man2/setfsuid.2

   1 .\" Copyright (C) 1995, Thomas K. Dyas <tdyas@eden.rutgers.edu>
   2 .\" and Copyright (C) 2013, 2019, Michael Kerrisk <mtk.manpages@gmail.com>
   3 .\"
   4 .\" SPDX-License-Identifier: Linux-man-pages-copyleft
   5 .\"
   6 .\" Created   1995-08-06 Thomas K. Dyas <tdyas@eden.rutgers.edu>
   7 .\" Modified  2000-07-01 aeb
   8 .\" Modified  2002-07-23 aeb
   9 .\" Modified, 27 May 2004, Michael Kerrisk <mtk.manpages@gmail.com>
  10 .\"     Added notes on capability requirements
  11 .\"
  12 .TH SETFSUID 2 2021-03-22 "Linux man-pages (unreleased)" "Linux Programmer's Manual"
  13 .SH NAME
  14 setfsuid \- set user identity used for filesystem checks
  15 .SH LIBRARY
  16 Standard C library
  17 .RI ( libc ", " \-lc )
  18 .SH SYNOPSIS
  19 .nf
  20 .B #include <sys/fsuid.h>
  21 .PP
  22 .BI "int setfsuid(uid_t " fsuid );
  23 .fi
  24 .SH DESCRIPTION
  25 On Linux, a process has both a filesystem user ID and an effective user ID.
  26 The (Linux-specific) filesystem user ID is used
  27 for permissions checking when accessing filesystem objects,
  28 while the effective user ID is used for various other kinds
  29 of permissions checks (see
  30 .BR credentials (7)).
  31 .PP
  32 Normally, the value of the process's filesystem user ID
  33 is the same as the value of its effective user ID.
  34 This is so, because whenever a process's effective user ID is changed,
  35 the kernel also changes the filesystem user ID to be the same as
  36 the new value of the effective user ID.
  37 A process can cause the value of its filesystem user ID to diverge
  38 from its effective user ID by using
  39 .BR setfsuid ()
  40 to change its filesystem user ID to the value given in
  41 .IR fsuid .
  42 .PP
  43 Explicit calls to
  44 .BR setfsuid ()
  45 and
  46 .BR setfsgid (2)
  47 are (were) usually used only by programs such as the Linux NFS server that
  48 need to change what user and group ID is used for file access without a
  49 corresponding change in the real and effective user and group IDs.
  50 A change in the normal user IDs for a program such as the NFS server
  51 is (was) a security hole that can expose it to unwanted signals.
  52 (However, this issue is historical; see below.)
  53 .PP
  54 .BR setfsuid ()
  55 will succeed only if the caller is the superuser or if
  56 .I fsuid
  57 matches either the caller's real user ID, effective user ID,
  58 saved set-user-ID, or current filesystem user ID.
  59 .SH RETURN VALUE
  60 On both success and failure,
  61 this call returns the previous filesystem user ID of the caller.
  62 .SH VERSIONS
  63 This system call is present in Linux since version 1.2.
  64 .\" This system call is present since Linux 1.1.44
  65 .\" and in libc since libc 4.7.6.
  66 .SH STANDARDS
  67 .BR setfsuid ()
  68 is Linux-specific and should not be used in programs intended
  69 to be portable.
  70 .SH NOTES
  71 At the time when this system call was introduced, one process
  72 could send a signal to another process with the same effective user ID.
  73 This meant that if a privileged process changed its effective user ID
  74 for the purpose of file permission checking,
  75 then it could become vulnerable to receiving signals
  76 sent by another (unprivileged) process with the same user ID.
  77 The filesystem user ID attribute was thus added to allow a process to
  78 change its user ID for the purposes of file permission checking without
  79 at the same time becoming vulnerable to receiving unwanted signals.
  80 Since Linux 2.0, signal permission handling is different (see
  81 .BR kill (2)),
  82 with the result that a process can change its effective user ID
  83 without being vulnerable to receiving signals from unwanted processes.
  84 Thus,
  85 .BR setfsuid ()
  86 is nowadays unneeded and should be avoided in new applications
  87 (likewise for
  88 .BR setfsgid (2)).
  89 .PP
  90 The original Linux
  91 .BR setfsuid ()
  92 system call supported only 16-bit user IDs.
  93 Subsequently, Linux 2.4 added
  94 .BR setfsuid32 ()
  95 supporting 32-bit IDs.
  96 The glibc
  97 .BR setfsuid ()
  98 wrapper function transparently deals with the variation across kernel versions.
  99 .SS C library/kernel differences
 100 In glibc 2.15 and earlier,
 101 when the wrapper for this system call determines that the argument can't be
 102 passed to the kernel without integer truncation (because the kernel
 103 is old and does not support 32-bit user IDs),
 104 it will return \-1 and set \fIerrno\fP to
 105 .B EINVAL
 106 without attempting
 107 the system call.
 108 .SH BUGS
 109 No error indications of any kind are returned to the caller,
 110 and the fact that both successful and unsuccessful calls return
 111 the same value makes it impossible to directly determine
 112 whether the call succeeded or failed.
 113 Instead, the caller must resort to looking at the return value
 114 from a further call such as
 115 .I setfsuid(\-1)
 116 (which will always fail), in order to determine if a preceding call to
 117 .BR setfsuid ()
 118 changed the filesystem user ID.
 119 At the very
 120 least,
 121 .B EPERM
 122 should be returned when the call fails (because the caller lacks the
 123 .B CAP_SETUID
 124 capability).
 125 .SH SEE ALSO
 126 .BR kill (2),
 127 .BR setfsgid (2),
 128 .BR capabilities (7),
 129 .BR credentials (7)