64 	 * applying stack alignment constraints: the AAPCS mandates a
   65 	 * 16-byte (i.e. 4-bit) aligned SP at function boundaries.
   66 	 *
   67 	 * The resulting 5 bits of entropy is seen in SP[8:4].
   68 	 */
   69 	choose_random_kstack_offset(get_random_u16() & 0x1FF);
   70 }
   71 
   72 static inline bool has_syscall_work(unsigned long flags)
   73 {
   74 	return unlikely(flags & _TIF_SYSCALL_WORK);
   75 }
   76 
   77 static void el0_svc_common(struct pt_regs *regs, int scno, int sc_nr,
   78 			   const syscall_fn_t syscall_table[])
   79 {
   80 	unsigned long flags = read_thread_flags();
   81 
   82 	regs->orig_x0 = regs->regs[0];
   83 	regs->syscallno = scno;
   84 
   85 	/*
   86 	 * BTI note:
   87 	 * The architecture does not guarantee that SPSR.BTYPE is zero
   88 	 * on taking an SVC, so we could return to userspace with a
   89 	 * non-zero BTYPE after the syscall.
   90 	 *
   91 	 * This shouldn't matter except when userspace is explicitly
   92 	 * doing something stupid, such as setting PROT_BTI on a page
   93 	 * that lacks conforming BTI/PACIxSP instructions, falling
   94 	 * through from one executable page to another with differing
   95 	 * PROT_BTI, or messing with BTYPE via ptrace: in such cases,
   96 	 * userspace should not be surprised if a SIGILL occurs on
   97 	 * syscall return.
   98 	 *
   99 	 * So, don't touch regs->pstate & PSR_BTYPE_MASK here.
  100 	 * (Similarly for HVC and SMC elsewhere.)
  101 	 */
  102 
  103 	if (flags & _TIF_MTE_ASYNC_FAULT) {
  104 		/*
  105 		 * Process the asynchronous tag check fault before the actual
  106 		 * syscall. do_notify_resume() will send a signal to userspace
  107 		 * before the syscall is restarted.
  108 		 */
  109 		syscall_set_return_value(current, regs, -ERESTARTNOINTR, 0);
  110 		return;
  111 	}
  112 
  113 	if (has_syscall_work(flags)) {
  114 		/*
  115 		 * The de-facto standard way to skip a system call using ptrace
  116 		 * is to set the system call to -1 (NO_SYSCALL) and set x0 to a
  117 		 * suitable error code for consumption by userspace. However,
  118 		 * this cannot be distinguished from a user-issued syscall(-1)
  119 		 * and so we must set x0 to -ENOSYS here in case the tracer doesn't
  120 		 * issue the skip and we fall into trace_exit with x0 preserved.
  121 		 *
  122 		 * This is slightly odd because it also means that if a tracer
  123 		 * sets the system call number to -1 but does not initialise x0,
  124 		 * then x0 will be preserved for all system calls apart from a
  125 		 * user-issued syscall(-1). However, requesting a skip and not
  126 		 * setting the return value is unlikely to do anything sensible
  127 		 * anyway.
  128 		 */
  129 		if (scno == NO_SYSCALL)
  130 			syscall_set_return_value(current, regs, -ENOSYS, 0);
  131 		scno = syscall_trace_enter(regs);
  132 		if (scno == NO_SYSCALL)
  133 			goto trace_exit;
  134 	}
  135 
  136 	invoke_syscall(regs, scno, sc_nr, syscall_table);
  137 
  138 	/*
  139 	 * The tracing status may have changed under our feet, so we have to
  140 	 * check again. However, if we were tracing entry, then we always trace
  141 	 * exit regardless, as the old entry assembly did.
  142 	 */
  143 	if (!has_syscall_work(flags) && !IS_ENABLED(CONFIG_DEBUG_RSEQ)) {
  144 		flags = read_thread_flags();
  145 		if (!has_syscall_work(flags) && !(flags & _TIF_SINGLESTEP))
  146 			return;
  147 	}
  148 
  149 trace_exit:
  150 	syscall_trace_exit(regs);
  151 }
  152 
  153 void do_el0_svc(struct pt_regs *regs)
  154 {
  155 	el0_svc_common(regs, regs->regs[8], __NR_syscalls, sys_call_table);
  156 }
  157 
  158 #ifdef CONFIG_COMPAT
  159 void do_el0_svc_compat(struct pt_regs *regs)
  160 {
  161 	el0_svc_common(regs, regs->regs[7], __NR_compat_syscalls,
  162 		       compat_sys_call_table);
  163 }
  164 #endif