Since the state patch introduction, we've been freeing uninitialized
memory in lib/utils_base.c::np_cleanup(), which caused coredumps
with check_snmp when illegal threshold ranges (for example) were
passed, or when we called 'die' without having read any state.
This patch fixes it by replacing the malloc() calls in there (all of
them, since using malloc() is almost always an error) with calloc().
malloc() either doesn't initialize the memory at all, or taints it
with a special marker so it can tell us when we're free()'ing memory
that hasn't been initialized. calloc() explicitly initializes the
allocated memory to nul bytes, which is a zero-cost operation when
we get the memory from the kernel (which alread does that) and almost
always desirable everywhere else.
Signed-off-by: Andreas Ericsson <ae@op5.se>
Various values (dused_pct, dfree_pct, inodes_free, inodes_total, ...) are now carried in
the parameter_list structure. Assignments have been moved to a subroutine preventing
code redundancies.
Group metrics are now calculated based on units rather than blocks. This fixes freespace calculation
when blocksizes differ within a group.
Some commands need the environment to function properly. One such
example is check_ssh and check_by_ssh when a SOCKS proxy is required.
This patch use setenv and extern char **environ to alter and pass the
new environment to the child process Those modules have been added to
Gnulib for portability.
Allocate the appropriate amount of memory for storing the thresholds
data. Before, we allocated the amount of memory required for storing a
_pointer_ to the thresholds data. This crashed (at least) check_mysql
when using its "-S" option on FreeBSD/amd64 (as reported and analyzed by
Nikita Kalabukhov - 2797757).
Signed-off-by: Holger Weiss <holger@zedat.fu-berlin.de>
This patch makes use of standard threshold functions. This allows using
doubles as thresholds.
Since SNMP supports only integers, double precision numbers are only
printed when parsed from a STRING type.
In addition, support for printing properly Timeticks type has been added,
and the code has been thoroughly cleaned.
Gnulib says: "$(LTLIBINTL) when linking with libtool, $(LIBINTL) otherwise"
Everywhere else we use LTLIBINTL so I assume the tests should use it as well.
In 538b97c I reverted a serie of libtap patch by Ton Voon; it included
the removal of the useless basename check.
Relevant portion of the changelog from which it was removed (f31cd40):
Removed basename test as we should use base_name from GNUlib which is
tested separately
Trying to see if this has any relation to the tinderbox failure. I doubt it
and will revert back as soon as it's tested.
This reverts commit a97e5fbda2.
Detects system library as pre-1.4.13 and does not install the built-in library anymore
Tested on:
- Linux x86_64, with and without local libtap (System libtap installed)
- Linux x86, with and without local libtap (System libtap NOT installed)
- Solaris SPARC, with and without local libtap, with and without system libtap installed
By reading the current list instead of the additional one when counting parameters, extra-opts calculation was allocating insuficient space for the final array if later sections contained more arguments than the sum of all previously parsed sections.
git-svn-id: https://nagiosplug.svn.sourceforge.net/svnroot/nagiosplug/nagiosplug/trunk@2069 f882894a-f735-0410-b71e-b25c423dba1c
N::P allows both '#' and ';' for comments. Extra-opts used to allow only '#', it now allows both (';' is the standard for ini files)
Extra-opts does not allow trailing comments anymore (like N::P)
git-svn-id: https://nagiosplug.svn.sourceforge.net/svnroot/nagiosplug/nagiosplug/trunk@2063 f882894a-f735-0410-b71e-b25c423dba1c
