mmap — Memory-mapped file support¶Availability: not WASI.
This module does not work or is not available on WebAssembly. See WebAssembly platforms for more information.
Memory-mapped file objects behave like both bytearray and like
file objects. You can use mmap objects in most places
where bytearray are expected; for example, you can use the re
module to search through a memory-mapped file. You can also change a single
byte by doing obj[index] = 97, or change a subsequence by assigning to a
slice: obj[i1:i2] = b'...'. You can also read and write data starting at
the current file position, and seek() through the file to different positions.
A memory-mapped file is created by the mmap constructor, which is
different on Unix and on Windows. In either case you must provide a file
descriptor for a file opened for update. If you wish to map an existing Python
file object, use its fileno() method to obtain the correct value for the
fileno parameter. Otherwise, you can open the file using the
os.open() function, which returns a file descriptor directly (the file
still needs to be closed when done).
Note
If you want to create a memory-mapping for a writable, buffered file, you
should flush() the file first. This is necessary to ensure
that local modifications to the buffers are actually available to the
mapping.
For both the Unix and Windows versions of the constructor, access may be
specified as an optional keyword parameter. access accepts one of four
values: ACCESS_READ, ACCESS_WRITE, or ACCESS_COPY to
specify read-only, write-through or copy-on-write memory respectively, or
ACCESS_DEFAULT to defer to prot. access can be used on both Unix
and Windows. If access is not specified, Windows mmap returns a
write-through mapping. The initial memory values for all three access types
are taken from the specified file. Assignment to an ACCESS_READ
memory map raises a TypeError exception. Assignment to an
ACCESS_WRITE memory map affects both memory and the underlying file.
Assignment to an ACCESS_COPY memory map affects memory but does not
update the underlying file.
Changed in version 3.7: Added ACCESS_DEFAULT constant.
To map anonymous memory, -1 should be passed as the fileno along with the length.
(Windows version) Maps length bytes from the file specified by the
file descriptor fileno, and creates a mmap object. If length is larger
than the current size of the file, the file is extended to contain length
bytes. If length is 0, the maximum length of the map is the current
size of the file, except that if the file is empty Windows raises an
exception (you cannot create an empty mapping on Windows).
tagname, if specified and not None, is a string giving a tag name for
the mapping. Windows allows you to have many different mappings against
the same file. If you specify the name of an existing tag, that tag is
opened, otherwise a new tag of this name is created. If this parameter is
omitted or None, the mapping is created without a name. Avoiding the
use of the tagname parameter will assist in keeping your code portable
between Unix and Windows.
offset may be specified as a non-negative integer offset. mmap references
will be relative to the offset from the beginning of the file. offset
defaults to 0. offset must be a multiple of the ALLOCATIONGRANULARITY.
If trackfd is False, the file handle corresponding to fileno will
not be duplicated, and the resulting mmap object will not
be associated with the map’s underlying file.
This means that the size() and resize()
methods will fail.
This mode is useful to limit the number of open file handles.
The original file can be renamed (but not deleted) after closing fileno.
Changed in version 3.15: The trackfd parameter was added.
Raises an auditing event mmap.__new__ with arguments fileno, length, access, offset.
(Unix version) Maps length bytes from the file specified by the file
descriptor fileno, and returns a mmap object. If length is 0, the
maximum length of the map will be the current size of the file when
mmap is called.
flags specifies the nature of the mapping. MAP_PRIVATE creates a
private copy-on-write mapping, so changes to the contents of the mmap
object will be private to this process, and MAP_SHARED creates a
mapping that’s shared with all other processes mapping the same areas of
the file. The default value is MAP_SHARED. Some systems have
additional possible flags with the full list specified in
MAP_* constants.
prot, if specified, gives the desired memory protection; the two most
useful values are PROT_READ and PROT_WRITE, to specify
that the pages may be read or written. prot defaults to
PROT_READ | PROT_WRITE.
access may be specified in lieu of flags and prot as an optional keyword parameter. It is an error to specify both flags, prot and access. See the description of access above for information on how to use this parameter.
offset may be specified as a non-negative integer offset. mmap references
will be relative to the offset from the beginning of the file. offset
defaults to 0. offset must be a multiple of ALLOCATIONGRANULARITY
which is equal to PAGESIZE on Unix systems.
If trackfd is False, the file descriptor specified by fileno will
not be duplicated, and the resulting mmap object will not
be associated with the map’s underlying file.
This means that the size() and resize()
methods will fail.
This mode is useful to limit the number of open file descriptors.
To ensure validity of the created memory mapping the file specified by the descriptor fileno is internally automatically synchronized with the physical backing store on macOS.
Changed in version 3.13: The trackfd parameter was added.
This example shows a simple way of using mmap:
import mmap
# write a simple example file
with open("hello.txt", "wb") as f:
f.write(b"Hello Python!\n")
with open("hello.txt", "r+b") as f:
# memory-map the file, size 0 means whole file
mm = mmap.mmap(f.fileno(), 0)
# read content via standard file methods
print(mm.readline()) # prints b"Hello Python!\n"
# read content via slice notation
print(mm[:5]) # prints b"Hello"
# update content using slice notation;
# note that new content must have same size
mm[6:] = b" world!\n"
# ... and read again using standard file methods
mm.seek(0)
print(mm.readline()) # prints b"Hello world!\n"
# close the map
mm.close()
mmap can also be used as a context manager in a with
statement:
import mmap
with mmap.mmap(-1, 13) as mm:
mm.write(b"Hello world!")
Added in version 3.2: Context manager support.
The next example demonstrates how to create an anonymous map and exchange data between the parent and child processes:
import mmap
import os
mm = mmap.mmap(-1, 13)
mm.write(b"Hello world!")
pid = os.fork()
if pid == 0: # In a child process
mm.seek(0)
print(mm.readline())
mm.close()
Raises an auditing event mmap.__new__ with arguments fileno, length, access, offset.
Memory-mapped file objects support the following methods:
Closes the mmap. Subsequent calls to other methods of the object will result in a ValueError exception being raised. This will not close the open file.
True if the file is closed.
Added in version 3.2.
Returns the lowest index in the object where the subsequence sub is
found, such that sub is contained in the range [start, end].
Optional arguments start and end are interpreted as in slice notation.
Returns -1 on failure.
Changed in version 3.5: Writable bytes-like object is now accepted.
Flushes changes made to the in-memory copy of a file back to disk. Without
use of this call there is no guarantee that changes are written back before
the object is destroyed. If offset and size are specified, only
changes to the given range of bytes will be flushed to disk; otherwise, the
whole extent of the mapping is flushed. offset must be a multiple of the
PAGESIZE or ALLOCATIONGRANULARITY.
The flags parameter specifies the synchronization behavior. flags must be one of the MS_* constants available on the system.
On Windows, the flags parameter is ignored.
None is returned to indicate success. An exception is raised when the
call failed.
Changed in version 3.8: Previously, a nonzero value was returned on success; zero was returned on error under Windows. A zero value was returned on success; an exception was raised on error under Unix.
Changed in version 3.15: Allow specifying offset without size. Previously, both offset and size parameters were required together. Now offset can be specified alone, and the flush operation will extend from offset to the end of the mmap.
Changed in version 3.15: Added flags parameter to control synchronization behavior.
Send advice option to the kernel about the memory region beginning at
start and extending length bytes. option must be one of the
MADV_* constants available on the system. If
start and length are omitted, the entire mapping is spanned. On
some systems (including Linux), start must be a multiple of the
PAGESIZE.
Availability: Systems with the madvise() system call.
Added in version 3.8.
Copy the count bytes starting at offset src to the destination index
dest. If the mmap was created with ACCESS_READ, then calls to
move will raise a TypeError exception.
Return a bytes containing up to n bytes starting from the
current file position. If the argument is omitted, None or negative,
return all bytes from the current file position to the end of the
mapping. The file position is updated to point after the bytes that were
returned.
Changed in version 3.3: Argument can be omitted or None.
Returns a byte at the current file position as an integer, and advances the file position by 1.
Returns a single line, starting at the current file position and up to the next newline. The file position is updated to point after the bytes that were returned.
Resizes the map and the underlying file, if any.
Resizing a map created with access of ACCESS_READ or
ACCESS_COPY, will raise a TypeError exception.
Resizing a map created with trackfd set to False,
will raise a ValueError exception.
On Windows: Resizing the map will raise an OSError if there are other
maps against the same named file. Resizing an anonymous map (ie against the
pagefile) will silently create a new map with the original data copied over
up to the length of the new size.
Availability: Windows and systems with the mremap() system call.
Changed in version 3.11: Correctly fails if attempting to resize when another map is held Allows resize against an anonymous map on Windows
Returns the highest index in the object where the subsequence sub is
found, such that sub is contained in the range [start, end].
Optional arguments start and end are interpreted as in slice notation.
Returns -1 on failure.
Changed in version 3.5: Writable bytes-like object is now accepted.
Set the file’s current position. whence argument is optional and
defaults to os.SEEK_SET or 0 (absolute file positioning); other
values are os.SEEK_CUR or 1 (seek relative to the current
position) and os.SEEK_END or 2 (seek relative to the file’s end).
Changed in version 3.13: Return the new absolute position instead of None.
Return whether the file supports seeking, and the return value is always True.
Added in version 3.13.
Annotate the memory mapping with the given name for easier identification
in /proc/<pid>/maps if the kernel supports the feature and -X dev is passed
to Python or if Python is built in debug mode.
The length of name must not exceed 67 bytes including the '\0' terminator.
Availability: Linux >= 5.17 (kernel built with CONFIG_ANON_VMA_NAME option)
Added in version 3.15.
Return the length of the file, which can be larger than the size of the memory-mapped area. For an anonymous mapping, return its size.
Changed in version 3.15: Anonymous mappings are now supported on Unix.
Returns the current position of the file pointer.
Write the bytes in bytes into memory at the current position of the
file pointer and return the number of bytes written (never less than
len(bytes), since if the write fails, a ValueError will be
raised). The file position is updated to point after the bytes that
were written. If the mmap was created with ACCESS_READ, then
writing to it will raise a TypeError exception.
Changed in version 3.5: Writable bytes-like object is now accepted.
Changed in version 3.6: The number of bytes written is now returned.
These options can be passed to mmap.madvise(). Not every option will
be present on every system.
Availability: Systems with the madvise() system call.
Added in version 3.8.
These are the various flags that can be passed to mmap.mmap(). MAP_ALIGNED_SUPER
is only available at FreeBSD and MAP_CONCEAL is only available at OpenBSD. Note
that some options might not be present on some systems.
Changed in version 3.10: Added MAP_POPULATE constant.
Added in version 3.11: Added MAP_STACK constant.
Added in version 3.12: Added MAP_ALIGNED_SUPER and MAP_CONCEAL constants.
Added in version 3.13: Added MAP_32BIT, MAP_HASSEMAPHORE, MAP_JIT,
MAP_NOCACHE, MAP_NOEXTEND, MAP_NORESERVE,
MAP_RESILIENT_CODESIGN, MAP_RESILIENT_MEDIA,
MAP_TPRO, MAP_TRANSLATED_ALLOW_EXECUTE, and
MAP_UNIX03 constants.
These flags control the synchronization behavior for mmap.flush():
MS_SYNC - Synchronous flush: writes are scheduled and the call
blocks until they are physically written to storage.
MS_ASYNC - Asynchronous flush: writes are scheduled but the call
returns immediately without waiting for completion.
MS_INVALIDATE - Invalidate cached data: invalidates other mappings
of the same file so they can see the changes.
Added in version 3.15.