Debian Codegraph @

upstream/debian-unstable/man2/shmget.2

raw
1 .\" Copyright 1993, Luigi P. Bai <[email protected]>
2 .\" Copyright, the authors of the Linux man-pages project
3 .\"
4 .\" SPDX-License-Identifier: Linux-man-pages-copyleft
5 .\"
6 .TH shmget 2 2026-02-08 "Linux man-pages 6.18"
7 .SH NAME
8 shmget \- allocates a System V shared memory segment
9 .SH LIBRARY
10 Standard C library
11 .RI ( libc ,\~ \-lc )
12 .SH SYNOPSIS
13 .nf
14 .B #include <sys/shm.h>
15 .P
16 .BI "int shmget(key_t " key ", size_t " size ", int " shmflg );
17 .fi
18 .SH DESCRIPTION
19 .BR shmget ()
20 returns the identifier of the System\ V shared memory segment
21 associated with the value of the argument
22 .IR key .
23 It may be used either to obtain the identifier of a previously created
24 shared memory segment (when
25 .I shmflg
26 is zero and
27 .I key
28 does not have the value
29 .BR IPC_PRIVATE ),
30 or to create a new set.
31 .P
32 A new shared memory segment, with size equal to the value of
33 .I size
34 rounded up to a multiple of
35 .BR PAGE_SIZE ,
36 is created if
37 .I key
38 has the value
39 .B IPC_PRIVATE
40 or
41 .I key
42 isn't
43 .BR IPC_PRIVATE ,
44 no shared memory segment corresponding to
45 .I key
46 exists, and
47 .B IPC_CREAT
48 is specified in
49 .IR shmflg .
50 .P
51 If
52 .I shmflg
53 specifies both
54 .B IPC_CREAT
55 and
56 .B IPC_EXCL
57 and a shared memory segment already exists for
58 .IR key ,
59 then
60 .BR shmget ()
61 fails with
62 .I errno
63 set to
64 .BR EEXIST .
65 (This is analogous to the effect of the combination
66 .B O_CREAT | O_EXCL
67 for
68 .BR open (2).)
69 .P
70 The value
71 .I shmflg
72 is composed of:
73 .TP
74 .B IPC_CREAT
75 Create a new segment.
76 If this flag is not used, then
77 .BR shmget ()
78 will find the segment associated with
79 .I key
80 and check to see if
81 the user has permission to access the segment.
82 .TP
83 .B IPC_EXCL
84 This flag is used with
85 .B IPC_CREAT
86 to ensure that this call creates the segment.
87 If the segment already exists, the call fails.
88 .TP
89 .BR SHM_HUGETLB " (since Linux 2.6)"
90 Allocate the segment using "huge" pages.
91 See the Linux kernel source file
92 .I Documentation/admin\-guide/mm/hugetlbpage.rst
93 for further information.
94 .TP
95 .B SHM_HUGE_2MB
96 .TQ
97 .BR SHM_HUGE_1GB " (since Linux 3.8)"
98 .\" See https://lwn.net/Articles/533499/
99 Used in conjunction with
100 .B SHM_HUGETLB
101 to select alternative hugetlb page sizes (respectively, 2\ MB and 1\ GB)
102 on systems that support multiple hugetlb page sizes.
103 .IP
104 More generally, the desired huge page size can be configured by encoding
105 the base-2 logarithm of the desired page size in the six bits at the offset
106 .BR SHM_HUGE_SHIFT .
107 Thus, the above two constants are defined as:
108 .IP
109 .in +4n
110 .EX
111 #define SHM_HUGE_2MB (21 << SHM_HUGE_SHIFT)
112 #define SHM_HUGE_1GB (30 << SHM_HUGE_SHIFT)
113 .EE
114 .in
115 .IP
116 For some additional details,
117 see the discussion of the similarly named constants in
118 .BR mmap (2).
119 .TP
120 .BR SHM_NORESERVE " (since Linux 2.6.15)"
121 This flag serves the same purpose as the
122 .BR mmap (2)
123 .B MAP_NORESERVE
124 flag.
125 Do not reserve swap space for this segment.
126 When swap space is reserved, one has the guarantee
127 that it is possible to modify the segment.
128 When swap space is not reserved one might get
129 .B SIGSEGV
130 upon a write
131 if no physical memory is available.
132 See also the discussion of the file
133 .I /proc/sys/vm/overcommit_memory
134 in
135 .BR proc_sys_vm (5).
136 .\" As at 2.6.17-rc2, this flag has no effect if SHM_HUGETLB was also
137 .\" specified.
138 .P
139 In addition to the above flags, the least significant 9 bits of
140 .I shmflg
141 specify the permissions granted to the owner, group, and others.
142 These bits have the same format, and the same
143 meaning, as the
144 .I mode
145 argument of
146 .BR open (2).
147 Presently, execute permissions are not used by the system.
148 .P
149 When a new shared memory segment is created,
150 its contents are initialized to zero values, and
151 its associated data structure,
152 .I shmid_ds
153 (see
154 .BR shmctl (2)),
155 is initialized as follows:
156 .IP \[bu] 3
157 .I shm_perm.cuid
158 and
159 .I shm_perm.uid
160 are set to the effective user ID of the calling process.
161 .IP \[bu]
162 .I shm_perm.cgid
163 and
164 .I shm_perm.gid
165 are set to the effective group ID of the calling process.
166 .IP \[bu]
167 The least significant 9 bits of
168 .I shm_perm.mode
169 are set to the least significant 9 bit of
170 .IR shmflg .
171 .IP \[bu]
172 .I shm_segsz
173 is set to the value of
174 .IR size .
175 .IP \[bu]
176 .IR shm_lpid ,
177 .IR shm_nattch ,
178 .IR shm_atime ,
179 and
180 .I shm_dtime
181 are set to 0.
182 .IP \[bu]
183 .I shm_ctime
184 is set to the current time.
185 .P
186 If the shared memory segment already exists, the permissions are
187 verified, and a check is made to see if it is marked for destruction.
188 .SH RETURN VALUE
189 On success, a valid shared memory identifier is returned.
190 On error, \-1 is returned, and
191 .I errno
192 is set to indicate the error.
193 .SH ERRORS
194 .TP
195 .B EACCES
196 The user does not have permission to access the
197 shared memory segment, and does not have the
198 .B CAP_IPC_OWNER
199 capability in the user namespace that governs its IPC namespace.
200 .TP
201 .B EEXIST
202 .B IPC_CREAT
203 and
204 .B IPC_EXCL
205 were specified in
206 .IR shmflg ,
207 but a shared memory segment already exists for
208 .IR key .
209 .TP
210 .B EINVAL
211 A new segment was to be created and
212 .I size
213 is less than
214 .B SHMMIN
215 or greater than
216 .BR SHMMAX .
217 .TP
218 .B EINVAL
219 A segment for the given
220 .I key
221 exists, but
222 .I size
223 is greater than the size
224 of that segment.
225 .TP
226 .B ENFILE
227 .\" [2.6.7] shmem_zero_setup()-->shmem_file_setup()-->get_empty_filp()
228 The system-wide limit on the total number of open files has been reached.
229 .TP
230 .B ENOENT
231 No segment exists for the given
232 .IR key ,
233 and
234 .B IPC_CREAT
235 was not specified.
236 .TP
237 .B ENOMEM
238 No memory could be allocated for segment overhead.
239 .TP
240 .B ENOSPC
241 All possible shared memory IDs have been taken
242 .RB ( SHMMNI ),
243 or allocating a segment of the requested
244 .I size
245 would cause the system to exceed the system-wide limit on shared memory
246 .RB ( SHMALL ).
247 .TP
248 .B EPERM
249 The
250 .B SHM_HUGETLB
251 flag was specified, but the caller was not privileged (did not have the
252 .B CAP_IPC_LOCK
253 capability)
254 and is not a member of the
255 .I hugetlb_shm_group
256 group;
257 see the description of
258 .I /proc/sys/vm/hugetlb_shm_group
259 in
260 .BR proc_sys_vm (5).
261 .SH STANDARDS
262 POSIX.1-2024.
263 .P
264 .B SHM_HUGETLB
265 and
266 .B SHM_NORESERVE
267 are Linux extensions.
268 .SH HISTORY
269 POSIX.1-2001, SVr4.
270 .\" SVr4 documents an additional error condition EEXIST.
271 .SH NOTES
272 .B IPC_PRIVATE
273 isn't a flag field but a
274 .I key_t
275 type.
276 If this special value is used for
277 .IR key ,
278 the system call ignores all but the least significant 9 bits of
279 .I shmflg
280 and creates a new shared memory segment.
281 .\"
282 .SS Shared memory limits
283 The following limits on shared memory segment resources affect the
284 .BR shmget ()
285 call:
286 .TP
287 .B SHMALL
288 System-wide limit on the total amount of shared memory,
289 measured in units of the system page size.
290 .IP
291 On Linux, this limit can be read and modified via
292 .IR /proc/sys/kernel/shmall .
293 Since Linux 3.16,
294 .\" commit 060028bac94bf60a65415d1d55a359c3a17d5c31
295 the default value for this limit is:
296 .IP
297 .in +4n
298 .EX
299 ULONG_MAX - 2\[ha]24
300 .EE
301 .in
302 .IP
303 The effect of this value
304 (which is suitable for both 32-bit and 64-bit systems)
305 is to impose no limitation on allocations.
306 This value, rather than
307 .BR ULONG_MAX ,
308 was chosen as the default to prevent some cases where historical
309 applications simply raised the existing limit without first checking
310 its current value.
311 Such applications would cause the value to overflow if the limit was set at
312 .BR ULONG_MAX .
313 .IP
314 From Linux 2.4 up to Linux 3.15,
315 the default value for this limit was:
316 .IP
317 .in +4n
318 .EX
319 SHMMAX / PAGE_SIZE * (SHMMNI / 16)
320 .EE
321 .in
322 .IP
323 If
324 .B SHMMAX
325 and
326 .B SHMMNI
327 were not modified, then multiplying the result of this formula
328 by the page size (to get a value in bytes) yielded a value of 8\ GB
329 as the limit on the total memory used by all shared memory segments.
330 .TP
331 .B SHMMAX
332 Maximum size in bytes for a shared memory segment.
333 .IP
334 On Linux, this limit can be read and modified via
335 .IR /proc/sys/kernel/shmmax .
336 Since Linux 3.16,
337 .\" commit 060028bac94bf60a65415d1d55a359c3a17d5c31
338 the default value for this limit is:
339 .IP
340 .in +4n
341 .EX
342 ULONG_MAX - 2\[ha]24
343 .EE
344 .in
345 .IP
346 The effect of this value
347 (which is suitable for both 32-bit and 64-bit systems)
348 is to impose no limitation on allocations.
349 See the description of
350 .B SHMALL
351 for a discussion of why this default value (rather than
352 .BR ULONG_MAX )
353 is used.
354 .IP
355 From Linux 2.2 up to Linux 3.15, the default value of
356 this limit was 0x2000000 (32\ MiB).
357 .IP
358 Because it is not possible to map just part of a shared memory segment,
359 the amount of virtual memory places another limit on the maximum size of a
360 usable segment:
361 for example, on i386 the largest segments that can be mapped have a
362 size of around 2.8\ GB, and on x86-64 the limit is around 127 TB.
363 .TP
364 .B SHMMIN
365 Minimum size in bytes for a shared memory segment: implementation
366 dependent (currently 1 byte, though
367 .B PAGE_SIZE
368 is the effective minimum size).
369 .TP
370 .B SHMMNI
371 System-wide limit on the number of shared memory segments.
372 In Linux 2.2, the default value for this limit was 128;
373 since Linux 2.4, the default value is 4096.
374 .IP
375 On Linux, this limit can be read and modified via
376 .IR /proc/sys/kernel/shmmni .
377 .\" Kernels between Linux 2.4.x and Linux 2.6.8 had an off-by-one error
378 .\" that meant that we could create one more segment than SHMMNI -- MTK
379 .\" This /proc file is not available in Linux 2.2 and earlier -- MTK
380 .P
381 The implementation has no specific limits for the per-process maximum
382 number of shared memory segments
383 .RB ( SHMSEG ).
384 .SS Linux notes
385 Until Linux 2.3.30, Linux would return
386 .B EIDRM
387 for a
388 .BR shmget ()
389 on a shared memory segment scheduled for deletion.
390 .SH BUGS
391 The name choice
392 .B IPC_PRIVATE
393 was perhaps unfortunate,
394 .B IPC_NEW
395 would more clearly show its function.
396 .SH EXAMPLES
397 See
398 .BR shmop (2).
399 .SH SEE ALSO
400 .BR memfd_create (2),
401 .BR shmat (2),
402 .BR shmctl (2),
403 .BR shmdt (2),
404 .BR ftok (3),
405 .BR capabilities (7),
406 .BR shm_overview (7),
407 .BR sysvipc (7)