Lucas De Marchi [Wed, 19 Mar 2014 12:15:59 +0000 (09:15 -0300)]
Add test for depmod using search dirs with same prefix
Test depmod with search dirs "foo" and "foobar". Previously to 49b33c1
("depmod: do not allow partial matches with "search" directive") we were
failing this test due to matching the prefix without checking if
it's the full dir name.
We are adding 2 tests here in order to catch the case we only pass the
test due to processing the directories in a favourable order.
Anssi Hannula [Tue, 18 Mar 2014 23:26:00 +0000 (01:26 +0200)]
depmod: do not allow partial matches with "search" directive
Currently e.g. "search foo foobar built-in" will cause unpredictable
results if baz.ko is in both foo/ and foobar/, since "foo" in search may
match both of those directories and the preferred module therefore
depends on processing order.
Fix the code to ensure that the match is performed on full pathname
components only.
Lucas De Marchi [Fri, 7 Mar 2014 04:17:10 +0000 (01:17 -0300)]
libkmod-elf: Fix check by class in get_modversions()
Commit 51c409b ("Cache the offset of crc") unintentinally changed the
comparison "if (elf->class & KMOD_ELF_32)" to
"if (elf->class == KMOD_ELF_32)".
This has been reported by Serge Voilokov <serge0x76@gmail.com>:
On Raspberry PI elf->class equals KMOD_ELF_32|KMOD_ELF_LSB so
valid condition should be (elf->class & KMOD_ELF_32) instead of
(elf->class == KMOD_ELF_32).
This fixes "modprobe --dump-modversions" failing on 32b systems.
Michal Marek [Thu, 6 Mar 2014 17:03:46 +0000 (18:03 +0100)]
testsuite: Fix uname() during glibc startup
In a specific configuration (chroot with the linux32 personality), the
modprobe_install_cmd_loop test failed, because the bash process handling
the install command segfaulted. The backtrace showed a uname() call
during libpthread initialization, at which point the environ pointer
hadn't been initialized yet:
Program terminated with signal SIGSEGV, Segmentation fault.
#0 0x080c1591 in getenv (name=<optimized out>,
name@entry=0xf775f850 "TESTSUITE_UNAME_R") at getenv.c:81
81 for (i = 0, len = strlen (name); environ[i]; i++)
(gdb) bt
#0 0x080c1591 in getenv (name=<optimized out>,
name@entry=0xf775f850 "TESTSUITE_UNAME_R") at getenv.c:81
#1 0xf775f754 in uname (u=u@entry=0xff946350) at testsuite/uname.c:32
#2 0xf74ffc6c in is_smp_system ()
at ../nptl/sysdeps/unix/sysv/linux/i386/smp.h:39
#3 __pthread_initialize_minimal_internal () at nptl-init.c:460
#4 0xf74fe32c in _init () at ../sysdeps/i386/crti.S:74
#5 0x00000000 in ?? ()
(gdb) p environ
$1 = (char **) 0x0
I don't know why it only happend in the chroot, but glibc can call its
own functions and impose any restrictions before main() is started, so
we have to adapt.
Also, do not return error if there is an environment, but the
environment variable is not found. If uname() is called by kmod, then
the respective test will simply fail later. If it's something else
calling uname(), then we do not want to disturb the program.
Lucas De Marchi [Thu, 6 Mar 2014 05:47:05 +0000 (02:47 -0300)]
man: use systemd as example instead of udev
Nowadays udev doesn't create nodes in /dev anymore. This role is rather
taken by systemd-tmpfiles on early boot so reference it generically as
systemd.
Lucas De Marchi [Thu, 6 Mar 2014 04:51:33 +0000 (01:51 -0300)]
build-sys: add small redirecting Makefiles
These redirecting makefiles simplifies compiling from some editors and
when CWD is not the root of the source tree. This is similar to what was
introduced in systemd in 340d89e ("build-sys: add small redirecting
Makefiles to simplify compilation from within emacs")
Lucas De Marchi [Thu, 6 Mar 2014 04:36:02 +0000 (01:36 -0300)]
testsuite: add test for '.' correctly parsed in param's value
In kcmdline it's possible to have a dot in the param's value. The
support for this was added in 66f3228 ("libkmod: Add support for '.' in
module parameter on kcmdline") and is needed to correctly support some
modules that depend on it.
This test was added in order to make sure the commit aa87854
("libkmod-config: Only match dot before '=' in /proc/cmdline") didn't
break it. Although that commit message says it's allowing to match a
dot before '=' it's actually enforcing the first part of the string to
be always in the format "<module-name>.param". Dots after '=' are still
correctly allowed.
Lucas De Marchi [Thu, 6 Mar 2014 04:14:12 +0000 (01:14 -0300)]
testsuite: give a more suitable description to test
We are not only checking if those options are correctly parsed from
kcmdline, but if in fact they are being passed to the final
(f)init_module call. This is why we use 'modprobe --show-depends'
instead of the simpler 'modprobe -c'.
Lucas De Marchi [Thu, 6 Mar 2014 04:09:30 +0000 (01:09 -0300)]
testsuite: add test for kcmdline params with no value
Use "modprobe -c" to dump the configuration. Since we configure our
rootfs to have only a /proc/cmdline file, this should dump the knowledge
we have from its parsed content.
Test if <module>.option, without any value is correctly parsed, as fixed
in commit 493dc65 ("libkmod: Fix getting param with no value from kcmdline")
Michal Marek [Fri, 28 Feb 2014 12:05:28 +0000 (13:05 +0100)]
testsuite: Check the list of loaded modules after a test
Add a ->modules_loaded member to struct test, which is a comma-separated
list of modules that should be present after the test finishes. Both
missing and excess modules cause an error.
Anders Olofsson [Mon, 11 Nov 2013 01:41:25 +0000 (23:41 -0200)]
build: Allow disabling maintainer mode
This allows make rules for generated build files (i.e. configure,
Makefile.in, ... ) to be skipped. This is useful when
the source is stored without timestamps (for example in CVS or GIT).
When the build rules trigger to regenerate the build files, it tries to
use the same autotools version (currently 1.14) as was originally used
for the release. Since many of our build machines run Debian Squeeze,
they only have autotools 1.11 available and the build fails.
Currently, we have to work around this by touching all the generated
files before building to avoid triggering the make rule. With this
patch, we would be able to just run configure with
--disable-maintainer-mode instead. The patch sets the default to enable
to not change the default behavior.
Saul Wold [Thu, 10 Oct 2013 06:49:48 +0000 (23:49 -0700)]
Makefile.am: add mkdir testsuite
If we are note building in the existing source tree and have disabled
dependency-tracking then the testsuite directory is not created during
the configure phase and will not exist when the cp of ROOTFS_PRISTINE
occurs, thus causing an error and fail.
Lucas De Marchi [Thu, 22 Aug 2013 04:36:45 +0000 (01:36 -0300)]
libkmod-hash: always align n_buckets to power of 2
By aligning n_buckets to power of 2 we can turn the "bucket = hashval %
n_buckets" into a less expensive bucket = hashval & (n_buckets - 1).
This removes the DIV instruction as shown below.
With a microbenchmark, measuring the time to locate the bucket (i.e.
time_to_calculate_hashval + time_to_calculate_bucket_position) we have
the results below (time in clock cycles):
As expected the gain is constant, regardless of the key length.
The time to clculate the hashval varies with the key length, which
explains the bigger gains for short keys.
Lucas De Marchi [Thu, 22 Aug 2013 04:10:13 +0000 (01:10 -0300)]
util: Add ALIGN_POWER2
Add static inline function to align a value to it's next power of 2.
This is commonly done by a SWAR like the one in:
http://aggregate.org/MAGIC/#Next Largest Power of 2
However a microbench shows that the implementation herer is a faster.
It doesn't really impact the possible user of this function, but it's
interesting nonetheless.
Using a x86_64 i7 Ivy Bridge it shows a ~4% advantage by using clz
instead instead of the OR and SHL chain. And this is by using a BSR
since Ivy Bridge doesn't have LZCNT. New Haswell processors have the
LZCNT instruction which can make this even better. ARM also has a CLZ
instruction so it should be better, too.
Code used to test:
...
v = val[i];
t1 = get_cycles(0);
a = ALIGN_POWER2(v);
t1 = get_cycles(t1);
t2 = get_cycles(0);
v = nlpo2(v);
t2 = get_cycles(t2);
printf("%u\t%llu\t%llu\t%d\n", v, t1, t2, v == a);
...
In which val is an array of 20 random unsigned int, nlop2 is the SWAR
implementation and get_cycles uses RDTSC to measure the performance.
Tom Gundersen [Mon, 9 Sep 2013 18:01:01 +0000 (20:01 +0200)]
depmod: warn on invalid devname specification
During the last merge window (3.12) a couple of modules gained devname
aliases, but without the necessary major and minor information. These were
then silently ignored when generating modules.devname.
Complain loudly to avoid such errors sneaking in undetected in the future:
depmod: ERROR: Module 'zram' has devname (zram) but lacks major and minor information. Ignoring.
depmod: ERROR: Module 'uhid' has devname (uhid) but lacks major and minor information. Ignoring.
Cc: Kay Sievers <kay@vrfy.org> Cc: Lucas De Marchi <lucas.demarchi@profusion.mobi>
Commit 8efede20ef ("Use _Static_assert") introduced the usage of
_Static_assert(). However, _Static_assert() is a fairly new thing,
since it was introduced only in gcc 4.6. In order to support older
compilers, this patch adds a configure.in test that checks whether
_Static_assert() is usable or not, and adjust the behavior of the
assert_cc() macro accordingly.
However in all places that we use readdir_r() we have no concerns about
reentrance nor we have problems with threads. Thus use the simpler
readdir() instead.
We also remove the error logging here (that could be added back by
checking errno), but it was not adding much value so it's gone.
John Spencer [Mon, 26 Aug 2013 23:38:11 +0000 (01:38 +0200)]
testsuite: fix usage of reserved names
stdout and stderr are names reserved for the implementation
and musl uses them rightfully as macro - and the expansion
causes (of course) unexpected results.
rename the struct members stdout to out and stderr
to err, to be 1) compliant 2) cause compilation to
succeed.
Grepping for the error message revealed that there might be a missing
"else" keyword here, since it is unusual to put an "if" directly after
closing brace.
Lucas De Marchi [Mon, 15 Jul 2013 04:21:27 +0000 (01:21 -0300)]
testsuite: Fix mkdir_p corner cases
- Fix infinite loop when path is relative
- Fix not considering EEXIST as a success
- General refactor to mkdir_p so it never calls mkdir for an existing
dir (given no one creates it from outside)
Lucas De Marchi [Thu, 4 Jul 2013 19:01:55 +0000 (16:01 -0300)]
tools: Do not link dynamically with libkmod
Instead of linking dynamically with libkmod, use libkmod-private.la. We
disallow creating a static libkmod because we can't hide symbols there
and it cause problems with external programs. However this should not
prevent users that are only interested in the tools we provide not being
able to ship only them keeping the library alone.
Other projects also do this to allow our tools to use certain functions
that should not be used outside of the project.
Lucas De Marchi [Wed, 3 Jul 2013 00:03:20 +0000 (21:03 -0300)]
tools: Use test/kmod instead of kmod-nolib
The reason to have a kmod-nolib binary is that we need to call kmod on
test cases (or a symlink to it) and for testing things in tree. Since
we are using libtool if we are dinamically linking to libkmod what we
end up having is a shell script that (depending on the version *)
changes argv[0] to contain an "lt-" prefix. Since this screws with our
compat stuff, we had a kmod-nolib that links statically.
This all workaround works fine iff we are using one of the compat
commands, i.e. we are using the symlinks insmod, rmmod, modprobe, etc.
However if we are actually trying the kmod binary, this doesn't work
because we can't create a kmod symlink since there's already a kmod
binary.
So, completely give up on libtool fixing their mess. Now we create a
tool/test/ directory and the symlinks and kmod is put there.
Lucas De Marchi [Thu, 6 Jun 2013 14:43:19 +0000 (11:43 -0300)]
build-sys: do not allow --enable static
Do the same as done in systemd by Cristian RodrÃguez
<crrodriguez@opensuse.org>. We use private symbols, not namespaced. So
don't pretend we support static linking.
Jan Luebbe [Thu, 2 May 2013 14:47:12 +0000 (16:47 +0200)]
libkmod: Avoid calling syscall() with -1
At least in qemu 1.4.1 for vexpress/arm-cortexa9, this resulted in an
illegal instruction error. Solve that by returning an error when
__NR_finit_module is -1.
Lucas De Marchi [Sat, 11 May 2013 03:50:32 +0000 (00:50 -0300)]
Revert "missing: Don't call syscall() with syscallno == -1"
This reverts commit 38829712e5c411bc250aeae142fc6bf06e794d58. It fixes
the problem, but it breaks the testsuite for those who don't have
__NR_finit_module. The testsuite would have to make the same check.
Instead, I'm reverting this change and I'm going to apply another patch
from Jan Luebbe who got this right from the beginning.
Chengwei Yang [Sat, 4 May 2013 09:07:03 +0000 (17:07 +0800)]
Add document for exported enums
There are several exported enums by libkmod without document, this patch
mainly added documentation for below enums like the way kmod_resources
be documented in.
* kmod_index
* kmod_remove
* kmod_insert
* kmod_probe
* kmod_filter
* kmod_module_initstate
This is not the best way to document these exported enums, however, it's
the simple way due to gtkdoc limits. It doesn't support export plain
enum like below: see https://bugzilla.gnome.org/show_bug.cgi?id=657444
---------8<-------head.h--------------8<-----------
...
enum foo {
...
};
...
---------8<-------end of head.h-------8<-----------
---------8<-------source.c------------8<-----------
...
/**
* document for foo here
*/
...
typedef enum foo foo;
...
---------8<-------end of source.c-----8<----------
Johannes Berg [Thu, 2 May 2013 13:23:28 +0000 (15:23 +0200)]
modprobe: don't check refcount with remove command
The modprobe.d (5) documentation for the "install" command
states that you could specify
install fred /sbin/modprobe barney; /sbin/modprobe --ignore-install fred
This makes some sense, but then the loading of "barney" is
hidden from the user who did only "modprobe fred". Thus,
it seems it should be possible to be able to unload the
"fred" module with "modprobe -r fred" by configuring the
"barney" module to also be removed:
remove fred /sbin/rmmod barney fred
(or similar.)
Make this possible by not checking the refcount when an
unload command was configured.
Reported-by: David Spinadel <david.spinadel@intel.com>
Lucas De Marchi [Sun, 21 Apr 2013 19:16:18 +0000 (16:16 -0300)]
Add format attribute and fix issues
Add __attribute__((format)) to log_filep() and _show() functions, fixing
the bugs they found in the source code.
For functions that receive va_list instead of being variadic functions
we put 0 in the last argument, so at least the string is checked and we
get warnings of -Wformat-nonliteral type. So, it's better than adding a
pragma here to shut up the warning.
Tom Gundersen [Tue, 16 Apr 2013 20:39:55 +0000 (22:39 +0200)]
tools: add static-nodes tool
This tool reads modules.devname from the current kernel directory and outputs
the information. By default in a human-readable format, and optionally in
machine-readable formats.
For now only the tmpfiles.d(5) format is supported, but more could easily be
added in the future if there is a need.
This means nothing but kmod needs to reads the private files under
/lib/modules/. In particular systemd-udevd can stop reading modules.devname.
Tools that used to read /lib/modules/`uname -r`/modules.devname directly, can
now move to reading 'kmod static-nodes devname'.
Lucas De Marchi [Tue, 9 Apr 2013 14:54:05 +0000 (11:54 -0300)]
testsuite: Fix checking __sysno
Use an if instead of a case statemente. If __NR_finit_module is not
defined in system headers we define it to -1, causing a "duplicate case
value" error. Yet, we don't want to actually call our finit_module()
function if -1 is passed.
This also fix errno being set with negative value.
Lucas De Marchi [Tue, 9 Apr 2013 08:21:42 +0000 (05:21 -0300)]
testsuite: Wrap syscall() to get calls to finit_module()
When we don't have finit_module() in libc (most likely because as of
today glibc didn't add it yet), we end up using
syscall(__NR_finit_module, ...). In this case we would not wrap the
function in the testsuite and thus having some tests failing:
TESTSUITE: ERR: could not insert module: Operation not permitted
This implementation relies on the fact that this is the only caller of
syscall(2), because we can't call libc's syscall(). There's an abort()
in place to be future safe: as soon as we need more calls to syscall(),
we can detect (and decide what to do).
Now we have all tests passing in the testsuite again.
Lucas De Marchi [Tue, 9 Apr 2013 07:16:57 +0000 (04:16 -0300)]
libkmod: Move finit_module() definition to missing.h
Check for finit_module() and don't use our own static inline function if
there's such function in libc (or another lib).
In testsuite we need to unconditionally define HAVE_FINIT_MODULE because
we want to override this function, and never use the static inline one
in missing.h