man2/mmap.2

   1 .\" Copyright (C) 1996 Andries Brouwer <aeb@cwi.nl>
   2 .\" and Copyright (C) 2006, 2007 Michael Kerrisk <mtk.manpages@gmail.com>
   3 .\"
   4 .\" %%%LICENSE_START(VERBATIM)
   5 .\" Permission is granted to make and distribute verbatim copies of this
   6 .\" manual provided the copyright notice and this permission notice are
   7 .\" preserved on all copies.
   8 .\"
   9 .\" Permission is granted to copy and distribute modified versions of this
  10 .\" manual under the conditions for verbatim copying, provided that the
  11 .\" entire resulting derived work is distributed under the terms of a
  12 .\" permission notice identical to this one.
  13 .\"
  14 .\" Since the Linux kernel and libraries are constantly changing, this
  15 .\" manual page may be incorrect or out-of-date.  The author(s) assume no
  16 .\" responsibility for errors or omissions, or for damages resulting from
  17 .\" the use of the information contained herein.  The author(s) may not
  18 .\" have taken the same level of care in the production of this manual,
  19 .\" which is licensed free of charge, as they might when working
  20 .\" professionally.
  21 .\"
  22 .\" Formatted or processed versions of this manual, if unaccompanied by
  23 .\" the source, must acknowledge the copyright and authors of this work.
  24 .\" %%%LICENSE_END
  25 .\"
  26 .\" Modified 1997-01-31 by Eric S. Raymond <esr@thyrsus.com>
  27 .\" Modified 2000-03-25 by Jim Van Zandt <jrv@vanzandt.mv.com>
  28 .\" Modified 2001-10-04 by John Levon <moz@compsoc.man.ac.uk>
  29 .\" Modified 2003-02-02 by Andi Kleen <ak@muc.de>
  30 .\" Modified 2003-05-21 by Michael Kerrisk <mtk.manpages@gmail.com>
  31 .\"     MAP_LOCKED works from 2.5.37
  32 .\" Modified 2004-06-17 by Michael Kerrisk <mtk.manpages@gmail.com>
  33 .\" Modified 2004-09-11 by aeb
  34 .\" Modified 2004-12-08, from Eric Estievenart <eric.estievenart@free.fr>
  35 .\" Modified 2004-12-08, mtk, formatting tidy-ups
  36 .\" Modified 2006-12-04, mtk, various parts rewritten
  37 .\" 2007-07-10, mtk, Added an example program.
  38 .\" 2008-11-18, mtk, document MAP_STACK
  39 .\"
  40 .TH MMAP 2 2020-04-11 "Linux" "Linux Programmer's Manual"
  41 .SH NAME
  42 mmap, munmap \- map or unmap files or devices into memory
  43 .SH SYNOPSIS
  44 .nf
  45 .B #include <sys/mman.h>
  46 .PP
  47 .BI "void *mmap(void *" addr ", size_t " length \
  48 ", int " prot ", int " flags ,
  49 .BI "           int " fd ", off_t " offset );
  50 .BI "int munmap(void *" addr ", size_t " length );
  51 .fi
  52 .PP
  53 See NOTES for information on feature test macro requirements.
  54 .SH DESCRIPTION
  55 .BR mmap ()
  56 creates a new mapping in the virtual address space of
  57 the calling process.
  58 The starting address for the new mapping is specified in
  59 .IR addr .
  60 The
  61 .I length
  62 argument specifies the length of the mapping (which must be greater than 0).
  63 .PP
  64 If
  65 .I addr
  66 is NULL,
  67 then the kernel chooses the (page-aligned) address
  68 at which to create the mapping;
  69 this is the most portable method of creating a new mapping.
  70 If
  71 .I addr
  72 is not NULL,
  73 then the kernel takes it as a hint about where to place the mapping;
  74 on Linux, the kernel will pick a nearby page boundary (but always above
  75 or equal to the value specified by
  76 .IR /proc/sys/vm/mmap_min_addr )
  77 and attempt to create the mapping there.
  78 If another mapping already exists there, the kernel picks a new address that
  79 may or may not depend on the hint.
  80 .\" Before Linux 2.6.24, the address was rounded up to the next page
  81 .\" boundary; since 2.6.24, it is rounded down!
  82 The address of the new mapping is returned as the result of the call.
  83 .PP
  84 The contents of a file mapping (as opposed to an anonymous mapping; see
  85 .B MAP_ANONYMOUS
  86 below), are initialized using
  87 .I length
  88 bytes starting at offset
  89 .I offset
  90 in the file (or other object) referred to by the file descriptor
  91 .IR fd .
  92 .I offset
  93 must be a multiple of the page size as returned by
  94 .IR sysconf(_SC_PAGE_SIZE) .
  95 .PP
  96 After the
  97 .BR mmap ()
  98 call has returned, the file descriptor,
  99 .IR fd ,
 100 can be closed immediately without invalidating the mapping.
 101 .PP
 102 The
 103 .I prot
 104 argument describes the desired memory protection of the mapping
 105 (and must not conflict with the open mode of the file).
 106 It is either
 107 .B PROT_NONE
 108 or the bitwise OR of one or more of the following flags:
 109 .TP 1.1i
 110 .B PROT_EXEC
 111 Pages may be executed.
 112 .TP
 113 .B PROT_READ
 114 Pages may be read.
 115 .TP
 116 .B PROT_WRITE
 117 Pages may be written.
 118 .TP
 119 .B PROT_NONE
 120 Pages may not be accessed.
 121 .\"
 122 .SS The flags argument
 123 The
 124 .I flags
 125 argument determines whether updates to the mapping
 126 are visible to other processes mapping the same region,
 127 and whether updates are carried through to the underlying file.
 128 This behavior is determined by including exactly one
 129 of the following values in
 130 .IR flags :
 131 .TP
 132 .B MAP_SHARED
 133 Share this mapping.
 134 Updates to the mapping are visible to other processes mapping the same region,
 135 and (in the case of file-backed mappings)
 136 are carried through to the underlying file.
 137 (To precisely control when updates are carried through
 138 to the underlying file requires the use of
 139 .BR msync (2).)
 140 .TP
 141 .BR MAP_SHARED_VALIDATE " (since Linux 4.15)"
 142 This flag provides the same behavior as
 143 .B MAP_SHARED
 144 except that
 145 .B MAP_SHARED
 146 mappings ignore unknown flags in
 147 .IR flags .
 148 By contrast, when creating a mapping using
 149 .BR MAP_SHARED_VALIDATE ,
 150 the kernel verifies all passed flags are known and fails the
 151 mapping with the error
 152 .BR EOPNOTSUPP
 153 for unknown flags.
 154 This mapping type is also required to be able to use some mapping flags
 155 (e.g.,
 156 .BR MAP_SYNC ).
 157 .TP
 158 .B MAP_PRIVATE
 159 Create a private copy-on-write mapping.
 160 Updates to the mapping are not visible to other processes
 161 mapping the same file, and are not carried through to
 162 the underlying file.
 163 It is unspecified whether changes made to the file after the
 164 .BR mmap ()
 165 call are visible in the mapped region.
 166 .PP
 167 Both
 168 .B MAP_SHARED
 169 and
 170 .B MAP_PRIVATE
 171 are described in POSIX.1-2001 and POSIX.1-2008.
 172 .B MAP_SHARED_VALIDATE
 173 is a Linux extension.
 174 .PP
 175 In addition, zero or more of the following values can be ORed in
 176 .IR flags :
 177 .TP
 178 .BR MAP_32BIT " (since Linux 2.4.20, 2.6)"
 179 Put the mapping into the first 2 Gigabytes of the process address space.
 180 This flag is supported only on x86-64, for 64-bit programs.
 181 It was added to allow thread stacks to be allocated somewhere
 182 in the first 2\ GB of memory,
 183 so as to improve context-switch performance on some early
 184 64-bit processors.
 185 .\" See http://lwn.net/Articles/294642 "Tangled up in threads", 19 Aug 08
 186 Modern x86-64 processors no longer have this performance problem,
 187 so use of this flag is not required on those systems.
 188 The
 189 .B MAP_32BIT
 190 flag is ignored when
 191 .B MAP_FIXED
 192 is set.
 193 .TP
 194 .B MAP_ANON
 195 Synonym for
 196 .BR MAP_ANONYMOUS ;
 197 provided for compatibility with other implementations.
 198 .TP
 199 .B MAP_ANONYMOUS
 200 The mapping is not backed by any file;
 201 its contents are initialized to zero.
 202 The
 203 .I fd
 204 argument is ignored;
 205 however, some implementations require
 206 .I fd
 207 to be \-1 if
 208 .B MAP_ANONYMOUS
 209 (or
 210 .BR MAP_ANON )
 211 is specified,
 212 and portable applications should ensure this.
 213 The
 214 .I offset
 215 argument should be zero.
 216 .\" See the pgoff overflow check in do_mmap().
 217 .\" See the offset check in sys_mmap in arch/x86/kernel/sys_x86_64.c.
 218 The use of
 219 .B MAP_ANONYMOUS
 220 in conjunction with
 221 .B MAP_SHARED
 222 is supported on Linux only since kernel 2.4.
 223 .TP
 224 .B MAP_DENYWRITE
 225 This flag is ignored.
 226 .\" Introduced in 1.1.36, removed in 1.3.24.
 227 (Long ago\(emLinux 2.0 and earlier\(emit signaled
 228 that attempts to write to the underlying file should fail with
 229 .BR ETXTBUSY .
 230 But this was a source of denial-of-service attacks.)
 231 .TP
 232 .B MAP_EXECUTABLE
 233 This flag is ignored.
 234 .\" Introduced in 1.1.38, removed in 1.3.24. Flag tested in proc_follow_link.
 235 .\" (Long ago, it signaled that the underlying file is an executable.
 236 .\" However, that information was not really used anywhere.)
 237 .\" Linus talked about DOS related to MAP_EXECUTABLE, but he was thinking of
 238 .\" MAP_DENYWRITE?
 239 .TP
 240 .B MAP_FILE
 241 Compatibility flag.
 242 Ignored.
 243 .\" On some systems, this was required as the opposite of
 244 .\" MAP_ANONYMOUS -- mtk, 1 May 2007
 245 .TP
 246 .B MAP_FIXED
 247 Don't interpret
 248 .I addr
 249 as a hint: place the mapping at exactly that address.
 250 .I addr
 251 must be suitably aligned: for most architectures a multiple of the page
 252 size is sufficient; however, some architectures may impose additional
 253 restrictions.
 254 If the memory region specified by
 255 .I addr
 256 and
 257 .I len
 258 overlaps pages of any existing mapping(s), then the overlapped
 259 part of the existing mapping(s) will be discarded.
 260 If the specified address cannot be used,
 261 .BR mmap ()
 262 will fail.
 263 .IP
 264 Software that aspires to be portable should use the
 265 .BR MAP_FIXED
 266 flag with care,
 267 keeping in mind that the exact layout of a process's memory mappings
 268 is allowed to change significantly between kernel versions,
 269 C library versions, and operating system releases.
 270 .IR "Carefully read the discussion of this flag in NOTES!"
 271 .TP
 272 .BR MAP_FIXED_NOREPLACE " (since Linux 4.17)"
 273 .\" commit a4ff8e8620d3f4f50ac4b41e8067b7d395056843
 274 This flag provides behavior that is similar to
 275 .B MAP_FIXED
 276 with respect to the
 277 .I addr
 278 enforcement, but differs in that
 279 .B MAP_FIXED_NOREPLACE
 280 never clobbers a preexisting mapped range.
 281 If the requested range would collide with an existing mapping,
 282 then this call fails with the error
 283 .B EEXIST.
 284 This flag can therefore be used as a way to atomically
 285 (with respect to other threads) attempt to map an address range:
 286 one thread will succeed; all others will report failure.
 287 .IP
 288 Note that older kernels which do not recognize the
 289 .BR MAP_FIXED_NOREPLACE
 290 flag will typically (upon detecting a collision with a preexisting mapping)
 291 fall back to a "non-\c
 292 .B MAP_FIXED\c
 293 " type of behavior:
 294 they will return an address that is different from the requested address.
 295 Therefore, backward-compatible software
 296 should check the returned address against the requested address.
 297 .TP
 298 .B MAP_GROWSDOWN
 299 This flag is used for stacks.
 300 It indicates to the kernel virtual memory system that the mapping
 301 should extend downward in memory.
 302 The return address is one page lower than the memory area that is
 303 actually created in the process's virtual address space.
 304 Touching an address in the "guard" page below the mapping will cause
 305 the mapping to grow by a page.
 306 This growth can be repeated until the mapping grows to within a
 307 page of the high end of the next lower mapping,
 308 at which point touching the "guard" page will result in a
 309 .B SIGSEGV
 310 signal.
 311 .TP
 312 .BR MAP_HUGETLB " (since Linux 2.6.32)"
 313 Allocate the mapping using "huge pages."
 314 See the Linux kernel source file
 315 .I Documentation/admin-guide/mm/hugetlbpage.rst
 316 for further information, as well as NOTES, below.
 317 .TP
 318 .BR MAP_HUGE_2MB ", " MAP_HUGE_1GB " (since Linux 3.8)"
 319 .\" See https://lwn.net/Articles/533499/
 320 Used in conjunction with
 321 .B MAP_HUGETLB
 322 to select alternative hugetlb page sizes (respectively, 2\ MB and 1\ GB)
 323 on systems that support multiple hugetlb page sizes.
 324 .IP
 325 More generally, the desired huge page size can be configured by encoding
 326 the base-2 logarithm of the desired page size in the six bits at the offset
 327 .BR MAP_HUGE_SHIFT .
 328 (A value of zero in this bit field provides the default huge page size;
 329 the default huge page size can be discovered via the
 330 .I Hugepagesize
 331 field exposed by
 332 .IR /proc/meminfo .)
 333 Thus, the above two constants are defined as:
 334 .IP
 335 .in +4n
 336 .EX
 337 #define MAP_HUGE_2MB    (21 << MAP_HUGE_SHIFT)
 338 #define MAP_HUGE_1GB    (30 << MAP_HUGE_SHIFT)
 339 .EE
 340 .in
 341 .IP
 342 The range of huge page sizes that are supported by the system
 343 can be discovered by listing the subdirectories in
 344 .IR /sys/kernel/mm/hugepages .
 345 .TP
 346 .BR MAP_LOCKED " (since Linux 2.5.37)"
 347 Mark the mapped region to be locked in the same way as
 348 .BR mlock (2).
 349 This implementation will try to populate (prefault) the whole range but the
 350 .BR mmap ()
 351 call doesn't fail with
 352 .B ENOMEM
 353 if this fails.
 354 Therefore major faults might happen later on.
 355 So the semantic is not as strong as
 356 .BR mlock (2).
 357 One should use
 358 .BR mmap ()
 359 plus
 360 .BR mlock (2)
 361 when major faults are not acceptable after the initialization of the mapping.
 362 The
 363 .BR MAP_LOCKED
 364 flag is ignored in older kernels.
 365 .\" If set, the mapped pages will not be swapped out.
 366 .TP
 367 .BR MAP_NONBLOCK " (since Linux 2.5.46)"
 368 This flag is meaningful only in conjunction with
 369 .BR MAP_POPULATE .
 370 Don't perform read-ahead:
 371 create page tables entries only for pages
 372 that are already present in RAM.
 373 Since Linux 2.6.23,
 374 .\" commit 54cb8821de07f2ffcd28c380ce9b93d5784b40d7
 375 this flag causes
 376 .BR MAP_POPULATE
 377 to do nothing.
 378 One day, the combination of
 379 .BR MAP_POPULATE
 380 and
 381 .BR MAP_NONBLOCK
 382 may be reimplemented.
 383 .TP
 384 .B MAP_NORESERVE
 385 Do not reserve swap space for this mapping.
 386 When swap space is reserved, one has the guarantee
 387 that it is possible to modify the mapping.
 388 When swap space is not reserved one might get
 389 .B SIGSEGV
 390 upon a write
 391 if no physical memory is available.
 392 See also the discussion of the file
 393 .I /proc/sys/vm/overcommit_memory
 394 in
 395 .BR proc (5).
 396 In kernels before 2.6, this flag had effect only for
 397 private writable mappings.
 398 .TP
 399 .BR MAP_POPULATE " (since Linux 2.5.46)"
 400 Populate (prefault) page tables for a mapping.
 401 For a file mapping, this causes read-ahead on the file.
 402 This will help to reduce blocking on page faults later.
 403 .BR MAP_POPULATE
 404 is supported for private mappings only since Linux 2.6.23.
 405 .TP
 406 .BR MAP_STACK " (since Linux 2.6.27)"
 407 Allocate the mapping at an address suitable for a process
 408 or thread stack.
 409 .IP
 410 This flag is currently a no-op on Linux.
 411 However, by employing this flag, applications can ensure that
 412 they transparently obtain support if the flag
 413 is implemented in the future.
 414 Thus, it is used in the glibc threading implementation to allow for
 415 the fact that some architectures may (later) require special treatment
 416 for stack allocations.
 417 .\" See http://lwn.net/Articles/294642 "Tangled up in threads", 19 Aug 08
 418 .\" commit cd98a04a59e2f94fa64d5bf1e26498d27427d5e7
 419 .\" http://thread.gmane.org/gmane.linux.kernel/720412
 420 .\" "pthread_create() slow for many threads; also time to revisit 64b
 421 .\"  context switch optimization?"
 422 A further reason to employ this flag is portability:
 423 .BR MAP_STACK
 424 exists (and has an effect) on some other systems (e.g., some of the BSDs).
 425 .TP
 426 .BR MAP_SYNC " (since Linux 4.15)"
 427 This flag is available only with the
 428 .B MAP_SHARED_VALIDATE
 429 mapping type;
 430 mappings of type
 431 .B MAP_SHARED
 432 will silently ignore this flag.
 433 This flag is supported only for files supporting DAX
 434 (direct mapping of persistent memory).
 435 For other files, creating a mapping with this flag results in an
 436 .B EOPNOTSUPP
 437 error.
 438 .IP
 439 Shared file mappings with this flag provide the guarantee that while
 440 some memory is writably mapped in the address space of the process,
 441 it will be visible in the same file at the same offset even after
 442 the system crashes or is rebooted.
 443 In conjunction with the use of appropriate CPU instructions,
 444 this provides users of such mappings with a more efficient way
 445 of making data modifications persistent.
 446 .TP
 447 .BR MAP_UNINITIALIZED " (since Linux 2.6.33)"
 448 Don't clear anonymous pages.
 449 This flag is intended to improve performance on embedded devices.
 450 This flag is honored only if the kernel was configured with the
 451 .B CONFIG_MMAP_ALLOW_UNINITIALIZED
 452 option.
 453 Because of the security implications,
 454 that option is normally enabled only on embedded devices
 455 (i.e., devices where one has complete control of the contents of user memory).
 456 .PP
 457 Of the above flags, only
 458 .B MAP_FIXED
 459 is specified in POSIX.1-2001 and POSIX.1-2008.
 460 However, most systems also support
 461 .B MAP_ANONYMOUS
 462 (or its synonym
 463 .BR MAP_ANON ).
 464 .\" FIXME . for later review when Issue 8 is one day released...
 465 .\" POSIX may add MAP_ANON in the future
 466 .\" http://austingroupbugs.net/tag_view_page.php?tag_id=8
 467 .\" http://austingroupbugs.net/view.php?id=850
 468 .SS munmap()
 469 The
 470 .BR munmap ()
 471 system call deletes the mappings for the specified address range, and
 472 causes further references to addresses within the range to generate
 473 invalid memory references.
 474 The region is also automatically unmapped
 475 when the process is terminated.
 476 On the other hand, closing the file
 477 descriptor does not unmap the region.
 478 .PP
 479 The address
 480 .I addr
 481 must be a multiple of the page size (but
 482 .I length
 483 need not be).
 484 All pages containing a part
 485 of the indicated range are unmapped, and subsequent references
 486 to these pages will generate
 487 .BR SIGSEGV .
 488 It is not an error if the
 489 indicated range does not contain any mapped pages.
 490 .SH RETURN VALUE
 491 On success,
 492 .BR mmap ()
 493 returns a pointer to the mapped area.
 494 On error, the value
 495 .B MAP_FAILED
 496 (that is,
 497 .IR "(void\ *)\ \-1" )
 498 is returned, and
 499 .I errno
 500 is set to indicate the cause of the error.
 501 .PP
 502 On success,
 503 .BR munmap ()
 504 returns 0.
 505 On failure, it returns \-1, and
 506 .I errno
 507 is set to indicate the cause of the error (probably to
 508 .BR EINVAL ).
 509 .SH ERRORS
 510 .TP
 511 .B EACCES
 512 A file descriptor refers to a non-regular file.
 513 Or a file mapping was requested, but
 514 .I fd
 515 is not open for reading.
 516 Or
 517 .B MAP_SHARED
 518 was requested and
 519 .B PROT_WRITE
 520 is set, but
 521 .I fd
 522 is not open in read/write
 523 .RB ( O_RDWR )
 524 mode.
 525 Or
 526 .B PROT_WRITE
 527 is set, but the file is append-only.
 528 .TP
 529 .B EAGAIN
 530 The file has been locked, or too much memory has been locked (see
 531 .BR setrlimit (2)).
 532 .TP
 533 .B EBADF
 534 .I fd
 535 is not a valid file descriptor (and
 536 .B MAP_ANONYMOUS
 537 was not set).
 538 .TP
 539 .B EEXIST
 540 .BR MAP_FIXED_NOREPLACE
 541 was specified in
 542 .IR flags ,
 543 and the range covered by
 544 .IR addr
 545 and
 546 .IR length
 547 clashes with an existing mapping.
 548 .TP
 549 .B EINVAL
 550 We don't like
 551 .IR addr ,
 552 .IR length ,
 553 or
 554 .I offset
 555 (e.g., they are too large, or not aligned on a page boundary).
 556 .TP
 557 .B EINVAL
 558 (since Linux 2.6.12)
 559 .I length
 560 was 0.
 561 .TP
 562 .B EINVAL
 563 .I flags
 564 contained none of
 565 .BR MAP_PRIVATE ,
 566 .B MAP_SHARED
 567 or
 568 .BR MAP_SHARED_VALIDATE .
 569 .TP
 570 .B ENFILE
 571 .\" This is for shared anonymous segments
 572 .\" [2.6.7] shmem_zero_setup()-->shmem_file_setup()-->get_empty_filp()
 573 The system-wide limit on the total number of open files has been reached.
 574 .\" .TP
 575 .\" .B ENOEXEC
 576 .\" A file could not be mapped for reading.
 577 .TP
 578 .B ENODEV
 579 The underlying filesystem of the specified file does not support
 580 memory mapping.
 581 .TP
 582 .B ENOMEM
 583 No memory is available.
 584 .TP
 585 .B ENOMEM
 586 The process's maximum number of mappings would have been exceeded.
 587 This error can also occur for
 588 .BR munmap (),
 589 when unmapping a region in the middle of an existing mapping,
 590 since this results in two smaller mappings on either side of
 591 the region being unmapped.
 592 .TP
 593 .B ENOMEM
 594 (since Linux 4.7)
 595 The process's
 596 .B RLIMIT_DATA
 597 limit, described in
 598 .BR getrlimit (2),
 599 would have been exceeded.
 600 .TP
 601 .B EOVERFLOW
 602 On 32-bit architecture together with the large file extension
 603 (i.e., using 64-bit
 604 .IR off_t ):
 605 the number of pages used for
 606 .I length
 607 plus number of pages used for
 608 .I offset
 609 would overflow
 610 .I "unsigned long"
 611 (32 bits).
 612 .TP
 613 .B EPERM
 614 The
 615 .I prot
 616 argument asks for
 617 .B PROT_EXEC
 618 but the mapped area belongs to a file on a filesystem that
 619 was mounted no-exec.
 620 .\" (Since 2.4.25 / 2.6.0.)
 621 .TP
 622 .B EPERM
 623 The operation was prevented by a file seal; see
 624 .BR fcntl (2).
 625 .TP
 626 .B ETXTBSY
 627 .B MAP_DENYWRITE
 628 was set but the object specified by
 629 .I fd
 630 is open for writing.
 631 .PP
 632 Use of a mapped region can result in these signals:
 633 .TP
 634 .B SIGSEGV
 635 Attempted write into a region mapped as read-only.
 636 .TP
 637 .B SIGBUS
 638 Attempted access to a portion of the buffer that does not correspond
 639 to the file (for example, beyond the end of the file, including the
 640 case where another process has truncated the file).
 641 .SH ATTRIBUTES
 642 For an explanation of the terms used in this section, see
 643 .BR attributes (7).
 644 .TS
 645 allbox;
 646 lbw18 lb lb
 647 l l l.
 648 Interface       Attribute       Value
 649 T{
 650 .BR mmap (),
 651 .BR munmap ()
 652 T}      Thread safety   MT-Safe
 653 .TE
 654 .SH CONFORMING TO
 655 POSIX.1-2001, POSIX.1-2008, SVr4, 4.4BSD.
 656 .\" SVr4 documents additional error codes ENXIO and ENODEV.
 657 .\" SUSv2 documents additional error codes EMFILE and EOVERFLOW.
 658 .PP
 659 On POSIX systems on which
 660 .BR mmap (),
 661 .BR msync (2),
 662 and
 663 .BR munmap ()
 664 are available,
 665 .B _POSIX_MAPPED_FILES
 666 is defined in \fI<unistd.h>\fP to a value greater than 0.
 667 (See also
 668 .BR sysconf (3).)
 669 .\" POSIX.1-2001: It shall be defined to -1 or 0 or 200112L.
 670 .\" -1: unavailable, 0: ask using sysconf().
 671 .\" glibc defines it to 1.
 672 .SH NOTES
 673 .PP
 674 Memory mapped by
 675 .BR mmap ()
 676 is preserved across
 677 .BR fork (2),
 678 with the same attributes.
 679 .PP
 680 A file is mapped in multiples of the page size.
 681 For a file that is not
 682 a multiple of the page size, the remaining memory is zeroed when mapped,
 683 and writes to that region are not written out to the file.
 684 The effect of
 685 changing the size of the underlying file of a mapping on the pages that
 686 correspond to added or removed regions of the file is unspecified.
 687 .PP
 688 On some hardware architectures (e.g., i386),
 689 .B PROT_WRITE
 690 implies
 691 .BR PROT_READ .
 692 It is architecture dependent whether
 693 .B PROT_READ
 694 implies
 695 .B PROT_EXEC
 696 or not.
 697 Portable programs should always set
 698 .B PROT_EXEC
 699 if they intend to execute code in the new mapping.
 700 .PP
 701 The portable way to create a mapping is to specify
 702 .I addr
 703 as 0 (NULL), and omit
 704 .B MAP_FIXED
 705 from
 706 .IR flags .
 707 In this case, the system chooses the address for the mapping;
 708 the address is chosen so as not to conflict with any existing mapping,
 709 and will not be 0.
 710 If the
 711 .B MAP_FIXED
 712 flag is specified, and
 713 .I addr
 714 is 0 (NULL), then the mapped address will be 0 (NULL).
 715 .PP
 716 Certain
 717 .I flags
 718 constants are defined only if suitable feature test macros are defined
 719 (possibly by default):
 720 .BR _DEFAULT_SOURCE
 721 with glibc 2.19 or later;
 722 or
 723 .BR _BSD_SOURCE
 724 or
 725 .BR _SVID_SOURCE
 726 in glibc 2.19 and earlier.
 727 (Employing
 728 .BR _GNU_SOURCE
 729 also suffices,
 730 and requiring that macro specifically would have been more logical,
 731 since these flags are all Linux-specific.)
 732 The relevant flags are:
 733 .BR MAP_32BIT ,
 734 .BR MAP_ANONYMOUS
 735 (and the synonym
 736 .BR MAP_ANON ),
 737 .BR MAP_DENYWRITE ,
 738 .BR MAP_EXECUTABLE ,
 739 .BR MAP_FILE ,
 740 .BR MAP_GROWSDOWN ,
 741 .BR MAP_HUGETLB ,
 742 .BR MAP_LOCKED ,
 743 .BR MAP_NONBLOCK ,
 744 .BR MAP_NORESERVE ,
 745 .BR MAP_POPULATE ,
 746 and
 747 .BR MAP_STACK .
 748 .PP
 749 An application can determine which pages of a mapping are
 750 currently resident in the buffer/page cache using
 751 .BR mincore (2).
 752 .\"
 753 .SS Using MAP_FIXED safely
 754 The only safe use for
 755 .BR MAP_FIXED
 756 is where the address range specified by
 757 .IR addr
 758 and
 759 .I length
 760 was previously reserved using another mapping;
 761 otherwise, the use of
 762 .BR MAP_FIXED
 763 is hazardous because it forcibly removes preexisting mappings,
 764 making it easy for a multithreaded process to corrupt its own address space.
 765 .PP
 766 For example, suppose that thread A looks through
 767 .I /proc/<pid>/maps
 768 and in order to locate an unused address range that it can map using
 769 .BR MAP_FIXED ,
 770 while thread B simultaneously acquires part or all of that same
 771 address range.
 772 When thread A subsequently employs
 773 .BR mmap(MAP_FIXED) ,
 774 it will effectively clobber the mapping that thread B created.
 775 In this scenario,
 776 thread B need not create a mapping directly; simply making a library call
 777 that, internally, uses
 778 .BR dlopen (3)
 779 to load some other shared library, will suffice.
 780 The
 781 .BR dlopen (3)
 782 call will map the library into the process's address space.
 783 Furthermore, almost any library call may be implemented in a way that
 784 adds memory mappings to the address space, either with this technique,
 785 or by simply allocating memory.
 786 Examples include
 787 .BR brk (2),
 788 .BR malloc (3),
 789 .BR pthread_create (3),
 790 and the PAM libraries
 791 .UR http://www.linux-pam.org
 792 .UE .
 793 .PP
 794 Since Linux 4.17, a multithreaded program can use the
 795 .BR MAP_FIXED_NOREPLACE
 796 flag to avoid the hazard described above
 797 when attempting to create a mapping at a fixed address
 798 that has not been reserved by a preexisting mapping.
 799 .\"
 800 .SS Timestamps changes for file-backed mappings
 801 For file-backed mappings, the
 802 .I st_atime
 803 field for the mapped file may be updated at any time between the
 804 .BR mmap ()
 805 and the corresponding unmapping; the first reference to a mapped
 806 page will update the field if it has not been already.
 807 .PP
 808 The
 809 .I st_ctime
 810 and
 811 .I st_mtime
 812 field for a file mapped with
 813 .B PROT_WRITE
 814 and
 815 .B MAP_SHARED
 816 will be updated after
 817 a write to the mapped region, and before a subsequent
 818 .BR msync (2)
 819 with the
 820 .B MS_SYNC
 821 or
 822 .B MS_ASYNC
 823 flag, if one occurs.
 824 .\"
 825 .SS Huge page (Huge TLB) mappings
 826 For mappings that employ huge pages, the requirements for the arguments of
 827 .BR mmap ()
 828 and
 829 .BR munmap ()
 830 differ somewhat from the requirements for mappings
 831 that use the native system page size.
 832 .PP
 833 For
 834 .BR mmap (),
 835 .I offset
 836 must be a multiple of the underlying huge page size.
 837 The system automatically aligns
 838 .I length
 839 to be a multiple of the underlying huge page size.
 840 .PP
 841 For
 842 .BR munmap (),
 843 .I addr
 844 and
 845 .I length
 846 must both be a multiple of the underlying huge page size.
 847 .\"
 848 .SS C library/kernel differences
 849 This page describes the interface provided by the glibc
 850 .BR mmap ()
 851 wrapper function.
 852 Originally, this function invoked a system call of the same name.
 853 Since kernel 2.4, that system call has been superseded by
 854 .BR mmap2 (2),
 855 and nowadays
 856 .\" Since around glibc 2.1/2.2, depending on the platform.
 857 the glibc
 858 .BR mmap ()
 859 wrapper function invokes
 860 .BR mmap2 (2)
 861 with a suitably adjusted value for
 862 .IR offset .
 863 .SH BUGS
 864 On Linux, there are no guarantees like those suggested above under
 865 .BR MAP_NORESERVE .
 866 By default, any process can be killed
 867 at any moment when the system runs out of memory.
 868 .PP
 869 In kernels before 2.6.7, the
 870 .B MAP_POPULATE
 871 flag has effect only if
 872 .I prot
 873 is specified as
 874 .BR PROT_NONE .
 875 .PP
 876 SUSv3 specifies that
 877 .BR mmap ()
 878 should fail if
 879 .I length
 880 is 0.
 881 However, in kernels before 2.6.12,
 882 .BR mmap ()
 883 succeeded in this case: no mapping was created and the call returned
 884 .IR addr .
 885 Since kernel 2.6.12,
 886 .BR mmap ()
 887 fails with the error
 888 .B EINVAL
 889 for this case.
 890 .PP
 891 POSIX specifies that the system shall always
 892 zero fill any partial page at the end
 893 of the object and that system will never write any modification of the
 894 object beyond its end.
 895 On Linux, when you write data to such partial page after the end
 896 of the object, the data stays in the page cache even after the file
 897 is closed and unmapped
 898 and even though the data is never written to the file itself,
 899 subsequent mappings may see the modified content.
 900 In some cases, this could be fixed by calling
 901 .BR msync (2)
 902 before the unmap takes place;
 903 however, this doesn't work on
 904 .BR tmpfs (5)
 905 (for example, when using the POSIX shared memory interface documented in
 906 .BR shm_overview (7)).
 907 .SH EXAMPLES
 908 .\" FIXME . Add an example here that uses an anonymous shared region for
 909 .\" IPC between parent and child.
 910 .PP
 911 The following program prints part of the file specified in
 912 its first command-line argument to standard output.
 913 The range of bytes to be printed is specified via offset and length
 914 values in the second and third command-line arguments.
 915 The program creates a memory mapping of the required
 916 pages of the file and then uses
 917 .BR write (2)
 918 to output the desired bytes.
 919 .SS Program source
 920 .EX
 921 #include <sys/mman.h>
 922 #include <sys/stat.h>
 923 #include <fcntl.h>
 924 #include <stdio.h>
 925 #include <stdlib.h>
 926 #include <unistd.h>
 927
 928 #define handle_error(msg) \e
 929     do { perror(msg); exit(EXIT_FAILURE); } while (0)
 930
 931 int
 932 main(int argc, char *argv[])
 933 {
 934     char *addr;
 935     int fd;
 936     struct stat sb;
 937     off_t offset, pa_offset;
 938     size_t length;
 939     ssize_t s;
 940
 941     if (argc < 3 || argc > 4) {
 942         fprintf(stderr, "%s file offset [length]\en", argv[0]);
 943         exit(EXIT_FAILURE);
 944     }
 945
 946     fd = open(argv[1], O_RDONLY);
 947     if (fd == \-1)
 948         handle_error("open");
 949
 950     if (fstat(fd, &sb) == \-1)           /* To obtain file size */
 951         handle_error("fstat");
 952
 953     offset = atoi(argv[2]);
 954     pa_offset = offset & ~(sysconf(_SC_PAGE_SIZE) \- 1);
 955         /* offset for mmap() must be page aligned */
 956
 957     if (offset >= sb.st_size) {
 958         fprintf(stderr, "offset is past end of file\en");
 959         exit(EXIT_FAILURE);
 960     }
 961
 962     if (argc == 4) {
 963         length = atoi(argv[3]);
 964         if (offset + length > sb.st_size)
 965             length = sb.st_size \- offset;
 966                 /* Can\(aqt display bytes past end of file */
 967
 968     } else {    /* No length arg ==> display to end of file */
 969         length = sb.st_size \- offset;
 970     }
 971
 972     addr = mmap(NULL, length + offset \- pa_offset, PROT_READ,
 973                 MAP_PRIVATE, fd, pa_offset);
 974     if (addr == MAP_FAILED)
 975         handle_error("mmap");
 976
 977     s = write(STDOUT_FILENO, addr + offset \- pa_offset, length);
 978     if (s != length) {
 979         if (s == \-1)
 980             handle_error("write");
 981
 982         fprintf(stderr, "partial write");
 983         exit(EXIT_FAILURE);
 984     }
 985
 986     munmap(addr, length + offset \- pa_offset);
 987     close(fd);
 988
 989     exit(EXIT_SUCCESS);
 990 }
 991 .EE
 992 .SH SEE ALSO
 993 .BR ftruncate (2),
 994 .BR getpagesize (2),
 995 .BR memfd_create (2),
 996 .BR mincore (2),
 997 .BR mlock (2),
 998 .BR mmap2 (2),
 999 .BR mprotect (2),
1000 .BR mremap (2),
1001 .BR msync (2),
1002 .BR remap_file_pages (2),
1003 .BR setrlimit (2),
1004 .BR shmat (2),
1005 .BR userfaultfd (2),
1006 .BR shm_open (3),
1007 .BR shm_overview (7)
1008 .PP
1009 The descriptions of the following files in
1010 .BR proc (5):
1011 .IR /proc/[pid]/maps ,
1012 .IR /proc/[pid]/map_files ,
1013 and
1014 .IR /proc/[pid]/smaps .
1015 .PP
1016 B.O. Gallmeister, POSIX.4, O'Reilly, pp. 128\(en129 and 389\(en391.
1017 .\"
1018 .\" Repeat after me: private read-only mappings are 100% equivalent to
1019 .\" shared read-only mappings. No ifs, buts, or maybes. -- Linus