xref: /freebsd-14.2/lib/libc/sys/mmap.2 (revision b2c76c41)
1.\" Copyright (c) 1991, 1993
2.\"	The Regents of the University of California.  All rights reserved.
3.\"
4.\" Redistribution and use in source and binary forms, with or without
5.\" modification, are permitted provided that the following conditions
6.\" are met:
7.\" 1. Redistributions of source code must retain the above copyright
8.\"    notice, this list of conditions and the following disclaimer.
9.\" 2. Redistributions in binary form must reproduce the above copyright
10.\"    notice, this list of conditions and the following disclaimer in the
11.\"    documentation and/or other materials provided with the distribution.
12.\" 3. Neither the name of the University nor the names of its contributors
13.\"    may be used to endorse or promote products derived from this software
14.\"    without specific prior written permission.
15.\"
16.\" THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
17.\" ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18.\" IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19.\" ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
20.\" FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21.\" DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22.\" OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23.\" HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24.\" LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25.\" OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26.\" SUCH DAMAGE.
27.\"
28.\"	@(#)mmap.2	8.4 (Berkeley) 5/11/95
29.\"
30.Dd August 14, 2023
31.Dt MMAP 2
32.Os
33.Sh NAME
34.Nm mmap
35.Nd allocate memory, or map files or devices into memory
36.Sh LIBRARY
37.Lb libc
38.Sh SYNOPSIS
39.In sys/mman.h
40.Ft void *
41.Fn mmap "void *addr" "size_t len" "int prot" "int flags" "int fd" "off_t offset"
42.Sh DESCRIPTION
43The
44.Fn mmap
45system call causes the pages starting at
46.Fa addr
47and continuing for at most
48.Fa len
49bytes to be mapped from the object described by
50.Fa fd ,
51starting at byte offset
52.Fa offset .
53If
54.Fa len
55is not a multiple of the page size, the mapped region may extend past the
56specified range.
57Any such extension beyond the end of the mapped object will be zero-filled.
58.Pp
59If
60.Fa fd
61references a regular file or a shared memory object, the range of
62bytes starting at
63.Fa offset
64and continuing for
65.Fa len
66bytes must be legitimate for the possible (not necessarily
67current) offsets in the object.
68In particular, the
69.Fa offset
70value cannot be negative.
71If the object is truncated and the process later accesses a page that
72is wholly within the truncated region, the access is aborted and a
73.Dv SIGBUS
74signal is delivered to the process.
75.Pp
76If
77.Fa fd
78references a device file, the interpretation of the
79.Fa offset
80value is device specific and defined by the device driver.
81The virtual memory subsystem does not impose any restrictions on the
82.Fa offset
83value in this case, passing it unchanged to the driver.
84.Pp
85If
86.Fa addr
87is non-zero, it is used as a hint to the system.
88(As a convenience to the system, the actual address of the region may differ
89from the address supplied.)
90If
91.Fa addr
92is zero, an address will be selected by the system.
93The actual starting address of the region is returned.
94A successful
95.Fa mmap
96deletes any previous mapping in the allocated address range.
97.Pp
98The protections (region accessibility) are specified in the
99.Fa prot
100argument by
101.Em or Ns 'ing
102the following values:
103.Pp
104.Bl -tag -width PROT_WRITE -compact
105.It Dv PROT_NONE
106Pages may not be accessed.
107.It Dv PROT_READ
108Pages may be read.
109.It Dv PROT_WRITE
110Pages may be written.
111.It Dv PROT_EXEC
112Pages may be executed.
113.El
114.Pp
115In addition to these protection flags,
116.Fx
117provides the ability to set the maximum protection of a region allocated by
118.Nm
119and later altered by
120.Xr mprotect 2 .
121This is accomplished by
122.Em or Ns 'ing
123one or more
124.Dv PROT_
125values wrapped in the
126.Dv PROT_MAX()
127macro into the
128.Fa prot
129argument.
130.Pp
131The
132.Fa flags
133argument specifies the type of the mapped object, mapping options and
134whether modifications made to the mapped copy of the page are private
135to the process or are to be shared with other references.
136Sharing, mapping type and options are specified in the
137.Fa flags
138argument by
139.Em or Ns 'ing
140the following values:
141.Bl -tag -width MAP_PREFAULT_READ
142.It Dv MAP_32BIT
143Request a region in the first 2GB of the current process's address space.
144If a suitable region cannot be found,
145.Fn mmap
146will fail.
147.It Dv MAP_ALIGNED Ns Pq Fa n
148Align the region on a requested boundary.
149If a suitable region cannot be found,
150.Fn mmap
151will fail.
152The
153.Fa n
154argument specifies the binary logarithm of the desired alignment.
155.It Dv MAP_ALIGNED_SUPER
156Align the region to maximize the potential use of large
157.Pq Dq super
158pages.
159If a suitable region cannot be found,
160.Fn mmap
161will fail.
162The system will choose a suitable page size based on the size of
163mapping.
164The page size used as well as the alignment of the region may both be
165affected by properties of the file being mapped.
166In particular,
167the physical address of existing pages of a file may require a specific
168alignment.
169The region is not guaranteed to be aligned on any specific boundary.
170.It Dv MAP_ANON
171Map anonymous memory not associated with any specific file.
172The file descriptor used for creating
173.Dv MAP_ANON
174must be \-1.
175The
176.Fa offset
177argument must be 0.
178.\".It Dv MAP_FILE
179.\"Mapped from a regular file or character-special device memory.
180.It Dv MAP_ANONYMOUS
181This flag is identical to
182.Dv MAP_ANON
183and is provided for compatibility.
184.It Dv MAP_EXCL
185This flag can only be used in combination with
186.Dv MAP_FIXED .
187Please see the definition of
188.Dv MAP_FIXED
189for the description of its effect.
190.It Dv MAP_FIXED
191Do not permit the system to select a different address than the one
192specified.
193If the specified address cannot be used,
194.Fn mmap
195will fail.
196If
197.Dv MAP_FIXED
198is specified,
199.Fa addr
200must be a multiple of the page size.
201If
202.Dv MAP_EXCL
203is not specified, a successful
204.Dv MAP_FIXED
205request replaces any previous mappings for the process'
206pages in the range from
207.Fa addr
208to
209.Fa addr
210+
211.Fa len .
212In contrast, if
213.Dv MAP_EXCL
214is specified, the request will fail if a mapping
215already exists within the range.
216.It Dv MAP_GUARD
217Instead of a mapping, create a guard of the specified size.
218Guards allow a process to create reservations in its address space,
219which can later be replaced by actual mappings.
220.Pp
221.Fa mmap
222will not create mappings in the address range of a guard unless
223the request specifies
224.Dv MAP_FIXED .
225Guards can be destroyed with
226.Xr munmap 2 .
227Any memory access by a thread to the guarded range results
228in the delivery of a
229.Dv SIGSEGV
230signal to that thread.
231.It Dv MAP_NOCORE
232Region is not included in a core file.
233.It Dv MAP_NOSYNC
234Causes data dirtied via this VM map to be flushed to physical media
235only when necessary (usually by the pager) rather than gratuitously.
236Typically this prevents the update daemons from flushing pages dirtied
237through such maps and thus allows efficient sharing of memory across
238unassociated processes using a file-backed shared memory map.
239Without
240this option any VM pages you dirty may be flushed to disk every so often
241(every 30-60 seconds usually) which can create performance problems if you
242do not need that to occur (such as when you are using shared file-backed
243mmap regions for IPC purposes).
244Dirty data will be flushed automatically when all mappings of an object are
245removed and all descriptors referencing the object are closed.
246Note that VM/file system coherency is
247maintained whether you use
248.Dv MAP_NOSYNC
249or not.
250This option is not portable
251across
252.Ux
253platforms (yet), though some may implement the same behavior
254by default.
255.Pp
256.Em WARNING !
257Extending a file with
258.Xr ftruncate 2 ,
259thus creating a big hole, and then filling the hole by modifying a shared
260.Fn mmap
261can lead to severe file fragmentation.
262In order to avoid such fragmentation you should always pre-allocate the
263file's backing store by
264.Fn write Ns ing
265zero's into the newly extended area prior to modifying the area via your
266.Fn mmap .
267The fragmentation problem is especially sensitive to
268.Dv MAP_NOSYNC
269pages, because pages may be flushed to disk in a totally random order.
270.Pp
271The same applies when using
272.Dv MAP_NOSYNC
273to implement a file-based shared memory store.
274It is recommended that you create the backing store by
275.Fn write Ns ing
276zero's to the backing file rather than
277.Fn ftruncate Ns ing
278it.
279You can test file fragmentation by observing the KB/t (kilobytes per
280transfer) results from an
281.Dq Li iostat 1
282while reading a large file sequentially, e.g.,\& using
283.Dq Li dd if=filename of=/dev/null bs=32k .
284.Pp
285The
286.Xr fsync 2
287system call will flush all dirty data and metadata associated with a file,
288including dirty NOSYNC VM data, to physical media.
289The
290.Xr sync 8
291command and
292.Xr sync 2
293system call generally do not flush dirty NOSYNC VM data.
294The
295.Xr msync 2
296system call is usually not needed since
297.Bx
298implements a coherent file system buffer cache.
299However, it may be
300used to associate dirty VM pages with file system buffers and thus cause
301them to be flushed to physical media sooner rather than later.
302.It Dv MAP_PREFAULT_READ
303Immediately update the calling process's lowest-level virtual address
304translation structures, such as its page table, so that every memory
305resident page within the region is mapped for read access.
306Ordinarily these structures are updated lazily.
307The effect of this option is to eliminate any soft faults that would
308otherwise occur on the initial read accesses to the region.
309Although this option does not preclude
310.Fa prot
311from including
312.Dv PROT_WRITE ,
313it does not eliminate soft faults on the initial write accesses to the
314region.
315.It Dv MAP_PRIVATE
316Modifications are private.
317.It Dv MAP_SHARED
318Modifications are shared.
319.It Dv MAP_STACK
320Creates both a mapped region that grows downward on demand and an
321adjoining guard that both reserves address space for the mapped region
322to grow into and limits the mapped region's growth.
323Together, the mapped region and the guard occupy
324.Fa len
325bytes of the address space.
326The guard starts at the returned address, and the mapped region ends at
327the returned address plus
328.Fa len
329bytes.
330Upon access to the guard, the mapped region automatically grows in size,
331and the guard shrinks by an equal amount.
332Essentially, the boundary between the guard and the mapped region moves
333downward so that the access falls within the enlarged mapped region.
334However, the guard will never shrink to less than the number of pages
335specified by the sysctl
336.Dv security.bsd.stack_guard_page ,
337thereby ensuring that a gap for detecting stack overflow always exists
338between the downward growing mapped region and the closest mapped region
339beneath it.
340.Pp
341.Dv MAP_STACK
342implies
343.Dv MAP_ANON
344and
345.Fa offset
346of 0.
347The
348.Fa fd
349argument
350must be -1 and
351.Fa prot
352must include at least
353.Dv PROT_READ
354and
355.Dv PROT_WRITE .
356The size of the guard, in pages, is specified by sysctl
357.Dv security.bsd.stack_guard_page .
358.El
359.Pp
360The
361.Xr close 2
362system call does not unmap pages, see
363.Xr munmap 2
364for further information.
365.Sh NOTES
366Although this implementation does not impose any alignment restrictions on
367the
368.Fa offset
369argument, a portable program must only use page-aligned values.
370.Pp
371Large page mappings require that the pages backing an object be
372aligned in matching blocks in both the virtual address space and RAM.
373The system will automatically attempt to use large page mappings when
374mapping an object that is already backed by large pages in RAM by
375aligning the mapping request in the virtual address space to match the
376alignment of the large physical pages.
377The system may also use large page mappings when mapping portions of an
378object that are not yet backed by pages in RAM.
379The
380.Dv MAP_ALIGNED_SUPER
381flag is an optimization that will align the mapping request to the
382size of a large page similar to
383.Dv MAP_ALIGNED ,
384except that the system will override this alignment if an object already
385uses large pages so that the mapping will be consistent with the existing
386large pages.
387This flag is mostly useful for maximizing the use of large pages on the
388first mapping of objects that do not yet have pages present in RAM.
389.Sh RETURN VALUES
390Upon successful completion,
391.Fn mmap
392returns a pointer to the mapped region.
393Otherwise, a value of
394.Dv MAP_FAILED
395is returned and
396.Va errno
397is set to indicate the error.
398.Sh ERRORS
399The
400.Fn mmap
401system call
402will fail if:
403.Bl -tag -width Er
404.It Bq Er EACCES
405The flag
406.Dv PROT_READ
407was specified as part of the
408.Fa prot
409argument and
410.Fa fd
411was not open for reading.
412The flags
413.Dv MAP_SHARED
414and
415.Dv PROT_WRITE
416were specified as part of the
417.Fa flags
418and
419.Fa prot
420argument and
421.Fa fd
422was not open for writing.
423.It Bq Er EBADF
424The
425.Fa fd
426argument
427is not a valid open file descriptor.
428.It Bq Er EINVAL
429An invalid (negative) value was passed in the
430.Fa offset
431argument, when
432.Fa fd
433referenced a regular file or shared memory.
434.It Bq Er EINVAL
435An invalid value was passed in the
436.Fa prot
437argument.
438.It Bq Er EINVAL
439An undefined option was set in the
440.Fa flags
441argument.
442.It Bq Er EINVAL
443Both
444.Dv MAP_PRIVATE
445and
446.Dv MAP_SHARED
447were specified.
448.It Bq Er EINVAL
449None of
450.Dv MAP_ANON ,
451.Dv MAP_GUARD ,
452.Dv MAP_PRIVATE ,
453.Dv MAP_SHARED ,
454or
455.Dv MAP_STACK
456was specified.
457At least one of these flags must be included.
458.It Bq Er EINVAL
459.Dv MAP_STACK
460was specified and
461.Va len
462is less than or equal to the guard size.
463.It Bq Er EINVAL
464.Dv MAP_FIXED
465was specified and the
466.Fa addr
467argument was not page aligned, or part of the desired address space
468resides out of the valid address space for a user process.
469.It Bq Er EINVAL
470Both
471.Dv MAP_FIXED
472and
473.Dv MAP_32BIT
474were specified and part of the desired address space resides outside
475of the first 2GB of user address space.
476.It Bq Er EINVAL
477The
478.Fa len
479argument
480was equal to zero.
481.It Bq Er EINVAL
482.Dv MAP_ALIGNED
483was specified and the desired alignment was either larger than the
484virtual address size of the machine or smaller than a page.
485.It Bq Er EINVAL
486.Dv MAP_ANON
487was specified and the
488.Fa fd
489argument was not -1.
490.It Bq Er EINVAL
491.Dv MAP_ANON
492was specified and the
493.Fa offset
494argument was not 0.
495.It Bq Er EINVAL
496Both
497.Dv MAP_FIXED
498and
499.Dv MAP_EXCL
500were specified, but the requested region is already used by a mapping.
501.It Bq Er EINVAL
502.Dv MAP_EXCL
503was specified, but
504.Dv MAP_FIXED
505was not.
506.It Bq Er EINVAL
507.Dv MAP_GUARD
508was specified, but the
509.Fa offset
510argument was not zero, the
511.Fa fd
512argument was not -1, or the
513.Fa prot
514argument was not
515.Dv PROT_NONE .
516.It Bq Er EINVAL
517.Dv MAP_GUARD
518was specified together with one of the flags
519.Dv MAP_ANON ,
520.Dv MAP_PREFAULT ,
521.Dv MAP_PREFAULT_READ ,
522.Dv MAP_PRIVATE ,
523.Dv MAP_SHARED ,
524.Dv MAP_STACK .
525.It Bq Er ENODEV
526.Dv MAP_ANON
527has not been specified and
528.Fa fd
529did not reference a regular or character special file.
530.It Bq Er ENOMEM
531.Dv MAP_FIXED
532was specified and the
533.Fa addr
534argument was not available.
535.Dv MAP_ANON
536was specified and insufficient memory was available.
537.It Bq Er ENOTSUP
538The
539.Fa prot
540argument contains protections which are not a subset of the specified
541maximum protections.
542.El
543.Sh SEE ALSO
544.Xr madvise 2 ,
545.Xr mincore 2 ,
546.Xr minherit 2 ,
547.Xr mlock 2 ,
548.Xr mprotect 2 ,
549.Xr msync 2 ,
550.Xr munlock 2 ,
551.Xr munmap 2 ,
552.Xr getpagesize 3 ,
553.Xr getpagesizes 3
554.Sh HISTORY
555The
556.Nm
557system call was first documented in
558.Bx 4.2
559and implemented in
560.Bx 4.4 .
561.\" XXX: lots of missing history of FreeBSD additions.
562.Pp
563The
564.Dv PROT_MAX
565functionality was introduced in
566.Fx 13 .
567