Improve maintainability and robustness of the TCP server by clarifying
locking/ownership, tightening invariants, and simplifying queueing.
Also fix a long-standing pragma macro typo across the tree.
Impact: Internal behavior only. EPOLL re-arm now occurs while holding
pSess->mut; starvation cap counts only successful reads.
Before/After:
Before: EPOLL re-arm happened after leaving the critical section; read
starvation cap counted loop iterations; closeSess() sometimes unlocked;
select_* helpers used on non-epoll path; enqueueWork() returned status.
After: EPOLLONESHOT is re-armed before unlocking; starvation cap counts
only RS_RET_OK reads; closeSess() never unlocks; poll_* helpers replace
select_*; enqueueWork() is void (best-effort).
Technical details:
- Replace notifyReArm() with rearmIoEvent() (EPOLL_CTL_MOD with
EPOLLONESHOT|EPOLLET; asserts efd/sock; logs on failure).
- doReceive(): explicit state machine; would-block path re-arms before
unlock; close path unlocks then calls closeSess(); starvation handoff
enqueues without re-arming.
- Initialize ioDirection for listener and session descriptors; add
assert(sock >= 0) and widespread ATTR_NONNULL annotations.
- startWrkrPool(): single finalize rollback (cancel/join partial
threads; destroy cond/mutex); stopWrkrPool(): destroy cond/mutex.
- enqueueWork(): FIFO append under lock and cond signal; returns void.
- Cleanup hardening on construct failure: free ppLstn, ppLstnPort,
ppioDescrPtr; free fromHostIP on SessAccept() error.
- Non-epoll: rename select_Add/Poll/IsReady -> poll_*; RunPoll() uses
poll_* and sets sane ioDirection defaults.
- Typo fix: standardize PRAGMA_IGNORE_Wswitch_enum in header and all
users (action.c, rainerscript.c, template.c, tcpsrv.c).
This implements brace-style env vars inside backticks with `echo` and
fixes termination of unbraced `$VAR`. Most importantly, variables can
now be adjacent to static text (e.g., `foo${ENV}bar`), which previously
did not work and caused confusion. This aligns behavior with common
bash expectations.
Why
- Users expect `${VAR}` and `$VAR!` to expand while keeping punctuation.
- Concatenations like `foo${ENV}bar` are common and should be valid.
What
- Add support for braced variables `${VAR}` with proper `}` handling.
- For unbraced `$VAR`, stop the name at the first non `[A-Za-z0-9_]`
char and emit that char (e.g., `!`, `.`) as literal output.
- Improve error handling for overlong env var names.
- Keep other shell features (e.g., `$(pwd)`) unsupported by design.
Docs
- Update `constant_strings.rst` to document `${VAR}`, the new
termination rules, and examples including `foo${ENV}bar` and `$VAR!`.
Tests
- Add `rscript_backticks_braces_envvar.sh` for `${VAR}` support.
- Add `rscript_backticks_static_text.sh` for `$VAR!` and adjacency.
- Convert `rscript_backticks_empty_envvar-vg.sh` to non-VG variant and
adjust `tests/Makefile.am`.
Compatibility
- Backward compatible. Changes affect only previously broken edge cases.
Misc
- Refresh lexer copyright years.
This commit applies the new canonical formatting style using `clang-format` with custom settings (notably 4-space indentation), as part of our shift toward automated formatting normalization.
⚠️ No functional changes are included — only whitespace and layout modifications as produced by `clang-format`.
This change is part of the formatting modernization strategy discussed in:
https://github.com/rsyslog/rsyslog/issues/5747
Key context:
- Formatting is now treated as a disposable view, normalized via tooling.
- The `.clang-format` file defines the canonical style.
- A fixup script (`devtools/format-code.sh`) handles remaining edge cases.
- Formatting commits are added to `.git-blame-ignore-revs` to reduce noise.
- Developers remain free to format code however they prefer locally.
Undefine common syslog severity and facility macros before
redefining them in rsyslog.h. As rsyslog, we need consistent
definitions. This resolves "redefined" warnings when other
system headers are included. Improves build cleanliness.
This Implements suffix comparison similar to startswith. Note that we
do intentionally not use libestr functions in order to speed up adaption.
It would otherwise probably take years for distros to upgrade libestr.
With some help of the Codex and Gemini AI Agents.
closes https://github.com/rsyslog/rsyslog/issues/1255
among others, remove some warning suppressions by "fixing" the
respective constructs with work-arounds (root cause is compilers
do not handle enums in switch well).
We have these expressions in rsyslogd.conf.
set $!rsyslog_FileFormat = exec_template("RSYSLOG_FileFormat")
set $!localheader = re_extract($!rsyslog_FileFormat, "[^ ]+.* +port[0-9]", 0, 0, "");
set $!localpattern = re_extract($!rsyslog_FileFormat, " [^ ]+ +[^ ]+ +port[0-9]", 0, 0, "");
set $!localheader = replace($!localheader, $!localpattern, " ");
We have a message like this arriving.
<30>Feb 24 22:08:21 hostname port03 'label' RXDATA: \n
It was observed that when 2 of these messages arrive in a row, rsyslogd
crashes. This is clearly due to memory corruption, as the crash comes
from within calloc.
Unlike the crash, valgrind only complained about the first message. It
reported that the 'find' variable was being accessed in the replace
function, reading past allocated data.
The localpattern variable ends up "empty" (null?), because the pattern
fails to match. This ends up passed into the replace function as an
es_str_t with a length and buffer length of 0. There is no string data,
not even a null terminator.
As a result, the 'find' pointer is invalid, and accessing it is an
error. Protect against accessing the 'find' pointer when the buffer is
empty by exiting the two loops when j == lfind and lfind == 0.
This removes the report from valgrind, and stops rsyslogd from crashing.
When using the backtick feature with cat and a file that does not exist,
an error message is placed where the file content would be. This error
message contained a typo in could as well as a missing space between
see and error, which have now been fixed.
Now, length can have a negative value -n to denote that the
substring should be build between startpos and the character
-n chars from the end. This is a shortcut for stripping charactes
on "both ends" of the string.
Also, some hardening against invalid startpos and length has
been added.
In rscript, comparison operations on strings did not work correctly
and returned false results. This is cause by a regression in commit
5cec5dd634e0. While it fixed number comparisons, it introduced new
problems in string comparisons, which were not present before. Note
that most items in rsyslog are strings, so this can actually cause
some problems.
This problem occurred when numbers were used in rsyslog.conf in
the set statement, e.g.
set $nbr = 1234;
In this case, during comparisons, the number was actually interpreted
as a string with digits. Thus numerical comparisons lead to unexpected
results. Even more so, as in other places of the code they were
treated as native numbers.
This is now fixed. We cannot outrule that this causes, in border cases,
change of behavior to existing configs. But it is unlikely and the
previous behaviour was a clear bug and very unintuitive. This in our
opinion it is justified to risk a breaking change for an expected
very minor subset of installations, if any such exists at all.
The fix was combined with code refactoring. We did this, because
the fix itself would have been quite hard to read, and the need
for refactoring became obvious.
closes https://github.com/rsyslog/rsyslog/issues/4770
Brief overview:
TO configure tracking percentile metrics in rainerscript:
User would need to define:
- which percentile to track, such as [p50, p99, etc.]
- window size - note, this correlates directly with memory usage to
track the percentiles.
To track a value, user would call built-in function `percentile_observe()` in their configurations to
record an integer value, and percentile metrics would be emitted every
impstats interval.
Provides ability to evaluate a rsyslog variable using dynamically
evaluated parameters.
1st param is the rsyslog param, 2nd param is a key, can be an array
index or key string.
Useful for accessing json sub-objects, where a key
needs to be evaluated at runtime. Can be used to access arrays as well.
see tests for examples
remove unnecessary escape char
This commit adds a check that include files are processed in the
proper order.
It also slightly changes some text that seemed to cause the wrong
impression that include files were processed in the wrong order.
Right the contrary is the case, as config files are being put on
a stack before they are processed.
closes https://github.com/rsyslog/rsyslog/issues/4271
A data race could happen when a lookup table was reloaded. We found
this while moving to newer version of TSAN, but have no matching
report from practice. However, there is a potential for this to cause
a segfault under "bad circumstances".
This implements a way to check if rsyslog variables (e.g. '$!path!var') is
currently set of not.
Sample: if exists($!somevar) then ...
closes https://github.com/rsyslog/rsyslog/issues/4385
When there are actions configured after a STOP, a warning should be
emitted. In fact, an error message is generated. This prevents the
construct, which may have some legit uses in exotic settings. It
may also break older configs, but as the message is an error
for so long now, this should be no longer of concern.
There was wrong negation in the method so it returned 0/1 in reverse
and also it did not mark the node to not be reported as unknown at all
times which is needed after all.
when an `cat <filename>` consruct is used in rsyslog.conf ang <filename>
can not be accessed (does not exist, no permissions, ...), rsyslog
segfaults.
Thanks to Michael Skeffington for notifying us and providing analysis
of root cause.
closes https://github.com/rsyslog/rsyslog/issues/4290
When the 'config.enabled="on"' config parameter an invalid error message
was emitted that this parameter is not supported. However, it was still
applied properly. This commit removes the invalid error message.
closes https://github.com/rsyslog/rsyslog/issues/4011
Using config.enabled="off" could lead to error messages on
"parameter xxx not known", which were invalid. They occured
because the config handler expected them to be used, which
was not the case due to being disabled.
This commit fixes that issue.
closes https://github.com/rsyslog/rsyslog/issues/2520
Introduces the capability to create an output config file that explodes
all "includes" into a single file. This provides a much better overview
of how exactly the configuration is crafted. That could often be a great
troubleshooting aid.
This commit also contains some slight not-really-related cleanup.
closes https://github.com/rsyslog/rsyslog/issues/3634
The bug was introduced in commit abe0434 (config: enhance backticks "echo"
capability). The getenv() result passed to strlen() and es_addBuf() may be
NULL if the environment variable does not exist, resulting in a segfault.
This commit fixes Github issue #3006.
