1mremap(2) System Calls Manual mremap(2)
2
3
4
6 mremap - remap a virtual memory address
7
9 Standard C library (libc, -lc)
10
12 #define _GNU_SOURCE /* See feature_test_macros(7) */
13 #include <sys/mman.h>
14
15 void *mremap(void old_address[.old_size], size_t old_size,
16 size_t new_size, int flags, ... /* void *new_address */);
17
19 mremap() expands (or shrinks) an existing memory mapping, potentially
20 moving it at the same time (controlled by the flags argument and the
21 available virtual address space).
22
23 old_address is the old address of the virtual memory block that you
24 want to expand (or shrink). Note that old_address has to be page
25 aligned. old_size is the old size of the virtual memory block.
26 new_size is the requested size of the virtual memory block after the
27 resize. An optional fifth argument, new_address, may be provided; see
28 the description of MREMAP_FIXED below.
29
30 If the value of old_size is zero, and old_address refers to a shareable
31 mapping (see mmap(2) MAP_SHARED), then mremap() will create a new map‐
32 ping of the same pages. new_size will be the size of the new mapping
33 and the location of the new mapping may be specified with new_address;
34 see the description of MREMAP_FIXED below. If a new mapping is re‐
35 quested via this method, then the MREMAP_MAYMOVE flag must also be
36 specified.
37
38 The flags bit-mask argument may be 0, or include the following flags:
39
40 MREMAP_MAYMOVE
41 By default, if there is not sufficient space to expand a mapping
42 at its current location, then mremap() fails. If this flag is
43 specified, then the kernel is permitted to relocate the mapping
44 to a new virtual address, if necessary. If the mapping is relo‐
45 cated, then absolute pointers into the old mapping location be‐
46 come invalid (offsets relative to the starting address of the
47 mapping should be employed).
48
49 MREMAP_FIXED (since Linux 2.3.31)
50 This flag serves a similar purpose to the MAP_FIXED flag of
51 mmap(2). If this flag is specified, then mremap() accepts a
52 fifth argument, void *new_address, which specifies a page-
53 aligned address to which the mapping must be moved. Any previ‐
54 ous mapping at the address range specified by new_address and
55 new_size is unmapped.
56
57 If MREMAP_FIXED is specified, then MREMAP_MAYMOVE must also be
58 specified.
59
60 MREMAP_DONTUNMAP (since Linux 5.7)
61 This flag, which must be used in conjunction with MREMAP_MAY‐
62 MOVE, remaps a mapping to a new address but does not unmap the
63 mapping at old_address.
64
65 The MREMAP_DONTUNMAP flag can be used only with private anony‐
66 mous mappings (see the description of MAP_PRIVATE and MAP_ANONY‐
67 MOUS in mmap(2)).
68
69 After completion, any access to the range specified by old_ad‐
70 dress and old_size will result in a page fault. The page fault
71 will be handled by a userfaultfd(2) handler if the address is in
72 a range previously registered with userfaultfd(2). Otherwise,
73 the kernel allocates a zero-filled page to handle the fault.
74
75 The MREMAP_DONTUNMAP flag may be used to atomically move a map‐
76 ping while leaving the source mapped. See NOTES for some possi‐
77 ble applications of MREMAP_DONTUNMAP.
78
79 If the memory segment specified by old_address and old_size is locked
80 (using mlock(2) or similar), then this lock is maintained when the seg‐
81 ment is resized and/or relocated. As a consequence, the amount of mem‐
82 ory locked by the process may change.
83
85 On success mremap() returns a pointer to the new virtual memory area.
86 On error, the value MAP_FAILED (that is, (void *) -1) is returned, and
87 errno is set to indicate the error.
88
90 EAGAIN The caller tried to expand a memory segment that is locked, but
91 this was not possible without exceeding the RLIMIT_MEMLOCK re‐
92 source limit.
93
94 EFAULT Some address in the range old_address to old_address+old_size is
95 an invalid virtual memory address for this process. You can
96 also get EFAULT even if there exist mappings that cover the
97 whole address space requested, but those mappings are of differ‐
98 ent types.
99
100 EINVAL An invalid argument was given. Possible causes are:
101
102 • old_address was not page aligned;
103
104 • a value other than MREMAP_MAYMOVE or MREMAP_FIXED or
105 MREMAP_DONTUNMAP was specified in flags;
106
107 • new_size was zero;
108
109 • new_size or new_address was invalid;
110
111 • the new address range specified by new_address and new_size
112 overlapped the old address range specified by old_address and
113 old_size;
114
115 • MREMAP_FIXED or MREMAP_DONTUNMAP was specified without also
116 specifying MREMAP_MAYMOVE;
117
118 • MREMAP_DONTUNMAP was specified, but one or more pages in the
119 range specified by old_address and old_size were not private
120 anonymous;
121
122 • MREMAP_DONTUNMAP was specified and old_size was not equal to
123 new_size;
124
125 • old_size was zero and old_address does not refer to a share‐
126 able mapping (but see BUGS);
127
128 • old_size was zero and the MREMAP_MAYMOVE flag was not speci‐
129 fied.
130
131 ENOMEM Not enough memory was available to complete the operation. Pos‐
132 sible causes are:
133
134 • The memory area cannot be expanded at the current virtual ad‐
135 dress, and the MREMAP_MAYMOVE flag is not set in flags. Or,
136 there is not enough (virtual) memory available.
137
138 • MREMAP_DONTUNMAP was used causing a new mapping to be created
139 that would exceed the (virtual) memory available. Or, it
140 would exceed the maximum number of allowed mappings.
141
143 Linux.
144
146 Prior to glibc 2.4, glibc did not expose the definition of
147 MREMAP_FIXED, and the prototype for mremap() did not allow for the
148 new_address argument.
149
151 mremap() changes the mapping between virtual addresses and memory
152 pages. This can be used to implement a very efficient realloc(3).
153
154 In Linux, memory is divided into pages. A process has (one or) several
155 linear virtual memory segments. Each virtual memory segment has one or
156 more mappings to real memory pages (in the page table). Each virtual
157 memory segment has its own protection (access rights), which may cause
158 a segmentation violation (SIGSEGV) if the memory is accessed incor‐
159 rectly (e.g., writing to a read-only segment). Accessing virtual mem‐
160 ory outside of the segments will also cause a segmentation violation.
161
162 If mremap() is used to move or expand an area locked with mlock(2) or
163 equivalent, the mremap() call will make a best effort to populate the
164 new area but will not fail with ENOMEM if the area cannot be populated.
165
166 MREMAP_DONTUNMAP use cases
167 Possible applications for MREMAP_DONTUNMAP include:
168
169 • Non-cooperative userfaultfd(2): an application can yank out a vir‐
170 tual address range using MREMAP_DONTUNMAP and then employ a user‐
171 faultfd(2) handler to handle the page faults that subsequently occur
172 as other threads in the process touch pages in the yanked range.
173
174 • Garbage collection: MREMAP_DONTUNMAP can be used in conjunction with
175 userfaultfd(2) to implement garbage collection algorithms (e.g., in
176 a Java virtual machine). Such an implementation can be cheaper (and
177 simpler) than conventional garbage collection techniques that in‐
178 volve marking pages with protection PROT_NONE in conjunction with
179 the use of a SIGSEGV handler to catch accesses to those pages.
180
182 Before Linux 4.14, if old_size was zero and the mapping referred to by
183 old_address was a private mapping (mmap(2) MAP_PRIVATE), mremap() cre‐
184 ated a new private mapping unrelated to the original mapping. This be‐
185 havior was unintended and probably unexpected in user-space applica‐
186 tions (since the intention of mremap() is to create a new mapping based
187 on the original mapping). Since Linux 4.14, mremap() fails with the
188 error EINVAL in this scenario.
189
191 brk(2), getpagesize(2), getrlimit(2), mlock(2), mmap(2), sbrk(2), mal‐
192 loc(3), realloc(3)
193
194 Your favorite text book on operating systems for more information on
195 paged memory (e.g., Modern Operating Systems by Andrew S. Tanenbaum,
196 Inside Linux by Randolph Bentson, The Design of the UNIX Operating Sys‐
197 tem by Maurice J. Bach)
198
199
200
201Linux man-pages 6.05 2023-03-30 mremap(2)