Add a separate function to load the file contents when it's needed.
When it's not needed on the path of loading modules via finit_module(),
there is no need to mmap the file. This will help support loading
modules with the in-kernel compression support.
This is done differently than the lazy initialization for
kmod_file_get_elf() because on the contents case there is also the
file->size to be updated. It would be a weird API to return the pointer
and have the size changed as a side-effect.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Fix the following warning reported by UBSan (as of gcc-13.1.1):
shared/hash.c:244:35: runtime error: null pointer passed as
argument 2, which is declared to never be null
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Dmitry Antipov <dmantipov@yandex.ru>
[ reshuffle the code to use return-early style ]
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
An attempt to pass too long module name to, say, rmmod, may
cause an out-of-bounds memory access (as repoted by UBSan):
$ rmmod $(for i in $(seq 0 4200); do echo -ne x; done)
libkmod/libkmod-module.c:1828:8: runtime error: index 4107 out of bounds for type 'char [4096]'
This is because 'snprintf(path, sizeof(path), ...)' may return the
value which exceeds 'sizeof(path)' (which happens when an output
gets truncated). To play it safe, such a suspicious output is
better to be rejected explicitly.
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Dmitry Antipov <dmantipov@yandex.ru>
Link: https://lore.kernel.org/r/20230519074638.402045-1-dmantipov@yandex.ru
When the module is definitely present but CONFIG_MODVERSIONS is
disabled, the following error message may be somewhat confusing:
modprobe --dump-modversions /path/to/module.ko.xz
modprobe: FATAL: could not get modversions of /path/to/module.ko.xz: No such file or directory
Choosing among the convenient errno values, I would suggest to use
ENODATA when the module lacks a particular ELF section (and vermagic
as well). So now it is expected to be:
modprobe: FATAL: could not get modversions of /path/to/module.ko.xz: No data available
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Signed-off-by: Dmitry Antipov <dmantipov@yandex.ru>
Link: https://lore.kernel.org/r/20230519093630.474185-1-dmantipov@yandex.ru
Add a missing kmod_module_unref_list() to fix a memory leak.
Fixes: 42b32d30c3 ("modprobe: Fix holders removal")
Signed-off-by: Nicolas Schier <n.schier@avm.de>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Add liblzma_LIBS to LIBS to avoid the following build failure when
building with a static-only liblzma.a:
/home/autobuild/autobuild/instance-5/output-1/host/lib/gcc/xtensa-buildroot-linux-uclibc/11.3.0/../../../../xtensa-buildroot-linux-uclibc/bin/ld: libkmod/.libs/libkmod-internal.a(libkmod-file.o):(.text.xz_uncompress+0x10): undefined reference to `lzma_code'
For consistency, also update libzstd, zlib and libcrypto
Fixes:
- http://autobuild.buildroot.org/results/83a4a7ecc77f39639d3e5bc8554bd01a62a3ede0
References: https://github.com/kmod-project/kmod/pull/25
Signed-off-by: Fabrice Fontaine <fontaine.fabrice@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Replace the manual sed command, build rules and dist/clean for using
AC_CONFIG_FILES. It does the exact same thing, with an added bonus...
Currently we're missing version.py.in in the EXTRA_DIST. Thus a simple
"touch Makefile" should retrigger the regeneration of version.py. Which
would presumably fail, since the input file isn't in the distribution
tarball.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
None of the API is used outside of the compilation unit.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
It's no longer used outside the compilation unit, as of last commit.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
The last and only user was removed with commit 0246e06 ("depmod: Stop
opening modules.modinfo once per module")
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Instead of skipping tests if sysconfdir isn't /etc, just handle it
during the rootfs setup logic.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
It's easier to implement the logic outside of the Makefile, so rename
the populate-modules.sh script to setup-rootfs.sh and move the
additional logic from the makefile to the script.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Currently if we see unknown algorithm, we'll do an OOB read in
pkey_hash_algo. This can happen for example if OPENSSL_NO_SM3 is set and
the kernel module uses a SM3 hash.
Cc: Mikhail Novosyolov <m.novosyolov@rosalinux.ru>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
The uname used across the tests is same, so drop "_ORDER" from the macro
name and use it throughout. Similarly - add respective LIB_MODULES
defines and use them in the tests.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
As the inline comment says - the declarations have been dropped with
glibc 2.32.9000, as a result the build throws a set of lovely warnings.
Inspired by umockdev, which bears the same license as this project.
f1b4164004
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
This option is equivalent to basedir, with the small difference being
that's where the meta-data files are generated. In other words, this
allows us to have read-only input modules and modules.dep, while still
being able to generate the meta-data files.
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
[ Move files to a different dir so input files (produced by kernel build
system is separate from the files generated by depmod (output) ]
Signed-off-by: Lucas De Marchi <lucas.demarchi@gmail.com>
Improve user experience by explaining the option so the user may
not search explanations in other manpages (modprobe).
Signed-off-by: Yauheni Kaliuta <ykaliuta@redhat.com>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
This is useful to kernel module developers for testing a just compiled
module: instead of using insmod, they can load the module from the path
while getting all the benefits of modprobe (e.g. module dependency
resolution).
v2:
- Add test for relative path as well. (Lucas)
- Add note warning about modules with dependencies not matching the
installed depmod database. (Lucas)
Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
That same logic will be used for enabling modprobe for paths in the next
patch. As such, prepare for that by extracting that block into its own
function.
Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
One of the tests in an upcoming patch will need to change into a
specific directory to test loading a module from a relative path.
Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Otherwise, an implicit functiona declaration is used, causing
a C99 compatibility issue.
Signed-off-by: Florian Weimer <fweimer@redhat.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Example kernel module:
https://file-store.rosalinux.ru/download/7281f97e0c04c0f818ad3f936706f4a407e8dc7e
(/lib/modules/5.15.67-generic-1rosa2021.1-x86_64/kernel/drivers/usb/host/xhci-pci.ko.zst)
It is signed with Streebog 512.
libkmod v30 crashed in libkmod-module.c:2413 in this code:
n = kmod_module_info_append(list,
"sig_hashalgo", strlen("sig_hashalgo"),
sig_info.hash_algo, strlen(sig_info.hash_algo));
because strlen() got null.
When _FILE_OFFSET_BITS is 64, glibc headers turn `stat` calls into
`stat64`, and our `stat` override into a `stat64` function. However,
because we use dlsym to get the address of libc's `stat`, we end up
calling into the "real" `stat` function, which deals with 32-bit off_t,
and we treat its result as if it were returned from stat64. On most
architectures this seems to have been harmless, but on 32-bit mips,
st_mode's offset in struct stat and struct stat64 are different, so we
read garbage.
To fix this, explicitly unset _FILE_OFFSET_BITS in path.c, to turn off
the redirect magic in glibc headers, and override both the 32-bit and
64-bit functions so each call ends up wrapping the right libc function.
Fixes#16 (https://github.com/kmod-project/kmod/issues/16)
The man page modprobe(8) states for the --syslog option:
"This is also automatically enabled when stderr is unavailable."
but it wasn't happening.
This commit now makes modprobe write to syslog if stderr is closed.
During dpkg build, in a subdir, it is currently not possible to run
tests. Building testsuite/modules due to non-existance of the
testsuite directory under the build dir. Thus create it, when it is
not there.
Signed-off-by: Dimitri John Ledkov <dimitri.ledkov@canonical.com>
Function ``test_backoff_time`` does not initialize ``delta``, and
``get_backoff_delta_msec`` then performs a read from uninitialized
memory with the ``!*delta`` expression.
Signed-off-by: Jan Engelhardt <jengelh@inai.de>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
SM3 has been supported in kernel and cryptographic libraries like openssl.
This patch adds support for the SM3 algorithm of kmod.
Signed-off-by: HuaxinLu <luhuaxin1@huawei.com>
Retry module removal if it fails due to EAGAIN. This allows user to pass
--wait <timeout>, during which `modprobe -r` will keep trying to remove
the module.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Add simple functions to put the current thread to sleep using
exponential backoff to split the interval in smaller pieces.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Add a debugfs file in mod-simple for manual tests: insert the module and
open the file to have its refcount increased.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Make sure return value in flush_stream_to() is the length written
if the value didn't the size. Fix warning on gcc 12.1:
tools/depmod.c: In function ‘output_builtin_alias_bin’:
tools/depmod.c:2465:24: warning: array subscript 4096 is above array bounds of ‘char[4096]’ [-Warray-bounds]
2465 | modname[len] = '\0';
| ~~~~~~~^~~~~
tools/depmod.c:2460:22: note: while referencing ‘modname’
2460 | char modname[PATH_MAX];
| ^~~~~~~
tools/depmod.c:2477:22: warning: array subscript 4096 is above array bounds of ‘char[4096]’ [-Warray-bounds]
2477 | value[len] = '\0';
| ~~~~~^~~~~
tools/depmod.c:2461:22: note: while referencing ‘value’
2461 | char value[PATH_MAX];
| ^~~~~
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Move the remaining part of the removal sequence dangling in
rmmod_do_remove_module() to rmmod_do_module() so we can consider this
function is the one controlling all the module removals.
While at it, add some comments about the removal order and normalize
coding style in this function.
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
Do not mix the flags with and additional boolean from arguments.
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
The idea behind --remove-dependencies was to remove other modules that
depend on the current module being removed. It's the reverse
dependency list, not the dependency list of the current module: that
never works since the current module would still hold a ref on it.
Fix it by replacing the call to kmod_module_get_dependencies() with
kmod_module_get_holders() when using that option. Also try to cleanup
the confusion by renaming the option to --remove-holders: "holder" is
the name used in sysfs and by libkmod to refer to a "live" reverse
dependency like what we are interested in.
Before:
./tools/modprobe -D -r --remove-dependencies video
rmmod video
After:
./tools/modprobe -D -r --remove-holders video
rmmod i915
rmmod thinkpad_acpi
rmmod video
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
The variable 'str_start' is not useful here.
Replace it with 'str'.
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
The number of new arguments is (i + argc - 1) as it is set to *p_argc
one line below.
The correct location of NULL termination is new_argv[i + argc - 1].
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
This adds support to depmod to enable a new exclude directive in
the depmod.d/*.conf configuration file. Currently depmod
already excludes directories named source or build. This change
will allow additional directories like .debug to be excluded also
via a new exclude directive.
depmod.d/exclude.conf example:
exclude .debug
Signed-off-by: Saul Wold <saul.wold@windriver.com>
[ Fix warnings and make should_exclude_dir() return bool ]
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
It's used not only for dependencies, but also for pre and post softdep.
Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>
Reviewed-by: Luis Chamberlain <mcgrof@kernel.org>
