| 1 |
.\" Copyright 2014, Michael Kerrisk <[email protected]> |
| 2 |
.\" Copyright 2014, David Herrmann <[email protected]> |
| 3 |
.\" Copyright 2021, International Business Machines Corp. |
| 4 |
.\" Copyright, the authors of the Linux man-pages project |
| 5 |
.\" |
| 6 |
.\" SPDX-License-Identifier: GPL-2.0-or-later |
| 7 |
.\" |
| 8 |
.TH memfd_secret 2 2026-02-08 "Linux man-pages 6.18" |
| 9 |
.SH NAME |
| 10 |
memfd_secret \- create an anonymous RAM-based file |
| 11 |
to access secret memory regions |
| 12 |
.SH LIBRARY |
| 13 |
Standard C library |
| 14 |
.RI ( libc ,\~ \-lc ) |
| 15 |
.SH SYNOPSIS |
| 16 |
.nf |
| 17 |
.P |
| 18 |
.BR "#include <sys/syscall.h>" " /* Definition of " SYS_* " constants */" |
| 19 |
.B #include <unistd.h> |
| 20 |
.P |
| 21 |
.BI "int syscall(SYS_memfd_secret, unsigned int " flags ); |
| 22 |
.fi |
| 23 |
.P |
| 24 |
.IR Note : |
| 25 |
glibc provides no wrapper for |
| 26 |
.BR memfd_secret (), |
| 27 |
necessitating the use of |
| 28 |
.BR syscall (2). |
| 29 |
.SH DESCRIPTION |
| 30 |
.BR memfd_secret () |
| 31 |
creates an anonymous RAM-based file and returns a file descriptor |
| 32 |
that refers to it. |
| 33 |
The file provides a way to create and access memory regions |
| 34 |
with stronger protection than usual RAM-based files and |
| 35 |
anonymous memory mappings. |
| 36 |
Once all open references to the file are closed, |
| 37 |
it is automatically released. |
| 38 |
The initial size of the file is set to 0. |
| 39 |
Following the call, the file size should be set using |
| 40 |
.BR ftruncate (2). |
| 41 |
.P |
| 42 |
The memory areas backing the file created with |
| 43 |
.BR memfd_secret (2) |
| 44 |
are visible only to the processes that have access to the file descriptor. |
| 45 |
The memory region is removed from the kernel page tables |
| 46 |
and only the page tables of the processes holding the file descriptor |
| 47 |
map the corresponding physical memory. |
| 48 |
(Thus, the pages in the region can't be accessed by the kernel itself, |
| 49 |
so that, for example, pointers to the region can't be passed to |
| 50 |
system calls.) |
| 51 |
.P |
| 52 |
The following values may be bitwise ORed in |
| 53 |
.I flags |
| 54 |
to control the behavior of |
| 55 |
.BR memfd_secret (): |
| 56 |
.TP |
| 57 |
.B FD_CLOEXEC |
| 58 |
Set the close-on-exec flag on the new file descriptor, |
| 59 |
which causes the region to be removed from the process on |
| 60 |
.BR execve (2). |
| 61 |
See the description of the |
| 62 |
.B O_CLOEXEC |
| 63 |
flag in |
| 64 |
.BR open (2) |
| 65 |
.P |
| 66 |
As its return value, |
| 67 |
.BR memfd_secret () |
| 68 |
returns a new file descriptor that refers to an anonymous file. |
| 69 |
This file descriptor is opened for both reading and writing |
| 70 |
.RB ( O_RDWR ) |
| 71 |
and |
| 72 |
.B O_LARGEFILE |
| 73 |
is set for the file descriptor. |
| 74 |
.P |
| 75 |
With respect to |
| 76 |
.BR fork (2) |
| 77 |
and |
| 78 |
.BR execve (2), |
| 79 |
the usual semantics apply for the file descriptor created by |
| 80 |
.BR memfd_secret (). |
| 81 |
A copy of the file descriptor is inherited by the child produced by |
| 82 |
.BR fork (2) |
| 83 |
and refers to the same file. |
| 84 |
The file descriptor is preserved across |
| 85 |
.BR execve (2), |
| 86 |
unless the close-on-exec flag has been set. |
| 87 |
.P |
| 88 |
The memory region is locked into memory in the same way as with |
| 89 |
.BR mlock (2), |
| 90 |
so that it will never be written into swap, |
| 91 |
and hibernation is inhibited for as long as any |
| 92 |
.BR memfd_secret () |
| 93 |
descriptions exist. |
| 94 |
However the implementation of |
| 95 |
.BR memfd_secret () |
| 96 |
will not try to populate the whole range during the |
| 97 |
.BR mmap (2) |
| 98 |
call that attaches the region into the process's address space; |
| 99 |
instead, the pages are only actually allocated |
| 100 |
as they are faulted in. |
| 101 |
The amount of memory allowed for memory mappings |
| 102 |
of the file descriptor obeys the same rules as |
| 103 |
.BR mlock (2) |
| 104 |
and cannot exceed |
| 105 |
.BR RLIMIT_MEMLOCK . |
| 106 |
.SH RETURN VALUE |
| 107 |
On success, |
| 108 |
.BR memfd_secret () |
| 109 |
returns a new file descriptor. |
| 110 |
On error, \-1 is returned and |
| 111 |
.I errno |
| 112 |
is set to indicate the error. |
| 113 |
.SH ERRORS |
| 114 |
.TP |
| 115 |
.B EINVAL |
| 116 |
.I flags |
| 117 |
included unknown bits. |
| 118 |
.TP |
| 119 |
.B EMFILE |
| 120 |
The per-process limit on the number of open file descriptors has been reached. |
| 121 |
.TP |
| 122 |
.B EMFILE |
| 123 |
The system-wide limit on the total number of open files has been reached. |
| 124 |
.TP |
| 125 |
.B ENOMEM |
| 126 |
There was insufficient memory to create a new anonymous file. |
| 127 |
.TP |
| 128 |
.B ENOSYS |
| 129 |
.BR memfd_secret () |
| 130 |
is not implemented on this architecture, |
| 131 |
or has not been enabled on the kernel command-line with |
| 132 |
.BR secretmem_enable =1. |
| 133 |
.SH STANDARDS |
| 134 |
Linux. |
| 135 |
.SH HISTORY |
| 136 |
Linux 5.14. |
| 137 |
.P |
| 138 |
Before Linux 6.5, |
| 139 |
.\" commit b758fe6df50daf68fef089d8f3c1cd49fc794ed2 |
| 140 |
.BR memfd_secret () |
| 141 |
was disabled by default and only available |
| 142 |
if the system administrator turned it on using |
| 143 |
"secretmem.enable=y" kernel parameter. |
| 144 |
.SH NOTES |
| 145 |
The |
| 146 |
.BR memfd_secret () |
| 147 |
system call is designed to allow a user-space process |
| 148 |
to create a range of memory that is inaccessible to anybody else - |
| 149 |
kernel included. |
| 150 |
There is no 100% guarantee that kernel won't be able to access |
| 151 |
memory ranges backed by |
| 152 |
.BR memfd_secret () |
| 153 |
in any circumstances, but nevertheless, |
| 154 |
it is much harder to exfiltrate data from these regions. |
| 155 |
.P |
| 156 |
.BR memfd_secret () |
| 157 |
provides the following protections: |
| 158 |
.IP \[bu] 3 |
| 159 |
Enhanced protection |
| 160 |
(in conjunction with all the other in-kernel attack prevention systems) |
| 161 |
against ROP attacks. |
| 162 |
Absence of any in-kernel primitive for accessing memory backed by |
| 163 |
.BR memfd_secret () |
| 164 |
means that one-gadget ROP attack |
| 165 |
can't work to perform data exfiltration. |
| 166 |
The attacker would need to find enough ROP gadgets |
| 167 |
to reconstruct the missing page table entries, |
| 168 |
which significantly increases difficulty of the attack, |
| 169 |
especially when other protections like the kernel stack size limit |
| 170 |
and address space layout randomization are in place. |
| 171 |
.IP \[bu] |
| 172 |
Prevent cross-process user-space memory exposures. |
| 173 |
Once a region for a |
| 174 |
.BR memfd_secret () |
| 175 |
memory mapping is allocated, |
| 176 |
the user can't accidentally pass it into the kernel |
| 177 |
to be transmitted somewhere. |
| 178 |
The memory pages in this region cannot be accessed via the direct map |
| 179 |
and they are disallowed in get_user_pages. |
| 180 |
.IP \[bu] |
| 181 |
Harden against exploited kernel flaws. |
| 182 |
In order to access memory areas backed by |
| 183 |
.BR memfd_secret (), |
| 184 |
a kernel-side attack would need to |
| 185 |
either walk the page tables and create new ones, |
| 186 |
or spawn a new privileged user-space process to perform |
| 187 |
secrets exfiltration using |
| 188 |
.BR ptrace (2). |
| 189 |
.P |
| 190 |
To prevent potential data leaks of memory regions backed by |
| 191 |
.BR memfd_secret () |
| 192 |
from a hybernation image, |
| 193 |
hybernation is prevented when there are active |
| 194 |
.BR memfd_secret () |
| 195 |
users. |
| 196 |
.SH SEE ALSO |
| 197 |
.BR fcntl (2), |
| 198 |
.BR ftruncate (2), |
| 199 |
.BR mlock (2), |
| 200 |
.BR memfd_create (2), |
| 201 |
.BR mmap (2), |
| 202 |
.BR setrlimit (2) |