1 .\" Copyright (C) 1996 Andries Brouwer <aeb@cwi.nl>
2 .\" and Copyright (C) 2006, 2007 Michael Kerrisk <mtk.manpages@gmail.com>
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.
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.
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15 .\" manual page may be incorrect or out-of-date.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
22 .\" Formatted or processed versions of this manual, if unaccompanied by
23 .\" the source, must acknowledge the copyright and authors of this work.
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
40 .TH MMAP 2 2017-09-15 "Linux" "Linux Programmer's Manual"
42 mmap, munmap \- map or unmap files or devices into memory
45 .B #include <sys/mman.h>
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 );
53 See NOTES for information on feature test macro requirements.
56 creates a new mapping in the virtual address space of
58 The starting address for the new mapping is specified in
62 argument specifies the length of the mapping (which must be greater than 0).
67 then the kernel chooses the address at which to create the mapping;
68 this is the most portable method of creating a new mapping.
72 then the kernel takes it as a hint about where to place the mapping;
73 on Linux, the mapping will be created at a nearby page boundary.
74 .\" Before Linux 2.6.24, the address was rounded up to the next page
75 .\" boundary; since 2.6.24, it is rounded down!
76 The address of the new mapping is returned as the result of the call.
78 The contents of a file mapping (as opposed to an anonymous mapping; see
80 below), are initialized using
82 bytes starting at offset
84 in the file (or other object) referred to by the file descriptor
87 must be a multiple of the page size as returned by
88 .IR sysconf(_SC_PAGE_SIZE) .
92 argument describes the desired memory protection of the mapping
93 (and must not conflict with the open mode of the file).
96 or the bitwise OR of one or more of the following flags:
99 Pages may be executed.
105 Pages may be written.
108 Pages may not be accessed.
112 argument determines whether updates to the mapping
113 are visible to other processes mapping the same region,
114 and whether updates are carried through to the underlying file.
115 This behavior is determined by including exactly one
116 of the following values in
121 Updates to the mapping are visible to other processes mapping the same region,
122 and (in the case of file-backed mappings)
123 are carried through to the underlying file.
124 (To precisely control when updates are carried through
125 to the underlying file requires the use of
129 Create a private copy-on-write mapping.
130 Updates to the mapping are not visible to other processes
131 mapping the same file, and are not carried through to
133 It is unspecified whether changes made to the file after the
135 call are visible in the mapped region.
137 Both of these flags are described in POSIX.1-2001 and POSIX.1-2008.
139 In addition, zero or more of the following values can be ORed in
142 .BR MAP_32BIT " (since Linux 2.4.20, 2.6)"
143 Put the mapping into the first 2 Gigabytes of the process address space.
144 This flag is supported only on x86-64, for 64-bit programs.
145 It was added to allow thread stacks to be allocated somewhere
146 in the first 2\ GB of memory,
147 so as to improve context-switch performance on some early
149 .\" See http://lwn.net/Articles/294642 "Tangled up in threads", 19 Aug 08
150 Modern x86-64 processors no longer have this performance problem,
151 so use of this flag is not required on those systems.
164 The mapping is not backed by any file;
165 its contents are initialized to zero.
169 however, some implementations require
176 and portable applications should ensure this.
179 argument should be zero.
180 .\" See the pgoff overflow check in do_mmap().
181 .\" See the offset check in sys_mmap in arch/x86/kernel/sys_x86_64.c.
186 is supported on Linux only since kernel 2.4.
189 This flag is ignored.
190 .\" Introduced in 1.1.36, removed in 1.3.24.
191 (Long ago, it signaled that attempts to write to the underlying file
194 But this was a source of denial-of-service attacks.)
197 This flag is ignored.
198 .\" Introduced in 1.1.38, removed in 1.3.24. Flag tested in proc_follow_link.
199 .\" (Long ago, it signaled that the underlying file is an executable.
200 .\" However, that information was not really used anywhere.)
201 .\" Linus talked about DOS related to MAP_EXECUTABLE, but he was thinking of
207 .\" On some systems, this was required as the opposite of
208 .\" MAP_ANONYMOUS -- mtk, 1 May 2007
213 as a hint: place the mapping at exactly that address.
215 must be a multiple of the page size.
216 If the memory region specified by
220 overlaps pages of any existing mapping(s), then the overlapped
221 part of the existing mapping(s) will be discarded.
222 If the specified address cannot be used,
225 Because requiring a fixed address for a mapping is less portable,
226 the use of this option is discouraged.
229 This flag is used for stacks.
230 It indicates to the kernel virtual memory system that the mapping
231 should extend downward in memory.
232 The return address is one page lower than the memory area that is
233 actually created in the process's virtual address space.
234 Touching an address in the "guard" page below the mapping will cause
235 the mapping to grow by a page.
236 This growth can be repeated until the mapping grows to within a
237 page of the high end of the next lower mapping,
238 at which point touching the "guard" page will result in a
242 .BR MAP_HUGETLB " (since Linux 2.6.32)"
243 Allocate the mapping using "huge pages."
244 See the Linux kernel source file
245 .I Documentation/vm/hugetlbpage.txt
246 for further information, as well as NOTES, below.
248 .BR MAP_HUGE_2MB ", " MAP_HUGE_1GB " (since Linux 3.8)"
249 .\" See https://lwn.net/Articles/533499/
250 Used in conjunction with
252 to select alternative hugetlb page sizes (respectively, 2\ MB and 1\ GB)
253 on systems that support multiple hugetlb page sizes.
255 More generally, the desired huge page size can be configured by encoding
256 the base-2 logarithm of the desired page size in the six bits at the offset
258 (A value of zero in this bit field provides the default huge page size;
259 the default huge page size can be discovered vie the
263 Thus, the above two constants are defined as:
267 #define MAP_HUGE_2MB (21 << MAP_HUGE_SHIFT)
268 #define MAP_HUGE_1GB (30 << MAP_HUGE_SHIFT)
272 The range of huge page sizes that are supported by the system
273 can be discovered by listing the subdirectories in
274 .IR /sys/kernel/mm/hugepages .
276 .BR MAP_LOCKED " (since Linux 2.5.37)"
277 Mark the mmaped region to be locked in the same way as
279 This implementation will try to populate (prefault) the whole range but
280 the mmap call doesn't fail with
283 Therefore major faults might happen later on.
284 So the semantic is not as strong as
290 when major faults are not acceptable after the initialization of the mapping.
293 flag is ignored in older kernels.
294 .\" If set, the mapped pages will not be swapped out.
296 .BR MAP_NONBLOCK " (since Linux 2.5.46)"
297 This flag is meaningful only in conjunction with
299 Don't perform read-ahead:
300 create page tables entries only for pages
301 that are already present in RAM.
302 Since Linux 2.6.23, this flag causes
305 One day, the combination of
309 may be reimplemented.
312 Do not reserve swap space for this mapping.
313 When swap space is reserved, one has the guarantee
314 that it is possible to modify the mapping.
315 When swap space is not reserved one might get
318 if no physical memory is available.
319 See also the discussion of the file
320 .I /proc/sys/vm/overcommit_memory
323 In kernels before 2.6, this flag had effect only for
324 private writable mappings.
326 .BR MAP_POPULATE " (since Linux 2.5.46)"
327 Populate (prefault) page tables for a mapping.
328 For a file mapping, this causes read-ahead on the file.
329 This will help to reduce blocking on page faults later.
331 is supported for private mappings only since Linux 2.6.23.
333 .BR MAP_STACK " (since Linux 2.6.27)"
334 Allocate the mapping at an address suitable for a process
336 This flag is currently a no-op,
337 but is used in the glibc threading implementation so that
338 if some architectures require special treatment for stack allocations,
339 support can later be transparently implemented for glibc.
340 .\" See http://lwn.net/Articles/294642 "Tangled up in threads", 19 Aug 08
341 .\" commit cd98a04a59e2f94fa64d5bf1e26498d27427d5e7
342 .\" http://thread.gmane.org/gmane.linux.kernel/720412
343 .\" "pthread_create() slow for many threads; also time to revisit 64b
344 .\" context switch optimization?"
346 .BR MAP_UNINITIALIZED " (since Linux 2.6.33)"
347 Don't clear anonymous pages.
348 This flag is intended to improve performance on embedded devices.
349 This flag is honored only if the kernel was configured with the
350 .B CONFIG_MMAP_ALLOW_UNINITIALIZED
352 Because of the security implications,
353 that option is normally enabled only on embedded devices
354 (i.e., devices where one has complete control of the contents of user memory).
356 Of the above flags, only
358 is specified in POSIX.1-2001 and POSIX.1-2008.
359 However, most systems also support
363 .\" FIXME . for later review when Issue 8 is one day released...
364 .\" POSIX may add MAP_ANON in the future
365 .\" http://austingroupbugs.net/tag_view_page.php?tag_id=8
366 .\" http://austingroupbugs.net/view.php?id=850
372 with the same attributes.
374 A file is mapped in multiples of the page size.
375 For a file that is not
376 a multiple of the page size, the remaining memory is zeroed when mapped,
377 and writes to that region are not written out to the file.
379 changing the size of the underlying file of a mapping on the pages that
380 correspond to added or removed regions of the file is unspecified.
384 system call deletes the mappings for the specified address range, and
385 causes further references to addresses within the range to generate
386 invalid memory references.
387 The region is also automatically unmapped
388 when the process is terminated.
389 On the other hand, closing the file
390 descriptor does not unmap the region.
394 must be a multiple of the page size (but
397 All pages containing a part
398 of the indicated range are unmapped, and subsequent references
399 to these pages will generate
401 It is not an error if the
402 indicated range does not contain any mapped pages.
406 returns a pointer to the mapped area.
410 .IR "(void\ *)\ \-1" )
413 is set to indicate the cause of the error.
418 On failure, it returns \-1, and
420 is set to indicate the cause of the error (probably to
425 A file descriptor refers to a non-regular file.
426 Or a file mapping was requested, but
428 is not open for reading.
435 is not open in read/write
440 is set, but the file is append-only.
443 The file has been locked, or too much memory has been locked (see
448 is not a valid file descriptor (and
458 (e.g., they are too large, or not aligned on a page boundary).
471 or contained both of these values.
474 .\" This is for shared anonymous segments
475 .\" [2.6.7] shmem_zero_setup()-->shmem_file_setup()-->get_empty_filp()
476 The system-wide limit on the total number of open files has been reached.
479 .\" A file could not be mapped for reading.
482 The underlying filesystem of the specified file does not support
486 No memory is available.
489 The process's maximum number of mappings would have been exceeded.
490 This error can also occur for
492 when unmapping a region in the middle of an existing mapping,
493 since this results in two smaller mappings on either side of
494 the region being unmapped.
502 would have been exceeded.
505 On 32-bit architecture together with the large file extension
508 the number of pages used for
510 plus number of pages used for
521 but the mapped area belongs to a file on a filesystem that
523 .\" (Since 2.4.25 / 2.6.0.)
526 The operation was prevented by a file seal; see
531 was set but the object specified by
535 Use of a mapped region can result in these signals:
538 Attempted write into a region mapped as read-only.
541 Attempted access to a portion of the buffer that does not correspond
542 to the file (for example, beyond the end of the file, including the
543 case where another process has truncated the file).
545 For an explanation of the terms used in this section, see
551 Interface Attribute Value
555 T} Thread safety MT-Safe
558 POSIX.1-2001, POSIX.1-2008, SVr4, 4.4BSD.
559 .\" SVr4 documents additional error codes ENXIO and ENODEV.
560 .\" SUSv2 documents additional error codes EMFILE and EOVERFLOW.
562 On POSIX systems on which
568 .B _POSIX_MAPPED_FILES
569 is defined in \fI<unistd.h>\fP to a value greater than 0.
572 .\" POSIX.1-2001: It shall be defined to -1 or 0 or 200112L.
573 .\" -1: unavailable, 0: ask using sysconf().
574 .\" glibc defines it to 1.
576 On some hardware architectures (e.g., i386),
580 It is architecture dependent whether
585 Portable programs should always set
587 if they intend to execute code in the new mapping.
589 The portable way to create a mapping is to specify
591 as 0 (NULL), and omit
595 In this case, the system chooses the address for the mapping;
596 the address is chosen so as not to conflict with any existing mapping,
600 flag is specified, and
602 is 0 (NULL), then the mapped address will be 0 (NULL).
606 constants are defined only if suitable feature test macros are defined
607 (possibly by default):
609 with glibc 2.19 or later;
614 in glibc 2.19 and earlier.
618 and requiring that macro specifically would have been more logical,
619 since these flags are all Linux-specific.)
620 The relevant flags are:
637 An application can determine which pages of a mapping are
638 currently resident in the buffer/page cache using
641 .SS Timestamps changes for file-backed mappings
642 For file-backed mappings, the
644 field for the mapped file may be updated at any time between the
646 and the corresponding unmapping; the first reference to a mapped
647 page will update the field if it has not been already.
653 field for a file mapped with
657 will be updated after
658 a write to the mapped region, and before a subsequent
666 .SS Huge page (Huge TLB) mappings
667 For mappings that employ huge pages, the requirements for the arguments of
671 differ somewhat from the requirements for mappings
672 that use the native system page size.
677 must be a multiple of the underlying huge page size.
678 The system automatically aligns
680 to be a multiple of the underlying huge page size.
687 must both be a multiple of the underlying huge page size.
689 .SS C library/kernel differences
690 This page describes the interface provided by the glibc
693 Originally, this function invoked a system call of the same name.
694 Since kernel 2.4, that system call has been superseded by
697 .\" Since around glibc 2.1/2.2, depending on the platform.
700 wrapper function invokes
702 with a suitably adjusted value for
705 On Linux, there are no guarantees like those suggested above under
707 By default, any process can be killed
708 at any moment when the system runs out of memory.
710 In kernels before 2.6.7, the
712 flag has effect only if
722 However, in kernels before 2.6.12,
724 succeeded in this case: no mapping was created and the call returned
732 POSIX specifies that the system shall always
733 zero fill any partial page at the end
734 of the object and that system will never write any modification of the
735 object beyond its end.
736 On Linux, when you write data to such partial page after the end
737 of the object, the data stays in the page cache even after the file
738 is closed and unmapped
739 and even though the data is never written to the file itself,
740 subsequent mappings may see the modified content.
741 In some cases, this could be fixed by calling
743 before the unmap takes place;
744 however, this doesn't work on
746 (for example, when using the POSIX shared memory interface documented in
747 .BR shm_overview (7)).
749 .\" FIXME . Add an example here that uses an anonymous shared region for
750 .\" IPC between parent and child.
752 The following program prints part of the file specified in
753 its first command-line argument to standard output.
754 The range of bytes to be printed is specified via offset and length
755 values in the second and third command-line arguments.
756 The program creates a memory mapping of the required
757 pages of the file and then uses
759 to output the desired bytes.
762 #include <sys/mman.h>
763 #include <sys/stat.h>
769 #define handle_error(msg) \\
770 do { perror(msg); exit(EXIT_FAILURE); } while (0)
773 main(int argc, char *argv[])
778 off_t offset, pa_offset;
782 if (argc < 3 || argc > 4) {
783 fprintf(stderr, "%s file offset [length]\\n", argv[0]);
787 fd = open(argv[1], O_RDONLY);
789 handle_error("open");
791 if (fstat(fd, &sb) == \-1) /* To obtain file size */
792 handle_error("fstat");
794 offset = atoi(argv[2]);
795 pa_offset = offset & ~(sysconf(_SC_PAGE_SIZE) \- 1);
796 /* offset for mmap() must be page aligned */
798 if (offset >= sb.st_size) {
799 fprintf(stderr, "offset is past end of file\\n");
804 length = atoi(argv[3]);
805 if (offset + length > sb.st_size)
806 length = sb.st_size \- offset;
807 /* Can\(aqt display bytes past end of file */
809 } else { /* No length arg ==> display to end of file */
810 length = sb.st_size \- offset;
813 addr = mmap(NULL, length + offset \- pa_offset, PROT_READ,
814 MAP_PRIVATE, fd, pa_offset);
815 if (addr == MAP_FAILED)
816 handle_error("mmap");
818 s = write(STDOUT_FILENO, addr + offset \- pa_offset, length);
821 handle_error("write");
823 fprintf(stderr, "partial write");
827 munmap(addr, length + offset \- pa_offset);
836 .BR memfd_create (2),
843 .BR remap_file_pages (2),
850 The descriptions of the following files in
852 .IR /proc/[pid]/maps ,
853 .IR /proc/[pid]/map_files ,
855 .IR /proc/[pid]/smaps .
857 B.O. Gallmeister, POSIX.4, O'Reilly, pp. 128\(en129 and 389\(en391.
859 .\" Repeat after me: private read-only mappings are 100% equivalent to
860 .\" shared read-only mappings. No ifs, buts, or maybes. -- Linus