Let check_linux_raid return a WARNING instead of an OK state during
volume recovery.
See: http://bugs.debian.org/574612
(Fixed by Christoph Martin, forwarded by Jan Wagner.)
| check_linux_raid malfunctions if system has software RAID devices with
| two or more digits. For example, for system having /dev/md10,
| /dev/md11 etc, the plugin returns 'UNKNOWN' in automatic mode (if RAID
| devices are manually specified it works). Also, if system has both
| one-digit, and two-digit RAID devices, the two-digit devices are
| silently ignored in checks, which is even more problematic.
[ http://bugs.debian.org/534604 ]
(Fixed by Matija Nalis, forwarded by Jan Wagner.)
If Perl's "shift" function is called outside of a subroutine and without
any argument, it usually shifts @ARGV. However, if a plugin is executed
via ePN, such a call will shift @_ instead, so we must explicitly
specify @ARGV for this to work as expected.
This fixes Debian bug #545940, see: http://bugs.debian.org/545940
Commit de7191e342 fixed the same issue for
check_disk_smb.
(Reported by Hendrik Jaeger, forwarded by Jan Wagner.)
| check_radius doesn't seem to provide any way to modify the
| NAS-IP-Address attribute that it uses in the packets it sends, but it
| does so for NAS-Identifier.
|
| Instead, it hardcodes the IP address that it gets from the
| rc_own_ipaddress() library call, and that in turn translates into
| calling gethostbyname() on the result of uname(). This call can easily
| fail, and its result can easily be unsuitable - for example when the
| Nagios instance uses its own virtual host, and you don't want the
| original system hostname leaked to the RADIUS servers you monitor with
| this.
|
| Furthermore, this behaviour is inconsistent with RFC 2865, which
| defines the two attributes as analogous and never suggests hardcoding
| the value of either of them in client software.
Therefore, this commit adds the "-N, --nas-ip-address" option which
allows for specifying the value of the NAS-IP-Address attribute.
| I've also noticed that the original code for NAS-IP-Address hardcoding
| is broken in its error handling - it does "return (ERROR_PC)", which
| is meaningless in the context of check_radius.c. That actually seems
| to be copy&waste from radiusclient-0.3.2/src/radexample.c. :) I fixed
| that.
|
| While debugging, I also took the opportunity to decouple the
| nas-identifier rc_avpair_add() instance from the initial three,
| because this is just bad practice to lump a fourth optional attribute
| into the same block with the required attributes, the error handling
| for which is throwing the same daft message "Out of Memory?"...
[ http://bugs.debian.org/482947 ]
(Contributed by Josip Rodin, forwarded by Jan Wagner.)
The check_ldap plugin did not allow for specifying an empty LDAP base.
See: http://bugs.debian.org/479984
(Fixed by Stephane Chazelas, forwarded by Jan Wagner.)
Fix some problems regarding the way check_disk_smb passes command line
arguments to smbclient(1).
| It runs:
|
| $res = qx/$smbclient "\/\/$host\/$share" $pass -W $workgroup \
| -U $user $smbclientoptions -I $address -c ls/;
|
| [...]
|
| The documentation says that if the password is not passed, it
| defaults to "". That is not true above, as $pass expands to
| nothing which leaves no argument at all (instead of an empty
| argument) so is different from providing with an empty password
| or with the -N option.
|
| Also, if the password starts with "-", you're in trouble, that's
| why -U $user%$pass may be prefered.
|
| Also, the doc says that if $user is not provided, then it
| defaults to "guest" but the problem is that if it is provided
| but empty, it is changed to "guest" as well, which prevents us
| from querying hosts that don't do user authentication.
[ http://bugs.debian.org/478942 ]
(Fixed by Stephane Chazelas, forwarded by Jan Wagner.)
| When perl plugin scripts are run with the embedded perl interpreter in
| nagios3, the "shift" perl command doesn't shift @ARGV, but @_ (which
| happens to contain the same thing as @ARGV at the time the script was
| started).
|
| [...]
|
| A fix is to replace all the instances of "shift" with "shift @ARGV".
[ http://bugs.debian.org/478906 ]
(Fixed by Stephane Chazelas, forwarded by Jan Wagner.)
| Our automated build log filter[1] detected a problem that is likely to
| cause your package to segfault on architectures where the size of a
| pointer is greater than the size of an integer, such as ia64 and
| amd64.
|
| This is often due to a missing function prototype definition. For
| more information, see [2].
|
| Function `ldap_init' implicitly converted to pointer at
| check_ldap.c:124
|
| [1] http://people.debian.org/~dannf/check-implicit-pointer-functions
| [2] http://wiki.debian.org/ImplicitPointerConversions
|
| The libldap API has been updated and many functions used by the ldap
| plugin are now deprecated. This package should either update to the
| new API or define LDAP_DEPRECATED to continue using the deprecated
| interfaces.
|
| This patch implements the lazy solution.
[ http://bugs.debian.org/463322 ]
(Contributed by Dann Frazier, forwarded by Jan Wagner.)
| The check_disk_smb plugin purports to support guest users, but it
| doesn't work out, because it doesn't specify the -N command line
| option to smbclient when no password is specified, making smbclient
| stop and ask.
[ http://bugs.debian.org/425129 ]
(Fixed by Josip Rodin, forwarded by Jan Wagner.)
Add the "-a, --address option" which allows for specifying the IP
address of the server to connect to. If this option is used, the IP
address will be handed over to smbclient(1)'s "-I" option. (Contributed
by Sean Finney, forwarded by Jan Wagner.)
Rework last patch, copying the converted portion of the string rather
than re-converting it. This is safer for backwards-compatibility as the
value is never modified.
Bug #1867716 fixed what it meant to fix: broken perfdata strings.
Unfortunately some users relied on half-broken perfdata string where at
least the first token was OK. This patch do a two-way conversion
(string to double then back to string) instead and use the conversion
result for the performance data.
A possible caveat is that the string may change where it normally
shouldn't but the result should be somewhat similar.
Add support for checking the number of usable time sources (i.e., the
number of peers which are classified as so-called "truechimers" by NTP's
intersection algorithm). The new "-m" and "-n" options allow for
specifying the according WARNING and CRITICAL thresholds (and thereby
activating the truechimers check), respectively.
Use sendmail(8) instead of mail(1) in order to be able to set the
"Content-Type" header field on systems where the available mail(1)
command doesn't allow for setting it. This makes the "-H" flag (cf.
commit 71350c5a) unnecessary.
sfsnapshotgit:
- Use fetch/reset instead to pull to avoid merges on forced updates
sfsnapshot-upload:
- Fix link deletion walking the entire home dir
- Allow CLEAN_TIME=0 (no retention)
- Re-add per-branch links when CLEAN_TIME > 0
- Add many comments
Now that we moved our Git repositories to SourceForge, we don't need to
maintain local clones for generating commit notifications anymore, as
SourceForge provides shell access to the repositories. Instead, we now
run git-notify as a post-receive hook on the SourceForge server.
Actually, we use a wrapper which executes git-notify with the desired
options and which makes it easy to add other post-receive hooks in the
future.
The Gitweb URLs for repositories hosted by SourceForge are slightly
different than other Gitweb URLs. The correct URL cannot be specified
via "-u" if we append "/$repos_name.git/?" to that URL as we usually do.
If the new "-S" flag is specified or "notify.sourceforge" is set, we'll
append "/$repos_name;" instead, which makes the "-u" option usable for
SourceForge repositories.
Not all mail(1) implementations support specifying additional header
fields via "-a": with some, this flag is used for attaching files,
others don't provide an "-a" flag at all (this is true for the /bin/mail
utility currently installed on the SourceForge servers, for example).
We now provide the "-H" flag and the "notify.legacyMail" configuration
key for these cases.
Distinguish between annotated tags and lightweight tags. In the former
case, send an annotated "tag notification", in the latter case, send a
"ref change notification" (as we did in both cases before).
If the number of commits included with a single push exceeds the maximum
specified via "-n", a single notification will be generated instead of
individual e-mails. For listing the commits within such a notification,
git-rev-list(1)'s "--pretty" option is used. This yields output which
the git_rev_list() subroutine didn't accept. That's now fixed.
If the new "-T" option is specified or "notify.emitRepository" is set,
the subject of e-mail notifications will be prefixed with [<tag>], where
<tag> is the name of the updated repository.
The SHA1 object name part of Gitweb URLs is now only shortened if the
user requested this by specifying the new "-z" option (or by setting
"notify.shortURLs").
While at it, also shorten the additional URL which references a diff in
e-mail notifications which don't include that diff inline because its
size exceeds the maximum number of bytes specified via "-s".
Note that while the abbreviated SHA1 object names will be unique at push
time, this cannot be guaranteed for the future, so the shortened URLs
might break some day.
Only the author's name and address will now be mentioned in a commit
notification by default. However, if the "-C" option is specified (or
"notify.showCommitter" is set), the committer's name and address will
also be included in the notification if the committer is not the author
of the commit (as we previously did by default).
Making use of a state file in order to prevent duplicate notifications
is now optional. The user must explicitly specify a file path via the
"-t" option or by setting the git-config(1) variable "notify.statefile"
to activate this functionality.
Now that we don't ignore empty commits anymore, there's no need to keep
track of the number of commits actually notified about, as that will
always be equal to the number of commits returned by get_new_commits().
This reverts commit db63fbfa03, as it is
no longer needed and we'd like to keep the diff against the git-notify
version maintained by the Wine people as small as possible. The purpose
of db63fbfa was to suppress notifications on empty merge commits, which
can now be requested directly by specifying git-notify's "-X" option.
(Our change was implemented before the "-X" option was available, even
though the Git history suggests otherwise.)
Conflicts:
tools/git-notify
This reverts commit 5445b9769f. Alexandre
Julliard pointed out that the code in question was used if git-notify
was explicitly called with the SHA1 name of an annotated tag object. At
the moment, the code in question actually _is_ unused due to later
modifications, but it wasn't at the time 5445b976 was committed, and
we'll add further changes so that the code will be used again in the
future.
Conflicts:
tools/git-notify
For shared repositories, the state file used by git-notify should
usually be group writable, so we now set the umask to 0002 by default.
This can be adjusted by setting the "notify.umask" configuration key or
by using the "-U" option on the command line.
