manual/memory.texi - master - githubmirror/glibc

manual/memory.texi
3 917 строк152 KB
Florian Weimer
manual: Fix madvise typo in mseal documentation
11 дек 2025, 13:57
11 дек 2025, 13:57522310e
@node Memory, Character Handling, Error Reporting, Top
@chapter Virtual Memory Allocation And Paging
@c %MENU% Allocating virtual memory and controlling paging
@cindex memory allocation
@cindex storage allocation

This chapter describes how processes manage and use memory in a system
that uses @theglibc{}.

@Theglibc{} has several functions for dynamically allocating
virtual memory in various ways.  They vary in generality and in
efficiency.  The library also provides functions for controlling paging
and allocation of real memory.


@menu
* Memory Concepts::             An introduction to concepts and terminology.
* Memory Allocation::           Allocating storage for your program data
* Resizing the Data Segment::   @code{brk}, @code{sbrk}
* Memory Protection::           Controlling access to memory regions.
* Locking Pages::               Preventing page faults
@end menu

Memory mapped I/O is not discussed in this chapter.  @xref{Memory-mapped I/O}.



@node Memory Concepts
@section Process Memory Concepts

One of the most basic resources a process has available to it is memory.
There are a lot of different ways systems organize memory, but in a
typical one, each process has one linear virtual address space, with
addresses running from zero to some huge maximum.  It need not be
contiguous; i.e., not all of these addresses actually can be used to
store data.

The virtual memory is divided into pages (4 kilobytes is typical).
Backing each page of virtual memory is a page of real memory (called a
@dfn{frame}) or some secondary storage, usually disk space.  The disk
space might be swap space or just some ordinary disk file.  Actually, a
page of all zeroes sometimes has nothing at all backing it -- there's
just a flag saying it is all zeroes.
@cindex page frame
@cindex frame, real memory
@cindex swap space
@cindex page, virtual memory

The same frame of real memory or backing store can back multiple virtual
pages belonging to multiple processes.  This is normally the case, for
example, with virtual memory occupied by @glibcadj{} code.  The same
real memory frame containing the @code{printf} function backs a virtual
memory page in each of the existing processes that has a @code{printf}
call in its program.

In order for a program to access any part of a virtual page, the page
must at that moment be backed by (``connected to'') a real frame.  But
because there is usually a lot more virtual memory than real memory, the
pages must move back and forth between real memory and backing store
regularly, coming into real memory when a process needs to access them
and then retreating to backing store when not needed anymore.  This
movement is called @dfn{paging}.

When a program attempts to access a page which is not at that moment
backed by real memory, this is known as a @dfn{page fault}.  When a page
fault occurs, the kernel suspends the process, places the page into a
real page frame (this is called ``paging in'' or ``faulting in''), then
resumes the process so that from the process' point of view, the page
was in real memory all along.  In fact, to the process, all pages always
seem to be in real memory.  Except for one thing: the elapsed execution
time of an instruction that would normally be a few nanoseconds is
suddenly much, much, longer (because the kernel normally has to do I/O
to complete the page-in).  For programs sensitive to that, the functions
described in @ref{Locking Pages} can control it.
@cindex page fault
@cindex paging

Within each virtual address space, a process has to keep track of what
is at which addresses, and that process is called memory allocation.
Allocation usually brings to mind meting out scarce resources, but in
the case of virtual memory, that's not a major goal, because there is
generally much more of it than anyone needs.  Memory allocation within a
process is mainly just a matter of making sure that the same byte of
memory isn't used to store two different things.

Processes allocate memory in two major ways: by exec and
programmatically.  Actually, forking is a third way, but it's not very
interesting.  @xref{Creating a Process}.

Exec is the operation of creating a virtual address space for a process,
loading its basic program into it, and executing the program.  It is
done by the ``exec'' family of functions (e.g. @code{execl}).  The
operation takes a program file (an executable), it allocates space to
load all the data in the executable, loads it, and transfers control to
it.  That data is most notably the instructions of the program (the
@dfn{text}), but also literals and constants in the program and even
some variables: C variables with the static storage class (@pxref{Memory
Allocation and C}).
@cindex executable
@cindex literals
@cindex constants

Once that program begins to execute, it uses programmatic allocation to
gain additional memory.  In a C program with @theglibc{}, there
are two kinds of programmatic allocation: automatic and dynamic.
@xref{Memory Allocation and C}.

Memory-mapped I/O is another form of dynamic virtual memory allocation.
Mapping memory to a file means declaring that the contents of certain
range of a process' addresses shall be identical to the contents of a
specified regular file.  The system makes the virtual memory initially
contain the contents of the file, and if you modify the memory, the
system writes the same modification to the file.  Note that due to the
magic of virtual memory and page faults, there is no reason for the
system to do I/O to read the file, or allocate real memory for its
contents, until the program accesses the virtual memory.
@xref{Memory-mapped I/O}.
@cindex memory mapped I/O
@cindex memory mapped file
@cindex files, accessing

Just as it programmatically allocates memory, the program can
programmatically deallocate (@dfn{free}) it.  You can't free the memory
that was allocated by exec.  When the program exits or execs, you might
say that all its memory gets freed, but since in both cases the address
space ceases to exist, the point is really moot.  @xref{Program
Termination}.
@cindex execing a program
@cindex freeing memory
@cindex exiting a program

A process' virtual address space is divided into segments.  A segment is
a contiguous range of virtual addresses.  Three important segments are:

@itemize @bullet

@item

The @dfn{text segment} contains a program's instructions and literals and
static constants.  It is allocated by exec and stays the same size for
the life of the virtual address space.

@item
The @dfn{data segment} is working storage for the program.  It can be
preallocated and preloaded by exec and the process can extend or shrink
it by calling functions as described in @xref{Resizing the Data
Segment}.  Its lower end is fixed.

@item
The @dfn{stack segment} contains a program stack.  It grows as the stack
grows, but doesn't shrink when the stack shrinks.

@end itemize



@node Memory Allocation
@section Allocating Storage For Program Data

This section covers how ordinary programs manage storage for their data,
including the famous @code{malloc} function and some fancier facilities
special to @theglibc{} and GNU Compiler.

@menu
* Memory Allocation and C::     How to get different kinds of allocation in C.
* The GNU Allocator::		An overview of the GNU @code{malloc}
				implementation.
* Unconstrained Allocation::    The @code{malloc} facility allows fully general
		 		 dynamic allocation.
* Allocation Debugging::        Finding memory leaks and not freed memory.
* Replacing malloc::            Using your own @code{malloc}-style allocator.
* Obstacks::                    Obstacks are less general than malloc
				 but more efficient and convenient.
* Variable Size Automatic::     Allocation of variable-sized blocks
				 of automatic storage that are freed when the
				 calling function returns.
@end menu


@node Memory Allocation and C
@subsection Memory Allocation in C Programs

The C language supports two kinds of memory allocation through the
variables in C programs:

@itemize @bullet
@item
@dfn{Static allocation} is what happens when you declare a static or
global variable.  Each static or global variable defines one block of
space, of a fixed size.  The space is allocated once, when your program
is started (part of the exec operation), and is never freed.
@cindex static memory allocation
@cindex static storage class

@item
@dfn{Automatic allocation} happens when you declare an automatic
variable, such as a function argument or a local variable.  The space
for an automatic variable is allocated when the compound statement
containing the declaration is entered, and is freed when that
compound statement is exited.
@cindex automatic memory allocation
@cindex automatic storage class

In GNU C, the size of the automatic storage can be an expression
that varies.  In other C implementations, it must be a constant.
@end itemize

A third important kind of memory allocation, @dfn{dynamic allocation},
is not supported by C variables but is available via @glibcadj{}
functions.
@cindex dynamic memory allocation

@subsubsection Dynamic Memory Allocation
@cindex dynamic memory allocation

@dfn{Dynamic memory allocation} is a technique in which programs
determine as they are running where to store some information.  You need
dynamic allocation when the amount of memory you need, or how long you
continue to need it, depends on factors that are not known before the
program runs.

For example, you may need a block to store a line read from an input
file; since there is no limit to how long a line can be, you must
allocate the memory dynamically and make it dynamically larger as you
read more of the line.

Or, you may need a block for each record or each definition in the input
data; since you can't know in advance how many there will be, you must
allocate a new block for each record or definition as you read it.

When you use dynamic allocation, the allocation of a block of memory is
an action that the program requests explicitly.  You call a function or
macro when you want to allocate space, and specify the size with an
argument.  If you want to free the space, you do so by calling another
function or macro.  You can do these things whenever you want, as often
as you want.

Dynamic allocation is not supported by C variables; there is no storage
class ``dynamic'', and there can never be a C variable whose value is
stored in dynamically allocated space.  The only way to get dynamically
allocated memory is via a system call (which is generally via a @glibcadj{}
function call), and the only way to refer to dynamically
allocated space is through a pointer.  Because it is less convenient,
and because the actual process of dynamic allocation requires more
computation time, programmers generally use dynamic allocation only when
neither static nor automatic allocation will serve.

For example, if you want to allocate dynamically some space to hold a
@code{struct foobar}, you cannot declare a variable of type @code{struct
foobar} whose contents are the dynamically allocated space.  But you can
declare a variable of pointer type @code{struct foobar *} and assign it the
address of the space.  Then you can use the operators @samp{*} and
@samp{->} on this pointer variable to refer to the contents of the space:

@smallexample
@{
  struct foobar *ptr = malloc (sizeof *ptr);
  ptr->name = x;
  ptr->next = current_foobar;
  current_foobar = ptr;
@}
@end smallexample

@node The GNU Allocator
@subsection The GNU Allocator
@cindex gnu allocator

The @code{malloc} implementation in @theglibc{} is derived from ptmalloc
(pthreads malloc), which in turn is derived from dlmalloc (Doug Lea malloc).
This @code{malloc} may allocate memory
in two different ways depending on their size
and certain parameters that may be controlled by users. The most common way is
to allocate portions of memory (called chunks) from a large contiguous area of
memory and manage these areas to optimize their use and reduce wastage in the
form of unusable chunks. Traditionally the system heap was set up to be the one
large memory area but the @glibcadj{} @code{malloc} implementation maintains
multiple such areas to optimize their use in multi-threaded applications.  Each
such area is internally referred to as an @dfn{arena}.

As opposed to other versions, the @code{malloc} in @theglibc{} does not round
up chunk sizes to powers of two, neither for large nor for small sizes.
Neighboring chunks can be coalesced on a @code{free} no matter what their size
is.  This makes the implementation suitable for all kinds of allocation
patterns without generally incurring high memory waste through fragmentation.
The presence of multiple arenas allows multiple threads to allocate
memory simultaneously in separate arenas, thus improving performance.

The other way of memory allocation is for very large blocks, i.e. much larger
than a page. These requests are allocated with @code{mmap} (anonymous or via
@file{/dev/zero}; @pxref{Memory-mapped I/O})). This has the great advantage
that these chunks are returned to the system immediately when they are freed.
Therefore, it cannot happen that a large chunk becomes ``locked'' in between
smaller ones and even after calling @code{free} wastes memory.  The size
threshold for @code{mmap} to be used is dynamic and gets adjusted according to
allocation patterns of the program.  @code{mallopt} can be used to statically
adjust the threshold using @code{M_MMAP_THRESHOLD} and the use of @code{mmap}
can be disabled completely with @code{M_MMAP_MAX};
@pxref{Malloc Tunable Parameters}.

A more detailed technical description of the GNU Allocator is maintained in
the @glibcadj{} wiki. See
@uref{https://sourceware.org/glibc/wiki/MallocInternals}.

It is possible to use your own custom @code{malloc} instead of the
built-in allocator provided by @theglibc{}.  @xref{Replacing malloc}.

@node Unconstrained Allocation
@subsection Unconstrained Allocation
@cindex unconstrained memory allocation
@cindex @code{malloc} function
@cindex heap, dynamic allocation from

The most general dynamic allocation facility is @code{malloc}.  It
allows you to allocate blocks of memory of any size at any time, make
them bigger or smaller at any time, and free the blocks individually at
any time (or never).

@menu
* Basic Allocation::            Simple use of @code{malloc}.
* Malloc Examples::             Examples of @code{malloc}.  @code{xmalloc}.
* Freeing after Malloc::        Use @code{free} to free a block you
				 got with @code{malloc}.
* Changing Block Size::         Use @code{realloc} to make a block
				 bigger or smaller.
* Allocating Cleared Space::    Use @code{calloc} to allocate a
				 block and clear it.
* Aligned Memory Blocks::       Allocating specially aligned memory.
* Malloc Tunable Parameters::   Use @code{mallopt} to adjust allocation
                                 parameters.
* Heap Consistency Checking::   Automatic checking for errors.
* Statistics of Malloc::        Getting information about how much
				 memory your program is using.
* Summary of Malloc::           Summary of @code{malloc} and related functions.
@end menu

@node Basic Allocation
@subsubsection Basic Memory Allocation
@cindex allocation of memory with @code{malloc}

To allocate a block of memory, call @code{malloc}.  The prototype for
this function is in @file{stdlib.h}.
@pindex stdlib.h

@deftypefun {void *} malloc (size_t @var{size})
@standards{ISO, malloc.h}
@standards{ISO, stdlib.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}
@c Malloc hooks and __morecore pointers, as well as such parameters as
@c max_n_mmaps and max_mmapped_mem, are accessed without guards, so they
@c could pose a thread safety issue; in order to not declare malloc
@c MT-unsafe, it's modifying the hooks and parameters while multiple
@c threads are active that is regarded as unsafe.  An arena's next field
@c is initialized and never changed again, except for main_arena's,
@c that's protected by list_lock; next_free is only modified while
@c list_lock is held too.  All other data members of an arena, as well
@c as the metadata of the memory areas assigned to it, are only modified
@c while holding the arena's mutex (fastbin pointers use atomic ops
@c because they may be modified by free without taking the arena's
@c lock).  Some reassurance was needed for fastbins, for it wasn't clear
@c how they were initialized.  It turns out they are always
@c zero-initialized: main_arena's, for being static data, and other
@c arena's, for being just-mmapped memory.

@c Leaking file descriptors and memory in case of cancellation is
@c unavoidable without disabling cancellation, but the lock situation is
@c a bit more complicated: we don't have fallback arenas for malloc to
@c be safe to call from within signal handlers.  Error-checking mutexes
@c or trylock could enable us to try and use alternate arenas, even with
@c -DPER_THREAD (enabled by default), but supporting interruption
@c (cancellation or signal handling) while holding the arena list mutex
@c would require more work; maybe blocking signals and disabling async
@c cancellation while manipulating the arena lists?

@c __libc_malloc @asulock @aculock @acsfd @acsmem
@c  force_reg ok
@c  *malloc_hook unguarded
@c  arena_lock @asulock @aculock @acsfd @acsmem
@c   mutex_lock @asulock @aculock
@c   arena_get2 @asulock @aculock @acsfd @acsmem
@c    get_free_list @asulock @aculock
@c     mutex_lock (list_lock) dup @asulock @aculock
@c     mutex_unlock (list_lock) dup @aculock
@c     mutex_lock (arena lock) dup @asulock @aculock [returns locked]
@c    __get_nprocs ext ok @acsfd
@c    NARENAS_FROM_NCORES ok
@c    atomic_compare_and_exchange_bool_acq ok
@c    _int_new_arena ok @asulock @aculock @acsmem
@c     new_heap ok @acsmem
@c      mmap ok @acsmem
@c      munmap ok @acsmem
@c      mprotect ok
@c     chunk2mem ok
@c     set_head ok
@c     tsd_setspecific dup ok
@c     mutex_init ok
@c     mutex_lock (just-created mutex) ok, returns locked
@c     mutex_lock (list_lock) dup @asulock @aculock
@c     atomic_write_barrier ok
@c     mutex_unlock (list_lock) @aculock
@c    atomic_fetch_add ok
@c    reused_arena @asulock @aculock
@c      reads&writes next_to_use and iterates over arena next without guards
@c      those are harmless as long as we don't drop arenas from the
@c      NEXT list, and we never do; when a thread terminates,
@c      __malloc_arena_thread_freeres prepends the arena to the free_list
@c      NEXT_FREE list, but NEXT is never modified, so it's safe!
@c     mutex_trylock (arena lock) @asulock @aculock
@c     mutex_lock (arena lock) dup @asulock @aculock
@c     tsd_setspecific dup ok
@c  _int_malloc @acsfd @acsmem
@c   checked_request2size ok
@c    REQUEST_OUT_OF_RANGE ok
@c    request2size ok
@c   get_max_fast ok
@c   fastbin_index ok
@c   fastbin ok
@c   atomic_compare_and_exhange_val_acq ok
@c   malloc_printerr dup @mtsenv
@c     if we get to it, we're toast already, undefined behavior must have
@c     been invoked before
@c    libc_message @mtsenv [no leaks with cancellation disabled]
@c     FATAL_PREPARE ok
@c      pthread_setcancelstate disable ok
@c     libc_secure_getenv @mtsenv
@c      getenv @mtsenv
@c     open_not_cancel_2 dup @acsfd
@c     strchrnul ok
@c     WRITEV_FOR_FATAL ok
@c      writev ok
@c     mmap ok @acsmem
@c     munmap ok @acsmem
@c     BEFORE_ABORT @acsfd
@c      backtrace ok
@c      write_not_cancel dup ok
@c      backtrace_symbols_fd @aculock
@c      open_not_cancel_2 dup @acsfd
@c      read_not_cancel dup ok
@c      close_not_cancel_no_status dup @acsfd
@c     abort ok
@c    itoa_word ok
@c    abort ok
@c   check_remalloced_chunk ok/disabled
@c   chunk2mem dup ok
@c   alloc_perturb ok
@c   in_smallbin_range ok
@c   smallbin_index ok
@c   bin_at ok
@c   last ok
@c   malloc_consolidate ok
@c    get_max_fast dup ok
@c    clear_fastchunks ok
@c    unsorted_chunks dup ok
@c    fastbin dup ok
@c    atomic_exchange_acquire ok
@c    check_inuse_chunk dup ok/disabled
@c    chunk_at_offset dup ok
@c    chunksize dup ok
@c    inuse_bit_at_offset dup ok
@c    unlink dup ok
@c    clear_inuse_bit_at_offset dup ok
@c    in_smallbin_range dup ok
@c    set_head dup ok
@c    malloc_init_state ok
@c     bin_at dup ok
@c     set_noncontiguous dup ok
@c     set_max_fast dup ok
@c     initial_top ok
@c      unsorted_chunks dup ok
@c    check_malloc_state ok/disabled
@c   set_inuse_bit_at_offset ok
@c   check_malloced_chunk ok/disabled
@c   largebin_index ok
@c   have_fastchunks ok
@c   unsorted_chunks ok
@c    bin_at ok
@c   chunksize ok
@c   chunk_at_offset ok
@c   set_head ok
@c   set_foot ok
@c   mark_bin ok
@c    idx2bit ok
@c   first ok
@c   unlink ok
@c    malloc_printerr dup ok
@c    in_smallbin_range dup ok
@c   idx2block ok
@c   idx2bit dup ok
@c   next_bin ok
@c   sysmalloc @acsfd @acsmem
@c    MMAP @acsmem
@c    set_head dup ok
@c    check_chunk ok/disabled
@c    chunk2mem dup ok
@c    chunksize dup ok
@c    chunk_at_offset dup ok
@c    heap_for_ptr ok
@c    grow_heap ok
@c     mprotect ok
@c    set_head dup ok
@c    new_heap @acsmem
@c     MMAP dup @acsmem
@c     munmap @acsmem
@c    top ok
@c    set_foot dup ok
@c    contiguous ok
@c    MORECORE ok
@c     *__morecore ok unguarded
@c      __default_morecore
@c       sbrk ok
@c    force_reg dup ok
@c    *__after_morecore_hook unguarded
@c    set_noncontiguous ok
@c    malloc_printerr dup ok
@c    _int_free (have_lock) @acsfd @acsmem [@asulock @aculock]
@c     chunksize dup ok
@c     mutex_unlock dup @aculock/!have_lock
@c     malloc_printerr dup ok
@c     check_inuse_chunk ok/disabled
@c     chunk_at_offset dup ok
@c     mutex_lock dup @asulock @aculock/@have_lock
@c     chunk2mem dup ok
@c     free_perturb ok
@c     set_fastchunks ok
@c     fastbin_index dup ok
@c     fastbin dup ok
@c     atomic_compare_and_exchange_val_rel ok
@c     chunk_is_mmapped ok
@c     contiguous dup ok
@c     prev_inuse ok
@c     unlink dup ok
@c     inuse_bit_at_offset dup ok
@c     clear_inuse_bit_at_offset ok
@c     unsorted_chunks dup ok
@c     in_smallbin_range dup ok
@c     set_head dup ok
@c     set_foot dup ok
@c     check_free_chunk ok/disabled
@c     check_chunk dup ok/disabled
@c     have_fastchunks dup ok
@c     malloc_consolidate dup ok
@c     systrim ok
@c      MORECORE dup ok
@c      *__after_morecore_hook dup unguarded
@c      set_head dup ok
@c      check_malloc_state ok/disabled
@c     top dup ok
@c     heap_for_ptr dup ok
@c     heap_trim @acsfd @acsmem
@c      top dup ok
@c      chunk_at_offset dup ok
@c      prev_chunk ok
@c      chunksize dup ok
@c      prev_inuse dup ok
@c      delete_heap @acsmem
@c       munmap dup @acsmem
@c      unlink dup ok
@c      set_head dup ok
@c      shrink_heap @acsfd
@c       check_may_shrink_heap @acsfd
@c        open_not_cancel_2 @acsfd
@c        read_not_cancel ok
@c        close_not_cancel_no_status @acsfd
@c       MMAP dup ok
@c       madvise ok
@c     munmap_chunk @acsmem
@c      chunksize dup ok
@c      chunk_is_mmapped dup ok
@c      chunk2mem dup ok
@c      malloc_printerr dup ok
@c      munmap dup @acsmem
@c    check_malloc_state ok/disabled
@c  arena_get_retry @asulock @aculock @acsfd @acsmem
@c   mutex_unlock dup @aculock
@c   mutex_lock dup @asulock @aculock
@c   arena_get2 dup @asulock @aculock @acsfd @acsmem
@c  mutex_unlock @aculock
@c  mem2chunk ok
@c  chunk_is_mmapped ok
@c  arena_for_chunk ok
@c   chunk_non_main_arena ok
@c   heap_for_ptr ok
This function returns a pointer to a newly allocated block @var{size}
bytes long, or a null pointer (setting @code{errno})
if the block could not be allocated.
@end deftypefun

The contents of the block are undefined; you must initialize it yourself
(or use @code{calloc} instead; @pxref{Allocating Cleared Space}).
Normally you would convert the value to a pointer to the kind of object
that you want to store in the block.  Here we show an example of doing
so, and of initializing the space with zeros using the library function
@code{memset} (@pxref{Copying Strings and Arrays}):

@smallexample
struct foo *ptr = malloc (sizeof *ptr);
if (ptr == 0) abort ();
memset (ptr, 0, sizeof (struct foo));
@end smallexample

You can store the result of @code{malloc} into any pointer variable
without a cast, because @w{ISO C} automatically converts the type
@code{void *} to another type of pointer when necessary.  However, a cast
is necessary if the type is needed but not specified by context.

Remember that when allocating space for a string, the argument to
@code{malloc} must be one plus the length of the string.  This is
because a string is terminated with a null character that doesn't count
in the ``length'' of the string but does need space.  For example:

@smallexample
char *ptr = malloc (length + 1);
@end smallexample

@noindent
@xref{Representation of Strings}, for more information about this.

@node Malloc Examples
@subsubsection Examples of @code{malloc}

If no more space is available, @code{malloc} returns a null pointer.
You should check the value of @emph{every} call to @code{malloc}.  It is
useful to write a subroutine that calls @code{malloc} and reports an
error if the value is a null pointer, returning only if the value is
nonzero.  This function is conventionally called @code{xmalloc}.  Here
it is:
@cindex @code{xmalloc} function

@smallexample
void *
xmalloc (size_t size)
@{
  void *value = malloc (size);
  if (value == 0)
    fatal ("virtual memory exhausted");
  return value;
@}
@end smallexample

Here is a real example of using @code{malloc} (by way of @code{xmalloc}).
The function @code{savestring} will copy a sequence of characters into
a newly allocated null-terminated string:

@smallexample
@group
char *
savestring (const char *ptr, size_t len)
@{
  char *value = xmalloc (len + 1);
  value[len] = '\0';
  return memcpy (value, ptr, len);
@}
@end group
@end smallexample

The block that @code{malloc} gives you is guaranteed to be aligned so
that it can hold any type of data.  On @gnusystems{}, the address is
always a multiple of eight on 32-bit systems, and a multiple of 16 on
64-bit systems.  Only rarely is any higher boundary (such as a page
boundary) necessary; for those cases, use @code{aligned_alloc} or
@code{posix_memalign} (@pxref{Aligned Memory Blocks}).

Note that the memory located after the end of the block is likely to be
in use for something else; perhaps a block already allocated by another
call to @code{malloc}.  If you attempt to treat the block as longer than
you asked for it to be, you are liable to destroy the data that
@code{malloc} uses to keep track of its blocks, or you may destroy the
contents of another block.  If you have already allocated a block and
discover you want it to be bigger, use @code{realloc} (@pxref{Changing
Block Size}).

@strong{Portability Notes:}

@itemize @bullet
@item
In @theglibc{}, a successful @code{malloc (0)}
returns a non-null pointer to a newly allocated size-zero block;
other implementations may return @code{NULL} instead.
POSIX and the ISO C standard allow both behaviors.

@item
In @theglibc{}, a failed @code{malloc} call sets @code{errno},
but ISO C does not require this and non-POSIX implementations
need not set @code{errno} when failing.

@item
In @theglibc{}, @code{malloc} always fails when @var{size} exceeds
@code{PTRDIFF_MAX}, to avoid problems with programs that subtract
pointers or use signed indexes.  Other implementations may succeed in
this case, leading to undefined behavior later.
@end itemize

@node Freeing after Malloc
@subsubsection Freeing Memory Allocated with @code{malloc}
@cindex freeing memory allocated with @code{malloc}
@cindex heap, freeing memory from

When you no longer need a block that you got with @code{malloc}, use the
function @code{free} to make the block available to be allocated again.
The prototype for this function is in @file{stdlib.h}.
@pindex stdlib.h

@deftypefun void free (void *@var{ptr})
@standards{ISO, malloc.h}
@standards{ISO, stdlib.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}
@c __libc_free @asulock @aculock @acsfd @acsmem
@c   releasing memory into fastbins modifies the arena without taking
@c   its mutex, but atomic operations ensure safety.  If two (or more)
@c   threads are running malloc and have their own arenas locked when
@c   each gets a signal whose handler free()s large (non-fastbin-able)
@c   blocks from each other's arena, we deadlock; this is a more general
@c   case of @asulock.
@c  *__free_hook unguarded
@c  mem2chunk ok
@c  chunk_is_mmapped ok, chunk bits not modified after allocation
@c  chunksize ok
@c  munmap_chunk dup @acsmem
@c  arena_for_chunk dup ok
@c  _int_free (!have_lock) dup @asulock @aculock @acsfd @acsmem
The @code{free} function deallocates the block of memory pointed at
by @var{ptr}.
@end deftypefun

Freeing a block alters the contents of the block.  @strong{Do not expect to
find any data (such as a pointer to the next block in a chain of blocks) in
the block after freeing it.}  Copy whatever you need out of the block before
freeing it!  Here is an example of the proper way to free all the blocks in
a chain, and the strings that they point to:

@smallexample
struct chain
  @{
    struct chain *next;
    char *name;
  @}

void
free_chain (struct chain *chain)
@{
  while (chain != 0)
    @{
      struct chain *next = chain->next;
      free (chain->name);
      free (chain);
      chain = next;
    @}
@}
@end smallexample

Occasionally, @code{free} can actually return memory to the operating
system and make the process smaller.  Usually, all it can do is allow a
later call to @code{malloc} to reuse the space.  In the meantime, the
space remains in your program as part of a free-list used internally by
@code{malloc}.

The @code{free} function preserves the value of @code{errno}, so that
cleanup code need not worry about saving and restoring @code{errno}
around a call to @code{free}.  Although neither @w{ISO C} nor
POSIX.1-2017 requires @code{free} to preserve @code{errno}, a future
version of POSIX is planned to require it.

There is no point in freeing blocks at the end of a program, because all
of the program's space is given back to the system when the process
terminates.

@deftypefun void free_sized (void *@var{ptr}, size_t @var{size})
@standards{ISO, stdlib.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}
The @code{free_sized} function deallocates the block of memory pointed at
by @var{ptr} that was previously allocated by @code{malloc}, @code{calloc}
or @code{realloc}.  The size @var{size} must match the previously requested
total size provided to @code{malloc}, @code{calloc} or @code{realloc}.
Attempting to deallocated memory allocated by @code{aligned_alloc},
@code{memalign}, @code{posix_memalign}, @code{valloc} or @code{pvalloc} is
undefined behavior.  For @code{aligned_alloc}, @code{memalign} or
@code{posix_memalign} use @code{free_aligned_sized} instead.  Additionally
it is also undefined behavior to call @code{free_sized} for allocations
which the caller did not directly allocate but must still deallocate, such
as the result of @code{strdup} or @code{strndup}.  Instead continue using
@code{free} for these cases.
@end deftypefun

@deftypefun void free_aligned_sized (void *@var{ptr}, size_t @var{alignment}, size_t @var{size})
@standards{ISO, stdlib.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}
The @code{free_aligned_sized} function deallocates the block of memory
pointed at by @var{ptr} that was previously allocated by
@code{aligned_alloc}, @code{memalign} or @code{posix_memalign}.
The size @var{size} and alignment @var{alignment} must match the
previously requested size and alignment provided to
@code{aligned_alloc}, @code{memalign} or @code{posix_memalign}.
@end deftypefun

@node Changing Block Size
@subsubsection Changing the Size of a Block
@cindex changing the size of a block (@code{malloc})

Often you do not know for certain how big a block you will ultimately need
at the time you must begin to use the block.  For example, the block might
be a buffer that you use to hold a line being read from a file; no matter
how long you make the buffer initially, you may encounter a line that is
longer.

You can make the block longer by calling @code{realloc} or
@code{reallocarray}.  These functions are declared in @file{stdlib.h}.
@pindex stdlib.h

@deftypefun {void *} realloc (void *@var{ptr}, size_t @var{newsize})
@standards{ISO, malloc.h}
@standards{ISO, stdlib.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}
@c It may call the implementations of malloc and free, so all of their
@c issues arise, plus the realloc hook, also accessed without guards.

@c __libc_realloc @asulock @aculock @acsfd @acsmem
@c  *__realloc_hook unguarded
@c  __libc_free dup @asulock @aculock @acsfd @acsmem
@c  __libc_malloc dup @asulock @aculock @acsfd @acsmem
@c  mem2chunk dup ok
@c  chunksize dup ok
@c  malloc_printerr dup ok
@c  checked_request2size dup ok
@c  chunk_is_mmapped dup ok
@c  mremap_chunk
@c   chunksize dup ok
@c   __mremap ok
@c   set_head dup ok
@c  MALLOC_COPY ok
@c   memcpy ok
@c  munmap_chunk dup @acsmem
@c  arena_for_chunk dup ok
@c  mutex_lock (arena mutex) dup @asulock @aculock
@c  _int_realloc @acsfd @acsmem
@c   malloc_printerr dup ok
@c   check_inuse_chunk dup ok/disabled
@c   chunk_at_offset dup ok
@c   chunksize dup ok
@c   set_head_size dup ok
@c   chunk_at_offset dup ok
@c   set_head dup ok
@c   chunk2mem dup ok
@c   inuse dup ok
@c   unlink dup ok
@c   _int_malloc dup @acsfd @acsmem
@c   mem2chunk dup ok
@c   MALLOC_COPY dup ok
@c   _int_free (have_lock) dup @acsfd @acsmem
@c   set_inuse_bit_at_offset dup ok
@c   set_head dup ok
@c  mutex_unlock (arena mutex) dup @aculock
@c  _int_free (!have_lock) dup @asulock @aculock @acsfd @acsmem

The @code{realloc} function changes the size of the block whose address is
@var{ptr} to be @var{newsize}.

Since the space after the end of the block may be in use, @code{realloc}
may find it necessary to copy the block to a new address where more free
space is available.  The value of @code{realloc} is the new address of the
block.  If the block needs to be moved, @code{realloc} copies the old
contents.

If you pass a null pointer for @var{ptr}, @code{realloc} behaves just
like @samp{malloc (@var{newsize})}.
Otherwise, if @var{newsize} is zero
@code{realloc} frees the block and returns @code{NULL}.
Otherwise, if @code{realloc} cannot reallocate the requested size
it returns @code{NULL} and sets @code{errno}; the original block
is left undisturbed.
@end deftypefun

@deftypefun {void *} reallocarray (void *@var{ptr}, size_t @var{nmemb}, size_t @var{size})
@standards{POSIX.1-2024, malloc.h}
@standards{POSIX.1-2024, stdlib.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}

The @code{reallocarray} function changes the size of the block whose address
is @var{ptr} to be long enough to contain a vector of @var{nmemb} elements,
each of size @var{size}.  It is equivalent to @samp{realloc (@var{ptr},
@var{nmemb} * @var{size})}, except that @code{reallocarray} fails safely if
the multiplication overflows, by setting @code{errno} to @code{ENOMEM},
returning a null pointer, and leaving the original block unchanged.

@code{reallocarray} should be used instead of @code{realloc} when the new size
of the allocated block is the result of a multiplication that might overflow.

This function was originally derived from OpenBSD 5.6, but was added in
POSIX.1-2024.
@end deftypefun

Like @code{malloc}, @code{realloc} and @code{reallocarray} may return a null
pointer if no memory space is available to make the block bigger.  When this
happens, the original block is untouched; it has not been modified or
relocated.

In most cases it makes no difference what happens to the original block
when @code{realloc} fails, because the application program cannot continue
when it is out of memory, and the only thing to do is to give a fatal error
message.  Often it is convenient to write and use subroutines,
conventionally called @code{xrealloc} and @code{xreallocarray},
that take care of the error message
as @code{xmalloc} does for @code{malloc}:
@cindex @code{xrealloc} and @code{xreallocarray} functions

@smallexample
void *
xreallocarray (void *ptr, size_t nmemb, size_t size)
@{
  void *value = reallocarray (ptr, nmemb, size);
  if (value == 0)
    fatal ("Virtual memory exhausted");
  return value;
@}

void *
xrealloc (void *ptr, size_t size)
@{
  return xreallocarray (ptr, 1, size);
@}
@end smallexample

You can also use @code{realloc} or @code{reallocarray} to make a block
smaller.  The reason you would do this is to avoid tying up a lot of memory
space when only a little is needed.
@comment The following is no longer true with the new malloc.
@comment But it seems wise to keep the warning for other implementations.
In several allocation implementations, making a block smaller sometimes
necessitates copying it, so it can fail if no other space is available.

@strong{Portability Notes:}

@itemize @bullet
@item
Portable programs should not attempt to reallocate blocks to be size zero.
On other implementations if @var{ptr} is non-null, @code{realloc (ptr, 0)}
might free the block and return a non-null pointer to a size-zero
object, or it might fail and return @code{NULL} without freeing the block.
The ISO C17 standard allows these variations.

@item
In @theglibc{}, reallocation fails if the resulting block
would exceed @code{PTRDIFF_MAX} in size, to avoid problems with programs
that subtract pointers or use signed indexes.  Other implementations may
succeed, leading to undefined behavior later.

@item
In @theglibc{}, if the new size is the same as the old, @code{realloc} and
@code{reallocarray} are guaranteed to change nothing and return the same
address that you gave.  However, POSIX and ISO C allow the functions
to relocate the object or fail in this situation.
@end itemize

@node Allocating Cleared Space
@subsubsection Allocating Cleared Space

The function @code{calloc} allocates memory and clears it to zero.  It
is declared in @file{stdlib.h}.
@pindex stdlib.h

@deftypefun {void *} calloc (size_t @var{count}, size_t @var{eltsize})
@standards{ISO, malloc.h}
@standards{ISO, stdlib.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}
@c Same caveats as malloc.

@c __libc_calloc @asulock @aculock @acsfd @acsmem
@c  *__malloc_hook dup unguarded
@c  memset dup ok
@c  arena_get @asulock @aculock @acsfd @acsmem
@c   arena_lock dup @asulock @aculock @acsfd @acsmem
@c  top dup ok
@c  chunksize dup ok
@c  heap_for_ptr dup ok
@c  _int_malloc dup @acsfd @acsmem
@c  arena_get_retry dup @asulock @aculock @acsfd @acsmem
@c  mutex_unlock dup @aculock
@c  mem2chunk dup ok
@c  chunk_is_mmapped dup ok
@c  MALLOC_ZERO ok
@c   memset dup ok
This function allocates a block long enough to contain a vector of
@var{count} elements, each of size @var{eltsize}.  Its contents are
cleared to zero before @code{calloc} returns.
@end deftypefun

You could define @code{calloc} as follows:

@smallexample
void *
calloc (size_t count, size_t eltsize)
@{
  void *value = reallocarray (0, count, eltsize);
  if (value != 0)
    memset (value, 0, count * eltsize);
  return value;
@}
@end smallexample

But in general, it is not guaranteed that @code{calloc} calls
@code{reallocarray} and @code{memset} internally.  For example, if the
@code{calloc} implementation knows for other reasons that the new
memory block is zero, it need not zero out the block again with
@code{memset}.  Also, if an application provides its own
@code{reallocarray} outside the C library, @code{calloc} might not use
that redefinition.  @xref{Replacing malloc}.

@node Aligned Memory Blocks
@subsubsection Allocating Aligned Memory Blocks

@cindex page boundary
@cindex alignment (with @code{malloc})
@pindex stdlib.h
The address of a block returned by @code{malloc} or @code{realloc} in
@gnusystems{} is always a multiple of eight (or sixteen on 64-bit
systems).  If you need a block whose address is a multiple of a higher
power of two than that, use @code{aligned_alloc} or @code{posix_memalign}.
@code{aligned_alloc} and @code{posix_memalign} are declared in
@file{stdlib.h}.

@deftypefun {void *} aligned_alloc (size_t @var{alignment}, size_t @var{size})
@standards{???, stdlib.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}
@c Alias to memalign.
The @code{aligned_alloc} function allocates a block of @var{size} bytes whose
address is a multiple of @var{alignment}.  The @var{alignment} must be a
power of two.

The @code{aligned_alloc} function returns a null pointer on error and sets
@code{errno} to one of the following values:

@table @code
@item ENOMEM
There was insufficient memory available to satisfy the request.

@item EINVAL
@var{alignment} is not a power of two.

This function was introduced in @w{ISO C11} and hence may have better
portability to modern non-POSIX systems than @code{posix_memalign}.
@end table

@end deftypefun

@deftypefun {void *} memalign (size_t @var{boundary}, size_t @var{size})
@standards{BSD, malloc.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}
@c Same issues as malloc.  The padding bytes are safely freed in
@c _int_memalign, with the arena still locked.

@c __libc_memalign @asulock @aculock @acsfd @acsmem
@c  *__memalign_hook dup unguarded
@c  __libc_malloc dup @asulock @aculock @acsfd @acsmem
@c  arena_get dup @asulock @aculock @acsfd @acsmem
@c  _int_memalign @acsfd @acsmem
@c   _int_malloc dup @acsfd @acsmem
@c   checked_request2size dup ok
@c   mem2chunk dup ok
@c   chunksize dup ok
@c   chunk_is_mmapped dup ok
@c   set_head dup ok
@c   chunk2mem dup ok
@c   set_inuse_bit_at_offset dup ok
@c   set_head_size dup ok
@c   _int_free (have_lock) dup @acsfd @acsmem
@c   chunk_at_offset dup ok
@c   check_inuse_chunk dup ok
@c  arena_get_retry dup @asulock @aculock @acsfd @acsmem
@c  mutex_unlock dup @aculock
The @code{memalign} function allocates a block of @var{size} bytes whose
address is a multiple of @var{boundary}.  The @var{boundary} must be a
power of two!  The function @code{memalign} works by allocating a
somewhat larger block, and then returning an address within the block
that is on the specified boundary.

The @code{memalign} function returns a null pointer on error and sets
@code{errno} to one of the following values:

@table @code
@item ENOMEM
There was insufficient memory available to satisfy the request.

@item EINVAL
@var{boundary} is not a power of two.

@end table

The @code{memalign} function is obsolete and @code{aligned_alloc} or
@code{posix_memalign} should be used instead.
@end deftypefun

@deftypefun int posix_memalign (void **@var{memptr}, size_t @var{alignment}, size_t @var{size})
@standards{POSIX, stdlib.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asulock{}}@acunsafe{@aculock{} @acsfd{} @acsmem{}}}
@c Calls memalign unless the requirements are not met (powerof2 macro is
@c safe given an automatic variable as an argument) or there's a
@c memalign hook (accessed unguarded, but safely).
The @code{posix_memalign} function is similar to the @code{memalign}
function in that it returns a buffer of @var{size} bytes aligned to a
multiple of @var{alignment}.  But it adds one requirement to the
parameter @var{alignment}: the value must be a power of two multiple of
@code{sizeof (void *)}.

If the function succeeds in allocation memory a pointer to the allocated
memory is returned in @code{*@var{memptr}} and the return value is zero.
Otherwise the function returns an error value indicating the problem.
The possible error values returned are:

@table @code
@item ENOMEM
There was insufficient memory available to satisfy the request.

@item EINVAL
@var{alignment} is not a power of two multiple of @code{sizeof (void *)}.

@end table

This function was introduced in POSIX 1003.1d.  Although this function is
superseded by @code{aligned_alloc}, it is more portable to older POSIX
systems that do not support @w{ISO C11}.
@end deftypefun

@deftypefun {void *} valloc (size_t @var{size})
@standards{BSD, malloc.h}
@standards{BSD, stdlib.h}
@safety{@prelim{}@mtunsafe{@mtuinit{}}@asunsafe{@asuinit{} @asulock{}}@acunsafe{@acuinit{} @aculock{} @acsfd{} @acsmem{}}}
@c __libc_valloc @mtuinit @asuinit @asulock @aculock @acsfd @acsmem
@c  ptmalloc_init (once) @mtsenv @asulock @aculock @acsfd @acsmem
@c   _dl_addr @asucorrupt? @aculock
@c    __rtld_lock_lock_recursive (dl_load_lock) @asucorrupt? @aculock
@c    _dl_find_dso_for_object ok, iterates over dl_ns and its _ns_loaded objs
@c      the ok above assumes no partial updates on dl_ns and _ns_loaded
@c      that could confuse a _dl_addr call in a signal handler
@c     _dl_addr_inside_object ok
@c    determine_info ok
@c    __rtld_lock_unlock_recursive (dl_load_lock) @aculock
@c   *_environ @mtsenv
@c   next_env_entry ok
@c   strcspn dup ok
@c   __libc_mallopt dup @mtasuconst:mallopt [setting mp_]
@c   *__malloc_initialize_hook unguarded, ok
@c  *__memalign_hook dup ok, unguarded
@c  arena_get dup @asulock @aculock @acsfd @acsmem
@c  _int_valloc @acsfd @acsmem
@c   malloc_consolidate dup ok
@c   _int_memalign dup @acsfd @acsmem
@c  arena_get_retry dup @asulock @aculock @acsfd @acsmem
@c  _int_memalign dup @acsfd @acsmem
@c  mutex_unlock dup @aculock
Using @code{valloc} is like using @code{memalign} and passing the page size
as the value of the first argument.  It is implemented like this:

@smallexample
void *
valloc (size_t size)
@{
  return memalign (getpagesize (), size);
@}
@end smallexample

@ref{Query Memory Parameters} for more information about the memory
subsystem.

The @code{valloc} function is obsolete and @code{aligned_alloc} or
@code{posix_memalign} should be used instead.
@end deftypefun

You can determine the alignment of a pointer with the
@code{memalignment} function.

@deftypefun size_t memalignment (void *@var{p})
@standards{C23, stdlib.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

This function, defined in C23, returns the alignment of @var{p}, as a
power of two.  If @var{p} is a null pointer, it returns zero.  C23
requires @var{p} to be a valid pointer to an object or a null pointer;
as a GNU extension, @theglibc{} supports this function on arbitrary
bit patterns of pointer type.

This function was added to the C23 standard to support unconventional
platforms where a pointer's low-order bits are unrelated to alignment.
For conventional platforms, one can instead cast the pointer to
@code{uintptr_t} and then test the low order bits:
this is portable to pre-C23 and is typically a bit faster.

For example, if you want to read an @code{int}
addressed by possibly-misaligned pointer @code{p},
the following pre-C23 code works on all conventional platforms:

@smallexample
int i;
if (((uintptr_t) p & (alignof (int) - 1)) != 0)
  memcpy (&i, p, sizeof i);
else
  i = *p;
@end smallexample

However, it might not work on unconventional platforms, where one
would need something like the following C23 code:

@smallexample
int i;
if (memalignment (p) < alignof (int))
  memcpy (&i, p, sizeof i);
else
  i = *p;
@end smallexample

However, for this particular case, performance does not improve if
different code is used for aligned and unaligned pointers,
and the following code is preferable:

@smallexample
int i;
memcpy (&i, p, sizeof i);
@end smallexample

The compiler will generate the most efficient way to access unaligned
data for the architecture, optimizing away the @code{memcpy} call.
@end deftypefun

@node Malloc Tunable Parameters
@subsubsection Malloc Tunable Parameters

You can adjust some parameters for dynamic memory allocation with the
@code{mallopt} function.  This function is the general SVID/XPG
interface, defined in @file{malloc.h}.
@pindex malloc.h

@deftypefun int mallopt (int @var{param}, int @var{value})
@safety{@prelim{}@mtunsafe{@mtuinit{} @mtasuconst{:mallopt}}@asunsafe{@asuinit{} @asulock{}}@acunsafe{@acuinit{} @aculock{}}}
@c __libc_mallopt @mtuinit @mtasuconst:mallopt @asuinit @asulock @aculock
@c  ptmalloc_init (once) dup @mtsenv @asulock @aculock @acsfd @acsmem
@c  mutex_lock (main_arena->mutex) @asulock @aculock
@c  malloc_consolidate dup ok
@c  set_max_fast ok
@c  mutex_unlock dup @aculock

When calling @code{mallopt}, the @var{param} argument specifies the
parameter to be set, and @var{value} the new value to be set.  Possible
choices for @var{param}, as defined in @file{malloc.h}, are:

@vtable @code
@item M_MMAP_MAX
The maximum number of chunks to allocate with @code{mmap}.  Setting this
to zero disables all use of @code{mmap}.

The default value of this parameter is @code{65536}.

This parameter can also be set for the process at startup by setting the
environment variable @env{MALLOC_MMAP_MAX_} to the desired value.

@item M_MMAP_THRESHOLD
All chunks larger than this value are allocated outside the normal
heap, using the @code{mmap} system call.  This way it is guaranteed
that the memory for these chunks can be returned to the system on
@code{free}.  Note that requests smaller than this threshold might still
be allocated via @code{mmap}.

If this parameter is not set, the default value is set as 128 KiB and the
threshold is adjusted dynamically to suit the allocation patterns of the
program. If the parameter is set, the dynamic adjustment is disabled and the
value is set statically to the input value.

This parameter can also be set for the process at startup by setting the
environment variable @env{MALLOC_MMAP_THRESHOLD_} to the desired value.
@comment TODO: @item M_MXFAST

@item M_PERTURB
If non-zero, memory blocks are filled with values depending on some
low order bits of this parameter when they are allocated (except when
allocated by @code{calloc}) and freed.  This can be used to debug the
use of uninitialized or freed heap memory.  Note that this option does not
guarantee that the freed block will have any specific values.  It only
guarantees that the content the block had before it was freed will be
overwritten.

The default value of this parameter is @code{0}.

This parameter can also be set for the process at startup by setting the
environment variable @env{MALLOC_PERTURB_} to the desired value.

@item M_TOP_PAD
This parameter determines the amount of extra memory to obtain from the system
when an arena needs to be extended.  It also specifies the number of bytes to
retain when shrinking an arena.  This provides the necessary hysteresis in heap
size such that excessive amounts of system calls can be avoided.

The default value of this parameter is @code{0}.

This parameter can also be set for the process at startup by setting the
environment variable @env{MALLOC_TOP_PAD_} to the desired value.

@item M_TRIM_THRESHOLD
This is the minimum size (in bytes) of the top-most, releasable chunk
that will trigger a system call in order to return memory to the system.

If this parameter is not set, the default value is set as 128 KiB and the
threshold is adjusted dynamically to suit the allocation patterns of the
program. If the parameter is set, the dynamic adjustment is disabled and the
value is set statically to the provided input.

This parameter can also be set for the process at startup by setting the
environment variable @env{MALLOC_TRIM_THRESHOLD_} to the desired value.

@item M_ARENA_TEST
This parameter specifies the number of arenas that can be created before the
test on the limit to the number of arenas is conducted. The value is ignored if
@code{M_ARENA_MAX} is set.

The default value of this parameter is 2 on 32-bit systems and 8 on 64-bit
systems.

This parameter can also be set for the process at startup by setting the
environment variable @env{MALLOC_ARENA_TEST} to the desired value.

@item M_ARENA_MAX
This parameter sets the number of arenas to use regardless of the number of
cores in the system.

The default value of this tunable is @code{0}, meaning that the limit on the
number of arenas is determined by the number of CPU cores online. For 32-bit
systems the limit is twice the number of cores online and on 64-bit systems, it
is eight times the number of cores online.  Note that the default value is not
derived from the default value of M_ARENA_TEST and is computed independently.

This parameter can also be set for the process at startup by setting the
environment variable @env{MALLOC_ARENA_MAX} to the desired value.
@end vtable

@end deftypefun

@node Heap Consistency Checking
@subsubsection Heap Consistency Checking

@cindex heap consistency checking
@cindex consistency checking, of heap

You can ask @code{malloc} to check the consistency of dynamic memory by
using the @code{mcheck} function and preloading the malloc debug library
@file{libc_malloc_debug} using the @var{LD_PRELOAD} environment variable.
This function is a GNU extension, declared in @file{mcheck.h}.
@pindex mcheck.h

@deftypefun int mcheck (void (*@var{abortfn}) (enum mcheck_status @var{status}))
@standards{GNU, mcheck.h}
@safety{@prelim{}@mtunsafe{@mtasurace{:mcheck} @mtasuconst{:malloc_hooks}}@asunsafe{@asucorrupt{}}@acunsafe{@acucorrupt{}}}
@c The hooks must be set up before malloc is first used, which sort of
@c implies @mtuinit/@asuinit but since the function is a no-op if malloc
@c was already used, that doesn't pose any safety issues.  The actual
@c problem is with the hooks, designed for single-threaded
@c fully-synchronous operation: they manage an unguarded linked list of
@c allocated blocks, and get temporarily overwritten before calling the
@c allocation functions recursively while holding the old hooks.  There
@c are no guards for thread safety, and inconsistent hooks may be found
@c within signal handlers or left behind in case of cancellation.

Calling @code{mcheck} tells @code{malloc} to perform occasional
consistency checks.  These will catch things such as writing
past the end of a block that was allocated with @code{malloc}.

The @var{abortfn} argument is the function to call when an inconsistency
is found.  If you supply a null pointer, then @code{mcheck} uses a
default function which prints a message and calls @code{abort}
(@pxref{Aborting a Program}).  The function you supply is called with
one argument, which says what sort of inconsistency was detected; its
type is described below.

It is too late to begin allocation checking once you have allocated
anything with @code{malloc}.  So @code{mcheck} does nothing in that
case.  The function returns @code{-1} if you call it too late, and
@code{0} otherwise (when it is successful).

The easiest way to arrange to call @code{mcheck} early enough is to use
the option @samp{-lmcheck} when you link your program; then you don't
need to modify your program source at all.  Alternatively you might use
a debugger to insert a call to @code{mcheck} whenever the program is
started, for example these gdb commands will automatically call @code{mcheck}
whenever the program starts:

@smallexample
(gdb) break main
Breakpoint 1, main (argc=2, argv=0xbffff964) at whatever.c:10
(gdb) command 1
Type commands for when breakpoint 1 is hit, one per line.
End with a line saying just "end".
>call mcheck(0)
>continue
>end
(gdb) @dots{}
@end smallexample

This will however only work if no initialization function of any object
involved calls any of the @code{malloc} functions since @code{mcheck}
must be called before the first such function.

@end deftypefun

@deftypefun {enum mcheck_status} mprobe (void *@var{pointer})
@safety{@prelim{}@mtunsafe{@mtasurace{:mcheck} @mtasuconst{:malloc_hooks}}@asunsafe{@asucorrupt{}}@acunsafe{@acucorrupt{}}}
@c The linked list of headers may be modified concurrently by other
@c threads, and it may find a partial update if called from a signal
@c handler.  It's mostly read only, so cancelling it might be safe, but
@c it will modify global state that, if cancellation hits at just the
@c right spot, may be left behind inconsistent.  This path is only taken
@c if checkhdr finds an inconsistency.  If the inconsistency could only
@c occur because of earlier undefined behavior, that wouldn't be an
@c additional safety issue problem, but because of the other concurrency
@c issues in the mcheck hooks, the apparent inconsistency could be the
@c result of mcheck's own internal data race.  So, AC-Unsafe it is.

The @code{mprobe} function lets you explicitly check for inconsistencies
in a particular allocated block.  You must have already called
@code{mcheck} at the beginning of the program, to do its occasional
checks; calling @code{mprobe} requests an additional consistency check
to be done at the time of the call.

The argument @var{pointer} must be a pointer returned by @code{malloc}
or @code{realloc}.  @code{mprobe} returns a value that says what
inconsistency, if any, was found.  The values are described below.
@end deftypefun

@deftp {Data Type} {enum mcheck_status}
This enumerated type describes what kind of inconsistency was detected
in an allocated block, if any.  Here are the possible values:

@table @code
@item MCHECK_DISABLED
@code{mcheck} was not called before the first allocation.
No consistency checking can be done.
@item MCHECK_OK
No inconsistency detected.
@item MCHECK_HEAD
The data immediately before the block was modified.
This commonly happens when an array index or pointer
is decremented too far.
@item MCHECK_TAIL
The data immediately after the block was modified.
This commonly happens when an array index or pointer
is incremented too far.
@item MCHECK_FREE
The block was already freed.
@end table
@end deftp

Another possibility to check for and guard against bugs in the use of
@code{malloc}, @code{realloc} and @code{free} is to set the environment
variable @code{MALLOC_CHECK_}.  When @code{MALLOC_CHECK_} is set to a
non-zero value less than 4, a special (less efficient) implementation is
used which is designed to be tolerant against simple errors, such as
double calls of @code{free} with the same argument, or overruns of a
single byte (off-by-one bugs).  Not all such errors can be protected
against, however, and memory leaks can result.  Like in the case of
@code{mcheck}, one would need to preload the @file{libc_malloc_debug}
library to enable @code{MALLOC_CHECK_} functionality.  Without this
preloaded library, setting @code{MALLOC_CHECK_} will have no effect.

Any detected heap corruption results in immediate termination of the
process.

There is one problem with @code{MALLOC_CHECK_}: in SUID or SGID binaries
it could possibly be exploited since diverging from the normal programs
behavior it now writes something to the standard error descriptor.
Therefore the use of @code{MALLOC_CHECK_} is disabled by default for
SUID and SGID binaries.

So, what's the difference between using @code{MALLOC_CHECK_} and linking
with @samp{-lmcheck}?  @code{MALLOC_CHECK_} is orthogonal with respect to
@samp{-lmcheck}.  @samp{-lmcheck} has been added for backward
compatibility.  Both @code{MALLOC_CHECK_} and @samp{-lmcheck} should
uncover the same bugs - but using @code{MALLOC_CHECK_} you don't need to
recompile your application.

@c __morecore, __after_morecore_hook are undocumented
@c It's not clear whether to document them.

@node Statistics of Malloc
@subsubsection Statistics for Memory Allocation with @code{malloc}

@cindex allocation statistics
You can get information about dynamic memory allocation by calling the
@code{mallinfo2} function.  This function and its associated data type
are declared in @file{malloc.h}; they are an extension of the standard
SVID/XPG version.
@pindex malloc.h

@deftp {Data Type} {struct mallinfo2}
@standards{GNU, malloc.h}
This structure type is used to return information about the dynamic
memory allocator.  It contains the following members:

@table @code
@item size_t arena
This is the total size of memory allocated with @code{sbrk} by
@code{malloc}, in bytes.

@item size_t ordblks
This is the number of chunks not in use.  (The memory allocator
internally gets chunks of memory from the operating system, and then
carves them up to satisfy individual @code{malloc} requests;
@pxref{The GNU Allocator}.)

@item size_t smblks
This field is unused.

@item size_t hblks
This is the total number of chunks allocated with @code{mmap}.

@item size_t hblkhd
This is the total size of memory allocated with @code{mmap}, in bytes.

@item size_t usmblks
This field is unused and always 0.

@item size_t fsmblks
This field is unused.

@item size_t uordblks
This is the total size of memory occupied by chunks handed out by
@code{malloc}.

@item size_t fordblks
This is the total size of memory occupied by free (not in use) chunks.

@item size_t keepcost
This is the size of the top-most releasable chunk that normally
borders the end of the heap (i.e., the high end of the virtual address
space's data segment).

@end table
@end deftp

@deftypefun {struct mallinfo2} mallinfo2 (void)
@standards{SVID, malloc.h}
@safety{@prelim{}@mtunsafe{@mtuinit{} @mtasuconst{:mallopt}}@asunsafe{@asuinit{} @asulock{}}@acunsafe{@acuinit{} @aculock{}}}
@c Accessing mp_.n_mmaps and mp_.max_mmapped_mem, modified with atomics
@c but non-atomically elsewhere, may get us inconsistent results.  We
@c mark the statistics as unsafe, rather than the fast-path functions
@c that collect the possibly inconsistent data.

@c __libc_mallinfo2 @mtuinit @mtasuconst:mallopt @asuinit @asulock @aculock
@c  ptmalloc_init (once) dup @mtsenv @asulock @aculock @acsfd @acsmem
@c  mutex_lock dup @asulock @aculock
@c  int_mallinfo @mtasuconst:mallopt [mp_ access on main_arena]
@c   malloc_consolidate dup ok
@c   check_malloc_state dup ok/disabled
@c   chunksize dup ok
@c   fastbin dupo ok
@c   bin_at dup ok
@c   last dup ok
@c  mutex_unlock @aculock

This function returns information about the current dynamic memory usage
in a structure of type @code{struct mallinfo2}.
@end deftypefun

@node Summary of Malloc
@subsubsection Summary of @code{malloc}-Related Functions

Here is a summary of the functions that work with @code{malloc}:

@table @code
@item void *malloc (size_t @var{size})
Allocate a block of @var{size} bytes.  @xref{Basic Allocation}.

@item void free (void *@var{addr})
Free a block previously allocated by @code{malloc}.  @xref{Freeing after
Malloc}.

@item void *realloc (void *@var{addr}, size_t @var{size})
Make a block previously allocated by @code{malloc} larger or smaller,
possibly by copying it to a new location.  @xref{Changing Block Size}.

@item void *reallocarray (void *@var{ptr}, size_t @var{nmemb}, size_t @var{size})
Change the size of a block previously allocated by @code{malloc} to
@code{@var{nmemb} * @var{size}} bytes as with @code{realloc}.  @xref{Changing
Block Size}.

@item void *calloc (size_t @var{count}, size_t @var{eltsize})
Allocate a block of @var{count} * @var{eltsize} bytes using
@code{malloc}, and set its contents to zero.  @xref{Allocating Cleared
Space}.

@item void *valloc (size_t @var{size})
Allocate a block of @var{size} bytes, starting on a page boundary.
@xref{Aligned Memory Blocks}.

@item void *aligned_alloc (size_t @var{alignment}, size_t @var{size})
Allocate a block of @var{size} bytes, starting on an address that is a
multiple of @var{alignment}.  @xref{Aligned Memory Blocks}.

@item int posix_memalign (void **@var{memptr}, size_t @var{alignment}, size_t @var{size})
Allocate a block of @var{size} bytes, starting on an address that is a
multiple of @var{alignment}.  @xref{Aligned Memory Blocks}.

@item void *memalign (size_t @var{boundary}, size_t @var{size})
Allocate a block of @var{size} bytes, starting on an address that is a
multiple of @var{boundary}.  @xref{Aligned Memory Blocks}.

@item int mallopt (int @var{param}, int @var{value})
Adjust a tunable parameter.  @xref{Malloc Tunable Parameters}.

@item int mcheck (void (*@var{abortfn}) (void))
Tell @code{malloc} to perform occasional consistency checks on
dynamically allocated memory, and to call @var{abortfn} when an
inconsistency is found.  @xref{Heap Consistency Checking}.

@item struct mallinfo2 mallinfo2 (void)
Return information about the current dynamic memory usage.
@xref{Statistics of Malloc}.
@end table

@node Allocation Debugging
@subsection Allocation Debugging
@cindex allocation debugging
@cindex malloc debugger

A complicated task when programming with languages which do not use
garbage collected dynamic memory allocation is to find memory leaks.
Long running programs must ensure that dynamically allocated objects are
freed at the end of their lifetime.  If this does not happen the system
runs out of memory, sooner or later.

The @code{malloc} implementation in @theglibc{} provides some
simple means to detect such leaks and obtain some information to find
the location.  To do this the application must be started in a special
mode which is enabled by an environment variable.  There are no speed
penalties for the program if the debugging mode is not enabled.

@menu
* Tracing malloc::               How to install the tracing functionality.
* Using the Memory Debugger::    Example programs excerpts.
* Tips for the Memory Debugger:: Some more or less clever ideas.
* Interpreting the traces::      What do all these lines mean?
@end menu

@node Tracing malloc
@subsubsection How to install the tracing functionality

@deftypefun void mtrace (void)
@standards{GNU, mcheck.h}
@safety{@prelim{}@mtunsafe{@mtsenv{} @mtasurace{:mtrace} @mtuinit{}}@asunsafe{@asuinit{} @ascuheap{} @asucorrupt{} @asulock{}}@acunsafe{@acuinit{} @acucorrupt{} @aculock{} @acsfd{} @acsmem{}}}
@c Like the mcheck hooks, these are not designed with thread safety in
@c mind, because the hook pointers are temporarily modified without
@c regard to other threads, signals or cancellation.

@c mtrace @mtuinit @mtasurace:mtrace @mtsenv @asuinit @ascuheap @asucorrupt @acuinit @acucorrupt @aculock @acsfd @acsmem
@c  __libc_secure_getenv dup @mtsenv
@c  malloc dup @ascuheap @acsmem
@c  fopen dup @ascuheap @asulock @aculock @acsmem @acsfd
@c  fcntl dup ok
@c  setvbuf dup @aculock
@c  fprintf dup (on newly-created stream) @aculock
@c  __cxa_atexit (once) dup @asulock @aculock @acsmem
@c  free dup @ascuheap @acsmem
The @code{mtrace} function provides a way to trace memory allocation
events in the program that calls it.  It is disabled by default in the
library and can be enabled by preloading the debugging library
@file{libc_malloc_debug} using the @code{LD_PRELOAD} environment
variable.

When the @code{mtrace} function is called it looks for an environment
variable named @code{MALLOC_TRACE}.  This variable is supposed to
contain a valid file name.  The user must have write access.  If the
file already exists it is truncated.  If the environment variable is not
set or it does not name a valid file which can be opened for writing
nothing is done.  The behavior of @code{malloc} etc. is not changed.
For obvious reasons this also happens if the application is installed
with the SUID or SGID bit set.

If the named file is successfully opened, @code{mtrace} installs special
handlers for the functions @code{malloc}, @code{realloc}, and
@code{free}.  From then on, all uses of these functions are traced and
protocolled into the file.  There is now of course a speed penalty for all
calls to the traced functions so tracing should not be enabled during normal
use.

This function is a GNU extension and generally not available on other
systems.  The prototype can be found in @file{mcheck.h}.
@end deftypefun

@deftypefun void muntrace (void)
@standards{GNU, mcheck.h}
@safety{@prelim{}@mtunsafe{@mtasurace{:mtrace} @mtslocale{}}@asunsafe{@asucorrupt{} @ascuheap{}}@acunsafe{@acucorrupt{} @acsmem{} @aculock{} @acsfd{}}}

@c muntrace @mtasurace:mtrace @mtslocale @asucorrupt @ascuheap @acucorrupt @acsmem @aculock @acsfd
@c  fprintf (fputs) dup @mtslocale @asucorrupt @ascuheap @acsmem @aculock @acucorrupt
@c  fclose dup @ascuheap @asulock @aculock @acsmem @acsfd
The @code{muntrace} function can be called after @code{mtrace} was used
to enable tracing the @code{malloc} calls.  If no (successful) call of
@code{mtrace} was made @code{muntrace} does nothing.

Otherwise it deinstalls the handlers for @code{malloc}, @code{realloc},
and @code{free} and then closes the protocol file.  No calls are
protocolled anymore and the program runs again at full speed.

This function is a GNU extension and generally not available on other
systems.  The prototype can be found in @file{mcheck.h}.
@end deftypefun

@node Using the Memory Debugger
@subsubsection Example program excerpts

Even though the tracing functionality does not influence the runtime
behavior of the program it is not a good idea to call @code{mtrace} in
all programs.  Just imagine that you debug a program using @code{mtrace}
and all other programs used in the debugging session also trace their
@code{malloc} calls.  The output file would be the same for all programs
and thus is unusable.  Therefore one should call @code{mtrace} only if
compiled for debugging.  A program could therefore start like this:

@example
#include <mcheck.h>

int
main (int argc, char *argv[])
@{
#ifdef DEBUGGING
  mtrace ();
#endif
  @dots{}
@}
@end example

This is all that is needed if you want to trace the calls during the
whole runtime of the program.  Alternatively you can stop the tracing at
any time with a call to @code{muntrace}.  It is even possible to restart
the tracing again with a new call to @code{mtrace}.  But this can cause
unreliable results since there may be calls of the functions which are
not called.  Please note that not only the application uses the traced
functions, also libraries (including the C library itself) use these
functions.

This last point is also why it is not a good idea to call @code{muntrace}
before the program terminates.  The libraries are informed about the
termination of the program only after the program returns from
@code{main} or calls @code{exit} and so cannot free the memory they use
before this time.

So the best thing one can do is to call @code{mtrace} as the very first
function in the program and never call @code{muntrace}.  So the program
traces almost all uses of the @code{malloc} functions (except those
calls which are executed by constructors of the program or used
libraries).

@node Tips for the Memory Debugger
@subsubsection Some more or less clever ideas

You know the situation.  The program is prepared for debugging and in
all debugging sessions it runs well.  But once it is started without
debugging the error shows up.  A typical example is a memory leak that
becomes visible only when we turn off the debugging.  If you foresee
such situations you can still win.  Simply use something equivalent to
the following little program:

@example
#include <mcheck.h>
#include <signal.h>

static void
enable (int sig)
@{
  mtrace ();
  signal (SIGUSR1, enable);
@}

static void
disable (int sig)
@{
  muntrace ();
  signal (SIGUSR2, disable);
@}

int
main (int argc, char *argv[])
@{
  @dots{}

  signal (SIGUSR1, enable);
  signal (SIGUSR2, disable);

  @dots{}
@}
@end example

I.e., the user can start the memory debugger any time s/he wants if the
program was started with @code{MALLOC_TRACE} set in the environment.
The output will of course not show the allocations which happened before
the first signal but if there is a memory leak this will show up
nevertheless.

@node Interpreting the traces
@subsubsection Interpreting the traces

If you take a look at the output it will look similar to this:

@example
= Start
@ [0x8048209] - 0x8064cc8
@ [0x8048209] - 0x8064ce0
@ [0x8048209] - 0x8064cf8
@ [0x80481eb] + 0x8064c48 0x14
@ [0x80481eb] + 0x8064c60 0x14
@ [0x80481eb] + 0x8064c78 0x14
@ [0x80481eb] + 0x8064c90 0x14
= End
@end example

What this all means is not really important since the trace file is not
meant to be read by a human.  Therefore no attention is given to
readability.  Instead there is a program which comes with @theglibc{}
which interprets the traces and outputs a summary in an
user-friendly way.  The program is called @code{mtrace} (it is in fact a
Perl script) and it takes one or two arguments.  In any case the name of
the file with the trace output must be specified.  If an optional
argument precedes the name of the trace file this must be the name of
the program which generated the trace.

@example
drepper$ mtrace tst-mtrace log
No memory leaks.
@end example

In this case the program @code{tst-mtrace} was run and it produced a
trace file @file{log}.  The message printed by @code{mtrace} shows there
are no problems with the code, all allocated memory was freed
afterwards.

If we call @code{mtrace} on the example trace given above we would get a
different output:

@example
drepper$ mtrace errlog
- 0x08064cc8 Free 2 was never alloc'd 0x8048209
- 0x08064ce0 Free 3 was never alloc'd 0x8048209
- 0x08064cf8 Free 4 was never alloc'd 0x8048209

Memory not freed:
-----------------
   Address     Size     Caller
0x08064c48     0x14  at 0x80481eb
0x08064c60     0x14  at 0x80481eb
0x08064c78     0x14  at 0x80481eb
0x08064c90     0x14  at 0x80481eb
@end example

We have called @code{mtrace} with only one argument and so the script
has no chance to find out what is meant with the addresses given in the
trace.  We can do better:

@example
drepper$ mtrace tst errlog
- 0x08064cc8 Free 2 was never alloc'd /home/drepper/tst.c:39
- 0x08064ce0 Free 3 was never alloc'd /home/drepper/tst.c:39
- 0x08064cf8 Free 4 was never alloc'd /home/drepper/tst.c:39

Memory not freed:
-----------------
   Address     Size     Caller
0x08064c48     0x14  at /home/drepper/tst.c:33
0x08064c60     0x14  at /home/drepper/tst.c:33
0x08064c78     0x14  at /home/drepper/tst.c:33
0x08064c90     0x14  at /home/drepper/tst.c:33
@end example

Suddenly the output makes much more sense and the user can see
immediately where the function calls causing the trouble can be found.

Interpreting this output is not complicated.  There are at most two
different situations being detected.  First, @code{free} was called for
pointers which were never returned by one of the allocation functions.
This is usually a very bad problem and what this looks like is shown in
the first three lines of the output.  Situations like this are quite
rare and if they appear they show up very drastically: the program
normally crashes.

The other situation which is much harder to detect are memory leaks.  As
you can see in the output the @code{mtrace} function collects all this
information and so can say that the program calls an allocation function
from line 33 in the source file @file{/home/drepper/tst-mtrace.c} four
times without freeing this memory before the program terminates.
Whether this is a real problem remains to be investigated.

@node Replacing malloc
@subsection Replacing @code{malloc}

@cindex @code{malloc} replacement
@cindex @code{LD_PRELOAD} and @code{malloc}
@cindex alternative @code{malloc} implementations
@cindex customizing @code{malloc}
@cindex interposing @code{malloc}
@cindex preempting @code{malloc}
@cindex replacing @code{malloc}
@Theglibc{} supports replacing the built-in @code{malloc} implementation
with a different allocator with the same interface.  For dynamically
linked programs, this happens through ELF symbol interposition, either
using shared object dependencies or @code{LD_PRELOAD}.  For static
linking, the @code{malloc} replacement library must be linked in before
linking against @code{libc.a} (explicitly or implicitly).

Care must be taken not to use functionality from @theglibc{} that uses
@code{malloc} internally.  For example, the @code{fopen},
@code{opendir}, @code{dlopen}, and @code{pthread_setspecific} functions
currently use the @code{malloc} subsystem internally.  If the
replacement @code{malloc} or its dependencies use thread-local storage
(TLS), it must use the initial-exec TLS model, and not one of the
dynamic TLS variants.

@strong{Note:} Failure to provide a complete set of replacement
functions (that is, all the functions used by the application,
@theglibc{}, and other linked-in libraries) can lead to static linking
failures, and, at run time, to heap corruption and application crashes.
Replacement functions should implement the behavior documented for
their counterparts in @theglibc{}; for example, the replacement
@code{free} should also preserve @code{errno}.

The minimum set of functions which has to be provided by a custom
@code{malloc} is given in the table below.

@table @code
@item malloc
@item free
@item calloc
@item realloc
@end table

These @code{malloc}-related functions are required for @theglibc{} to
work.@footnote{Versions of @theglibc{} before 2.25 required that a
custom @code{malloc} defines @code{__libc_memalign} (with the same
interface as the @code{memalign} function).}

The @code{malloc} implementation in @theglibc{} provides additional
functionality not used by the library itself, but which is often used by
other system libraries and applications.  A general-purpose replacement
@code{malloc} implementation should provide definitions of these
functions, too.  Their names are listed in the following table.

@table @code
@item aligned_alloc
@item free_aligned_sized
@item free_sized
@item malloc_usable_size
@item memalign
@item posix_memalign
@item pvalloc
@item valloc
@end table

In addition, very old applications may use the obsolete @code{cfree}
function.

Further @code{malloc}-related functions such as @code{mallopt} or
@code{mallinfo2} will not have any effect or return incorrect statistics
when a replacement @code{malloc} is in use.  However, failure to replace
these functions typically does not result in crashes or other incorrect
application behavior, but may result in static linking failures.

There are other functions (@code{reallocarray}, @code{strdup}, etc.) in
@theglibc{} that are not listed above but return newly allocated memory to
callers.  Replacement of these functions is not supported and may produce
incorrect results.  @Theglibc{} implementations of these functions call
the replacement allocator functions whenever available, so they will work
correctly with @code{malloc} replacement.

@node Obstacks
@subsection Obstacks
@cindex obstacks

An @dfn{obstack} is a pool of memory containing a stack of objects.  You
can create any number of separate obstacks, and then allocate objects in
specified obstacks.  Within each obstack, the last object allocated must
always be the first one freed, but distinct obstacks are independent of
each other.

Aside from this one constraint of order of freeing, obstacks are totally
general: an obstack can contain any number of objects of any size.  They
are implemented with macros, so allocation is usually very fast as long as
the objects are usually small.  And the only space overhead per object is
the padding needed to start each object on a suitable boundary.

@menu
* Creating Obstacks::		How to declare an obstack in your program.
* Preparing for Obstacks::	Preparations needed before you can
				 use obstacks.
* Allocation in an Obstack::    Allocating objects in an obstack.
* Freeing Obstack Objects::     Freeing objects in an obstack.
* Obstack Functions::		The obstack functions are both
				 functions and macros.
* Growing Objects::             Making an object bigger by stages.
* Extra Fast Growing::		Extra-high-efficiency (though more
				 complicated) growing objects.
* Status of an Obstack::        Inquiries about the status of an obstack.
* Obstacks Data Alignment::     Controlling alignment of objects in obstacks.
* Obstack Chunks::              How obstacks obtain and release chunks;
				 efficiency considerations.
* Summary of Obstacks::
@end menu

@node Creating Obstacks
@subsubsection Creating Obstacks

The utilities for manipulating obstacks are declared in the header
file @file{obstack.h}.
@pindex obstack.h

@deftp {Data Type} {struct obstack}
@standards{GNU, obstack.h}
An obstack is represented by a data structure of type @code{struct
obstack}.  This structure has a small fixed size; it records the status
of the obstack and how to find the space in which objects are allocated.
It does not contain any of the objects themselves.  You should not try
to access the contents of the structure directly; use only the functions
described in this chapter.
@end deftp

You can declare variables of type @code{struct obstack} and use them as
obstacks, or you can allocate obstacks dynamically like any other kind
of object.  Dynamic allocation of obstacks allows your program to have a
variable number of different stacks.  (You can even allocate an
obstack structure in another obstack, but this is rarely useful.)

All the functions that work with obstacks require you to specify which
obstack to use.  You do this with a pointer of type @code{struct obstack
*}.  In the following, we often say ``an obstack'' when strictly
speaking the object at hand is such a pointer.

The objects in the obstack are packed into large blocks called
@dfn{chunks}.  The @code{struct obstack} structure points to a chain of
the chunks currently in use.

The obstack library obtains a new chunk whenever you allocate an object
that won't fit in the previous chunk.  Since the obstack library manages
chunks automatically, you don't need to pay much attention to them, but
you do need to supply a function which the obstack library should use to
get a chunk.  Usually you supply a function which uses @code{malloc}
directly or indirectly.  You must also supply a function to free a chunk.
These matters are described in the following section.

@node Preparing for Obstacks
@subsubsection Preparing for Using Obstacks

Each source file in which you plan to use the obstack functions
must include the header file @file{obstack.h}, like this:

@smallexample
#include <obstack.h>
@end smallexample

@findex obstack_chunk_alloc
@findex obstack_chunk_free
Also, if the source file uses the macro @code{obstack_init}, it must
declare or define two functions or macros that will be called by the
obstack library.  One, @code{obstack_chunk_alloc}, is used to allocate
the chunks of memory into which objects are packed.  The other,
@code{obstack_chunk_free}, is used to return chunks when the objects in
them are freed.  These macros should appear before any use of obstacks
in the source file.

Usually these are defined to use @code{malloc} via the intermediary
@code{xmalloc} (@pxref{Unconstrained Allocation}).  This is done with
the following pair of macro definitions:

@smallexample
#define obstack_chunk_alloc xmalloc
#define obstack_chunk_free free
@end smallexample

@noindent
Though the memory you get using obstacks really comes from @code{malloc},
using obstacks is faster because @code{malloc} is called less often, for
larger blocks of memory.  @xref{Obstack Chunks}, for full details.

At run time, before the program can use a @code{struct obstack} object
as an obstack, it must initialize the obstack by calling
@code{obstack_init}.

@deftypefun int obstack_init (struct obstack *@var{obstack-ptr})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acsafe{@acsmem{}}}
@c obstack_init @mtsrace:obstack-ptr @acsmem
@c  _obstack_begin @acsmem
@c    chunkfun = obstack_chunk_alloc (suggested malloc)
@c    freefun = obstack_chunk_free (suggested free)
@c   *chunkfun @acsmem
@c    obstack_chunk_alloc user-supplied
@c   *obstack_alloc_failed_handler user-supplied
@c    -> print_and_abort (default)
@c
@c print_and_abort
@c  _ dup @ascuintl
@c  fxprintf dup @asucorrupt @aculock @acucorrupt
@c  exit @acucorrupt?
Initialize obstack @var{obstack-ptr} for allocation of objects.  This
function calls the obstack's @code{obstack_chunk_alloc} function.  If
allocation of memory fails, the function pointed to by
@code{obstack_alloc_failed_handler} is called.  The @code{obstack_init}
function always returns 1 (Compatibility notice: Former versions of
obstack returned 0 if allocation failed).
@end deftypefun

Here are two examples of how to allocate the space for an obstack and
initialize it.  First, an obstack that is a static variable:

@smallexample
static struct obstack myobstack;
@dots{}
obstack_init (&myobstack);
@end smallexample

@noindent
Second, an obstack that is itself dynamically allocated:

@smallexample
struct obstack *myobstack_ptr
  = (struct obstack *) xmalloc (sizeof (struct obstack));

obstack_init (myobstack_ptr);
@end smallexample

@defvar obstack_alloc_failed_handler
@standards{GNU, obstack.h}
The value of this variable is a pointer to a function that
@code{obstack} uses when @code{obstack_chunk_alloc} fails to allocate
memory.  The default action is to print a message and abort.
You should supply a function that either calls @code{exit}
(@pxref{Program Termination}) or @code{longjmp} (@pxref{Non-Local
Exits}) and doesn't return.

@smallexample
void my_obstack_alloc_failed (void)
@dots{}
obstack_alloc_failed_handler = &my_obstack_alloc_failed;
@end smallexample

@end defvar

@node Allocation in an Obstack
@subsubsection Allocation in an Obstack
@cindex allocation (obstacks)

The most direct way to allocate an object in an obstack is with
@code{obstack_alloc}, which is invoked almost like @code{malloc}.

@deftypefun {void *} obstack_alloc (struct obstack *@var{obstack-ptr}, int @var{size})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_alloc @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  obstack_blank dup @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  obstack_finish dup @mtsrace:obstack-ptr @acucorrupt
This allocates an uninitialized block of @var{size} bytes in an obstack
and returns its address.  Here @var{obstack-ptr} specifies which obstack
to allocate the block in; it is the address of the @code{struct obstack}
object which represents the obstack.  Each obstack function or macro
requires you to specify an @var{obstack-ptr} as the first argument.

This function calls the obstack's @code{obstack_chunk_alloc} function if
it needs to allocate a new chunk of memory; it calls
@code{obstack_alloc_failed_handler} if allocation of memory by
@code{obstack_chunk_alloc} failed.
@end deftypefun

For example, here is a function that allocates a copy of a string @var{str}
in a specific obstack, which is in the variable @code{string_obstack}:

@smallexample
struct obstack string_obstack;

char *
copystring (char *string)
@{
  size_t len = strlen (string) + 1;
  char *s = (char *) obstack_alloc (&string_obstack, len);
  memcpy (s, string, len);
  return s;
@}
@end smallexample

To allocate a block with specified contents, use the function
@code{obstack_copy}, declared like this:

@deftypefun {void *} obstack_copy (struct obstack *@var{obstack-ptr}, void *@var{address}, int @var{size})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_copy @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  obstack_grow dup @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  obstack_finish dup @mtsrace:obstack-ptr @acucorrupt
This allocates a block and initializes it by copying @var{size}
bytes of data starting at @var{address}.  It calls
@code{obstack_alloc_failed_handler} if allocation of memory by
@code{obstack_chunk_alloc} failed.
@end deftypefun

@deftypefun {void *} obstack_copy0 (struct obstack *@var{obstack-ptr}, void *@var{address}, int @var{size})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_copy0 @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  obstack_grow0 dup @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  obstack_finish dup @mtsrace:obstack-ptr @acucorrupt
Like @code{obstack_copy}, but appends an extra byte containing a null
character.  This extra byte is not counted in the argument @var{size}.
@end deftypefun

The @code{obstack_copy0} function is convenient for copying a sequence
of characters into an obstack as a null-terminated string.  Here is an
example of its use:

@smallexample
char *
obstack_savestring (char *addr, int size)
@{
  return obstack_copy0 (&myobstack, addr, size);
@}
@end smallexample

@noindent
Contrast this with the previous example of @code{savestring} using
@code{malloc} (@pxref{Basic Allocation}).

@node Freeing Obstack Objects
@subsubsection Freeing Objects in an Obstack
@cindex freeing (obstacks)

To free an object allocated in an obstack, use the function
@code{obstack_free}.  Since the obstack is a stack of objects, freeing
one object automatically frees all other objects allocated more recently
in the same obstack.

@deftypefun void obstack_free (struct obstack *@var{obstack-ptr}, void *@var{object})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{}}}
@c obstack_free @mtsrace:obstack-ptr @acucorrupt
@c  (obstack_free) @mtsrace:obstack-ptr @acucorrupt
@c   *freefun dup user-supplied
If @var{object} is a null pointer, everything allocated in the obstack
is freed.  Otherwise, @var{object} must be the address of an object
allocated in the obstack.  Then @var{object} is freed, along with
everything allocated in @var{obstack-ptr} since @var{object}.
@end deftypefun

Note that if @var{object} is a null pointer, the result is an
uninitialized obstack.  To free all memory in an obstack but leave it
valid for further allocation, call @code{obstack_free} with the address
of the first object allocated on the obstack:

@smallexample
obstack_free (obstack_ptr, first_object_allocated_ptr);
@end smallexample

Recall that the objects in an obstack are grouped into chunks.  When all
the objects in a chunk become free, the obstack library automatically
frees the chunk (@pxref{Preparing for Obstacks}).  Then other
obstacks, or non-obstack allocation, can reuse the space of the chunk.

@node Obstack Functions
@subsubsection Obstack Functions and Macros
@cindex macros

The interfaces for using obstacks may be defined either as functions or
as macros, depending on the compiler.  The obstack facility works with
all C compilers, including both @w{ISO C} and traditional C, but there are
precautions you must take if you plan to use compilers other than GNU C.

If you are using an old-fashioned @w{non-ISO C} compiler, all the obstack
``functions'' are actually defined only as macros.  You can call these
macros like functions, but you cannot use them in any other way (for
example, you cannot take their address).

Calling the macros requires a special precaution: namely, the first
operand (the obstack pointer) may not contain any side effects, because
it may be computed more than once.  For example, if you write this:

@smallexample
obstack_alloc (get_obstack (), 4);
@end smallexample

@noindent
you will find that @code{get_obstack} may be called several times.
If you use @code{*obstack_list_ptr++} as the obstack pointer argument,
you will get very strange results since the incrementation may occur
several times.

In @w{ISO C}, each function has both a macro definition and a function
definition.  The function definition is used if you take the address of the
function without calling it.  An ordinary call uses the macro definition by
default, but you can request the function definition instead by writing the
function name in parentheses, as shown here:

@smallexample
char *x;
void *(*funcp) ();
/* @r{Use the macro}.  */
x = (char *) obstack_alloc (obptr, size);
/* @r{Call the function}.  */
x = (char *) (obstack_alloc) (obptr, size);
/* @r{Take the address of the function}.  */
funcp = obstack_alloc;
@end smallexample

@noindent
This is the same situation that exists in @w{ISO C} for the standard library
functions.  @xref{Macro Definitions}.

@strong{Warning:} When you do use the macros, you must observe the
precaution of avoiding side effects in the first operand, even in @w{ISO C}.

If you use the GNU C compiler, this precaution is not necessary, because
various language extensions in GNU C permit defining the macros so as to
compute each argument only once.

@node Growing Objects
@subsubsection Growing Objects
@cindex growing objects (in obstacks)
@cindex changing the size of a block (obstacks)

Because memory in obstack chunks is used sequentially, it is possible to
build up an object step by step, adding one or more bytes at a time to the
end of the object.  With this technique, you do not need to know how much
data you will put in the object until you come to the end of it.  We call
this the technique of @dfn{growing objects}.  The special functions
for adding data to the growing object are described in this section.

You don't need to do anything special when you start to grow an object.
Using one of the functions to add data to the object automatically
starts it.  However, it is necessary to say explicitly when the object is
finished.  This is done with the function @code{obstack_finish}.

The actual address of the object thus built up is not known until the
object is finished.  Until then, it always remains possible that you will
add so much data that the object must be copied into a new chunk.

While the obstack is in use for a growing object, you cannot use it for
ordinary allocation of another object.  If you try to do so, the space
already added to the growing object will become part of the other object.

@deftypefun void obstack_blank (struct obstack *@var{obstack-ptr}, int @var{size})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_blank @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  _obstack_newchunk @mtsrace:obstack-ptr @acucorrupt @acsmem
@c   *chunkfun dup @acsmem
@c   *obstack_alloc_failed_handler dup user-supplied
@c   *freefun
@c  obstack_blank_fast dup @mtsrace:obstack-ptr
The most basic function for adding to a growing object is
@code{obstack_blank}, which adds space without initializing it.
@end deftypefun

@deftypefun void obstack_grow (struct obstack *@var{obstack-ptr}, void *@var{data}, int @var{size})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_grow @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  _obstack_newchunk dup @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  memcpy ok
To add a block of initialized space, use @code{obstack_grow}, which is
the growing-object analogue of @code{obstack_copy}.  It adds @var{size}
bytes of data to the growing object, copying the contents from
@var{data}.
@end deftypefun

@deftypefun void obstack_grow0 (struct obstack *@var{obstack-ptr}, void *@var{data}, int @var{size})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_grow0 @mtsrace:obstack-ptr @acucorrupt @acsmem
@c   (no sequence point between storing NUL and incrementing next_free)
@c   (multiple changes to next_free => @acucorrupt)
@c  _obstack_newchunk dup @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  memcpy ok
This is the growing-object analogue of @code{obstack_copy0}.  It adds
@var{size} bytes copied from @var{data}, followed by an additional null
character.
@end deftypefun

@deftypefun void obstack_1grow (struct obstack *@var{obstack-ptr}, char @var{c})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_1grow @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  _obstack_newchunk dup @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  obstack_1grow_fast dup @mtsrace:obstack-ptr @acucorrupt @acsmem
To add one character at a time, use the function @code{obstack_1grow}.
It adds a single byte containing @var{c} to the growing object.
@end deftypefun

@deftypefun void obstack_ptr_grow (struct obstack *@var{obstack-ptr}, void *@var{data})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_ptr_grow @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  _obstack_newchunk dup @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  obstack_ptr_grow_fast dup @mtsrace:obstack-ptr
Adding the value of a pointer one can use the function
@code{obstack_ptr_grow}.  It adds @code{sizeof (void *)} bytes
containing the value of @var{data}.
@end deftypefun

@deftypefun void obstack_int_grow (struct obstack *@var{obstack-ptr}, int @var{data})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_int_grow @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  _obstack_newchunk dup @mtsrace:obstack-ptr @acucorrupt @acsmem
@c  obstack_int_grow_fast dup @mtsrace:obstack-ptr
A single value of type @code{int} can be added by using the
@code{obstack_int_grow} function.  It adds @code{sizeof (int)} bytes to
the growing object and initializes them with the value of @var{data}.
@end deftypefun

@deftypefun {void *} obstack_finish (struct obstack *@var{obstack-ptr})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{}}}
@c obstack_finish @mtsrace:obstack-ptr @acucorrupt
When you are finished growing the object, use the function
@code{obstack_finish} to close it off and return its final address.

Once you have finished the object, the obstack is available for ordinary
allocation or for growing another object.

This function can return a null pointer under the same conditions as
@code{obstack_alloc} (@pxref{Allocation in an Obstack}).
@end deftypefun

When you build an object by growing it, you will probably need to know
afterward how long it became.  You need not keep track of this as you grow
the object, because you can find out the length from the obstack just
before finishing the object with the function @code{obstack_object_size},
declared as follows:

@deftypefun int obstack_object_size (struct obstack *@var{obstack-ptr})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acsafe{}}
This function returns the current size of the growing object, in bytes.
Remember to call this function @emph{before} finishing the object.
After it is finished, @code{obstack_object_size} will return zero.
@end deftypefun

If you have started growing an object and wish to cancel it, you should
finish it and then free it, like this:

@smallexample
obstack_free (obstack_ptr, obstack_finish (obstack_ptr));
@end smallexample

@noindent
This has no effect if no object was growing.

@cindex shrinking objects
You can use @code{obstack_blank} with a negative size argument to make
the current object smaller.  Just don't try to shrink it beyond zero
length---there's no telling what will happen if you do that.

@node Extra Fast Growing
@subsubsection Extra Fast Growing Objects
@cindex efficiency and obstacks

The usual functions for growing objects incur overhead for checking
whether there is room for the new growth in the current chunk.  If you
are frequently constructing objects in small steps of growth, this
overhead can be significant.

You can reduce the overhead by using special ``fast growth''
functions that grow the object without checking.  In order to have a
robust program, you must do the checking yourself.  If you do this checking
in the simplest way each time you are about to add data to the object, you
have not saved anything, because that is what the ordinary growth
functions do.  But if you can arrange to check less often, or check
more efficiently, then you make the program faster.

The function @code{obstack_room} returns the amount of room available
in the current chunk.  It is declared as follows:

@deftypefun int obstack_room (struct obstack *@var{obstack-ptr})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acsafe{}}
This returns the number of bytes that can be added safely to the current
growing object (or to an object about to be started) in obstack
@var{obstack-ptr} using the fast growth functions.
@end deftypefun

While you know there is room, you can use these fast growth functions
for adding data to a growing object:

@deftypefun void obstack_1grow_fast (struct obstack *@var{obstack-ptr}, char @var{c})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acunsafe{@acucorrupt{} @acsmem{}}}
@c obstack_1grow_fast @mtsrace:obstack-ptr @acucorrupt @acsmem
@c   (no sequence point between copying c and incrementing next_free)
The function @code{obstack_1grow_fast} adds one byte containing the
character @var{c} to the growing object in obstack @var{obstack-ptr}.
@end deftypefun

@deftypefun void obstack_ptr_grow_fast (struct obstack *@var{obstack-ptr}, void *@var{data})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acsafe{}}
@c obstack_ptr_grow_fast @mtsrace:obstack-ptr
The function @code{obstack_ptr_grow_fast} adds @code{sizeof (void *)}
bytes containing the value of @var{data} to the growing object in
obstack @var{obstack-ptr}.
@end deftypefun

@deftypefun void obstack_int_grow_fast (struct obstack *@var{obstack-ptr}, int @var{data})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acsafe{}}
@c obstack_int_grow_fast @mtsrace:obstack-ptr
The function @code{obstack_int_grow_fast} adds @code{sizeof (int)} bytes
containing the value of @var{data} to the growing object in obstack
@var{obstack-ptr}.
@end deftypefun

@deftypefun void obstack_blank_fast (struct obstack *@var{obstack-ptr}, int @var{size})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acsafe{}}
@c obstack_blank_fast @mtsrace:obstack-ptr
The function @code{obstack_blank_fast} adds @var{size} bytes to the
growing object in obstack @var{obstack-ptr} without initializing them.
@end deftypefun

When you check for space using @code{obstack_room} and there is not
enough room for what you want to add, the fast growth functions
are not safe.  In this case, simply use the corresponding ordinary
growth function instead.  Very soon this will copy the object to a
new chunk; then there will be lots of room available again.

So, each time you use an ordinary growth function, check afterward for
sufficient space using @code{obstack_room}.  Once the object is copied
to a new chunk, there will be plenty of space again, so the program will
start using the fast growth functions again.

Here is an example:

@smallexample
@group
void
add_string (struct obstack *obstack, const char *ptr, int len)
@{
  while (len > 0)
    @{
      int room = obstack_room (obstack);
      if (room == 0)
        @{
          /* @r{Not enough room.  Add one character slowly,}
             @r{which may copy to a new chunk and make room.}  */
          obstack_1grow (obstack, *ptr++);
          len--;
        @}
      else
        @{
          if (room > len)
            room = len;
          /* @r{Add fast as much as we have room for.} */
          len -= room;
          while (room-- > 0)
            obstack_1grow_fast (obstack, *ptr++);
        @}
    @}
@}
@end group
@end smallexample

@node Status of an Obstack
@subsubsection Status of an Obstack
@cindex obstack status
@cindex status of obstack

Here are functions that provide information on the current status of
allocation in an obstack.  You can use them to learn about an object while
still growing it.

@deftypefun {void *} obstack_base (struct obstack *@var{obstack-ptr})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asucorrupt{}}@acsafe{}}
This function returns the tentative address of the beginning of the
currently growing object in @var{obstack-ptr}.  If you finish the object
immediately, it will have that address.  If you make it larger first, it
may outgrow the current chunk---then its address will change!

If no object is growing, this value says where the next object you
allocate will start (once again assuming it fits in the current
chunk).
@end deftypefun

@deftypefun {void *} obstack_next_free (struct obstack *@var{obstack-ptr})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{}@asunsafe{@asucorrupt{}}@acsafe{}}
This function returns the address of the first free byte in the current
chunk of obstack @var{obstack-ptr}.  This is the end of the currently
growing object.  If no object is growing, @code{obstack_next_free}
returns the same value as @code{obstack_base}.
@end deftypefun

@deftypefun int obstack_object_size (struct obstack *@var{obstack-ptr})
@standards{GNU, obstack.h}
@c dup
@safety{@prelim{}@mtsafe{@mtsrace{:obstack-ptr}}@assafe{}@acsafe{}}
This function returns the size in bytes of the currently growing object.
This is equivalent to

@smallexample
obstack_next_free (@var{obstack-ptr}) - obstack_base (@var{obstack-ptr})
@end smallexample
@end deftypefun

@node Obstacks Data Alignment
@subsubsection Alignment of Data in Obstacks
@cindex alignment (in obstacks)

Each obstack has an @dfn{alignment boundary}; each object allocated in
the obstack automatically starts on an address that is a multiple of the
specified boundary.  By default, this boundary is aligned so that
the object can hold any type of data.

To access an obstack's alignment boundary, use the macro
@code{obstack_alignment_mask}, whose function prototype looks like
this:

@deftypefn Macro int obstack_alignment_mask (struct obstack *@var{obstack-ptr})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
The value is a bit mask; a bit that is 1 indicates that the corresponding
bit in the address of an object should be 0.  The mask value should be one
less than a power of 2; the effect is that all object addresses are
multiples of that power of 2.  The default value of the mask is a value
that allows aligned objects to hold any type of data: for example, if
its value is 3, any type of data can be stored at locations whose
addresses are multiples of 4.  A mask value of 0 means an object can start
on any multiple of 1 (that is, no alignment is required).

The expansion of the macro @code{obstack_alignment_mask} is an lvalue,
so you can alter the mask by assignment.  For example, this statement:

@smallexample
obstack_alignment_mask (obstack_ptr) = 0;
@end smallexample

@noindent
has the effect of turning off alignment processing in the specified obstack.
@end deftypefn

Note that a change in alignment mask does not take effect until
@emph{after} the next time an object is allocated or finished in the
obstack.  If you are not growing an object, you can make the new
alignment mask take effect immediately by calling @code{obstack_finish}.
This will finish a zero-length object and then do proper alignment for
the next object.

@node Obstack Chunks
@subsubsection Obstack Chunks
@cindex efficiency of chunks
@cindex chunks

Obstacks work by allocating space for themselves in large chunks, and
then parceling out space in the chunks to satisfy your requests.  Chunks
are normally 4096 bytes long unless you specify a different chunk size.
The chunk size includes 8 bytes of overhead that are not actually used
for storing objects.  Regardless of the specified size, longer chunks
will be allocated when necessary for long objects.

The obstack library allocates chunks by calling the function
@code{obstack_chunk_alloc}, which you must define.  When a chunk is no
longer needed because you have freed all the objects in it, the obstack
library frees the chunk by calling @code{obstack_chunk_free}, which you
must also define.

These two must be defined (as macros) or declared (as functions) in each
source file that uses @code{obstack_init} (@pxref{Creating Obstacks}).
Most often they are defined as macros like this:

@smallexample
#define obstack_chunk_alloc malloc
#define obstack_chunk_free free
@end smallexample

Note that these are simple macros (no arguments).  Macro definitions with
arguments will not work!  It is necessary that @code{obstack_chunk_alloc}
or @code{obstack_chunk_free}, alone, expand into a function name if it is
not itself a function name.

If you allocate chunks with @code{malloc}, the chunk size should be a
power of 2.  The default chunk size, 4096, was chosen because it is long
enough to satisfy many typical requests on the obstack yet short enough
not to waste too much memory in the portion of the last chunk not yet used.

@deftypefn Macro int obstack_chunk_size (struct obstack *@var{obstack-ptr})
@standards{GNU, obstack.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
This returns the chunk size of the given obstack.
@end deftypefn

Since this macro expands to an lvalue, you can specify a new chunk size by
assigning it a new value.  Doing so does not affect the chunks already
allocated, but will change the size of chunks allocated for that particular
obstack in the future.  It is unlikely to be useful to make the chunk size
smaller, but making it larger might improve efficiency if you are
allocating many objects whose size is comparable to the chunk size.  Here
is how to do so cleanly:

@smallexample
if (obstack_chunk_size (obstack_ptr) < @var{new-chunk-size})
  obstack_chunk_size (obstack_ptr) = @var{new-chunk-size};
@end smallexample

@node Summary of Obstacks
@subsubsection Summary of Obstack Functions

Here is a summary of all the functions associated with obstacks.  Each
takes the address of an obstack (@code{struct obstack *}) as its first
argument.

@table @code
@item void obstack_init (struct obstack *@var{obstack-ptr})
Initialize use of an obstack.  @xref{Creating Obstacks}.

@item void *obstack_alloc (struct obstack *@var{obstack-ptr}, int @var{size})
Allocate an object of @var{size} uninitialized bytes.
@xref{Allocation in an Obstack}.

@item void *obstack_copy (struct obstack *@var{obstack-ptr}, void *@var{address}, int @var{size})
Allocate an object of @var{size} bytes, with contents copied from
@var{address}.  @xref{Allocation in an Obstack}.

@item void *obstack_copy0 (struct obstack *@var{obstack-ptr}, void *@var{address}, int @var{size})
Allocate an object of @var{size}+1 bytes, with @var{size} of them copied
from @var{address}, followed by a null character at the end.
@xref{Allocation in an Obstack}.

@item void obstack_free (struct obstack *@var{obstack-ptr}, void *@var{object})
Free @var{object} (and everything allocated in the specified obstack
more recently than @var{object}).  @xref{Freeing Obstack Objects}.

@item void obstack_blank (struct obstack *@var{obstack-ptr}, int @var{size})
Add @var{size} uninitialized bytes to a growing object.
@xref{Growing Objects}.

@item void obstack_grow (struct obstack *@var{obstack-ptr}, void *@var{address}, int @var{size})
Add @var{size} bytes, copied from @var{address}, to a growing object.
@xref{Growing Objects}.

@item void obstack_grow0 (struct obstack *@var{obstack-ptr}, void *@var{address}, int @var{size})
Add @var{size} bytes, copied from @var{address}, to a growing object,
and then add another byte containing a null character.  @xref{Growing
Objects}.

@item void obstack_1grow (struct obstack *@var{obstack-ptr}, char @var{data-char})
Add one byte containing @var{data-char} to a growing object.
@xref{Growing Objects}.

@item void *obstack_finish (struct obstack *@var{obstack-ptr})
Finalize the object that is growing and return its permanent address.
@xref{Growing Objects}.

@item int obstack_object_size (struct obstack *@var{obstack-ptr})
Get the current size of the currently growing object.  @xref{Growing
Objects}.

@item void obstack_blank_fast (struct obstack *@var{obstack-ptr}, int @var{size})
Add @var{size} uninitialized bytes to a growing object without checking
that there is enough room.  @xref{Extra Fast Growing}.

@item void obstack_1grow_fast (struct obstack *@var{obstack-ptr}, char @var{data-char})
Add one byte containing @var{data-char} to a growing object without
checking that there is enough room.  @xref{Extra Fast Growing}.

@item int obstack_room (struct obstack *@var{obstack-ptr})
Get the amount of room now available for growing the current object.
@xref{Extra Fast Growing}.

@item int obstack_alignment_mask (struct obstack *@var{obstack-ptr})
The mask used for aligning the beginning of an object.  This is an
lvalue.  @xref{Obstacks Data Alignment}.

@item int obstack_chunk_size (struct obstack *@var{obstack-ptr})
The size for allocating chunks.  This is an lvalue.  @xref{Obstack Chunks}.

@item void *obstack_base (struct obstack *@var{obstack-ptr})
Tentative starting address of the currently growing object.
@xref{Status of an Obstack}.

@item void *obstack_next_free (struct obstack *@var{obstack-ptr})
Address just after the end of the currently growing object.
@xref{Status of an Obstack}.
@end table

@node Variable Size Automatic
@subsection Automatic Storage with Variable Size
@cindex automatic freeing
@cindex @code{alloca} function
@cindex automatic storage with variable size

The function @code{alloca} supports a kind of half-dynamic allocation in
which blocks are allocated dynamically but freed automatically.

Allocating a block with @code{alloca} is an explicit action; you can
allocate as many blocks as you wish, and compute the size at run time.  But
all the blocks are freed when you exit the function that @code{alloca} was
called from, just as if they were automatic variables declared in that
function.  There is no way to free the space explicitly.

The prototype for @code{alloca} is in @file{stdlib.h}.  This function is
a BSD extension.
@pindex stdlib.h

@deftypefun {void *} alloca (size_t @var{size})
@standards{GNU, stdlib.h}
@standards{BSD, stdlib.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
The return value of @code{alloca} is the address of a block of @var{size}
bytes of memory, allocated in the stack frame of the calling function.
@end deftypefun

Do not use @code{alloca} inside the arguments of a function call---you
will get unpredictable results, because the stack space for the
@code{alloca} would appear on the stack in the middle of the space for
the function arguments.  An example of what to avoid is @code{foo (x,
alloca (4), y)}.
@c This might get fixed in future versions of GCC, but that won't make
@c it safe with compilers generally.

@menu
* Alloca Example::              Example of using @code{alloca}.
* Advantages of Alloca::        Reasons to use @code{alloca}.
* Disadvantages of Alloca::     Reasons to avoid @code{alloca}.
* GNU C Variable-Size Arrays::  Only in GNU C, here is an alternative
				 method of allocating dynamically and
				 freeing automatically.
@end menu

@node Alloca Example
@subsubsection @code{alloca} Example

As an example of the use of @code{alloca}, here is a function that opens
a file name made from concatenating two argument strings, and returns a
file descriptor or minus one signifying failure:

@smallexample
int
open2 (char *str1, char *str2, int flags, int mode)
@{
  char *name = (char *) alloca (strlen (str1) + strlen (str2) + 1);
  stpcpy (stpcpy (name, str1), str2);
  return open (name, flags, mode);
@}
@end smallexample

@noindent
Here is how you would get the same results with @code{malloc} and
@code{free}:

@smallexample
int
open2 (char *str1, char *str2, int flags, int mode)
@{
  char *name = malloc (strlen (str1) + strlen (str2) + 1);
  int desc;
  if (name == 0)
    fatal ("virtual memory exceeded");
  stpcpy (stpcpy (name, str1), str2);
  desc = open (name, flags, mode);
  free (name);
  return desc;
@}
@end smallexample

As you can see, it is simpler with @code{alloca}.  But @code{alloca} has
other, more important advantages, and some disadvantages.

@node Advantages of Alloca
@subsubsection Advantages of @code{alloca}

Here are the reasons why @code{alloca} may be preferable to @code{malloc}:

@itemize @bullet
@item
Using @code{alloca} wastes very little space and is very fast.  (It is
open-coded by the GNU C compiler.)

@item
Since @code{alloca} does not have separate pools for different sizes of
blocks, space used for any size block can be reused for any other size.
@code{alloca} does not cause memory fragmentation.

@item
@cindex longjmp
Nonlocal exits done with @code{longjmp} (@pxref{Non-Local Exits})
automatically free the space allocated with @code{alloca} when they exit
through the function that called @code{alloca}.  This is the most
important reason to use @code{alloca}.

To illustrate this, suppose you have a function
@code{open_or_report_error} which returns a descriptor, like
@code{open}, if it succeeds, but does not return to its caller if it
fails.  If the file cannot be opened, it prints an error message and
jumps out to the command level of your program using @code{longjmp}.
Let's change @code{open2} (@pxref{Alloca Example}) to use this
subroutine:

@smallexample
int
open2 (char *str1, char *str2, int flags, int mode)
@{
  char *name = (char *) alloca (strlen (str1) + strlen (str2) + 1);
  stpcpy (stpcpy (name, str1), str2);
  return open_or_report_error (name, flags, mode);
@}
@end smallexample

@noindent
Because of the way @code{alloca} works, the memory it allocates is
freed even when an error occurs, with no special effort required.

By contrast, the previous definition of @code{open2} (which uses
@code{malloc} and @code{free}) would develop a memory leak if it were
changed in this way.  Even if you are willing to make more changes to
fix it, there is no easy way to do so.
@end itemize

@node Disadvantages of Alloca
@subsubsection Disadvantages of @code{alloca}

@cindex @code{alloca} disadvantages
@cindex disadvantages of @code{alloca}
These are the disadvantages of @code{alloca} in comparison with
@code{malloc}:

@itemize @bullet
@item
If you try to allocate more memory than the machine can provide, you
don't get a clean error message.  Instead you get a fatal signal like
the one you would get from an infinite recursion; probably a
segmentation violation (@pxref{Program Error Signals}).

@item
Some @nongnusystems{} fail to support @code{alloca}, so it is less
portable.  However, a slower emulation of @code{alloca} written in C
is available for use on systems with this deficiency.
@end itemize

@node GNU C Variable-Size Arrays
@subsubsection GNU C Variable-Size Arrays
@cindex variable-sized arrays

In GNU C, you can replace most uses of @code{alloca} with an array of
variable size.  Here is how @code{open2} would look then:

@smallexample
int open2 (char *str1, char *str2, int flags, int mode)
@{
  char name[strlen (str1) + strlen (str2) + 1];
  stpcpy (stpcpy (name, str1), str2);
  return open (name, flags, mode);
@}
@end smallexample

But @code{alloca} is not always equivalent to a variable-sized array, for
several reasons:

@itemize @bullet
@item
A variable size array's space is freed at the end of the scope of the
name of the array.  The space allocated with @code{alloca}
remains until the end of the function.

@item
It is possible to use @code{alloca} within a loop, allocating an
additional block on each iteration.  This is impossible with
variable-sized arrays.
@end itemize

@strong{NB:} If you mix use of @code{alloca} and variable-sized arrays
within one function, exiting a scope in which a variable-sized array was
declared frees all blocks allocated with @code{alloca} during the
execution of that scope.


@node Resizing the Data Segment
@section Resizing the Data Segment

The symbols in this section are declared in @file{unistd.h}.

You will not normally use the functions in this section, because the
functions described in @ref{Memory Allocation} are easier to use.  Those
are interfaces to a @glibcadj{} memory allocator that uses the
functions below itself.  The functions below are simple interfaces to
system calls.

@deftypefun int brk (void *@var{addr})
@standards{BSD, unistd.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

@code{brk} sets the high end of the calling process' data segment to
@var{addr}.

The address of the end of a segment is defined to be the address of the
last byte in the segment plus 1.

The function has no effect if @var{addr} is lower than the low end of
the data segment.  (This is considered success, by the way.)

The function fails if it would cause the data segment to overlap another
segment or exceed the process' data storage limit (@pxref{Limits on
Resources}).

The function is named for a common historical case where data storage
and the stack are in the same segment.  Data storage allocation grows
upward from the bottom of the segment while the stack grows downward
toward it from the top of the segment and the curtain between them is
called the @dfn{break}.

The return value is zero on success.  On failure, the return value is
@code{-1} and @code{errno} is set accordingly.  The following @code{errno}
values are specific to this function:

@table @code
@item ENOMEM
The request would cause the data segment to overlap another segment or
exceed the process' data storage limit.
@end table

@c The Brk system call in Linux (as opposed to the GNU C Library function)
@c is considerably different.  It always returns the new end of the data
@c segment, whether it succeeds or fails.  The GNU C library Brk determines
@c it's a failure if and only if the system call returns an address less
@c than the address requested.

@end deftypefun


@deftypefun {void *} sbrk (ptrdiff_t @var{delta})
@standards{BSD, unistd.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

This function is the same as @code{brk} except that you specify the new
end of the data segment as an offset @var{delta} from the current end
and on success the return value is the address of the resulting end of
the data segment instead of zero.

This means you can use @samp{sbrk(0)} to find out what the current end
of the data segment is.

@end deftypefun

@node Memory Protection
@section Memory Protection
@cindex memory protection
@cindex page protection
@cindex protection flags

When a page is mapped using @code{mmap}, page protection flags can be
specified using the protection flags argument.  @xref{Memory-mapped
I/O}.

The following flags are available:

@vtable @code
@item PROT_WRITE
@standards{POSIX, sys/mman.h}
The memory can be written to.

@item PROT_READ
@standards{POSIX, sys/mman.h}
The memory can be read.  On some architectures, this flag implies that
the memory can be executed as well (as if @code{PROT_EXEC} had been
specified at the same time).

@item PROT_EXEC
@standards{POSIX, sys/mman.h}
The memory can be used to store instructions which can then be executed.
On most architectures, this flag implies that the memory can be read (as
if @code{PROT_READ} had been specified).

@item PROT_NONE
@standards{POSIX, sys/mman.h}
This flag must be specified on its own.

The memory is reserved, but cannot be read, written, or executed.  If
this flag is specified in a call to @code{mmap}, a virtual memory area
will be set aside for future use in the process, and @code{mmap} calls
without the @code{MAP_FIXED} flag will not use it for subsequent
allocations.  For anonymous mappings, the kernel will not reserve any
physical memory for the allocation at the time the mapping is created.
@end vtable

The operating system may keep track of these flags separately even if
the underlying hardware treats them the same for the purposes of access
checking (as happens with @code{PROT_READ} and @code{PROT_EXEC} on some
platforms).  On GNU systems, @code{PROT_EXEC} always implies
@code{PROT_READ}, so that users can view the machine code which is
executing on their system.

Inappropriate access will cause a segfault (@pxref{Program Error
Signals}).

After allocation, protection flags can be changed using the
@code{mprotect} function.

@deftypefun int mprotect (void *@var{address}, size_t @var{length}, int @var{protection})
@standards{POSIX, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

A successful call to the @code{mprotect} function changes the protection
flags of at least @var{length} bytes of memory, starting at
@var{address}.

@var{address} must be aligned to the page size for the mapping.  The
system page size can be obtained by calling @code{sysconf} with the
@code{_SC_PAGESIZE} parameter (@pxref{Sysconf Definition}).  The system
page size is the granularity in which the page protection of anonymous
memory mappings and most file mappings can be changed.  Memory which is
mapped from special files or devices may have larger page granularity
than the system page size and may require larger alignment.

@var{length} is the number of bytes whose protection flags must be
changed.  It is automatically rounded up to the next multiple of the
system page size.

@var{protection} is a combination of the @code{PROT_*} flags described
above.

The @code{mprotect} function returns @math{0} on success and @math{-1}
on failure.

The following @code{errno} error conditions are defined for this
function:

@table @code
@item ENOMEM
The system was not able to allocate resources to fulfill the request.
This can happen if there is not enough physical memory in the system for
the allocation of backing storage.  The error can also occur if the new
protection flags would cause the memory region to be split from its
neighbors, and the process limit for the number of such distinct memory
regions would be exceeded.

@item EINVAL
@var{address} is not properly aligned to a page boundary for the
mapping, or @var{length} (after rounding up to the system page size) is
not a multiple of the applicable page size for the mapping, or the
combination of flags in @var{protection} is not valid.

@item EACCES
The file for a file-based mapping was not opened with open flags which
are compatible with @var{protection}.

@item EPERM
The system security policy does not allow a mapping with the specified
flags.  For example, mappings which are both @code{PROT_EXEC} and
@code{PROT_WRITE} at the same time might not be allowed.
@end table
@end deftypefun

If the @code{mprotect} function is used to make a region of memory
inaccessible by specifying the @code{PROT_NONE} protection flag and
access is later restored, the memory retains its previous contents.

On some systems, it may not be possible to specify additional flags
which were not present when the mapping was first created.  For example,
an attempt to make a region of memory executable could fail if the
initial protection flags were @samp{PROT_READ | PROT_WRITE}.

In general, the @code{mprotect} function can be used to change any
process memory, no matter how it was allocated.  However, portable use
of the function requires that it is only used with memory regions
returned by @code{mmap} or @code{mmap64}.

@deftypefun int mseal (void *@var{address}, size_t @var{length}, unsigned long @var{flags})
@standards{Linux, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

A successful call to the @code {mseal} function protects the memory
range @var{address} of @var{length} bytes, previously allocated with
@code{mmap} or @code{mremap}, against further metadata changes, such
as:

@itemize @bullet
@item
Unmapping, moving to another location, extending or shrinking the size,
via @code{munmap} and @code{mremap}.

@item
Moving or expanding a different VMA into the current location, via
@code{mremap}.

@item
Modifying the memory range with @code{mmap} along with the flag @code{MAP_FIXED}.

@item
Change the protection flags with @code{mprotect} or @code{pkey_mprotect}.  Also
for certain destructive @code{madvise} behaviours (@code{MADV_DONTNEED},
@code{MADV_FREE}, @code{MADV_DONTNEED_LOCKED}, and @code{MADV_WIPEONFORK}),
@code{mseal} only blocks the operation if the protection key associated with
the memory denies write.

@item
Destructive behaviors on anonymous memory, such as @code{madvise} with
@code{MADV_DONTNEED}.
@end itemize

The @var{address} must be an allocated virtual memory done by @code{mmap}
or @code{mremap}, and it must be page-aligned.  The end address (@var{address}
plus @var{length}) must be within an allocated virtual memory range.  There
should be no unallocated memory between the start and end of the address range.

The @var{flags} is currently unused.

The @code{mseal} function returns @math{0} on success and @math{-1} on
failure.

The following @code{errno} error conditions are defined for this
function:

@table @code
@item EPERM
The system blocked the operation, and the given address range is unmodified
without a partial update.  This error is also returned when @code{mseal} is
issued on a 32-bit CPU (sealing is currently supported only on 64-bit CPUs,
although 32-bit binaries running on a 64-bit kernel are supported).

@item ENOMEM
Either the @var{address} is not allocated, or the end address is not within the
allocation, or there is unallocated memory between the start and end address.

@item ENOSYS
The kernel does not support the @code{mseal} syscall.

@end table
@end deftypefun

@strong{NB:} The memory sealing changes the lifetime of a mapping, where the
sealing memory could not be unmapped until the process terminates or replaces
the process image through @code{execve} function.  The sealed mappings are
inherited through @code{fork}.


@subsection Memory Protection Keys

@cindex memory protection key
@cindex protection key
@cindex MPK
On some systems, further access restrictions can be added to specific pages
using @dfn{memory protection keys}.  These restrictions work as follows:

@itemize @bullet
@item
All memory pages are associated with a protection key.  The default
protection key does not cause any additional protections to be applied
during memory accesses.  New keys can be allocated with the
@code{pkey_alloc} function, and applied to pages using
@code{pkey_mprotect}.

@item
Each thread has a set of separate access restrictions for each
protection key.  These access restrictions can be manipulated using the
@code{pkey_set} and @code{pkey_get} functions.

@item
During a memory access, the system obtains the protection key for the
accessed page and uses that to determine the applicable access restrictions,
as configured for the current thread.  If the access is restricted, a
segmentation fault is the result ((@pxref{Program Error Signals}).
These checks happen in addition to the @code{PROT_}* protection flags
set by @code{mprotect} or @code{pkey_mprotect}.
@end itemize

New threads and subprocesses inherit the access restrictions of the current
thread.  If a protection key is allocated subsequently, existing threads
(except the current) will use an unspecified system default for the
access restrictions associated with newly allocated keys.

Upon entering a signal handler, the system resets the access restrictions of
the current thread so that pages with the default key can be accessed,
but the access restrictions for other protection keys are unspecified.

Applications are expected to allocate a key once using
@code{pkey_alloc}, and apply the key to memory regions which need
special protection with @code{pkey_mprotect}:

@smallexample
  int key = pkey_alloc (0, PKEY_DISABLE_ACCESS);
  if (key < 0)
    /* Perform error checking, including fallback for lack of support.  */
    ...;

  /* Apply the key to a special memory region used to store critical
     data.  */
  if (pkey_mprotect (region, region_length,
                     PROT_READ | PROT_WRITE, key) < 0)
    ...; /* Perform error checking (generally fatal).  */
@end smallexample

If the key allocation fails due to lack of support for memory protection
keys, the @code{pkey_mprotect} call can usually be skipped.  In this
case, the region will not be protected by default.  It is also possible
to call @code{pkey_mprotect} with a key value of @math{-1}, in which
case it will behave in the same way as @code{mprotect}.

After key allocation assignment to memory pages, @code{pkey_set} can be
used to temporarily acquire access to the memory region and relinquish
it again:

@smallexample
  if (key >= 0 && pkey_set (key, PKEY_UNRESTRICTED) < 0)
    ...; /* Perform error checking (generally fatal).  */
  /* At this point, the current thread has read-write access to the
     memory region.  */
  ...
  /* Revoke access again.  */
  if (key >= 0 && pkey_set (key, PKEY_DISABLE_ACCESS) < 0)
    ...; /* Perform error checking (generally fatal).  */
@end smallexample

In this example, a negative key value indicates that no key had been
allocated, which means that the system lacks support for memory
protection keys and it is not necessary to change the the access restrictions
of the current thread (because it always has access).

Compared to using @code{mprotect} to change the page protection flags,
this approach has two advantages: It is thread-safe in the sense that
the access restrictions are only changed for the current thread, so another
thread which changes its own access restrictions concurrently to gain access
to the mapping will not suddenly see its access restrictions updated.  And
@code{pkey_set} typically does not involve a call into the kernel and a
context switch, so it is more efficient.

@deftypefun int pkey_alloc (unsigned int @var{flags}, unsigned int @var{access_restrictions})
@standards{Linux, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acunsafe{@acucorrupt{}}}
Allocate a new protection key.  The @var{flags} argument is reserved and
must be zero.  The @var{access_restrictions} argument specifies access restrictions
which are applied to the current thread (as if with @code{pkey_set}
below).  Access restrictions of other threads are not changed.

The function returns the new protection key, a non-negative number, or
@math{-1} on error.

The following @code{errno} error conditions are defined for this
function:

@table @code
@item ENOSYS
The system does not implement memory protection keys.

@item EINVAL
The @var{flags} argument is not zero.

The @var{access_restrictions} argument is invalid.

The system does not implement memory protection keys or runs in a mode
in which memory protection keys are disabled.

@item ENOSPC
All available protection keys already have been allocated.

The system does not implement memory protection keys or runs in a mode
in which memory protection keys are disabled.

@end table
@end deftypefun

@deftypefun int pkey_free (int @var{key})
@standards{Linux, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
Deallocate the protection key, so that it can be reused by
@code{pkey_alloc}.

Calling this function does not change the access restrictions of the freed
protection key.  The calling thread and other threads may retain access
to it, even if it is subsequently allocated again.  For this reason, it
is not recommended to call the @code{pkey_free} function.

@table @code
@item ENOSYS
The system does not implement memory protection keys.

@item EINVAL
The @var{key} argument is not a valid protection key.
@end table
@end deftypefun

@deftypefun int pkey_mprotect (void *@var{address}, size_t @var{length}, int @var{protection}, int @var{key})
@standards{Linux, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
Similar to @code{mprotect}, but also set the memory protection key for
the memory region to @code{key}.

Some systems use memory protection keys to emulate certain combinations
of @var{protection} flags.  Under such circumstances, specifying an
explicit protection key may behave as if additional flags have been
specified in @var{protection}, even though this does not happen with the
default protection key.  For example, some systems can support
@code{PROT_EXEC}-only mappings only with a default protection key, and
memory with a key which was allocated using @code{pkey_alloc} will still
be readable if @code{PROT_EXEC} is specified without @code{PROT_READ}.

If @var{key} is @math{-1}, the default protection key is applied to the
mapping, just as if @code{mprotect} had been called.

The @code{pkey_mprotect} function returns @math{0} on success and
@math{-1} on failure.  The same @code{errno} error conditions as for
@code{mprotect} are defined for this function, with the following
addition:

@table @code
@item EINVAL
The @var{key} argument is not @math{-1} or a valid memory protection
key allocated using @code{pkey_alloc}.

@item ENOSYS
The system does not implement memory protection keys, and @var{key} is
not @math{-1}.
@end table
@end deftypefun

@deftypefun int pkey_set (int @var{key}, unsigned int @var{access_restrictions})
@standards{Linux, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
Change the access restrictions of the current thread for memory pages with
the protection key @var{key} to @var{access_restrictions}.  If
@var{access_restrictions} is @code{PKEY_UNRESTRICTED} (zero), no additional
access restrictions on top of the page protection flags are applied.  Otherwise,
@var{access_restrictions} is a combination of the following flags:

@vtable @code
@item PKEY_DISABLE_READ
@standards{Linux, sys/mman.h}
Subsequent attempts to read from memory with the specified protection
key will fault.  At present only AArch64 platforms with enabled Stage 1
permission overlays feature support this type of restriction.

@item PKEY_DISABLE_WRITE
@standards{Linux, sys/mman.h}
Subsequent attempts to write to memory with the specified protection
key will fault.

@item PKEY_DISABLE_ACCESS
@standards{Linux, sys/mman.h}
Subsequent attempts to write to or read from memory with the specified
protection key will fault.  On AArch64 platforms with enabled Stage 1
permission overlays feature this restriction value has the same effect
as combination of @code{PKEY_DISABLE_READ} and @code{PKEY_DISABLE_WRITE}.

@item PKEY_DISABLE_EXECUTE
@standards{Linux, sys/mman.h}
Subsequent attempts to execute from memory with the specified protection
key will fault.  At present only AArch64 platforms with enabled Stage 1
permission overlays feature support this type of restriction.
@end vtable

Operations not specified as flags are not restricted.  In particular,
this means that the memory region will remain executable if it was
mapped with the @code{PROT_EXEC} protection flag and
@code{PKEY_DISABLE_ACCESS} has been specified.

Calling the @code{pkey_set} function with a protection key which was not
allocated by @code{pkey_alloc} results in undefined behavior.  This
means that calling this function on systems which do not support memory
protection keys is undefined.

The @code{pkey_set} function returns @math{0} on success and @math{-1}
on failure.

The following @code{errno} error conditions are defined for this
function:

@table @code
@item EINVAL
The system does not support the access restrictions expressed in
the @var{access_restrictions} argument.
@end table
@end deftypefun

@deftypefun int pkey_get (int @var{key})
@standards{Linux, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}
Return the access restrictions of the current thread for memory pages
with protection key @var{key}.  The return value is zero or a combination of
the @code{PKEY_DISABLE_}* flags; see the @code{pkey_set} function.

The returned value should be checked for presence or absence of specific flags
using bitwise operations.  Comparing the returned value with any of the flags
or their combination using equals will almost certainly fail.

Calling the @code{pkey_get} function with a protection key which was not
allocated by @code{pkey_alloc} results in undefined behavior.  This
means that calling this function on systems which do not support memory
protection keys is undefined.
@end deftypefun

@node Locking Pages
@section Locking Pages
@cindex locking pages
@cindex memory lock
@cindex paging

You can tell the system to associate a particular virtual memory page
with a real page frame and keep it that way --- i.e., cause the page to
be paged in if it isn't already and mark it so it will never be paged
out and consequently will never cause a page fault.  This is called
@dfn{locking} a page.

The functions in this chapter lock and unlock the calling process'
pages.

@menu
* Why Lock Pages::                Reasons to read this section.
* Locked Memory Details::         Everything you need to know locked
                                    memory
* Page Lock Functions::           Here's how to do it.
@end menu

@node Why Lock Pages
@subsection Why Lock Pages

Because page faults cause paged out pages to be paged in transparently,
a process rarely needs to be concerned about locking pages.  However,
there are two reasons people sometimes are:

@itemize @bullet

@item
Speed.  A page fault is transparent only insofar as the process is not
sensitive to how long it takes to do a simple memory access.  Time-critical
processes, especially realtime processes, may not be able to wait or
may not be able to tolerate variance in execution speed.
@cindex realtime processing
@cindex speed of execution

A process that needs to lock pages for this reason probably also needs
priority among other processes for use of the CPU.  @xref{Priority}.

In some cases, the programmer knows better than the system's demand
paging allocator which pages should remain in real memory to optimize
system performance.  In this case, locking pages can help.

@item
Privacy.  If you keep secrets in virtual memory and that virtual memory
gets paged out, that increases the chance that the secrets will get out.
If a passphrase gets written out to disk swap space, for example, it might
still be there long after virtual and real memory have been wiped clean.

@end itemize

Be aware that when you lock a page, that's one fewer page frame that can
be used to back other virtual memory (by the same or other processes),
which can mean more page faults, which means the system runs more
slowly.  In fact, if you lock enough memory, some programs may not be
able to run at all for lack of real memory.

@node Locked Memory Details
@subsection Locked Memory Details

A memory lock is associated with a virtual page, not a real frame.  The
paging rule is: If a frame backs at least one locked page, don't page it
out.

Memory locks do not stack.  I.e., you can't lock a particular page twice
so that it has to be unlocked twice before it is truly unlocked.  It is
either locked or it isn't.

A memory lock persists until the process that owns the memory explicitly
unlocks it.  (But process termination and exec cause the virtual memory
to cease to exist, which you might say means it isn't locked any more).

Memory locks are not inherited by child processes.  (But note that on a
modern Unix system, immediately after a fork, the parent's and the
child's virtual address space are backed by the same real page frames,
so the child enjoys the parent's locks).  @xref{Creating a Process}.

Because of its ability to impact other processes, only the superuser can
lock a page.  Any process can unlock its own page.

The system sets limits on the amount of memory a process can have locked
and the amount of real memory it can have dedicated to it.  @xref{Limits
on Resources}.

In Linux, locked pages aren't as locked as you might think.
Two virtual pages that are not shared memory can nonetheless be backed
by the same real frame.  The kernel does this in the name of efficiency
when it knows both virtual pages contain identical data, and does it
even if one or both of the virtual pages are locked.

But when a process modifies one of those pages, the kernel must get it a
separate frame and fill it with the page's data.  This is known as a
@dfn{copy-on-write page fault}.  It takes a small amount of time and in
a pathological case, getting that frame may require I/O.
@cindex copy-on-write page fault
@cindex page fault, copy-on-write

To make sure this doesn't happen to your program, don't just lock the
pages.  Write to them as well, unless you know you won't write to them
ever.  And to make sure you have pre-allocated frames for your stack,
enter a scope that declares a C automatic variable larger than the
maximum stack size you will need, set it to something, then return from
its scope.

@node Page Lock Functions
@subsection Functions To Lock And Unlock Pages

The symbols in this section are declared in @file{sys/mman.h}.  These
functions are defined by POSIX.1b, but their availability depends on
your kernel.  If your kernel doesn't allow these functions, they exist
but always fail.  They @emph{are} available with a Linux kernel.

@strong{Portability Note:} POSIX.1b requires that when the @code{mlock}
and @code{munlock} functions are available, the file @file{unistd.h}
define the macro @code{_POSIX_MEMLOCK_RANGE} and the file
@code{limits.h} define the macro @code{PAGESIZE} to be the size of a
memory page in bytes.  It requires that when the @code{mlockall} and
@code{munlockall} functions are available, the @file{unistd.h} file
define the macro @code{_POSIX_MEMLOCK}.  @Theglibc{} conforms to
this requirement.

@deftypefun int mlock (const void *@var{addr}, size_t @var{len})
@standards{POSIX.1b, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

@code{mlock} locks a range of the calling process' virtual pages.

The range of memory starts at address @var{addr} and is @var{len} bytes
long.  Actually, since you must lock whole pages, it is the range of
pages that include any part of the specified range.

When the function returns successfully, each of those pages is backed by
(connected to) a real frame (is resident) and is marked to stay that
way.  This means the function may cause page-ins and have to wait for
them.

When the function fails, it does not affect the lock status of any
pages.

The return value is zero if the function succeeds.  Otherwise, it is
@code{-1} and @code{errno} is set accordingly.  @code{errno} values
specific to this function are:

@table @code
@item ENOMEM
@itemize @bullet
@item
At least some of the specified address range does not exist in the
calling process' virtual address space.
@item
The locking would cause the process to exceed its locked page limit.
@end itemize

@item EPERM
The calling process is not superuser.

@item EINVAL
@var{len} is not positive.

@item ENOSYS
The kernel does not provide @code{mlock} capability.

@end table
@end deftypefun

@deftypefun int mlock2 (const void *@var{addr}, size_t @var{len}, unsigned int @var{flags})
@standards{Linux, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

This function is similar to @code{mlock}.  If @var{flags} is zero, a
call to @code{mlock2} behaves exactly as the equivalent call to @code{mlock}.

The @var{flags} argument must be a combination of zero or more of the
following flags:

@vtable @code
@item MLOCK_ONFAULT
@standards{Linux, sys/mman.h}
Only those pages in the specified address range which are already in
memory are locked immediately.  Additional pages in the range are
automatically locked in case of a page fault and allocation of memory.
@end vtable

Like @code{mlock}, @code{mlock2} returns zero on success and @code{-1}
on failure, setting @code{errno} accordingly.  Additional @code{errno}
values defined for @code{mlock2} are:

@table @code
@item EINVAL
The specified (non-zero) @var{flags} argument is not supported by this
system.
@end table
@end deftypefun

You can lock @emph{all} a process' memory with @code{mlockall}.  You
unlock memory with @code{munlock} or @code{munlockall}.

To avoid all page faults in a C program, you have to use
@code{mlockall}, because some of the memory a program uses is hidden
from the C code, e.g. the stack and automatic variables, and you
wouldn't know what address to tell @code{mlock}.

@deftypefun int munlock (const void *@var{addr}, size_t @var{len})
@standards{POSIX.1b, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

@code{munlock} unlocks a range of the calling process' virtual pages.

@code{munlock} is the inverse of @code{mlock} and functions completely
analogously to @code{mlock}, except that there is no @code{EPERM}
failure.

@end deftypefun

@deftypefun int mlockall (int @var{flags})
@standards{POSIX.1b, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

@code{mlockall} locks all the pages in a process' virtual memory address
space, and/or any that are added to it in the future.  This includes the
pages of the code, data and stack segment, as well as shared libraries,
user space kernel data, shared memory, and memory mapped files.

@var{flags} is a string of single bit flags represented by the following
macros.  They tell @code{mlockall} which of its functions you want.  All
other bits must be zero.

@vtable @code

@item MCL_CURRENT
Lock all pages which currently exist in the calling process' virtual
address space.

@item MCL_FUTURE
Set a mode such that any pages added to the process' virtual address
space in the future will be locked from birth.  This mode does not
affect future address spaces owned by the same process so exec, which
replaces a process' address space, wipes out @code{MCL_FUTURE}.
@xref{Executing a File}.

@end vtable

When the function returns successfully, and you specified
@code{MCL_CURRENT}, all of the process' pages are backed by (connected
to) real frames (they are resident) and are marked to stay that way.
This means the function may cause page-ins and have to wait for them.

When the process is in @code{MCL_FUTURE} mode because it successfully
executed this function and specified @code{MCL_CURRENT}, any system call
by the process that requires space be added to its virtual address space
fails with @code{errno} = @code{ENOMEM} if locking the additional space
would cause the process to exceed its locked page limit.  In the case
that the address space addition that can't be accommodated is stack
expansion, the stack expansion fails and the kernel sends a
@code{SIGSEGV} signal to the process.

When the function fails, it does not affect the lock status of any pages
or the future locking mode.

The return value is zero if the function succeeds.  Otherwise, it is
@code{-1} and @code{errno} is set accordingly.  @code{errno} values
specific to this function are:

@table @code
@item ENOMEM
@itemize @bullet
@item
At least some of the specified address range does not exist in the
calling process' virtual address space.
@item
The locking would cause the process to exceed its locked page limit.
@end itemize

@item EPERM
The calling process is not superuser.

@item EINVAL
Undefined bits in @var{flags} are not zero.

@item ENOSYS
The kernel does not provide @code{mlockall} capability.

@end table

You can lock just specific pages with @code{mlock}.  You unlock pages
with @code{munlockall} and @code{munlock}.

@end deftypefun


@deftypefun int munlockall (void)
@standards{POSIX.1b, sys/mman.h}
@safety{@prelim{}@mtsafe{}@assafe{}@acsafe{}}

@code{munlockall} unlocks every page in the calling process' virtual
address space and turns off @code{MCL_FUTURE} future locking mode.

The return value is zero if the function succeeds.  Otherwise, it is
@code{-1} and @code{errno} is set accordingly.  The only way this
function can fail is for generic reasons that all functions and system
calls can fail, so there are no specific @code{errno} values.

@end deftypefun




@ignore
@c This was never actually implemented.  -zw
@node Relocating Allocator
@section Relocating Allocator

@cindex relocating memory allocator
Any system of dynamic memory allocation has overhead: the amount of
space it uses is more than the amount the program asks for.  The
@dfn{relocating memory allocator} achieves very low overhead by moving
blocks in memory as necessary, on its own initiative.

@c @menu
@c * Relocator Concepts::		How to understand relocating allocation.
@c * Using Relocator::		Functions for relocating allocation.
@c @end menu

@node Relocator Concepts
@subsection Concepts of Relocating Allocation

@ifinfo
The @dfn{relocating memory allocator} achieves very low overhead by
moving blocks in memory as necessary, on its own initiative.
@end ifinfo

When you allocate a block with @code{malloc}, the address of the block
never changes unless you use @code{realloc} to change its size.  Thus,
you can safely store the address in various places, temporarily or
permanently, as you like.  This is not safe when you use the relocating
memory allocator, because any and all relocatable blocks can move
whenever you allocate memory in any fashion.  Even calling @code{malloc}
or @code{realloc} can move the relocatable blocks.

@cindex handle
For each relocatable block, you must make a @dfn{handle}---a pointer
object in memory, designated to store the address of that block.  The
relocating allocator knows where each block's handle is, and updates the
address stored there whenever it moves the block, so that the handle
always points to the block.  Each time you access the contents of the
block, you should fetch its address anew from the handle.

To call any of the relocating allocator functions from a signal handler
is almost certainly incorrect, because the signal could happen at any
time and relocate all the blocks.  The only way to make this safe is to
block the signal around any access to the contents of any relocatable
block---not a convenient mode of operation.  @xref{Nonreentrancy}.

@node Using Relocator
@subsection Allocating and Freeing Relocatable Blocks

@pindex malloc.h
In the descriptions below, @var{handleptr} designates the address of the
handle.  All the functions are declared in @file{malloc.h}; all are GNU
extensions.

@comment malloc.h
@comment GNU
@c @deftypefun {void *} r_alloc (void **@var{handleptr}, size_t @var{size})
This function allocates a relocatable block of size @var{size}.  It
stores the block's address in @code{*@var{handleptr}} and returns
a non-null pointer to indicate success.

If @code{r_alloc} can't get the space needed, it stores a null pointer
in @code{*@var{handleptr}}, and returns a null pointer.
@end deftypefun

@comment malloc.h
@comment GNU
@c @deftypefun void r_alloc_free (void **@var{handleptr})
This function is the way to free a relocatable block.  It frees the
block that @code{*@var{handleptr}} points to, and stores a null pointer
in @code{*@var{handleptr}} to show it doesn't point to an allocated
block any more.
@end deftypefun

@comment malloc.h
@comment GNU
@c @deftypefun {void *} r_re_alloc (void **@var{handleptr}, size_t @var{size})
The function @code{r_re_alloc} adjusts the size of the block that
@code{*@var{handleptr}} points to, making it @var{size} bytes long.  It
stores the address of the resized block in @code{*@var{handleptr}} and
returns a non-null pointer to indicate success.

If enough memory is not available, this function returns a null pointer
and does not modify @code{*@var{handleptr}}.
@end deftypefun
@end ignore




@ignore
@comment No longer available...

@comment @node Memory Warnings
@comment @section Memory Usage Warnings
@comment @cindex memory usage warnings
@comment @cindex warnings of memory almost full

@pindex malloc.c
You can ask for warnings as the program approaches running out of memory
space, by calling @code{memory_warnings}.  This tells @code{malloc} to
check memory usage every time it asks for more memory from the operating
system.  This is a GNU extension declared in @file{malloc.h}.

@comment malloc.h
@comment GNU
@comment @deftypefun void memory_warnings (void *@var{start}, void (*@var{warn-func}) (const char *))
Call this function to request warnings for nearing exhaustion of virtual
memory.

The argument @var{start} says where data space begins, in memory.  The
allocator compares this against the last address used and against the
limit of data space, to determine the fraction of available memory in
use.  If you supply zero for @var{start}, then a default value is used
which is right in most circumstances.

For @var{warn-func}, supply a function that @code{malloc} can call to
warn you.  It is called with a string (a warning message) as argument.
Normally it ought to display the string for the user to read.
@end deftypefun

The warnings come when memory becomes 75% full, when it becomes 85%
full, and when it becomes 95% full.  Above 95% you get another warning
each time memory usage increases.

@end ignore