man2/memfd_create.2

   1 .\" Copyright (C) 2014 Michael Kerrisk <mtk.manpages@gmail.com>
   2 .\" and Copyright (C) 2014 David Herrmann <dh.herrmann@gmail.com>
   3 .\"
   4 .\" %%%LICENSE_START(GPLv2+)
   5 .\"
   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.
  10 .\"
  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.
  15 .\"
  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/>.
  19 .\" %%%LICENSE_END
  20 .\"
  21 .TH MEMFD_CREATE 2 2018-02-02 Linux "Linux Programmer's Manual"
  22 .SH NAME
  23 memfd_create \- create an anonymous file
  24 .SH SYNOPSIS
  25 .nf
  26 .BR "#define _GNU_SOURCE" "         /* See feature_test_macros(7) */"
  27 .B #include <sys/mman.h>
  28 .PP
  29 .BI "int memfd_create(const char *" name ", unsigned int " flags ");"
  30 .SH DESCRIPTION
  31 .BR memfd_create ()
  32 creates an anonymous file and returns a file descriptor that refers to it.
  33 The file behaves like a regular file, and so can be modified,
  34 truncated, memory-mapped, and so on.
  35 However, unlike a regular file,
  36 it lives in RAM and has a volatile backing storage.
  37 Once all references to the file are dropped, it is automatically released.
  38 Anonymous memory is used for all backing pages of the file.
  39 Therefore, files created by
  40 .BR memfd_create ()
  41 have the same semantics as other anonymous
  42 .\" David Herrmann:
  43 .\"     memfd uses VM_NORESERVE so each page is accounted on first access.
  44 .\"     This means, the overcommit-limits (see __vm_enough_memory()) and the
  45 .\"     memory-cgroup limits (mem_cgroup_try_charge()) are applied. Note that
  46 .\"     those are accounted on "current" and "current->mm", that is, the
  47 .\"     process doing the first page access.
  48 memory allocations such as those allocated using
  49 .BR mmap (2)
  50 with the
  51 .BR MAP_ANONYMOUS
  52 flag.
  53 .PP
  54 The initial size of the file is set to 0.
  55 Following the call, the file size should be set using
  56 .BR ftruncate (2).
  57 (Alternatively, the file may be populated by calls to
  58 .BR write (2)
  59 or similar.)
  60 .PP
  61 The name supplied in
  62 .I name
  63 is used as a filename and will be displayed
  64 as the target of the corresponding symbolic link in the directory
  65 .IR /proc/self/fd/ .
  66 The displayed name is always prefixed with
  67 .IR memfd:
  68 and serves only for debugging purposes.
  69 Names do not affect the behavior of the file descriptor,
  70 and as such multiple files can have the same name without any side effects.
  71 .PP
  72 The following values may be bitwise ORed in
  73 .IR flags
  74 to change the behavior of
  75 .BR memfd_create ():
  76 .TP
  77 .BR MFD_CLOEXEC
  78 Set the close-on-exec
  79 .RB ( FD_CLOEXEC )
  80 flag on the new file descriptor.
  81 See the description of the
  82 .B O_CLOEXEC
  83 flag in
  84 .BR open (2)
  85 for reasons why this may be useful.
  86 .TP
  87 .BR MFD_ALLOW_SEALING
  88 Allow sealing operations on this file.
  89 See the discussion of the
  90 .B F_ADD_SEALS
  91 and
  92 .BR F_GET_SEALS
  93 operations in
  94 .BR fcntl (2),
  95 and also NOTES, below.
  96 The initial set of seals is empty.
  97 If this flag is not set, the initial set of seals will be
  98 .BR F_SEAL_SEAL ,
  99 meaning that no other seals can be set on the file.
 100 .\" FIXME Why is the MFD_ALLOW_SEALING behavior not simply the default?
 101 .\" Is it worth adding some text explaining this?
 102 .TP
 103 .BR MFD_HUGETLB " (since Linux 4.14)"
 104 .\" commit 749df87bd7bee5a79cef073f5d032ddb2b211de8
 105 The anonymous file will be created in the hugetlbfs filesystem using
 106 huge pages.
 107 See the Linux kernel source file
 108 .I Documentation/vm/hugetlbpage.txt
 109 for more information about hugetlbfs.
 110 .\" commit 47b9012ecdc747f6936395265e677d41e11a31ff
 111 Specifying both
 112 .B MFD_HUGETLB
 113 and
 114 .B MFD_ALLOW_SEALING
 115 in
 116 .I flags
 117 is supported since Linux 4.16.
 118 .TP
 119 .BR MFD_HUGE_2MB ", " MFD_HUGE_1GB ", " "..."
 120 Used in conjunction with
 121 .B MFD_HUGETLB
 122 to select alternative hugetlb page sizes (respectively, 2\ MB, 1\ GB, ...)
 123 on systems that support multiple hugetlb page sizes.
 124 Definitions for known
 125 huge page sizes are included in the header file
 126 .I <linux/memfd.h>.
 127 .IP
 128 For details on encoding huge page sizes not included in the header file,
 129 see the discussion of the similarly named constants in
 130 .BR mmap (2).
 131 .PP
 132 Unused bits in
 133 .I flags
 134 must be 0.
 135 .PP
 136 As its return value,
 137 .BR memfd_create ()
 138 returns a new file descriptor that can be used to refer to the file.
 139 This file descriptor is opened for both reading and writing
 140 .RB ( O_RDWR )
 141 and
 142 .B O_LARGEFILE
 143 is set for the file descriptor.
 144 .PP
 145 With respect to
 146 .BR fork (2)
 147 and
 148 .BR execve (2),
 149 the usual semantics apply for the file descriptor created by
 150 .BR memfd_create ().
 151 A copy of the file descriptor is inherited by the child produced by
 152 .BR fork (2)
 153 and refers to the same file.
 154 The file descriptor is preserved across
 155 .BR execve (2),
 156 unless the close-on-exec flag has been set.
 157 .SH RETURN VALUE
 158 On success,
 159 .BR memfd_create ()
 160 returns a new file descriptor.
 161 On error, \-1 is returned and
 162 .I errno
 163 is set to indicate the error.
 164 .SH ERRORS
 165 .TP
 166 .B EFAULT
 167 The address in
 168 .IR name
 169 points to invalid memory.
 170 .TP
 171 .B EINVAL
 172 .I flags
 173 included unknown bits.
 174 .TP
 175 .B EINVAL
 176 .I name
 177 was too long.
 178 (The limit is
 179 .\" NAME_MAX - strlen("memfd:")
 180 249 bytes, excluding the terminating null byte.)
 181 .TP
 182 .B EINVAL
 183 Both
 184 .B MFD_HUGETLB
 185 and
 186 .B MFD_ALLOW_SEALING
 187 were specified in
 188 .IR flags .
 189 .TP
 190 .B EMFILE
 191 The per-process limit on the number of open file descriptors has been reached.
 192 .TP
 193 .B ENFILE
 194 The system-wide limit on the total number of open files has been reached.
 195 .TP
 196 .B ENOMEM
 197 There was insufficient memory to create a new anonymous file.
 198 .SH VERSIONS
 199 The
 200 .BR memfd_create ()
 201 system call first appeared in Linux 3.17;
 202 glibc support was added in version 2.27.
 203 .SH CONFORMING TO
 204 The
 205 .BR memfd_create ()
 206 system call is Linux-specific.
 207 .SH NOTES
 208 .PP
 209 .\" See also http://lwn.net/Articles/593918/
 210 .\" and http://lwn.net/Articles/594919/ and http://lwn.net/Articles/591108/
 211 The
 212 .BR memfd_create ()
 213 system call provides a simple alternative to manually mounting a
 214 .BR tmpfs (5)
 215 filesystem and creating and opening a file in that filesystem.
 216 The primary purpose of
 217 .BR memfd_create ()
 218 is to create files and associated file descriptors that are
 219 used with the file-sealing APIs provided by
 220 .BR fcntl (2).
 221 .PP
 222 The
 223 .BR memfd_create ()
 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
 227 flag).
 228 In particular, it can be used as an alternative to creating files in
 229 .IR tmp
 230 or as an alternative to using the
 231 .BR  open (2)
 232 .B O_TMPFILE
 233 in cases where there is no intention to actually link the
 234 resulting file into the filesystem.
 235 .SS File sealing
 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
 247 .BR SIGBUS
 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
 251 .BR SIGBUS
 252 signal.)
 253 .PP
 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.
 259 .PP
 260 An example of the usage of the sealing mechanism is as follows:
 261 .IP 1. 3
 262 The first process creates a
 263 .BR tmpfs (5)
 264 file using
 265 .BR memfd_create ().
 266 The call yields a file descriptor used in subsequent steps.
 267 .IP 2.
 268 The first process
 269 sizes the file created in the previous step using
 270 .BR ftruncate (2),
 271 maps it using
 272 .BR mmap (2),
 273 and populates the shared memory with the desired data.
 274 .IP 3.
 275 The first process uses the
 276 .BR fcntl (2)
 277 .B F_ADD_SEALS
 278 operation to place one or more seals on the file,
 279 in order to restrict further modifications on the file.
 280 (If placing the seal
 281 .BR F_SEAL_WRITE ,
 282 then it will be necessary to first unmap the shared writable mapping
 283 created in the previous step.)
 284 .IP 4.
 285 A second process obtains a file descriptor for the
 286 .BR tmpfs (5)
 287 file and maps it.
 288 Among the possible ways in which this could happen are the following:
 289 .RS
 290 .IP * 3
 291 The process that called
 292 .BR memfd_create ()
 293 could transfer the resulting file descriptor to the second process
 294 via a UNIX domain socket (see
 295 .BR unix (7)
 296 and
 297 .BR cmsg (3)).
 298 The second process then maps the file using
 299 .BR mmap (2).
 300 .IP *
 301 The second process is created via
 302 .BR fork (2)
 303 and thus automatically inherits the file descriptor and mapping.
 304 (Note that in this case and the next,
 305 there is a natural trust relationship between the two processes,
 306 since they are running under the same user ID.
 307 Therefore, file sealing would not normally be necessary.)
 308 .IP *
 309 The second process opens the file
 310 .IR /proc/<pid>/fd/<fd> ,
 311 where
 312 .I <pid>
 313 is the PID of the first process (the one that called
 314 .BR memfd_create ()),
 315 and
 316 .I <fd>
 317 is the number of the file descriptor returned by the call to
 318 .BR memfd_create ()
 319 in that process.
 320 The second process then maps the file using
 321 .BR mmap (2).
 322 .RE
 323 .IP 5.
 324 The second process uses the
 325 .BR fcntl (2)
 326 .B F_GET_SEALS
 327 operation to retrieve the bit mask of seals
 328 that has been applied to the file.
 329 This bit mask can be inspected in order to determine
 330 what kinds of restrictions have been placed on file modifications.
 331 If desired, the second process can apply further seals
 332 to impose additional restrictions (so long as the
 333 .BR F_SEAL_SEAL
 334 seal has not yet been applied).
 335 .SH EXAMPLE
 336 Below are shown two example programs that demonstrate the use of
 337 .BR memfd_create ()
 338 and the file sealing API.
 339 .PP
 340 The first program,
 341 .IR t_memfd_create.c ,
 342 creates a
 343 .BR tmpfs (5)
 344 file using
 345 .BR memfd_create (),
 346 sets a size for the file, maps it into memory,
 347 and optionally places some seals on the file.
 348 The program accepts up to three command-line arguments,
 349 of which the first two are required.
 350 The first argument is the name to associate with the file,
 351 the second argument is the size to be set for the file,
 352 and the optional third argument is a string of characters that specify
 353 seals to be set on file.
 354 .PP
 355 The second program,
 356 .IR t_get_seals.c ,
 357 can be used to open an existing file that was created via
 358 .BR memfd_create ()
 359 and inspect the set of seals that have been applied to that file.
 360 .PP
 361 The following shell session demonstrates the use of these programs.
 362 First we create a
 363 .BR tmpfs (5)
 364 file and set some seals on it:
 365 .PP
 366 .in +4n
 367 .EX
 368 $ \fB./t_memfd_create my_memfd_file 4096 sw &\fP
 369 [1] 11775
 370 PID: 11775; fd: 3; /proc/11775/fd/3
 371 .EE
 372 .in
 373 .PP
 374 At this point, the
 375 .I t_memfd_create
 376 program continues to run in the background.
 377 From another program, we can obtain a file descriptor for the
 378 file created by
 379 .BR memfd_create ()
 380 by opening the
 381 .IR /proc/[pid]/fd
 382 file that corresponds to the file descriptor opened by
 383 .BR memfd_create ().
 384 Using that pathname, we inspect the content of the
 385 .IR /proc/[pid]/fd
 386 symbolic link, and use our
 387 .I t_get_seals
 388 program to view the seals that have been placed on the file:
 389 .PP
 390 .in +4n
 391 .EX
 392 $ \fBreadlink /proc/11775/fd/3\fP
 393 /memfd:my_memfd_file (deleted)
 394 $ \fB./t_get_seals /proc/11775/fd/3\fP
 395 Existing seals: WRITE SHRINK
 396 .EE
 397 .in
 398 .SS Program source: t_memfd_create.c
 399 \&
 400 .EX
 401 #define _GNU_SOURCE
 402 #include <sys/mman.h>
 403 #include <fcntl.h>
 404 #include <stdlib.h>
 405 #include <unistd.h>
 406 #include <string.h>
 407 #include <stdio.h>
 408
 409 #define errExit(msg)    do { perror(msg); exit(EXIT_FAILURE); \\
 410                         } while (0)
 411
 412 int
 413 main(int argc, char *argv[])
 414 {
 415     int fd;
 416     unsigned int seals;
 417     char *addr;
 418     char *name, *seals_arg;
 419     ssize_t len;
 420
 421     if (argc < 3) {
 422         fprintf(stderr, "%s name size [seals]\\n", argv[0]);
 423         fprintf(stderr, "\\t\(aqseals\(aq can contain any of the "
 424                 "following characters:\\n");
 425         fprintf(stderr, "\\t\\tg \- F_SEAL_GROW\\n");
 426         fprintf(stderr, "\\t\\ts \- F_SEAL_SHRINK\\n");
 427         fprintf(stderr, "\\t\\tw \- F_SEAL_WRITE\\n");
 428         fprintf(stderr, "\\t\\tS \- F_SEAL_SEAL\\n");
 429         exit(EXIT_FAILURE);
 430     }
 431
 432     name = argv[1];
 433     len = atoi(argv[2]);
 434     seals_arg = argv[3];
 435
 436     /* Create an anonymous file in tmpfs; allow seals to be
 437        placed on the file */
 438
 439     fd = memfd_create(name, MFD_ALLOW_SEALING);
 440     if (fd == \-1)
 441         errExit("memfd_create");
 442
 443     /* Size the file as specified on the command line */
 444
 445     if (ftruncate(fd, len) == \-1)
 446         errExit("truncate");
 447
 448     printf("PID: %ld; fd: %d; /proc/%ld/fd/%d\\n",
 449             (long) getpid(), fd, (long) getpid(), fd);
 450
 451     /* Code to map the file and populate the mapping with data
 452        omitted */
 453
 454     /* If a \(aqseals\(aq command\-line argument was supplied, set some
 455        seals on the file */
 456
 457     if (seals_arg != NULL) {
 458         seals = 0;
 459
 460         if (strchr(seals_arg, \(aqg\(aq) != NULL)
 461             seals |= F_SEAL_GROW;
 462         if (strchr(seals_arg, \(aqs\(aq) != NULL)
 463             seals |= F_SEAL_SHRINK;
 464         if (strchr(seals_arg, \(aqw\(aq) != NULL)
 465             seals |= F_SEAL_WRITE;
 466         if (strchr(seals_arg, \(aqS\(aq) != NULL)
 467             seals |= F_SEAL_SEAL;
 468
 469         if (fcntl(fd, F_ADD_SEALS, seals) == \-1)
 470             errExit("fcntl");
 471     }
 472
 473     /* Keep running, so that the file created by memfd_create()
 474        continues to exist */
 475
 476     pause();
 477
 478     exit(EXIT_SUCCESS);
 479 }
 480 .EE
 481 .SS Program source: t_get_seals.c
 482 \&
 483 .EX
 484 #define _GNU_SOURCE
 485 #include <sys/mman.h>
 486 #include <fcntl.h>
 487 #include <unistd.h>
 488 #include <stdlib.h>
 489 #include <string.h>
 490 #include <stdio.h>
 491
 492 #define errExit(msg)    do { perror(msg); exit(EXIT_FAILURE); \\
 493                         } while (0)
 494
 495 int
 496 main(int argc, char *argv[])
 497 {
 498     int fd;
 499     unsigned int seals;
 500
 501     if (argc != 2) {
 502         fprintf(stderr, "%s /proc/PID/fd/FD\\n", argv[0]);
 503         exit(EXIT_FAILURE);
 504     }
 505
 506     fd = open(argv[1], O_RDWR);
 507     if (fd == \-1)
 508         errExit("open");
 509
 510     seals = fcntl(fd, F_GET_SEALS);
 511     if (seals == \-1)
 512         errExit("fcntl");
 513
 514     printf("Existing seals:");
 515     if (seals & F_SEAL_SEAL)
 516         printf(" SEAL");
 517     if (seals & F_SEAL_GROW)
 518         printf(" GROW");
 519     if (seals & F_SEAL_WRITE)
 520         printf(" WRITE");
 521     if (seals & F_SEAL_SHRINK)
 522         printf(" SHRINK");
 523     printf("\\n");
 524
 525     /* Code to map the file and access the contents of the
 526        resulting mapping omitted */
 527
 528     exit(EXIT_SUCCESS);
 529 }
 530 .EE
 531 .SH SEE ALSO
 532 .BR fcntl (2),
 533 .BR ftruncate (2),
 534 .BR mmap (2),
 535 .BR shmget (2),
 536 .BR shm_open (3)