Not all architectures use a /vmlinuz symlink in Debian. For instance,
armhf systems may only provide a symlink in /boot/vmlinuz. Fall back to
testing /boot/vmlinuz if /vmlinuz is not found.
Signed-off-by: Timo Sigurdsson <public_timo.s@silentcreek.de>
* Check if the "locale" binary is available before using it.
This is no functional change as it will still fall back to english
when the locale can't be determined. This fix gets rid of the
following error when running on systems without the locale binary:
./lynis: line 112: locale: command not found
Signed-off-by: Daniel Romell <daro@hms.se>
* tests_kernel: KRNL-5677: Fix invalid use of shell test.
This fixes an issue (syntax error) triggered on systems with no PAE or
NX extensions:
- Checking CPU support (NX/PAE)
/usr/libexec/lynis/include/tests_kernel: line 126: [: too many arguments
/usr/libexec/lynis/include/tests_kernel: line 132: [: too many arguments
No need to use [] when only looking at function return values.
Signed-off-by: Daniel Romell <daro@hms.se>
* Work around Solaris' /bin/sh not being POSIX.
If /usr/xpg4/bin/sh is present, we are (definitely?) on Solaris or
a derivative, and /bin/sh cannot be trusted to support POSIX, but
/usr/xpg4/bin/sh can be. Exec it right away.
* Work around Solaris 'which' command oddity.
Solaris' (at least) 'which' command outputs not-found errors to STDOUT
instead of STDERR.
This makes "did we get any output from which" checks insufficient;
piping to grep -v the "no foo in ..." message should work.
Note that this patch set includes all such uses of which that I could
find, including ones that should never be reached on Solaris (i.e. only
executed on some other OS) just for consistency.
* Improved alternate-sh exec to avoid looping.
* Solaris' /usr/ucb/echo supports -n.
* Check for the best hash type that openssl supports.
When using openssl to generate hashes, do not assume it supports
sha256; try that, then sha1, then give up and use md5.
* Solaris does not support sed -i; use a tempfile.
* Use the full path for modinfo.
When running as non-root, /usr/sbin/ might not be in PATH.
include/tests_accounting already calls modinfo by full path, but
include/tests_kernel did not.
* Solaris find does not support -maxdepth.
This mirrors the logic already in tests_homedirs.
* Use PSBINARY instead of ps.
* Work around Solaris' date not supporting +%s.
Printing nawk's srand value is a bizarre but apparently once popular
workaround for there being no normal userland command to print
UNIX epoch seconds. A perl one-liner is the other common approach,
but nawk may be more reliably present on Solaris than perl.
* Revert to using sha1 for HOSTID.
* Whitespace cleanup for openssl hash tests.
* Description fix: SafePerms works on files not dirs.
All uses of SafePerms are on files (and indeed, it would reject
directories which would have +x set).
* Lots of whitespace cleanups.
Enforce everywhere(?) the same indentations for if/fi blocks.
The standard for the Lynis codebase is 4 spaces. But sometimes
it's 1, sometimes 3, sometimes 8.
These patches standardize all(?) if blocks but _not_ else's (which
are usually indented 2, but sometimes zero); I was too lazy to
identify those (see below).
This diff is giant, but should not change code behavior at all;
diff -w shows no changes apart from whitespace.
FWIW I identified instances to check by using:
perl -ne 'if ($oldfile ne $ARGV) { $.=1; $oldfile=$ARGV; }; chomp; if ($spaces) { next unless /^( *)([^ ]+)/; $newspaces=length($1); $firsttok = $2; next unless defined($firsttok); $offset = ($firsttok eq "elif" ? 0 : 4); if ($newspaces != $spaces + $offset) { print "$ARGV:$ifline\n$ARGV:$.:$_\n\n" }; $ifline=""; $spaces=""; } if (/^( *)if (?!.*[; ]fi)/) { $ifline = "$.:$_"; $spaces = length($1); }' $(find . -type f -print0 | xargs -0 file | egrep shell | cut -d: -f1)
Which produced output like:
./extras/build-lynis.sh:217: if [ ${VERSION_IN_SPECFILE} = "" -o ! "${VERSION_IN_SPECFILE}" = "${LYNIS_VERSION}" ]; then
./extras/build-lynis.sh:218: echo "[X] Version in specfile is outdated"
./plugins/plugin_pam_phase1:69: if [ -d ${PAM_DIRECTORY} ]; then
./plugins/plugin_pam_phase1:70: LogText "Result: /etc/pam.d exists"
...There's probably formal shellscript-beautification tools that
I'm oblivious about.
* More whitespace standardization.
* Fix a syntax error.
This looks like an if [ foo -o bar ]; was converted to if .. elif,
but incompletely.
* Add whitespace before closing ].
Without it, the shell thinks the ] is part of the last string, and
emits warnings like:
.../lynis/include/tests_authentication: line 1028: [: missing `]'
* Typo fix.
* Style change: always use $(), never ``.
The Lynis code already mostly used $(), but backticks were sprinkled
around. Converted all of them.
* Lots of minor spelling/typo fixes.
FWIW these were found with:
find . -type f -print0 | xargs -0 cat | aspell list | sort -u | egrep '^[a-z]+$' | less
And then reviewing the list to pick out things that looked like
misspelled words as opposed to variables, etc., and then manual
inspection of context to determine the intention.
* Update facter location for BSDs
BSDs tend to place third party binaries in /usr/local rather than /usr
* Add support for DragonFly boot loader detection
DragonFly BSD has the same file paths for the bootloader as FreeBSD
* Add kernel module checking for DragonFly
DragonFly BSD checks kernel modules the same way as FreeBSD
* Add DragonFly check for login shells
DragonFly's login files are the same as FreeBSD's
* Add HAMMER PFS Detection
All PFS mounts in HAMMER systems for DragonFly will be detected now
* If grsec installed, build FINDKERNEL from uname -r
When running a grsecurity-patched custom kernel, the /vmlinuz link is often missing. If this link is missing, and grsecurity is installed, then we can calculate the location of FINDKERNEL with the words "linux-image-", plus the output of "uname -r".
* Suggest manually checking kernel if grsec installed
We can't rely on the apt-cache output when running grsecurity. This is because apt-cache can't tell us if we're running an up-to-date kernel, when it's a custom kernel with grsecurity. Instead of confirming that the kernel is OK, we instead should remind the auditor to double-check themselves.
