Add setup action for Alpine and start building it. Unfortunately the
testsuite is not ready yet and fails to build like this:
../testsuite/path.c:219:14: error: conflicting types for '__xstat64';
have 'int(int, const char *, struct stat64 *)'
219 | WRAP_VERSTAT(__x,64);
| ^~~
Add everything needed on the setup action, so when the underlying issue
is fixed, the testsuite can be enabled in CI.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Reviewed-by: Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Link: https://github.com/kmod-project/kmod/pull/44
Add an action to setup latest Fedora and use it for the initial setup.
Add hack on /usr/lib/modules so /usr/lib/modules/*/build can still be
used in the rest of the workflow as the KDIR.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Reviewed-by: Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Link: https://github.com/kmod-project/kmod/pull/44
Add an action to setup Archlinux and use it for the initial setup.
The rest of the workflow can stay the same.
Currently the Arch container in docker hub is not very up-to-date with
some expired keys and other keys not yet in the package. Workaround
that by wiping the current db and reinstalling the keyring. As requested
by Emil, add a comment in this part of the github-action about why
that is done.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Reviewed-by: Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Link: https://github.com/kmod-project/kmod/pull/44
Use containers to build/test so it's not restricted to the OS versions
supported by Github runners. A few changes are needed to the workflow to
support using containers:
1) No need for sudo, so remove it
2) Explicitly install missing zlib
3) Explicitly pass KDIR= while building kernel modules, since it
won't match `uname -r`. This assumes the container has just
one kernel installed and so /usr/lib/modules/*/build can be
used as the single symlink to the kernel headers. This should
be common to other distros to be added, too.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Reviewed-by: Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Link: https://github.com/kmod-project/kmod/pull/44
Separate the package installation step with a local action so
it's easier to support more versions and distros.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Reviewed-by: Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Link: https://github.com/kmod-project/kmod/pull/44
- State support for clang and other libc's
- Fix typos
- Reword the compatibility with module-init-tools section,
removing most of the specific examples as we didn't keep
track of all of them, and they are not important anymore
in year 2024
Reviewed-by: Emil Velikov <emil.l.velikov@gmail.com>
Link: https://lore.kernel.org/r/20240712132449.780421-1-lucas.de.marchi@gmail.com
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
The allocator in glibc has a particular quirk that successive
reallocs on the same pointer are cheap, at the cost of excess
memory fragmentation. Other allocators generally do not do this,
so excessive reallocs become relatively expensive.
Reducing the number of reallocations by using a more agressive
strategy for buffer size increase makes performance better on
those setups, e.g. musl libc, or generally any other allocator;
on my Chimera Linux setup with Scudo allocator (LLVM) it doubles
to triples the performance of running e.g. depmod.
Signed-off-by: q66 <q66@chimera-linux.org>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Fix distcheck failure:
make[5]: *** No rule to make target
'/home/runner/work/kmod/kmod/build/kmod-32/_build/sub/testsuite/module-playground/mod-weakdep.o',
needed by
'/home/runner/work/kmod/kmod/build/kmod-32/_build/sub/testsuite/module-playground/'.
Stop.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Somewhat inspired by my selfish use of VIM as man pager. Namely, when
there are multiple options on the same line, only the first one gets
properly rendered.
A good bonus point is that very long instances, like modinfo's legacy
"--author, --description ..." look a bit neater now.
With this is also more consistently handle short/long options which take
an argument.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Replace the partial listing with suggestion to check through git
shortlog and git blame.
Explicitly spell out the current maintainer, alongside giving Jon
attribution for the original project.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
At a glance through my system, around 2% of the man pages include such
statement.
Looking through git log, Jon has been active in a while and presumably
have moved on.
Most importantly the Copyright section isn't the best place to reference
the maintainer/contact person.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Point the users to modprobe.d(5) instead.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
The environment variable may alter config file ordering, so mention it
in the man page.
In addition, highlight that the format is intentionally undocumented
since the use by third parties is discouraged.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
The modprobe.d configuration order/handling aligns with existing tools
such as sysctl.d, even though there is no mention in the manual.
Reorder the list in SYNOPSIS and add a bit of verbiage describing things.
Section is inspired by sysctl.d(5) and sysctl(8).
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Point the users to depmod.d(5) instead.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Use a slightly longer, more gradual introduction.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
The depmod.d configuration order/handling aligns with existing tools
such as sysctl.d, even though there is no mention in the manual.
Reorder the list in SYNOPSIS and add a bit of verbiage describing things.
Section is inspired by sysctl.d(5) and sysctl(8).
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Presently when DISTCONFDIR is identical to /lib we remove the whole
line, since the man pages already have an /lib instance.
At the same time, there are in-text DISTCONFDIR entries, and removing
the whole line outright breaks the documentation.
Drop the removal line - worst case scenario we get a duplicate entry in
the synopsis.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
For example modinfo -F requires the field name, although the
documentation was missing the "field".
Similarly modprobe has omissions, so let's fix those as well.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Some commas and a full stop was missing ;-)
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
These were present in the original xml files and I opted to keep them
separate fix to make the transition/comparison easier.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Instead of creating invalid man pages, make sure the configure step
fails if scdoc is not available.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Note that scdoc does not handle natively handle the dummy
modules.dep.bin.5 entry, so we need to create one manually.
Not a big deal, since it's single static line anyway.
Also: pkg-config --variable=scdoc scdoc, produces the full executable
and path, although for now we stick with the AC_PATH_PROG approach.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
[ Do not add the stub man page (modules.dep.bin.5) to CLEANFILES
to avoid removing a file that is tracked by git. Also fix a typo
s/AM_V_SCDOR_/AM_V_SCDOC_/ to make silent rules to work ]
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Working with xml is quite fiddly. Swap that for scdoc, which has very
trivial markup-like syntax.
We have opted for scdoc since it's available on practically any linux
distribution. Implementation-wise it's a trivial C99 project, at 1k LoC.
If using scdoc proves to be a burden, we can trivially port these to
pandoc or similar - 90% of the content will stay the same.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
The following tests to verify weak dependencies have been implemented:
1) modprobe test to check that related weakdep modules are not loaded
due to being a weakdep.
2) depmod test to check weakdep output.
3) user test to check that configuration files with weakdep are parsed
correctly and related weakdep modules can be read correctly from user
applications.
Signed-off-by: Jose Ignacio Tornos Martinez <jtornosm@redhat.com>
Link: https://lore.kernel.org/r/20240530070836.9438-1-jtornosm@redhat.com
[ Minor whitespace issues and define MODULE_WEAKDEP if it's not defined
already ]
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
basename() moved to missing.h when the libc doesn't provide it, but
testsuite is not including it. Add missing include.
Fixes: 11eb9bc67c ("Use portable implementation for basename API")
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Printing time_t, suseconds_t, useconds_t in 32bits with -D_TIME_BITS=64
leads to the following warning:
../tools/depmod.c:2641:61: warning: format ‘%li’ expects argument of type ‘long int’, but argument 6 has type ‘__suseconds64_t’ {aka ‘long long int’} [-Wformat=]
2641 | snprintf(tmp, sizeof(tmp), "%s.%i.%li.%li", itr->name, getpid(),
| ~~^
| |
| long int
| %lli
2642 | tv.tv_usec, tv.tv_sec);
| ~~~~~~~~~~
Paper it over by casting the argument to 64 bits and switching to long long.
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Fix building with -D_FILE_OFFSET_BITS=64 -D_TIME_BITS=64 on 32bit arch.
Closes: https://github.com/kmod-project/kmod/issues/37
Bug: https://bugs.debian.org/1065973
Co-authored-by: Jochen Sprickerhof <github@jochen.sprickerhof.de>
Signed-off-by: Shengjing Zhu <shengjing.zhu@canonical.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
musl has removed the non-prototype declaration of basename from
string.h [1] which now results in build errors with clang-17+ compiler
Implement GNU basename behavior using strchr which is portable across libcs
Fixes
../git/tools/kmod.c:71:19: error: call to undeclared function 'basename'; ISO C99 and later do not support implicit function declarations [-Wimplicit-function-declaration]
71 | "Commands:\n", basename(argv[0]));
| ^
[1] https://git.musl-libc.org/cgit/musl/commit/?id=725e17ed6dff4d0cd22487bb64470881e86a92e7
Suggested-by: Rich Felker
Signed-off-by: Khem Raj <raj.khem@gmail.com>
[ Implement a basename() function in missing.h and ensure we always use
the right include rather than having a separate gnu_basename() ]
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
It has been seen that for some network mac drivers (i.e. lan78xx) the
related module for the phy is loaded dynamically depending on the current
hardware. In this case, the associated phy is read using mdio bus and then
the associated phy module is loaded during runtime (kernel function
phy_request_driver_module). However, no software dependency is defined, so
the user tools will no be able to get this dependency. For example, if
dracut is used and the hardware is present, lan78xx will be included but no
phy module will be added, and in the next restart the device will not work
from boot because no related phy will be found during initramfs stage.
In order to solve this, we could define a normal 'pre' software dependency
in lan78xx module with all the possible phy modules (there may be some),
but proceeding in that way, all the possible phy modules would be loaded
while only one is necessary.
The idea is to create a new type of dependency, that we are going to call
'weak' to be used only by the user tools that need to detect this situation.
In that way, for example, dracut could check the 'weak' dependency of the
modules involved in order to install these dependencies in initramfs too.
That is, for the commented lan78xx module, defining the 'weak' dependency
with the possible phy modules list, only the necessary phy would be loaded
on demand keeping the same behavior, but all the possible phy modules would
be available from initramfs.
A new function 'kmod_module_get_weakdeps' in libkmod will be added for
this to avoid breaking the API and maintain backward compatibility. This
general procedure could be useful for other similar cases (not only for
dynamic phy loading).
Signed-off-by: Jose Ignacio Tornos Martinez <jtornosm@redhat.com>
Link: https://lore.kernel.org/r/20240327141116.97587-1-jtornosm@redhat.com
It's cleaner to handle all compression types and load functions in the
same style.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
We're about to reference it in comp_types with next commit.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
This commit cleans up the indentation and the error path of the
function. It bears no functional changes.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
[ Move assert to avoid warning with -Wdeclaration-after-statement ]
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Since all the compression magic is always available now, we don't need
to loop at runtime nor use alloca - latter of which comes with a handful
of caveats.
Simply throw in a few assert_cc(), which will trigger at build-time.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Currently, when built w/o given compression we'll incorrectly report a
"compression_none".
As we reach do_finit_module(), we'll naively assume that the kernel can
handle the compressed module, yet omit the MODULE_INIT_COMPRESSED_FILE
flag.
As result the kernel will barf at us, do_finit_module will fail with non
-ENOSYS and we won't end in the do_init_module codepath (which will also
fail).
In other words: with this change, you can build kmod without zstd, xz
and zlib support and the kernel will load the modules, assuming it
supports the format \o/
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
When dealing with an elf, we don't know or care about loading the file.
The kmod_elf subsystem/API will deal with the required parts itself.
Which in this case, already calls kmod_file_load_contents() as
applicable.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
Propagate any errors during decompression further up the call stack.
Without this we could easily pass NULL as mem to init_module(2).
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
With the previous commits, we removed the need for a distinct unload
callback.
So nuke the struct all together and only use/keep the load one around.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
On mmap failure file->memory is set to -1, which we'll happily pass down
to munmap later on.
More importantly, since we do a NULL check in kmod_file_load_contents()
we will exit the function without (re)attempting the load again.
Since we ignore the return code for the load function(s), one can end up
calling kmod_elf_get_memory() and feed that -1 into init_module.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
These are used to protect a free(file->memory), within their respective
unload functions. Where the sole caller of the unload function already
does a NULL check prior.
Even so, free(NULL) is guaranteed to be safe by the standard.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
There is no need to keep the root gzFile context open for the whole
duration. Once we've copied the decompressed module to file->memory we
can close the handle.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
The gzdopen() API used, takes ownership of the fd. To make that more
explicit we clear it (-1) as applicable.
Yet again, kmod has explicit API to return the fd to the user - which
currently is used solely when uncompressed, so we're safe.
Regardless - simply duplicate the fd locally and use that with zlib.
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Lucas De Marchi <lucas.de.marchi@gmail.com>
