1GETRLIMIT(2) Linux Programmer's Manual GETRLIMIT(2)
2
6 getrlimit, setrlimit, prlimit - get/set resource limits
7
9 #include <sys/resource.h>
10 int getrlimit(int resource, struct rlimit *rlim);
12 int setrlimit(int resource, const struct rlimit *rlim);
13 int prlimit(pid_t pid, int resource, const struct rlimit *new_limit,
15 struct rlimit *old_limit);
16 Feature Test Macro Requirements for glibc (see feature_test_macros(7)):
18 prlimit():
20 _GNU_SOURCE
21
23 The getrlimit() and setrlimit() system calls get and set resource lim‐
24 its. Each resource has an associated soft and hard limit, as defined
25 by the rlimit structure:
26 struct rlimit {
28 rlim_t rlim_cur; /* Soft limit */
29 rlim_t rlim_max; /* Hard limit (ceiling for rlim_cur) */
30 };
31 The soft limit is the value that the kernel enforces for the corre‐
33 sponding resource. The hard limit acts as a ceiling for the soft
34 limit: an unprivileged process may set only its soft limit to a value
35 in the range from 0 up to the hard limit, and (irreversibly) lower its
36 hard limit. A privileged process (under Linux: one with the
37 CAP_SYS_RESOURCE capability in the initial user namespace) may make ar‐
38 bitrary changes to either limit value.
39 The value RLIM_INFINITY denotes no limit on a resource (both in the
41 structure returned by getrlimit() and in the structure passed to setr‐
42 limit()).
43 The resource argument must be one of:
45 RLIMIT_AS
47 This is the maximum size of the process's virtual memory (ad‐
48 dress space). The limit is specified in bytes, and is rounded
49 down to the system page size. This limit affects calls to
50 brk(2), mmap(2), and mremap(2), which fail with the error ENOMEM
51 upon exceeding this limit. In addition, automatic stack expan‐
52 sion fails (and generates a SIGSEGV that kills the process if no
53 alternate stack has been made available via sigaltstack(2)).
54 Since the value is a long, on machines with a 32-bit long either
55 this limit is at most 2 GiB, or this resource is unlimited.
56 RLIMIT_CORE
58 This is the maximum size of a core file (see core(5)) in bytes
59 that the process may dump. When 0 no core dump files are cre‐
60 ated. When nonzero, larger dumps are truncated to this size.
61 RLIMIT_CPU
63 This is a limit, in seconds, on the amount of CPU time that the
64 process can consume. When the process reaches the soft limit,
65 it is sent a SIGXCPU signal. The default action for this signal
66 is to terminate the process. However, the signal can be caught,
67 and the handler can return control to the main program. If the
68 process continues to consume CPU time, it will be sent SIGXCPU
69 once per second until the hard limit is reached, at which time
70 it is sent SIGKILL. (This latter point describes Linux behav‐
71 ior. Implementations vary in how they treat processes which
72 continue to consume CPU time after reaching the soft limit.
73 Portable applications that need to catch this signal should per‐
74 form an orderly termination upon first receipt of SIGXCPU.)
75 RLIMIT_DATA
77 This is the maximum size of the process's data segment (initial‐
78 ized data, uninitialized data, and heap). The limit is speci‐
79 fied in bytes, and is rounded down to the system page size.
80 This limit affects calls to brk(2), sbrk(2), and (since Linux
81 4.7) mmap(2), which fail with the error ENOMEM upon encountering
82 the soft limit of this resource.
83 RLIMIT_FSIZE
85 This is the maximum size in bytes of files that the process may
86 create. Attempts to extend a file beyond this limit result in
87 delivery of a SIGXFSZ signal. By default, this signal termi‐
88 nates a process, but a process can catch this signal instead, in
89 which case the relevant system call (e.g., write(2), trun‐
90 cate(2)) fails with the error EFBIG.
91 RLIMIT_LOCKS (Linux 2.4.0 to 2.4.24)
93 This is a limit on the combined number of flock(2) locks and fc‐
94 ntl(2) leases that this process may establish.
95 RLIMIT_MEMLOCK
97 This is the maximum number of bytes of memory that may be locked
98 into RAM. This limit is in effect rounded down to the nearest
99 multiple of the system page size. This limit affects mlock(2),
100 mlockall(2), and the mmap(2) MAP_LOCKED operation. Since Linux
101 2.6.9, it also affects the shmctl(2) SHM_LOCK operation, where
102 it sets a maximum on the total bytes in shared memory segments
103 (see shmget(2)) that may be locked by the real user ID of the
104 calling process. The shmctl(2) SHM_LOCK locks are accounted for
105 separately from the per-process memory locks established by
106 mlock(2), mlockall(2), and mmap(2) MAP_LOCKED; a process can
107 lock bytes up to this limit in each of these two categories.
108 In Linux kernels before 2.6.9, this limit controlled the amount
110 of memory that could be locked by a privileged process. Since
111 Linux 2.6.9, no limits are placed on the amount of memory that a
112 privileged process may lock, and this limit instead governs the
113 amount of memory that an unprivileged process may lock.
114 RLIMIT_MSGQUEUE (since Linux 2.6.8)
116 This is a limit on the number of bytes that can be allocated for
117 POSIX message queues for the real user ID of the calling
118 process. This limit is enforced for mq_open(3). Each message
119 queue that the user creates counts (until it is removed) against
120 this limit according to the formula:
121 Since Linux 3.5:
123 bytes = attr.mq_maxmsg * sizeof(struct msg_msg) +
125 min(attr.mq_maxmsg, MQ_PRIO_MAX) *
126 sizeof(struct posix_msg_tree_node)+
127 /* For overhead */
128 attr.mq_maxmsg * attr.mq_msgsize;
129 /* For message data */
130 Linux 3.4 and earlier:
132 bytes = attr.mq_maxmsg * sizeof(struct msg_msg *) +
134 /* For overhead */
135 attr.mq_maxmsg * attr.mq_msgsize;
136 /* For message data */
137 where attr is the mq_attr structure specified as the fourth ar‐
139 gument to mq_open(3), and the msg_msg and posix_msg_tree_node
140 structures are kernel-internal structures.
141 The "overhead" addend in the formula accounts for overhead bytes
143 required by the implementation and ensures that the user cannot
144 create an unlimited number of zero-length messages (such mes‐
145 sages nevertheless each consume some system memory for bookkeep‐
146 ing overhead).
147 RLIMIT_NICE (since Linux 2.6.12, but see BUGS below)
149 This specifies a ceiling to which the process's nice value can
150 be raised using setpriority(2) or nice(2). The actual ceiling
151 for the nice value is calculated as 20 - rlim_cur. The useful
152 range for this limit is thus from 1 (corresponding to a nice
153 value of 19) to 40 (corresponding to a nice value of -20). This
154 unusual choice of range was necessary because negative numbers
155 cannot be specified as resource limit values, since they typi‐
156 cally have special meanings. For example, RLIM_INFINITY typi‐
157 cally is the same as -1. For more detail on the nice value, see
158 sched(7).
159 RLIMIT_NOFILE
161 This specifies a value one greater than the maximum file de‐
162 scriptor number that can be opened by this process. Attempts
163 (open(2), pipe(2), dup(2), etc.) to exceed this limit yield the
164 error EMFILE. (Historically, this limit was named RLIMIT_OFILE
165 on BSD.)
166 Since Linux 4.5, this limit also defines the maximum number of
168 file descriptors that an unprivileged process (one without the
169 CAP_SYS_RESOURCE capability) may have "in flight" to other pro‐
170 cesses, by being passed across UNIX domain sockets. This limit
171 applies to the sendmsg(2) system call. For further details, see
172 unix(7).
173 RLIMIT_NPROC
175 This is a limit on the number of extant process (or, more pre‐
176 cisely on Linux, threads) for the real user ID of the calling
177 process. So long as the current number of processes belonging
178 to this process's real user ID is greater than or equal to this
179 limit, fork(2) fails with the error EAGAIN.
180 The RLIMIT_NPROC limit is not enforced for processes that have
182 either the CAP_SYS_ADMIN or the CAP_SYS_RESOURCE capability.
183 RLIMIT_RSS
185 This is a limit (in bytes) on the process's resident set (the
186 number of virtual pages resident in RAM). This limit has effect
187 only in Linux 2.4.x, x < 30, and there affects only calls to
188 madvise(2) specifying MADV_WILLNEED.
189 RLIMIT_RTPRIO (since Linux 2.6.12, but see BUGS)
191 This specifies a ceiling on the real-time priority that may be
192 set for this process using sched_setscheduler(2) and sched_set‐
193 param(2).
194 For further details on real-time scheduling policies, see
196 sched(7)
197 RLIMIT_RTTIME (since Linux 2.6.25)
199 This is a limit (in microseconds) on the amount of CPU time that
200 a process scheduled under a real-time scheduling policy may con‐
201 sume without making a blocking system call. For the purpose of
202 this limit, each time a process makes a blocking system call,
203 the count of its consumed CPU time is reset to zero. The CPU
204 time count is not reset if the process continues trying to use
205 the CPU but is preempted, its time slice expires, or it calls
206 sched_yield(2).
207 Upon reaching the soft limit, the process is sent a SIGXCPU sig‐
209 nal. If the process catches or ignores this signal and contin‐
210 ues consuming CPU time, then SIGXCPU will be generated once each
211 second until the hard limit is reached, at which point the
212 process is sent a SIGKILL signal.
213 The intended use of this limit is to stop a runaway real-time
215 process from locking up the system.
216 For further details on real-time scheduling policies, see
218 sched(7)
219 RLIMIT_SIGPENDING (since Linux 2.6.8)
221 This is a limit on the number of signals that may be queued for
222 the real user ID of the calling process. Both standard and
223 real-time signals are counted for the purpose of checking this
224 limit. However, the limit is enforced only for sigqueue(3); it
225 is always possible to use kill(2) to queue one instance of any
226 of the signals that are not already queued to the process.
227 RLIMIT_STACK
229 This is the maximum size of the process stack, in bytes. Upon
230 reaching this limit, a SIGSEGV signal is generated. To handle
231 this signal, a process must employ an alternate signal stack
232 (sigaltstack(2)).
233 Since Linux 2.6.23, this limit also determines the amount of
235 space used for the process's command-line arguments and environ‐
236 ment variables; for details, see execve(2).
237 prlimit()
239 The Linux-specific prlimit() system call combines and extends the func‐
240 tionality of setrlimit() and getrlimit(). It can be used to both set
241 and get the resource limits of an arbitrary process.
242 The resource argument has the same meaning as for setrlimit() and getr‐
244 limit().
245 If the new_limit argument is a not NULL, then the rlimit structure to
247 which it points is used to set new values for the soft and hard limits
248 for resource. If the old_limit argument is a not NULL, then a success‐
249 ful call to prlimit() places the previous soft and hard limits for re‐
250 source in the rlimit structure pointed to by old_limit.
251 The pid argument specifies the ID of the process on which the call is
253 to operate. If pid is 0, then the call applies to the calling process.
254 To set or get the resources of a process other than itself, the caller
255 must have the CAP_SYS_RESOURCE capability in the user namespace of the
256 process whose resource limits are being changed, or the real, effec‐
257 tive, and saved set user IDs of the target process must match the real
258 user ID of the caller and the real, effective, and saved set group IDs
259 of the target process must match the real group ID of the caller.
260
262 On success, these system calls return 0. On error, -1 is returned, and
263 errno is set to indicate the error.
264
266 EFAULT A pointer argument points to a location outside the accessible
267 address space.
268 EINVAL The value specified in resource is not valid; or, for setr‐
270 limit() or prlimit(): rlim->rlim_cur was greater than
271 rlim->rlim_max.
272 EPERM An unprivileged process tried to raise the hard limit; the
274 CAP_SYS_RESOURCE capability is required to do this.
275 EPERM The caller tried to increase the hard RLIMIT_NOFILE limit above
277 the maximum defined by /proc/sys/fs/nr_open (see proc(5))
278 EPERM (prlimit()) The calling process did not have permission to set
280 limits for the process specified by pid.
281 ESRCH Could not find a process with the ID specified in pid.
283
285 The prlimit() system call is available since Linux 2.6.36. Library
286 support is available since glibc 2.13.
287
289 For an explanation of the terms used in this section, see at‐
290 tributes(7).
291 ┌────────────────────────────────────────────┬───────────────┬─────────┐
293 │Interface │ Attribute │ Value │
294 ├────────────────────────────────────────────┼───────────────┼─────────┤
295 │getrlimit(), setrlimit(), prlimit() │ Thread safety │ MT-Safe │
296 └────────────────────────────────────────────┴───────────────┴─────────┘
297
299 getrlimit(), setrlimit(): POSIX.1-2001, POSIX.1-2008, SVr4, 4.3BSD.
300 prlimit(): Linux-specific.
302 RLIMIT_MEMLOCK and RLIMIT_NPROC derive from BSD and are not specified
304 in POSIX.1; they are present on the BSDs and Linux, but on few other
305 implementations. RLIMIT_RSS derives from BSD and is not specified in
306 POSIX.1; it is nevertheless present on most implementations. RLIM‐
307 IT_MSGQUEUE, RLIMIT_NICE, RLIMIT_RTPRIO, RLIMIT_RTTIME, and RLIMIT_SIG‐
308 PENDING are Linux-specific.
309
311 A child process created via fork(2) inherits its parent's resource lim‐
312 its. Resource limits are preserved across execve(2).
313 Resource limits are per-process attributes that are shared by all of
315 the threads in a process.
316 Lowering the soft limit for a resource below the process's current con‐
318 sumption of that resource will succeed (but will prevent the process
319 from further increasing its consumption of the resource).
320 One can set the resource limits of the shell using the built-in ulimit
322 command (limit in csh(1)). The shell's resource limits are inherited
323 by the processes that it creates to execute commands.
324 Since Linux 2.6.24, the resource limits of any process can be inspected
326 via /proc/[pid]/limits; see proc(5).
327 Ancient systems provided a vlimit() function with a similar purpose to
329 setrlimit(). For backward compatibility, glibc also provides vlimit().
330 All new applications should be written using setrlimit().
331 C library/kernel ABI differences
333 Since version 2.13, the glibc getrlimit() and setrlimit() wrapper func‐
334 tions no longer invoke the corresponding system calls, but instead em‐
335 ploy prlimit(), for the reasons described in BUGS.
336 The name of the glibc wrapper function is prlimit(); the underlying
338 system call is prlimit64().
339
341 In older Linux kernels, the SIGXCPU and SIGKILL signals delivered when
342 a process encountered the soft and hard RLIMIT_CPU limits were deliv‐
343 ered one (CPU) second later than they should have been. This was fixed
344 in kernel 2.6.8.
345 In 2.6.x kernels before 2.6.17, a RLIMIT_CPU limit of 0 is wrongly
347 treated as "no limit" (like RLIM_INFINITY). Since Linux 2.6.17, set‐
348 ting a limit of 0 does have an effect, but is actually treated as a
349 limit of 1 second.
350 A kernel bug means that RLIMIT_RTPRIO does not work in kernel 2.6.12;
352 the problem is fixed in kernel 2.6.13.
353 In kernel 2.6.12, there was an off-by-one mismatch between the priority
355 ranges returned by getpriority(2) and RLIMIT_NICE. This had the effect
356 that the actual ceiling for the nice value was calculated as
357 19 - rlim_cur. This was fixed in kernel 2.6.13.
358 Since Linux 2.6.12, if a process reaches its soft RLIMIT_CPU limit and
360 has a handler installed for SIGXCPU, then, in addition to invoking the
361 signal handler, the kernel increases the soft limit by one second.
362 This behavior repeats if the process continues to consume CPU time, un‐
363 til the hard limit is reached, at which point the process is killed.
364 Other implementations do not change the RLIMIT_CPU soft limit in this
365 manner, and the Linux behavior is probably not standards conformant;
366 portable applications should avoid relying on this Linux-specific be‐
367 havior. The Linux-specific RLIMIT_RTTIME limit exhibits the same be‐
368 havior when the soft limit is encountered.
369 Kernels before 2.4.22 did not diagnose the error EINVAL for setrlimit()
371 when rlim->rlim_cur was greater than rlim->rlim_max.
372 Linux doesn't return an error when an attempt to set RLIMIT_CPU has
374 failed, for compatibility reasons.
375 Representation of "large" resource limit values on 32-bit platforms
377 The glibc getrlimit() and setrlimit() wrapper functions use a 64-bit
378 rlim_t data type, even on 32-bit platforms. However, the rlim_t data
379 type used in the getrlimit() and setrlimit() system calls is a (32-bit)
380 unsigned long. Furthermore, in Linux, the kernel represents resource
381 limits on 32-bit platforms as unsigned long. However, a 32-bit data
382 type is not wide enough. The most pertinent limit here is RLIM‐
383 IT_FSIZE, which specifies the maximum size to which a file can grow: to
384 be useful, this limit must be represented using a type that is as wide
385 as the type used to represent file offsets—that is, as wide as a 64-bit
386 off_t (assuming a program compiled with _FILE_OFFSET_BITS=64).
387 To work around this kernel limitation, if a program tried to set a re‐
389 source limit to a value larger than can be represented in a 32-bit un‐
390 signed long, then the glibc setrlimit() wrapper function silently con‐
391 verted the limit value to RLIM_INFINITY. In other words, the requested
392 resource limit setting was silently ignored.
393 Since version 2.13, glibc works around the limitations of the getrlim‐
395 it() and setrlimit() system calls by implementing setrlimit() and getr‐
396 limit() as wrapper functions that call prlimit().
397
399 The program below demonstrates the use of prlimit().
400 #define _GNU_SOURCE
402 #define _FILE_OFFSET_BITS 64
403 #include <stdint.h>
404 #include <stdio.h>
405 #include <time.h>
406 #include <stdlib.h>
407 #include <unistd.h>
408 #include <sys/resource.h>
409 #define errExit(msg) do { perror(msg); exit(EXIT_FAILURE); \
411 } while (0)
412 int
414 main(int argc, char *argv[])
415 {
416 struct rlimit old, new;
417 struct rlimit *newp;
418 pid_t pid;
419 if (!(argc == 2 || argc == 4)) {
421 fprintf(stderr, "Usage: %s <pid> [<new-soft-limit> "
422 "<new-hard-limit>]\n", argv[0]);
423 exit(EXIT_FAILURE);
424 }
425 pid = atoi(argv[1]); /* PID of target process */
427 newp = NULL;
429 if (argc == 4) {
430 new.rlim_cur = atoi(argv[2]);
431 new.rlim_max = atoi(argv[3]);
432 newp = &new;
433 }
434 /* Set CPU time limit of target process; retrieve and display
436 previous limit */
437 if (prlimit(pid, RLIMIT_CPU, newp, &old) == -1)
439 errExit("prlimit-1");
440 printf("Previous limits: soft=%jd; hard=%jd\n",
441 (intmax_t) old.rlim_cur, (intmax_t) old.rlim_max);
442 /* Retrieve and display new CPU time limit */
444 if (prlimit(pid, RLIMIT_CPU, NULL, &old) == -1)
446 errExit("prlimit-2");
447 printf("New limits: soft=%jd; hard=%jd\n",
448 (intmax_t) old.rlim_cur, (intmax_t) old.rlim_max);
449 exit(EXIT_SUCCESS);
451 }
452
454 prlimit(1), dup(2), fcntl(2), fork(2), getrusage(2), mlock(2), mmap(2),
455 open(2), quotactl(2), sbrk(2), shmctl(2), malloc(3), sigqueue(3),
456 ulimit(3), core(5), capabilities(7), cgroups(7), credentials(7), sig‐
457 nal(7)
458
460 This page is part of release 5.12 of the Linux man-pages project. A
461 description of the project, information about reporting bugs, and the
462 latest version of this page, can be found at
463 https://www.kernel.org/doc/man-pages/.
464Linux 2021-03-22 GETRLIMIT(2)