1 .\" Copyright (C) 2014 Michael Kerrisk <mtk.manpages@gmail.com>
2 .\" and Copyright (C) 2014 David Herrmann <dh.herrmann@gmail.com>
4 .\" %%%LICENSE_START(GPLv2+)
6 .\" This program is free software; you can redistribute it and/or modify
7 .\" it under the terms of the GNU General Public License as published by
8 .\" the Free Software Foundation; either version 2 of the License, or
9 .\" (at your option) any later version.
11 .\" This program is distributed in the hope that it will be useful,
12 .\" but WITHOUT ANY WARRANTY; without even the implied warranty of
13 .\" MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 .\" GNU General Public License for more details.
16 .\" You should have received a copy of the GNU General Public
17 .\" License along with this manual; if not, see
18 .\" <http://www.gnu.org/licenses/>.
21 .TH MEMFD_CREATE 2 2021-03-22 Linux "Linux Programmer's Manual"
23 memfd_create \- create an anonymous file
26 .BR "#define _GNU_SOURCE" " /* See feature_test_macros(7) */"
27 .B #include <sys/mman.h>
29 .BI "int memfd_create(const char *" name ", unsigned int " flags ");"
33 creates an anonymous file and returns a file descriptor that refers to it.
34 The file behaves like a regular file, and so can be modified,
35 truncated, memory-mapped, and so on.
36 However, unlike a regular file,
37 it lives in RAM and has a volatile backing storage.
38 Once all references to the file are dropped, it is automatically released.
39 Anonymous memory is used for all backing pages of the file.
40 Therefore, files created by
42 have the same semantics as other anonymous
44 .\" memfd uses VM_NORESERVE so each page is accounted on first access.
45 .\" This means, the overcommit-limits (see __vm_enough_memory()) and the
46 .\" memory-cgroup limits (mem_cgroup_try_charge()) are applied. Note that
47 .\" those are accounted on "current" and "current->mm", that is, the
48 .\" process doing the first page access.
49 memory allocations such as those allocated using
55 The initial size of the file is set to 0.
56 Following the call, the file size should be set using
58 (Alternatively, the file may be populated by calls to
64 is used as a filename and will be displayed
65 as the target of the corresponding symbolic link in the directory
67 The displayed name is always prefixed with
69 and serves only for debugging purposes.
70 Names do not affect the behavior of the file descriptor,
71 and as such multiple files can have the same name without any side effects.
73 The following values may be bitwise ORed in
75 to change the behavior of
81 flag on the new file descriptor.
82 See the description of the
86 for reasons why this may be useful.
89 Allow sealing operations on this file.
90 See the discussion of the
96 and also NOTES, below.
97 The initial set of seals is empty.
98 If this flag is not set, the initial set of seals will be
100 meaning that no other seals can be set on the file.
101 .\" FIXME Why is the MFD_ALLOW_SEALING behavior not simply the default?
102 .\" Is it worth adding some text explaining this?
104 .BR MFD_HUGETLB " (since Linux 4.14)"
105 .\" commit 749df87bd7bee5a79cef073f5d032ddb2b211de8
106 The anonymous file will be created in the hugetlbfs filesystem using
108 See the Linux kernel source file
109 .I Documentation/admin\-guide/mm/hugetlbpage.rst
110 for more information about hugetlbfs.
111 .\" commit 47b9012ecdc747f6936395265e677d41e11a31ff
118 is supported since Linux 4.16.
120 .BR MFD_HUGE_2MB ", " MFD_HUGE_1GB ", " "..."
121 Used in conjunction with
123 to select alternative hugetlb page sizes (respectively, 2\ MB, 1\ GB, ...)
124 on systems that support multiple hugetlb page sizes.
125 Definitions for known
126 huge page sizes are included in the header file
129 For details on encoding huge page sizes not included in the header file,
130 see the discussion of the similarly named constants in
139 returns a new file descriptor that can be used to refer to the file.
140 This file descriptor is opened for both reading and writing
144 is set for the file descriptor.
150 the usual semantics apply for the file descriptor created by
152 A copy of the file descriptor is inherited by the child produced by
154 and refers to the same file.
155 The file descriptor is preserved across
157 unless the close-on-exec flag has been set.
161 returns a new file descriptor.
162 On error, \-1 is returned and
164 is set to indicate the error.
170 points to invalid memory.
174 included unknown bits.
180 .\" NAME_MAX - strlen("memfd:")
181 249 bytes, excluding the terminating null byte.)
192 The per-process limit on the number of open file descriptors has been reached.
195 The system-wide limit on the total number of open files has been reached.
198 There was insufficient memory to create a new anonymous file.
202 system call first appeared in Linux 3.17;
203 glibc support was added in version 2.27.
207 system call is Linux-specific.
209 .\" See also http://lwn.net/Articles/593918/
210 .\" and http://lwn.net/Articles/594919/ and http://lwn.net/Articles/591108/
213 system call provides a simple alternative to manually mounting a
215 filesystem and creating and opening a file in that filesystem.
216 The primary purpose of
218 is to create files and associated file descriptors that are
219 used with the file-sealing APIs provided by
224 system call also has uses without file sealing
225 (which is why file-sealing is disabled, unless explicitly requested with the
226 .BR MFD_ALLOW_SEALING
228 In particular, it can be used as an alternative to creating files in
230 or as an alternative to using the
233 in cases where there is no intention to actually link the
234 resulting file into the filesystem.
236 In the absence of file sealing,
237 processes that communicate via shared memory must either trust each other,
238 or take measures to deal with the possibility that an untrusted peer
239 may manipulate the shared memory region in problematic ways.
240 For example, an untrusted peer might modify the contents of the
241 shared memory at any time, or shrink the shared memory region.
242 The former possibility leaves the local process vulnerable to
243 time-of-check-to-time-of-use race conditions
244 (typically dealt with by copying data from
245 the shared memory region before checking and using it).
246 The latter possibility leaves the local process vulnerable to
248 signals when an attempt is made to access a now-nonexistent
249 location in the shared memory region.
250 (Dealing with this possibility necessitates the use of a handler for the
254 Dealing with untrusted peers imposes extra complexity on
255 code that employs shared memory.
256 Memory sealing enables that extra complexity to be eliminated,
257 by allowing a process to operate secure in the knowledge that
258 its peer can't modify the shared memory in an undesired fashion.
260 An example of the usage of the sealing mechanism is as follows:
262 The first process creates a
266 The call yields a file descriptor used in subsequent steps.
269 sizes the file created in the previous step using
273 and populates the shared memory with the desired data.
275 The first process uses the
278 operation to place one or more seals on the file,
279 in order to restrict further modifications on the file.
282 then it will be necessary to first unmap the shared writable mapping
283 created in the previous step.
284 Otherwise, behavior similar to
286 can be achieved by using
287 .BR F_SEAL_FUTURE_WRITE ,
288 which will prevent future writes via
292 from succeeding while keeping existing shared writable mappings).
294 A second process obtains a file descriptor for the
297 Among the possible ways in which this could happen are the following:
300 The process that called
302 could transfer the resulting file descriptor to the second process
303 via a UNIX domain socket (see
307 The second process then maps the file using
310 The second process is created via
312 and thus automatically inherits the file descriptor and mapping.
313 (Note that in this case and the next,
314 there is a natural trust relationship between the two processes,
315 since they are running under the same user ID.
316 Therefore, file sealing would not normally be necessary.)
318 The second process opens the file
319 .IR /proc/<pid>/fd/<fd> ,
322 is the PID of the first process (the one that called
323 .BR memfd_create ()),
326 is the number of the file descriptor returned by the call to
329 The second process then maps the file using
333 The second process uses the
336 operation to retrieve the bit mask of seals
337 that has been applied to the file.
338 This bit mask can be inspected in order to determine
339 what kinds of restrictions have been placed on file modifications.
340 If desired, the second process can apply further seals
341 to impose additional restrictions (so long as the
343 seal has not yet been applied).
345 Below are shown two example programs that demonstrate the use of
347 and the file sealing API.
350 .IR t_memfd_create.c ,
355 sets a size for the file, maps it into memory,
356 and optionally places some seals on the file.
357 The program accepts up to three command-line arguments,
358 of which the first two are required.
359 The first argument is the name to associate with the file,
360 the second argument is the size to be set for the file,
361 and the optional third argument is a string of characters that specify
362 seals to be set on file.
366 can be used to open an existing file that was created via
368 and inspect the set of seals that have been applied to that file.
370 The following shell session demonstrates the use of these programs.
373 file and set some seals on it:
377 $ \fB./t_memfd_create my_memfd_file 4096 sw &\fP
379 PID: 11775; fd: 3; /proc/11775/fd/3
385 program continues to run in the background.
386 From another program, we can obtain a file descriptor for the
391 file that corresponds to the file descriptor opened by
393 Using that pathname, we inspect the content of the
395 symbolic link, and use our
397 program to view the seals that have been placed on the file:
401 $ \fBreadlink /proc/11775/fd/3\fP
402 /memfd:my_memfd_file (deleted)
403 $ \fB./t_get_seals /proc/11775/fd/3\fP
404 Existing seals: WRITE SHRINK
407 .SS Program source: t_memfd_create.c
412 #include <sys/mman.h>
419 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \e
423 main(int argc, char *argv[])
428 char *name, *seals_arg;
432 fprintf(stderr, "%s name size [seals]\en", argv[0]);
433 fprintf(stderr, "\et\(aqseals\(aq can contain any of the "
434 "following characters:\en");
435 fprintf(stderr, "\et\etg \- F_SEAL_GROW\en");
436 fprintf(stderr, "\et\ets \- F_SEAL_SHRINK\en");
437 fprintf(stderr, "\et\etw \- F_SEAL_WRITE\en");
438 fprintf(stderr, "\et\etW \- F_SEAL_FUTURE_WRITE\en");
439 fprintf(stderr, "\et\etS \- F_SEAL_SEAL\en");
447 /* Create an anonymous file in tmpfs; allow seals to be
448 placed on the file. */
450 fd = memfd_create(name, MFD_ALLOW_SEALING);
452 errExit("memfd_create");
454 /* Size the file as specified on the command line. */
456 if (ftruncate(fd, len) == \-1)
459 printf("PID: %jd; fd: %d; /proc/%jd/fd/%d\en",
460 (intmax_t) getpid(), fd, (intmax_t) getpid(), fd);
462 /* Code to map the file and populate the mapping with data
465 /* If a \(aqseals\(aq command\-line argument was supplied, set some
466 seals on the file. */
468 if (seals_arg != NULL) {
471 if (strchr(seals_arg, \(aqg\(aq) != NULL)
472 seals |= F_SEAL_GROW;
473 if (strchr(seals_arg, \(aqs\(aq) != NULL)
474 seals |= F_SEAL_SHRINK;
475 if (strchr(seals_arg, \(aqw\(aq) != NULL)
476 seals |= F_SEAL_WRITE;
477 if (strchr(seals_arg, \(aqW\(aq) != NULL)
478 seals |= F_SEAL_FUTURE_WRITE;
479 if (strchr(seals_arg, \(aqS\(aq) != NULL)
480 seals |= F_SEAL_SEAL;
482 if (fcntl(fd, F_ADD_SEALS, seals) == \-1)
486 /* Keep running, so that the file created by memfd_create()
487 continues to exist. */
494 .SS Program source: t_get_seals.c
498 #include <sys/mman.h>
505 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \e
509 main(int argc, char *argv[])
515 fprintf(stderr, "%s /proc/PID/fd/FD\en", argv[0]);
519 fd = open(argv[1], O_RDWR);
523 seals = fcntl(fd, F_GET_SEALS);
527 printf("Existing seals:");
528 if (seals & F_SEAL_SEAL)
530 if (seals & F_SEAL_GROW)
532 if (seals & F_SEAL_WRITE)
534 if (seals & F_SEAL_FUTURE_WRITE)
535 printf(" FUTURE_WRITE");
536 if (seals & F_SEAL_SHRINK)
540 /* Code to map the file and access the contents of the
541 resulting mapping omitted. */