On my non-representative hardware, the full build using autotools
(./autogen.sh && CFLAGS=-O2 ./configure && make -j`nproc` && make -j`nproc` install)
takes about 45 seconds.
On the same hardware, the full build using meson
(meson setup -Doptimization=2 dir && meson compile -C dir && meson install -C dir)
takes just about 7.5 seconds.
The utmp database is unreliable for counting logged in users, since
there is no standard which defines who should create an entry at which
time for which reason. And it has a Y2038 problem with glibc/x86-64.
Query systemd-logind for the number of user sessions instead.
This way it is visibly different from the configure variable SCONFIGDIR,
which is helpful, because their values are slightly different:
the macro is quoted while the configure variable is not quoted,
and this difference may cause problems with other build systems.
When limits.d is processed, pl->conf_file consequently points to
elements of filename_list array which are freed later even though
it can still be used in the logging.
Signed-off-by: Shaleen Bathla <shaleen.bathla@oracle.com>
Co-authored-by: Dmitry V. Levin <ldv@strace.io>
Fixes: f6b973b922 ("Add support for limits.d directory to pam_limits")
The line itself is modified while parsing its content. Print the line
number and file name instead so an administrator can investigate it.
Signed-off-by: Tobias Stoeckmann <tobias@stoeckmann.org>
Move function bodies from headers into dedicated object files stored in
libpam_internal. This library won't be installed.
Keep the debug function body in header, even though disabled when
building Linux-PAM, to stay API compatible with previous versions.
Signed-off-by: Tobias Stoeckmann <tobias@stoeckmann.org>
The pam client library libpamc is only needed if libpam_misc is in use.
But libpam_misc is only used by an SELinux helper binary.
Remove the libpamc includes from the search path in all other cases.
Signed-off-by: Tobias Stoeckmann <tobias@stoeckmann.org>
Building outside of source directory fails if --disable-doc is not
explicitly chosen.
This happens because generated files are sometimes expected in the
source directory, where they won't exist.
Signed-off-by: Tobias Stoeckmann <tobias@stoeckmann.org>
The strcpy is not safe anymore because input lines can be very long.
Use strdup instead. If allocation fails, treat the error exactly
like a memory allocation issue in pam_modutil functions.
Signed-off-by: Tobias Stoeckmann <tobias@stoeckmann.org>
Even though sscanf is easy to use for scanning strings, it has the
drawback that the required memory has to be allocated beforehand.
Since variable "line" is not accessed anymore after sscanf, it is
easier to point into the memory assigned to "line". The function
strtok_r can be used as a replacement for sscanf.
Signed-off-by: Tobias Stoeckmann <tobias@stoeckmann.org>
Even if this module is supported officially on Linux, make sure it can
still build fine on non-Linux OSes, to ease its testing/fixing a bit:
- build parse_kernel_limits() and stuff needed for it only on Linux,
as it is called already only on Linux
- limit the code needed to apply the 'nonewprivs' options to Linux only,
as it uses a Linux-specific way to set it; add a syslog message for
other OSes
GCC and Clang only define the macro `linux` when using the GNU dialect
of C (e.g. -std=gnu11 instead of -std=c11). Since `linux` is also not
in a reserved namespace it might be target of collisions.
Use the canonical macro `__linux__` instead (already used in
pam_limits.c).
Otherwise the corresponding files are still installed in /etc/security.
* configure.ac (AC_SUBST): Add VENDOR_SCONFIGDIR.
(AM_CONDITIONAL): Add HAVE_VENDORDIR.
* modules/*/Makefile.am (secureconfdir): Set to VENDOR_SCONFIGDIR
if HAVE_VENDORDIR has been set, otherwise to SCONFIGDIR.
gcc-12 insists on issuing the following warning:
In file included from /usr/include/string.h:535,
from pam_limits.c:24:
In function 'strncat',
inlined from 'check_logins' at pam_limits.c:287:6,
inlined from 'setup_limits' at pam_limits.c:1066:13,
inlined from 'pam_sm_open_session' at pam_limits.c:1267:14:
/usr/include/x86_64-linux-gnu/bits/string_fortified.h:138:10: error: '__builtin___strncat_chk' argument 2 declared attribute 'nonstring' [-Werror=stringop-overread]
138 | return __builtin___strncat_chk (__dest, __src, __len,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
139 | __glibc_objsize (__dest));
| ~~~~~~~~~~~~~~~~~~~~~~~~~
In file included from /usr/include/utmp.h:29,
from pam_limits.c:37:
/usr/include/x86_64-linux-gnu/bits/utmp.h: In function 'pam_sm_open_session':
/usr/include/x86_64-linux-gnu/bits/utmp.h:66:8: note: argument 'ut_user' declared here
66 | char ut_user[UT_NAMESIZE]
| ^~~~~~~
* modules/pam_limits/pam_limits.c (check_logins): Use memcpy instead of
strncat to pacify the compiler.
Changed files
--------------
Make.xml.rules.in:
- Using RNG file instead of DTD file for checking XML files.
- Taking the correct stylesheet for README files.
doc/sag/Makefile.am, doc/adg/Makefile.am, doc/mwg/Makefile.am:
- Using RNG file instead of DTD file for checking XML files.
configure.ac:
- Adding a new option for selecting RNG check file (-enable-docbook-rng)
- Switching stylesheets to docbook 5
- Checking DocBook 5 environment instead of DocBook 4 environment
*.xml:
Update from DockBook 4 to DocBook 5
Use the vendor directory as fallback for a distribution provided default
config if there is no configuration in /etc.
pam_limits.c: Take care about the fallback configuration in vendor directory.
pam_limits.8.xml: Add description for vendor directory.
LIMITS_FILE_DIR used to define a glob pattern instead of a directory
name, fix that inconsistency.
* modules/pam_limits/Makefile.am (AM_CFLAGS): Move "/*.conf" ending of
LIMITS_FILE_DIR macro ...
* modules/pam_limits/pam_limits.c (LIMITS_CONF_GLOB): ... here.
As PR_SET_NO_NEW_PRIVS was introduced by Linux kernel commit
v3.5-rc1~161^2~37, provide a fallback definition to fix build
with older Linux kernel headers.
* modules/pam_limits/pam_limits.c [!PR_SET_NO_NEW_PRIVS]
(PR_SET_NO_NEW_PRIVS): New macro.
Resolves: https://github.com/linux-pam/linux-pam/issues/406
Fixes: dd9cf929 ("modules/pam_limits: add support for nonewprivs")
Replace it with a value obtained from /proc/sys/fs/nr_open
* modules/pam_limits/limits.conf.5.xml: Document the replacement.
* modules/pam_limits/pam_limits.c: Replace unlimited RLIMIT_NOFILE
value with a value obtained from /proc/sys/fs/nr_open
Expose prctl(PR_SET_NO_NEW_PRIVS) as "nonewprivs" item.
The valid values are a boolean toggle 0/1 to keep semi-consistent
with the other numeric limits. It's slightly awkward as this is
an oddball relative to the other items in pam_limits but outside
of the item value itself this does seem at home in pam_limits.
Resolves: https://github.com/linux-pam/linux-pam/issues/224
Resolves: https://github.com/linux-pam/linux-pam/pull/225
Starting with commit a684595c0b aka
Linux-PAM-1.3.0~14 (Remove "--enable-static-modules" option and support
from Linux-PAM), PAM_SM_* macros have no effect.
... and remove $(TESTS) from EXTRA_DIST.
The change is performed automatically using the following script:
sed -i -e 's/^TESTS = \(tst.*\)/dist_check_SCRIPTS = \1\nTESTS = $(dist_check_SCRIPTS)/' \
-e '/^EXTRA_DIST/ s/ \$(TESTS)//' modules/*/Makefile.am
... and remove $(MANS) from EXTRA_DIST.
The change is performed automatically using the following script:
sed -i 's/^man_MANS/dist_&/; /^EXTRA_DIST/ s/ \$(MANS)//' modules/*/Makefile.am
... and remove $(DATA) from EXTRA_DIST.
The change is performed automatically using the following script:
sed -i 's/^[a-z]*_DATA/dist_&/; /^EXTRA_DIST/ s/ \$(DATA)//' modules/*/Makefile.am
Since the GNU Automake distributes README files by default, the only
reason why README had to be listed in EXTRA_DIST was to make these
README files generated.
Since README is also listed in noinst_DATA, we can safely replace
README in EXTRA_DIST with $(DATA), this also opens the way for
further EXTRA_DIST cleanup.
* modules/*/Makefile.am (EXTRA_DIST): Replace README with $(DATA).
