Support the faccessat2 syscall

diff --git a/tapset/linux/aux_syscalls.stp b/tapset/linux/aux_syscalls.stp
index 4a4bac4dfa992db4b7e93efa4dac8ebccd7c6288..af6b0aefd8d2b4a6157759151486672469e84fcf 100644 (file)
--- a/tapset/linux/aux_syscalls.stp
+++ b/tapset/linux/aux_syscalls.stp
@@ -1422,6 +1422,22 @@ function _access_mode_str:string(mode:long)
                           MAXSTRINGLEN);
 %}
 
+%{
+static const _stp_val_array _stp_faccessat2_flags_list[] = {
+       {0x100, "AT_SYMLINK_NOFOLLOW"},
+       {0x200, "AT_EACCESS"},
+       {0, NULL}
+};
+%}
+
+function _faccessat2_flags_str:string(flags:long)
+%{ /* pure */ /* unprivileged */
+       unsigned int flags = (unsigned int)STAP_ARG_flags;
+       _stp_lookup_or_str(_stp_faccessat2_flags_list, flags, STAP_RETVALUE,
+                          MAXSTRINGLEN);
+%}
+
+
 /* `man 2 open` for more information */
 @__private30 function _sys_open_mode_str(f)
 {

diff --git a/tapset/linux/sysc_faccessat2.stp b/tapset/linux/sysc_faccessat2.stp
new file mode 100644 (file)
index 0000000..c3c7121
--- /dev/null
+++ b/tapset/linux/sysc_faccessat2.stp
@@ -0,0 +1,143 @@
+# faccessat2 __________________________________________________
+# new function faccessat2() was added to Linux in version 5.8.
+# with commit c8ffd8bcdd28296a198f237cc595148a8d4adfbe
+#
+# from man faccessat2:  the raw Linux faccessat() system call does not have a
+# flags argument.  To allow  for  a  proper implementation,  Linux  5.8 added
+# the faccessat2() system call, which supports the flags argument and allows
+# a correct implementation of the faccessat() wrapper function (in glibc)
+#
+# long sys_faccessat2(int dfd, const char __user *filename, int mode, int flags)
+
+@define _SYSCALL_FACCESSAT2_NAME
+%(
+       name = "faccessat2"
+%)
+
+@define _SYSCALL_FACCESSAT2_ARGSTR
+%(
+       argstr = sprintf("%s, %s, %s, %s", dirfd_str, pathname, mode_str, flags_str)
+%)
+
+@define _SYSCALL_FACCESSAT2_REGARGS
+%(
+       dirfd = int_arg(1)
+       dirfd_str = _dfd_str(dirfd)
+       pathname = user_string_quoted(pointer_arg(2))
+       pathname_unquoted = user_string_nofault(pointer_arg(2))
+       mode = int_arg(3)
+       mode_str = _access_mode_str(mode)
+       flags = int_arg(4)
+       flags_str = _faccessat2_flags_str(flags)
+%)
+
+@define _SYSCALL_FACCESSAT2_REGARGS_STORE
+%(
+        if (@probewrite(dirfd))
+          set_int_arg(1, dirfd)
+
+       if (@probewrite(pathname_unquoted))
+         set_user_string_arg(pointer_arg(2), pathname_unquoted)
+
+       if (@probewrite(mode))
+         set_int_arg(3, mode)
+
+        if (@probewrite(flags))
+          set_int_arg(4, flags)
+%)
+
+probe syscall.faccessat2 = dw_syscall.faccessat2 !, nd_syscall.faccessat2 ? {}
+probe syscall.faccessat2.return = dw_syscall.faccessat2.return !, nd_syscall.faccessat2.return ? {}
+
+# dw_faccessat2 _____________________________________________________
+
+probe dw_syscall.faccessat2 = kernel.function("sys_faccessat2").call ?
+{
+       @__syscall_compat_gate(@const("__NR_faccessat2"),
+                              @const("__NR_compat_faccessat2"))
+       @_SYSCALL_FACCESSAT2_NAME
+       dirfd = __int32($dfd)
+       dirfd_str = _dfd_str(__int32($dfd))
+       pathname = user_string_quoted($filename)
+       mode = __int32($mode)
+       mode_str = _access_mode_str(__int32($mode))
+       flags = __int32($flags)
+       flags_str = _faccessat2_flags_str(__int32($flags))
+       @_SYSCALL_FACCESSAT2_ARGSTR
+}
+probe dw_syscall.faccessat2.return = kernel.function("sys_faccessat2").return ?
+{
+       @__syscall_compat_gate(@const("__NR_faccessat2"),
+                              @const("__NR_compat_faccessat2"))
+       @_SYSCALL_FACCESSAT2_NAME
+       @SYSC_RETVALSTR($return)
+}
+
+# nd_faccessat2 _____________________________________________________
+
+probe nd_syscall.faccessat2 = nd1_syscall.faccessat2!, nd2_syscall.faccessat2!, tp_syscall.faccessat2
+  { }
+
+probe nd1_syscall.faccessat2 = kprobe.function("sys_faccessat2") ?
+{
+       @__syscall_compat_gate(@const("__NR_faccessat2"),
+                              @const("__NR_compat_faccessat2"))
+       @_SYSCALL_FACCESSAT2_NAME
+       asmlinkage()
+       @_SYSCALL_FACCESSAT2_REGARGS
+       @_SYSCALL_FACCESSAT2_ARGSTR
+}
+
+/* kernel 4.17+ */
+probe nd2_syscall.faccessat2 = kprobe.function(@arch_syscall_prefix "sys_faccessat2") ?
+{
+       __set_syscall_pt_regs(pointer_arg(1))
+        @_SYSCALL_FACCESSAT2_NAME
+        @_SYSCALL_FACCESSAT2_REGARGS
+        @_SYSCALL_FACCESSAT2_ARGSTR
+},
+{
+        %( @_IS_SREG_KERNEL %? @_SYSCALL_FACCESSAT2_REGARGS_STORE %)
+}
+
+/* kernel 3.5+, but undesirable because it affects all syscalls */
+probe tp_syscall.faccessat2 = kernel.trace("sys_enter")
+{
+        __set_syscall_pt_regs($regs)
+       @__syscall_compat_gate(@const("__NR_faccessat2"),
+                              @const("__NR_compat_faccessat2"))
+        @_SYSCALL_FACCESSAT2_NAME
+        @_SYSCALL_FACCESSAT2_REGARGS
+        @_SYSCALL_FACCESSAT2_ARGSTR
+},
+{
+        %( @_IS_SREG_KERNEL %? @_SYSCALL_FACCESSAT2_REGARGS_STORE %)
+}
+
+probe nd_syscall.faccessat2.return = nd1_syscall.faccessat2.return!, nd2_syscall.faccessat2.return!, tp_syscall.faccessat2.return
+  { }
+
+probe nd1_syscall.faccessat2.return = kprobe.function("sys_faccessat2").return ?
+{
+       @__syscall_compat_gate(@const("__NR_faccessat2"),
+                              @const("__NR_compat_faccessat2"))
+       @_SYSCALL_FACCESSAT2_NAME
+       @SYSC_RETVALSTR(returnval())
+}
+
+/* kernel 4.17+ */
+probe nd2_syscall.faccessat2.return = kprobe.function(@arch_syscall_prefix "sys_faccessat2").return ?
+{
+       @_SYSCALL_FACCESSAT2_NAME
+       @SYSC_RETVALSTR(returnval())
+}
+
+/* kernel 3.5+, but undesirable because it affects all syscalls */
+probe tp_syscall.faccessat2.return = kernel.trace("sys_exit")
+{
+       __set_syscall_pt_regs($regs)
+       @__syscall_compat_gate(@const("__NR_faccessat2"),
+                              @const("__NR_compat_faccessat2"))
+       @_SYSCALL_FACCESSAT2_NAME
+       @SYSC_RETVALSTR($ret)
+}

diff --git a/testsuite/systemtap.syscall/access.c b/testsuite/systemtap.syscall/access.c
index 2b510a35ee587897be1ad26200061c8d6584cbe5..6d17eaaf614f41b853df0eee5d0dbb90bb8af8ad 100644 (file)
--- a/testsuite/systemtap.syscall/access.c
+++ b/testsuite/systemtap.syscall/access.c
@@ -28,7 +28,7 @@ int main()
 
 #if GLIBC_SUPPORT
   faccessat(AT_FDCWD, "foobar1", F_OK, 0);
-  //staptest// faccessat (AT_FDCWD, "foobar1", F_OK) = 0
+  //staptest// [[[[faccessat (AT_FDCWD, "foobar1", F_OK)!!!!faccessat2 (AT_FDCWD, "foobar1", F_OK, 0x0)]]]] = 0
 #endif
 
   access("foobar1", R_OK);
@@ -36,7 +36,7 @@ int main()
 
 #if GLIBC_SUPPORT
   faccessat(AT_FDCWD, "foobar1", R_OK, 0);
-  //staptest// faccessat (AT_FDCWD, "foobar1", R_OK) = 0
+  //staptest// [[[[faccessat (AT_FDCWD, "foobar1", R_OK)!!!!faccessat2 (AT_FDCWD, "foobar1", R_OK, 0x0)]]]] = 0
 #endif
 
   access("foobar1", W_OK);
@@ -44,7 +44,7 @@ int main()
 
 #if GLIBC_SUPPORT
   faccessat(AT_FDCWD, "foobar1", W_OK, 0);
-  //staptest// faccessat (AT_FDCWD, "foobar1", W_OK) = 0
+  //staptest// [[[[faccessat (AT_FDCWD, "foobar1", W_OK)!!!!faccessat2 (AT_FDCWD, "foobar1", W_OK, 0x0)]]]] = 0
 #endif
 
   access("foobar1", X_OK);
@@ -52,7 +52,7 @@ int main()
 
 #if GLIBC_SUPPORT
   faccessat(AT_FDCWD, "foobar1", X_OK, 0);
-  //staptest// faccessat (AT_FDCWD, "foobar1", X_OK) = -NNNN (EACCES)
+  //staptest// [[[[faccessat (AT_FDCWD, "foobar1", X_OK)!!!!faccessat2 (AT_FDCWD, "foobar1", X_OK, 0x0)]]]] = -NNNN (EACCES)
 #endif
 
   access("foobar1", R_OK|W_OK);
@@ -60,7 +60,7 @@ int main()
 
 #if GLIBC_SUPPORT
   faccessat(AT_FDCWD, "foobar1", R_OK|W_OK, 0);
-  //staptest// faccessat (AT_FDCWD, "foobar1", R_OK|W_OK) = 0
+  //staptest// [[[[faccessat (AT_FDCWD, "foobar1", R_OK|W_OK)!!!!faccessat2 (AT_FDCWD, "foobar1", R_OK|W_OK, 0x0)]]]] = 0
 #endif
 
   access("foobar1", R_OK|W_OK|X_OK);
@@ -68,7 +68,7 @@ int main()
 
 #if GLIBC_SUPPORT
   faccessat(AT_FDCWD, "foobar1", R_OK|W_OK|X_OK, 0);
-  //staptest// faccessat (AT_FDCWD, "foobar1", R_OK|W_OK|X_OK) = -NNNN (EACCES)
+  //staptest// [[[[faccessat (AT_FDCWD, "foobar1", R_OK|W_OK|X_OK)!!!!faccessat2 (AT_FDCWD, "foobar1", R_OK|W_OK|X_OK, 0x0)]]]] = -NNNN (EACCES)
 #endif
 
   access((char *)-1, F_OK);
@@ -83,17 +83,17 @@ int main()
 
 #if GLIBC_SUPPORT
   faccessat(-1, "foobar1", F_OK, 0);
-  //staptest// faccessat (-1, "foobar1", F_OK) = -NNNN (EBADF)
+  //staptest// [[[[faccessat (-1, "foobar1", F_OK)!!!!faccessat2 (-1, "foobar1", F_OK, 0x0)]]]] = -NNNN (EBADF)
 
   faccessat(AT_FDCWD, (char *)-1, F_OK, 0);
 #ifdef __s390__
   //staptest// faccessat (AT_FDCWD, 0x[7]?[f]+, F_OK) = -NNNN (EFAULT)
 #else
-  //staptest// faccessat (AT_FDCWD, 0x[f]+, F_OK) = -NNNN (EFAULT)
+  //staptest// [[[[faccessat (AT_FDCWD, 0x[f]+, F_OK)!!!!faccessat2 (AT_FDCWD, 0x[f]+, F_OK, 0x0)]]]] = -NNNN (EFAULT)
 #endif
 
   faccessat(AT_FDCWD, "foobar1", -1, 0);
-  //staptest// faccessat (AT_FDCWD, "foobar1", R_OK|W_OK|X_OK|0x[f]+8) = -NNNN (EINVAL)
+  //staptest// [[[[faccessat (AT_FDCWD, "foobar1", R_OK|W_OK|X_OK|0x[f]+8)!!!!faccessat2 (AT_FDCWD, "foobar1", R_OK|W_OK|X_OK|0x[f]+8, 0x0)]]]] = -NNNN (EINVAL)
 
   // We can't test the last argument to faccessat() as a -1, since
   // glibc will realize that's wrong and not issue a syscall.