xref: /freebsd/lib/libsys/mmap.2 (revision 1edb7116f450c1a1793f2fd25f6bdc16735ef888)
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.Dd August 14, 2023
29.Dt MMAP 2
30.Os
31.Sh NAME
32.Nm mmap
33.Nd allocate memory, or map files or devices into memory
34.Sh LIBRARY
35.Lb libc
36.Sh SYNOPSIS
37.In sys/mman.h
38.Ft void *
39.Fn mmap "void *addr" "size_t len" "int prot" "int flags" "int fd" "off_t offset"
40.Sh DESCRIPTION
41The
42.Fn mmap
43system call causes the pages starting at
44.Fa addr
45and continuing for at most
46.Fa len
47bytes to be mapped from the object described by
48.Fa fd ,
49starting at byte offset
50.Fa offset .
51If
52.Fa len
53is not a multiple of the page size, the mapped region may extend past the
54specified range.
55Any such extension beyond the end of the mapped object will be zero-filled.
56.Pp
57If
58.Fa fd
59references a regular file or a shared memory object, the range of
60bytes starting at
61.Fa offset
62and continuing for
63.Fa len
64bytes must be legitimate for the possible (not necessarily
65current) offsets in the object.
66In particular, the
67.Fa offset
68value cannot be negative.
69If the object is truncated and the process later accesses a page that
70is wholly within the truncated region, the access is aborted and a
71.Dv SIGBUS
72signal is delivered to the process.
73.Pp
74If
75.Fa fd
76references a device file, the interpretation of the
77.Fa offset
78value is device specific and defined by the device driver.
79The virtual memory subsystem does not impose any restrictions on the
80.Fa offset
81value in this case, passing it unchanged to the driver.
82.Pp
83If
84.Fa addr
85is non-zero, it is used as a hint to the system.
86(As a convenience to the system, the actual address of the region may differ
87from the address supplied.)
88If
89.Fa addr
90is zero, an address will be selected by the system.
91The actual starting address of the region is returned.
92A successful
93.Fa mmap
94deletes any previous mapping in the allocated address range.
95.Pp
96The protections (region accessibility) are specified in the
97.Fa prot
98argument by
99.Em or Ns 'ing
100the following values:
101.Pp
102.Bl -tag -width PROT_WRITE -compact
103.It Dv PROT_NONE
104Pages may not be accessed.
105.It Dv PROT_READ
106Pages may be read.
107.It Dv PROT_WRITE
108Pages may be written.
109.It Dv PROT_EXEC
110Pages may be executed.
111.El
112.Pp
113In addition to these protection flags,
114.Fx
115provides the ability to set the maximum protection of a region allocated by
116.Nm
117and later altered by
118.Xr mprotect 2 .
119This is accomplished by
120.Em or Ns 'ing
121one or more
122.Dv PROT_
123values wrapped in the
124.Dv PROT_MAX()
125macro into the
126.Fa prot
127argument.
128.Pp
129The
130.Fa flags
131argument specifies the type of the mapped object, mapping options and
132whether modifications made to the mapped copy of the page are private
133to the process or are to be shared with other references.
134Sharing, mapping type and options are specified in the
135.Fa flags
136argument by
137.Em or Ns 'ing
138the following values:
139.Bl -tag -width MAP_PREFAULT_READ
140.It Dv MAP_32BIT
141Request a region in the first 2GB of the current process's address space.
142If a suitable region cannot be found,
143.Fn mmap
144will fail.
145.It Dv MAP_ALIGNED Ns Pq Fa n
146Align the region on a requested boundary.
147If a suitable region cannot be found,
148.Fn mmap
149will fail.
150The
151.Fa n
152argument specifies the binary logarithm of the desired alignment.
153.It Dv MAP_ALIGNED_SUPER
154Align the region to maximize the potential use of large
155.Pq Dq super
156pages.
157If a suitable region cannot be found,
158.Fn mmap
159will fail.
160The system will choose a suitable page size based on the size of
161mapping.
162The page size used as well as the alignment of the region may both be
163affected by properties of the file being mapped.
164In particular,
165the physical address of existing pages of a file may require a specific
166alignment.
167The region is not guaranteed to be aligned on any specific boundary.
168.It Dv MAP_ANON
169Map anonymous memory not associated with any specific file.
170The file descriptor used for creating
171.Dv MAP_ANON
172must be \-1.
173The
174.Fa offset
175argument must be 0.
176.\".It Dv MAP_FILE
177.\"Mapped from a regular file or character-special device memory.
178.It Dv MAP_ANONYMOUS
179This flag is identical to
180.Dv MAP_ANON
181and is provided for compatibility.
182.It Dv MAP_EXCL
183This flag can only be used in combination with
184.Dv MAP_FIXED .
185Please see the definition of
186.Dv MAP_FIXED
187for the description of its effect.
188.It Dv MAP_FIXED
189Do not permit the system to select a different address than the one
190specified.
191If the specified address cannot be used,
192.Fn mmap
193will fail.
194If
195.Dv MAP_FIXED
196is specified,
197.Fa addr
198must be a multiple of the page size.
199If
200.Dv MAP_EXCL
201is not specified, a successful
202.Dv MAP_FIXED
203request replaces any previous mappings for the process'
204pages in the range from
205.Fa addr
206to
207.Fa addr
208+
209.Fa len .
210In contrast, if
211.Dv MAP_EXCL
212is specified, the request will fail if a mapping
213already exists within the range.
214.It Dv MAP_GUARD
215Instead of a mapping, create a guard of the specified size.
216Guards allow a process to create reservations in its address space,
217which can later be replaced by actual mappings.
218.Pp
219.Fa mmap
220will not create mappings in the address range of a guard unless
221the request specifies
222.Dv MAP_FIXED .
223Guards can be destroyed with
224.Xr munmap 2 .
225Any memory access by a thread to the guarded range results
226in the delivery of a
227.Dv SIGSEGV
228signal to that thread.
229.It Dv MAP_NOCORE
230Region is not included in a core file.
231.It Dv MAP_NOSYNC
232Causes data dirtied via this VM map to be flushed to physical media
233only when necessary (usually by the pager) rather than gratuitously.
234Typically this prevents the update daemons from flushing pages dirtied
235through such maps and thus allows efficient sharing of memory across
236unassociated processes using a file-backed shared memory map.
237Without
238this option any VM pages you dirty may be flushed to disk every so often
239(every 30-60 seconds usually) which can create performance problems if you
240do not need that to occur (such as when you are using shared file-backed
241mmap regions for IPC purposes).
242Dirty data will be flushed automatically when all mappings of an object are
243removed and all descriptors referencing the object are closed.
244Note that VM/file system coherency is
245maintained whether you use
246.Dv MAP_NOSYNC
247or not.
248This option is not portable
249across
250.Ux
251platforms (yet), though some may implement the same behavior
252by default.
253.Pp
254.Em WARNING !
255Extending a file with
256.Xr ftruncate 2 ,
257thus creating a big hole, and then filling the hole by modifying a shared
258.Fn mmap
259can lead to severe file fragmentation.
260In order to avoid such fragmentation you should always pre-allocate the
261file's backing store by
262.Fn write Ns ing
263zero's into the newly extended area prior to modifying the area via your
264.Fn mmap .
265The fragmentation problem is especially sensitive to
266.Dv MAP_NOSYNC
267pages, because pages may be flushed to disk in a totally random order.
268.Pp
269The same applies when using
270.Dv MAP_NOSYNC
271to implement a file-based shared memory store.
272It is recommended that you create the backing store by
273.Fn write Ns ing
274zero's to the backing file rather than
275.Fn ftruncate Ns ing
276it.
277You can test file fragmentation by observing the KB/t (kilobytes per
278transfer) results from an
279.Dq Li iostat 1
280while reading a large file sequentially, e.g.,\& using
281.Dq Li dd if=filename of=/dev/null bs=32k .
282.Pp
283The
284.Xr fsync 2
285system call will flush all dirty data and metadata associated with a file,
286including dirty NOSYNC VM data, to physical media.
287The
288.Xr sync 8
289command and
290.Xr sync 2
291system call generally do not flush dirty NOSYNC VM data.
292The
293.Xr msync 2
294system call is usually not needed since
295.Bx
296implements a coherent file system buffer cache.
297However, it may be
298used to associate dirty VM pages with file system buffers and thus cause
299them to be flushed to physical media sooner rather than later.
300.It Dv MAP_PREFAULT_READ
301Immediately update the calling process's lowest-level virtual address
302translation structures, such as its page table, so that every memory
303resident page within the region is mapped for read access.
304Ordinarily these structures are updated lazily.
305The effect of this option is to eliminate any soft faults that would
306otherwise occur on the initial read accesses to the region.
307Although this option does not preclude
308.Fa prot
309from including
310.Dv PROT_WRITE ,
311it does not eliminate soft faults on the initial write accesses to the
312region.
313.It Dv MAP_PRIVATE
314Modifications are private.
315.It Dv MAP_SHARED
316Modifications are shared.
317.It Dv MAP_STACK
318Creates both a mapped region that grows downward on demand and an
319adjoining guard that both reserves address space for the mapped region
320to grow into and limits the mapped region's growth.
321Together, the mapped region and the guard occupy
322.Fa len
323bytes of the address space.
324The guard starts at the returned address, and the mapped region ends at
325the returned address plus
326.Fa len
327bytes.
328Upon access to the guard, the mapped region automatically grows in size,
329and the guard shrinks by an equal amount.
330Essentially, the boundary between the guard and the mapped region moves
331downward so that the access falls within the enlarged mapped region.
332However, the guard will never shrink to less than the number of pages
333specified by the sysctl
334.Dv security.bsd.stack_guard_page ,
335thereby ensuring that a gap for detecting stack overflow always exists
336between the downward growing mapped region and the closest mapped region
337beneath it.
338.Pp
339.Dv MAP_STACK
340implies
341.Dv MAP_ANON
342and
343.Fa offset
344of 0.
345The
346.Fa fd
347argument
348must be -1 and
349.Fa prot
350must include at least
351.Dv PROT_READ
352and
353.Dv PROT_WRITE .
354The size of the guard, in pages, is specified by sysctl
355.Dv security.bsd.stack_guard_page .
356.El
357.Pp
358The
359.Xr close 2
360system call does not unmap pages, see
361.Xr munmap 2
362for further information.
363.Sh NOTES
364Although this implementation does not impose any alignment restrictions on
365the
366.Fa offset
367argument, a portable program must only use page-aligned values.
368.Pp
369Large page mappings require that the pages backing an object be
370aligned in matching blocks in both the virtual address space and RAM.
371The system will automatically attempt to use large page mappings when
372mapping an object that is already backed by large pages in RAM by
373aligning the mapping request in the virtual address space to match the
374alignment of the large physical pages.
375The system may also use large page mappings when mapping portions of an
376object that are not yet backed by pages in RAM.
377The
378.Dv MAP_ALIGNED_SUPER
379flag is an optimization that will align the mapping request to the
380size of a large page similar to
381.Dv MAP_ALIGNED ,
382except that the system will override this alignment if an object already
383uses large pages so that the mapping will be consistent with the existing
384large pages.
385This flag is mostly useful for maximizing the use of large pages on the
386first mapping of objects that do not yet have pages present in RAM.
387.Sh RETURN VALUES
388Upon successful completion,
389.Fn mmap
390returns a pointer to the mapped region.
391Otherwise, a value of
392.Dv MAP_FAILED
393is returned and
394.Va errno
395is set to indicate the error.
396.Sh ERRORS
397The
398.Fn mmap
399system call
400will fail if:
401.Bl -tag -width Er
402.It Bq Er EACCES
403The flag
404.Dv PROT_READ
405was specified as part of the
406.Fa prot
407argument and
408.Fa fd
409was not open for reading.
410The flags
411.Dv MAP_SHARED
412and
413.Dv PROT_WRITE
414were specified as part of the
415.Fa flags
416and
417.Fa prot
418argument and
419.Fa fd
420was not open for writing.
421.It Bq Er EBADF
422The
423.Fa fd
424argument
425is not a valid open file descriptor.
426.It Bq Er EINVAL
427An invalid (negative) value was passed in the
428.Fa offset
429argument, when
430.Fa fd
431referenced a regular file or shared memory.
432.It Bq Er EINVAL
433An invalid value was passed in the
434.Fa prot
435argument.
436.It Bq Er EINVAL
437An undefined option was set in the
438.Fa flags
439argument.
440.It Bq Er EINVAL
441Both
442.Dv MAP_PRIVATE
443and
444.Dv MAP_SHARED
445were specified.
446.It Bq Er EINVAL
447None of
448.Dv MAP_ANON ,
449.Dv MAP_GUARD ,
450.Dv MAP_PRIVATE ,
451.Dv MAP_SHARED ,
452or
453.Dv MAP_STACK
454was specified.
455At least one of these flags must be included.
456.It Bq Er EINVAL
457.Dv MAP_STACK
458was specified and
459.Va len
460is less than or equal to the guard size.
461.It Bq Er EINVAL
462.Dv MAP_FIXED
463was specified and the
464.Fa addr
465argument was not page aligned, or part of the desired address space
466resides out of the valid address space for a user process.
467.It Bq Er EINVAL
468Both
469.Dv MAP_FIXED
470and
471.Dv MAP_32BIT
472were specified and part of the desired address space resides outside
473of the first 2GB of user address space.
474.It Bq Er EINVAL
475The
476.Fa len
477argument
478was equal to zero.
479.It Bq Er EINVAL
480.Dv MAP_ALIGNED
481was specified and the desired alignment was either larger than the
482virtual address size of the machine or smaller than a page.
483.It Bq Er EINVAL
484.Dv MAP_ANON
485was specified and the
486.Fa fd
487argument was not -1.
488.It Bq Er EINVAL
489.Dv MAP_ANON
490was specified and the
491.Fa offset
492argument was not 0.
493.It Bq Er EINVAL
494Both
495.Dv MAP_FIXED
496and
497.Dv MAP_EXCL
498were specified, but the requested region is already used by a mapping.
499.It Bq Er EINVAL
500.Dv MAP_EXCL
501was specified, but
502.Dv MAP_FIXED
503was not.
504.It Bq Er EINVAL
505.Dv MAP_GUARD
506was specified, but the
507.Fa offset
508argument was not zero, the
509.Fa fd
510argument was not -1, or the
511.Fa prot
512argument was not
513.Dv PROT_NONE .
514.It Bq Er EINVAL
515.Dv MAP_GUARD
516was specified together with one of the flags
517.Dv MAP_ANON ,
518.Dv MAP_PREFAULT ,
519.Dv MAP_PREFAULT_READ ,
520.Dv MAP_PRIVATE ,
521.Dv MAP_SHARED ,
522.Dv MAP_STACK .
523.It Bq Er ENODEV
524.Dv MAP_ANON
525has not been specified and
526.Fa fd
527did not reference a regular or character special file.
528.It Bq Er ENOMEM
529.Dv MAP_FIXED
530was specified and the
531.Fa addr
532argument was not available.
533.Dv MAP_ANON
534was specified and insufficient memory was available.
535.It Bq Er ENOTSUP
536The
537.Fa prot
538argument contains protections which are not a subset of the specified
539maximum protections.
540.El
541.Sh SEE ALSO
542.Xr madvise 2 ,
543.Xr mincore 2 ,
544.Xr minherit 2 ,
545.Xr mlock 2 ,
546.Xr mprotect 2 ,
547.Xr msync 2 ,
548.Xr munlock 2 ,
549.Xr munmap 2 ,
550.Xr getpagesize 3 ,
551.Xr getpagesizes 3
552.Sh HISTORY
553The
554.Nm
555system call was first documented in
556.Bx 4.2
557and implemented in
558.Bx 4.4 .
559.\" XXX: lots of missing history of FreeBSD additions.
560.Pp
561The
562.Dv PROT_MAX
563functionality was introduced in
564.Fx 13 .
565