1CORE(5) Linux Programmer's Manual CORE(5)
2
6 core - core dump file
7
9 The default action of certain signals is to cause a process to termi‐
10 nate and produce a core dump file, a file containing an image of the
11 process's memory at the time of termination. This image can be used in
12 a debugger (e.g., gdb(1)) to inspect the state of the program at the
13 time that it terminated. A list of the signals which cause a process
14 to dump core can be found in signal(7).
15 A process can set its soft RLIMIT_CORE resource limit to place an upper
17 limit on the size of the core dump file that will be produced if it re‐
18 ceives a "core dump" signal; see getrlimit(2) for details.
19 There are various circumstances in which a core dump file is not pro‐
21 duced:
22 * The process does not have permission to write the core file. (By
24 default, the core file is called core or core.pid, where pid is the
25 ID of the process that dumped core, and is created in the current
26 working directory. See below for details on naming.) Writing the
27 core file fails if the directory in which it is to be created is not
28 writable, or if a file with the same name exists and is not writable
29 or is not a regular file (e.g., it is a directory or a symbolic
30 link).
31 * A (writable, regular) file with the same name as would be used for
33 the core dump already exists, but there is more than one hard link
34 to that file.
35 * The filesystem where the core dump file would be created is full; or
37 has run out of inodes; or is mounted read-only; or the user has
38 reached their quota for the filesystem.
39 * The directory in which the core dump file is to be created does not
41 exist.
42 * The RLIMIT_CORE (core file size) or RLIMIT_FSIZE (file size) re‐
44 source limits for the process are set to zero; see getrlimit(2) and
45 the documentation of the shell's ulimit command (limit in csh(1)).
46 * The binary being executed by the process does not have read permis‐
48 sion enabled. (This is a security measure to ensure that an exe‐
49 cutable whose contents are not readable does not produce a—possibly
50 readable—core dump containing an image of the executable.)
51 * The process is executing a set-user-ID (set-group-ID) program that
53 is owned by a user (group) other than the real user (group) ID of
54 the process, or the process is executing a program that has file ca‐
55 pabilities (see capabilities(7)). (However, see the description of
56 the prctl(2) PR_SET_DUMPABLE operation, and the description of the
57 /proc/sys/fs/suid_dumpable file in proc(5).)
58 * /proc/sys/kernel/core_pattern is empty and /proc/sys/ker‐
60 nel/core_uses_pid contains the value 0. (These files are described
61 below.) Note that if /proc/sys/kernel/core_pattern is empty and
62 /proc/sys/kernel/core_uses_pid contains the value 1, core dump files
63 will have names of the form .pid, and such files are hidden unless
64 one uses the ls(1) -a option.
65 * (Since Linux 3.7) The kernel was configured without the CONFIG_CORE‐
67 DUMP option.
68 In addition, a core dump may exclude part of the address space of the
70 process if the madvise(2) MADV_DONTDUMP flag was employed.
71 On systems that employ systemd(1) as the init framework, core dumps may
73 instead be placed in a location determined by systemd(1). See below
74 for further details.
75 Naming of core dump files
77 By default, a core dump file is named core, but the /proc/sys/ker‐
78 nel/core_pattern file (since Linux 2.6 and 2.4.21) can be set to define
79 a template that is used to name core dump files. The template can con‐
80 tain % specifiers which are substituted by the following values when a
81 core file is created:
82 %% A single % character.
84 %c Core file size soft resource limit of crashing process (since
85 Linux 2.6.24).
86 %d Dump mode—same as value returned by prctl(2) PR_GET_DUMPABLE
87 (since Linux 3.7).
88 %e The process or thread's comm value, which typically is the same
89 as the executable filename (without path prefix, and truncated
90 to a maximum of 15 characters), but may have been modified to
91 be something different; see the discussion of /proc/[pid]/comm
92 and /proc/[pid]/task/[tid]/comm in proc(5).
93 %E Pathname of executable, with slashes ('/') replaced by exclama‐
94 tion marks ('!') (since Linux 3.0).
95 %g Numeric real GID of dumped process.
96 %h Hostname (same as nodename returned by uname(2)).
97 %i TID of thread that triggered core dump, as seen in the PID
98 namespace in which the thread resides (since Linux 3.18).
99 %I TID of thread that triggered core dump, as seen in the initial
100 PID namespace (since Linux 3.18).
101 %p PID of dumped process, as seen in the PID namespace in which
102 the process resides.
103 %P PID of dumped process, as seen in the initial PID namespace
104 (since Linux 3.12).
105 %s Number of signal causing dump.
106 %t Time of dump, expressed as seconds since the Epoch, 1970-01-01
107 00:00:00 +0000 (UTC).
108 %u Numeric real UID of dumped process.
109 A single % at the end of the template is dropped from the core file‐
111 name, as is the combination of a % followed by any character other than
112 those listed above. All other characters in the template become a lit‐
113 eral part of the core filename. The template may include '/' charac‐
114 ters, which are interpreted as delimiters for directory names. The
115 maximum size of the resulting core filename is 128 bytes (64 bytes in
116 kernels before 2.6.19). The default value in this file is "core". For
117 backward compatibility, if /proc/sys/kernel/core_pattern does not in‐
118 clude %p and /proc/sys/kernel/core_uses_pid (see below) is nonzero,
119 then .PID will be appended to the core filename.
120 Paths are interpreted according to the settings that are active for the
122 crashing process. That means the crashing process's mount namespace
123 (see mount_namespaces(7)), its current working directory (found via
124 getcwd(2)), and its root directory (see chroot(2)).
125 Since version 2.4, Linux has also provided a more primitive method of
127 controlling the name of the core dump file. If the /proc/sys/ker‐
128 nel/core_uses_pid file contains the value 0, then a core dump file is
129 simply named core. If this file contains a nonzero value, then the
130 core dump file includes the process ID in a name of the form core.PID.
131 Since Linux 3.6, if /proc/sys/fs/suid_dumpable is set to 2 ("suid‐
133 safe"), the pattern must be either an absolute pathname (starting with
134 a leading '/' character) or a pipe, as defined below.
135 Piping core dumps to a program
137 Since kernel 2.6.19, Linux supports an alternate syntax for the
138 /proc/sys/kernel/core_pattern file. If the first character of this
139 file is a pipe symbol (|), then the remainder of the line is inter‐
140 preted as the command-line for a user-space program (or script) that is
141 to be executed.
142 Since kernel 5.3.0, the pipe template is split on spaces into an argu‐
144 ment list before the template parameters are expanded. In earlier ker‐
145 nels, the template parameters are expanded first and the resulting
146 string is split on spaces into an argument list. This means that in
147 earlier kernels executable names added by the %e and %E template param‐
148 eters could get split into multiple arguments. So the core dump han‐
149 dler needs to put the executable names as the last argument and ensure
150 it joins all parts of the executable name using spaces. Executable
151 names with multiple spaces in them are not correctly represented in
152 earlier kernels, meaning that the core dump handler needs to use mecha‐
153 nisms to find the executable name.
154 Instead of being written to a file, the core dump is given as standard
156 input to the program. Note the following points:
157 * The program must be specified using an absolute pathname (or a path‐
159 name relative to the root directory, /), and must immediately follow
160 the '|' character.
161 * The command-line arguments can include any of the % specifiers
163 listed above. For example, to pass the PID of the process that is
164 being dumped, specify %p in an argument.
165 * The process created to run the program runs as user and group root.
167 * Running as root does not confer any exceptional security bypasses.
169 Namely, LSMs (e.g., SELinux) are still active and may prevent the
170 handler from accessing details about the crashed process via
171 /proc/[pid].
172 * The program pathname is interpreted with respect to the initial
174 mount namespace as it is always executed there. It is not affected
175 by the settings (e.g., root directory, mount namespace, current
176 working directory) of the crashing process.
177 * The process runs in the initial namespaces (PID, mount, user, and so
179 on) and not in the namespaces of the crashing process. One can uti‐
180 lize specifiers such as %P to find the right /proc/[pid] directory
181 and probe/enter the crashing process's namespaces if needed.
182 * The process starts with its current working directory as the root
184 directory. If desired, it is possible change to the working direc‐
185 tory of the dumping process by employing the value provided by the
186 %P specifier to change to the location of the dumping process via
187 /proc/[pid]/cwd.
188 * Command-line arguments can be supplied to the program (since Linux
190 2.6.24), delimited by white space (up to a total line length of 128
191 bytes).
192 * The RLIMIT_CORE limit is not enforced for core dumps that are piped
194 to a program via this mechanism.
195 /proc/sys/kernel/core_pipe_limit
197 When collecting core dumps via a pipe to a user-space program, it can
198 be useful for the collecting program to gather data about the crashing
199 process from that process's /proc/[pid] directory. In order to do this
200 safely, the kernel must wait for the program collecting the core dump
201 to exit, so as not to remove the crashing process's /proc/[pid] files
202 prematurely. This in turn creates the possibility that a misbehaving
203 collecting program can block the reaping of a crashed process by simply
204 never exiting.
205 Since Linux 2.6.32, the /proc/sys/kernel/core_pipe_limit can be used to
207 defend against this possibility. The value in this file defines how
208 many concurrent crashing processes may be piped to user-space programs
209 in parallel. If this value is exceeded, then those crashing processes
210 above this value are noted in the kernel log and their core dumps are
211 skipped.
212 A value of 0 in this file is special. It indicates that unlimited pro‐
214 cesses may be captured in parallel, but that no waiting will take place
215 (i.e., the collecting program is not guaranteed access to /proc/<crash‐
216 ing-PID>). The default value for this file is 0.
217 Controlling which mappings are written to the core dump
219 Since kernel 2.6.23, the Linux-specific /proc/[pid]/coredump_filter
220 file can be used to control which memory segments are written to the
221 core dump file in the event that a core dump is performed for the
222 process with the corresponding process ID.
223 The value in the file is a bit mask of memory mapping types (see
225 mmap(2)). If a bit is set in the mask, then memory mappings of the
226 corresponding type are dumped; otherwise they are not dumped. The bits
227 in this file have the following meanings:
228 bit 0 Dump anonymous private mappings.
230 bit 1 Dump anonymous shared mappings.
231 bit 2 Dump file-backed private mappings.
232 bit 3 Dump file-backed shared mappings.
233 bit 4 (since Linux 2.6.24)
234 Dump ELF headers.
235 bit 5 (since Linux 2.6.28)
236 Dump private huge pages.
237 bit 6 (since Linux 2.6.28)
238 Dump shared huge pages.
239 bit 7 (since Linux 4.4)
240 Dump private DAX pages.
241 bit 8 (since Linux 4.4)
242 Dump shared DAX pages.
243 By default, the following bits are set: 0, 1, 4 (if the CON‐
245 FIG_CORE_DUMP_DEFAULT_ELF_HEADERS kernel configuration option is en‐
246 abled), and 5. This default can be modified at boot time using the
247 coredump_filter boot option.
248 The value of this file is displayed in hexadecimal. (The default value
250 is thus displayed as 33.)
251 Memory-mapped I/O pages such as frame buffer are never dumped, and vir‐
253 tual DSO (vdso(7)) pages are always dumped, regardless of the core‐
254 dump_filter value.
255 A child process created via fork(2) inherits its parent's coredump_fil‐
257 ter value; the coredump_filter value is preserved across an execve(2).
258 It can be useful to set coredump_filter in the parent shell before run‐
260 ning a program, for example:
261 $ echo 0x7 > /proc/self/coredump_filter
263 $ ./some_program
264 This file is provided only if the kernel was built with the CON‐
266 FIG_ELF_CORE configuration option.
267 Core dumps and systemd
269 On systems using the systemd(1) init framework, core dumps may be
270 placed in a location determined by systemd(1). To do this, systemd(1)
271 employs the core_pattern feature that allows piping core dumps to a
272 program. One can verify this by checking whether core dumps are being
273 piped to the systemd-coredump(8) program:
274 $ cat /proc/sys/kernel/core_pattern
276 |/usr/lib/systemd/systemd-coredump %P %u %g %s %t %c %e
277 In this case, core dumps will be placed in the location configured for
279 systemd-coredump(8), typically as lz4(1) compressed files in the direc‐
280 tory /var/lib/systemd/coredump/. One can list the core dumps that have
281 been recorded by systemd-coredump(8) using coredumpctl(1):
282 $ coredumpctl list | tail -5
284 Wed 2017-10-11 22:25:30 CEST 2748 1000 1000 3 present /usr/bin/sleep
285 Thu 2017-10-12 06:29:10 CEST 2716 1000 1000 3 present /usr/bin/sleep
286 Thu 2017-10-12 06:30:50 CEST 2767 1000 1000 3 present /usr/bin/sleep
287 Thu 2017-10-12 06:37:40 CEST 2918 1000 1000 3 present /usr/bin/cat
288 Thu 2017-10-12 08:13:07 CEST 2955 1000 1000 3 present /usr/bin/cat
289 The information shown for each core dump includes the date and time of
291 the dump, the PID, UID, and GID of the dumping process, the signal
292 number that caused the core dump, and the pathname of the executable
293 that was being run by the dumped process. Various options to core‐
294 dumpctl(1) allow a specified coredump file to be pulled from the sys‐
295 temd(1) location into a specified file. For example, to extract the
296 core dump for PID 2955 shown above to a file named core in the current
297 directory, one could use:
298 $ coredumpctl dump 2955 -o core
300 For more extensive details, see the coredumpctl(1) manual page.
302 To (persistently) disable the systemd(1) mechanism that archives core
304 dumps, restoring to something more like traditional Linux behavior, one
305 can set an override for the systemd(1) mechanism, using something like:
306 # echo "kernel.core_pattern=core.%p" > \
308 /etc/sysctl.d/50-coredump.conf
309 # /lib/systemd/systemd-sysctl
310 It is also possible to temporarily (i.e., until the next reboot) change
312 the core_pattern setting using a command such as the following (which
313 causes the names of core dump files to include the executable name as
314 well as the number of the signal which triggered the core dump):
315 # sysctl -w kernel.core_pattern="%e-%s.core"
317
319 The gdb(1) gcore command can be used to obtain a core dump of a running
320 process.
321 In Linux versions up to and including 2.6.27, if a multithreaded
323 process (or, more precisely, a process that shares its memory with an‐
324 other process by being created with the CLONE_VM flag of clone(2))
325 dumps core, then the process ID is always appended to the core file‐
326 name, unless the process ID was already included elsewhere in the file‐
327 name via a %p specification in /proc/sys/kernel/core_pattern. (This is
328 primarily useful when employing the obsolete LinuxThreads implementa‐
329 tion, where each thread of a process has a different PID.)
330
332 The program below can be used to demonstrate the use of the pipe syntax
333 in the /proc/sys/kernel/core_pattern file. The following shell session
334 demonstrates the use of this program (compiled to create an executable
335 named core_pattern_pipe_test):
336 $ cc -o core_pattern_pipe_test core_pattern_pipe_test.c
338 $ su
339 Password:
340 # echo "|$PWD/core_pattern_pipe_test %p UID=%u GID=%g sig=%s" > \
341 /proc/sys/kernel/core_pattern
342 # exit
343 $ sleep 100
344 ^\ # type control-backslash
345 Quit (core dumped)
346 $ cat core.info
347 argc=5
348 argc[0]=</home/mtk/core_pattern_pipe_test>
349 argc[1]=<20575>
350 argc[2]=<UID=1000>
351 argc[3]=<GID=100>
352 argc[4]=<sig=3>
353 Total bytes in core dump: 282624
354 Program source
356 /* core_pattern_pipe_test.c */
358 #define _GNU_SOURCE
360 #include <sys/stat.h>
361 #include <fcntl.h>
362 #include <limits.h>
363 #include <stdio.h>
364 #include <stdlib.h>
365 #include <unistd.h>
366 #define BUF_SIZE 1024
368 int
370 main(int argc, char *argv[])
371 {
372 ssize_t nread, tot;
373 char buf[BUF_SIZE];
374 FILE *fp;
375 char cwd[PATH_MAX];
376 /* Change our current working directory to that of the
378 crashing process. */
379 snprintf(cwd, PATH_MAX, "/proc/%s/cwd", argv[1]);
381 chdir(cwd);
382 /* Write output to file "core.info" in that directory. */
384 fp = fopen("core.info", "w+");
386 if (fp == NULL)
387 exit(EXIT_FAILURE);
388 /* Display command-line arguments given to core_pattern
390 pipe program. */
391 fprintf(fp, "argc=%d\n", argc);
393 for (int j = 0; j < argc; j++)
394 fprintf(fp, "argc[%d]=<%s>\n", j, argv[j]);
395 /* Count bytes in standard input (the core dump). */
397 tot = 0;
399 while ((nread = read(STDIN_FILENO, buf, BUF_SIZE)) > 0)
400 tot += nread;
401 fprintf(fp, "Total bytes in core dump: %zd\n", tot);
402 fclose(fp);
404 exit(EXIT_SUCCESS);
405 }
406
408 bash(1), coredumpctl(1), gdb(1), getrlimit(2), mmap(2), prctl(2),
409 sigaction(2), elf(5), proc(5), pthreads(7), signal(7), systemd-core‐
410 dump(8)
411
413 This page is part of release 5.13 of the Linux man-pages project. A
414 description of the project, information about reporting bugs, and the
415 latest version of this page, can be found at
416 https://www.kernel.org/doc/man-pages/.
417Linux 2021-03-22 CORE(5)