2 Commits
Author SHA1 Message Date
TheWitnessandCopilot 9b217e0e3a INFO: set compat to 1.2.29 (#838)
* INFO: set compat to 1.2.32

* docs: update Copilot instructions Cacti baseline to 1.2.32 (matches INFO compat)

* INFO: back off compat to 1.2.29

* Add manifest.json + upgrade-time file pruning

* Prune phpunit.xml on upgrade and ignore markdown-lint configs

phpunit.xml ships in the repo for CI but is a dev-only test artifact, so it
is removed from manifest.json 'expected', pruned from installs on upgrade
(like tests/), and excluded from the manifest drift check.

The retired markdown-lint configs (.mdlrc, .md_style.rb) are deleted, and
.md* is now ignored like .git*: protected from pruning and excluded from
drift if it reappears.

* Move thold_functions.php and thold_webapi.php into includes/; flatten css

The main threshold library becomes includes/functions.php and the web API
helper becomes includes/webapi.php; every caller (setup.php, includes/*, the
CLI/poller/notify entry points and the test suite) loads them from includes/.
Each per-theme stylesheet css/<theme>/main.css becomes css/<theme>.css so users
can drop their own css/<theme>.css into the css base, and the page_head hook
looks there. Old paths are tombstoned so existing installs drop the stale
copies on upgrade.

* Regenerate cacti.pot for the includes/ file relocations

Moving the plugin's library files into includes/ changed the source-reference
lines in the translation template; regenerate it so the 'Verify translation
template' CI check passes.

* Harden the hook-include test and make RPN integer ops explicit

The HookFunctionsIncludeTest invokes real hooks with placeholder input purely
to execute their relocated include_once lines; it now swallows engine
notices/warnings/deprecations during those calls (and restores the error
handler and config afterward) so it cannot trip the suite's failOnDeprecation
or leak state into other tests. The RPN % and ^ operators now cast their
operands to int explicitly, matching their long-standing integer-truncation
behaviour while silencing the implicit float-to-int conversion deprecation.

* Harden prune against tampered-manifest path traversal

The upgrade-time prune now rejects any tombstone containing '.'/'..' segments
(which could escape the plugin directory or resolve to its root) and treats
ancestors of whitelist entries as protected, so a tombstone on a parent
directory can no longer delete a whitelisted file beneath it.

* docs: update and align the Project Structure block

Point the structure tree at the relocated files (libraries now under
includes/, data files under docs/, stylesheets under css/) and align the
trailing '# ...' comments to a single column so they no longer drift right.

* docs/tests: standardize test header and correct manifest-contract notes

- Use the full license header (from the plugin's own setup.php) in
  tests/Unit/PruneFilesTest.php instead of the abbreviated copyright banner.
- Document that the manifest drift check and the upgrade-time prune also
  handle phpunit.xml and .md* files, matching the implemented behavior.

* refactor: rename prune helpers to the plugin-slug function prefix

The repository naming contract reserves the plugin_<name>_ prefix for plugin
lifecycle / hook-registration functions; all other functions use the plain
<name>_ prefix. Rename the internal upgrade helpers accordingly:

  plugin_<name>_prune_files() -> <name>_prune_files()
  plugin_<name>_rmtree()      -> <name>_rmtree()

The call site, unit tests, and the copilot-instructions.md references are
updated to match. No behavioral change.

* chore: set the Cacti compatibility floor to 1.2.29

Align the compat metadata (and the contributor-guide references) with the
team decision to standardize the minimum supported Cacti version at 1.2.29;
the previously proposed 1.2.32 floor was not adopted.

---------

Co-authored-by: Copilot <copilot@users.noreply.github.com>
2026-10-01 01:40:00 -05:00
aba15fd779 security: replace eval() in RPN binary math with safe dispatch (GHSA-vr4v-qvqm-gm9j) (#807)
* security: replace eval() in RPN math binary ops with safe dispatch (S1523)

Replace the eval()-based binary arithmetic evaluation at line 378
with a new thold_rpn_math_binary() helper function that uses a
switch statement for +, -, *, /, %, ^ operators.

The operator was already whitelisted via in_array(, , true)
and operands validated with is_numeric(), but removing eval() eliminates
the code injection risk entirely (CWE-95).

Addresses SonarQube finding: php:S1523 at thold_functions.php:378

* ci: retry transient dependency downloads

* test(rpn): cover safe binary dispatch

* fix: restore SQL binding, trigger-command quoting, and RPN safety fixes

Same set of pre-existing gaps as develop (covered by tests but never
implemented in thold_functions.php), plus one bug specific to this PR:

- get_allowed_thresholds()/get_allowed_threshold_logs(): bind graph_id and
  caller-supplied params instead of interpolating them into the SQL text,
  and use the *_prepared fetchers.
- thold_replace_threshold_tags(): add a \ flag that quotes
  device/threshold free-text values with cacti_escapeshellarg() when the
  substituted text is a trigger command.
- thold_command_execution(): pass shell=true for the three trigger
  commands, and fix a topic-string typo ('thold' vs 'thold_cmd') that
  silently suppressed exit-status logging for inline command runs.
- thold_expression_math_rpn(): remove a stray break that exited the
  operator switch before pushing the 0/0 result, flag modulo-by-zero
  instead of letting PHP throw DivisionByZeroError, reject non-numeric
  unary operands before eval(), and flag SQRT/LOG results that are NAN or
  INF instead of pushing them onto the stack (unary math here is still
  eval()-based; unrelated to this PR's binary-math dispatcher).
- thold_rpn_math_binary(): this PR's new dispatcher mapped '^' to pow(),
  silently changing it from the pre-existing bitwise XOR behaviour
  (eval('... ^ ...') always computed XOR). Restored XOR to avoid breaking
  existing user thresholds, and corrected this PR's own
  TholdRpnBinaryMathTest 'power' dataset, which had asserted the
  incorrect pow() behaviour.

* fix(tests): stop includes/arrays.php loading from silently no-op'ing

thold_functions.php includes includes/arrays.php with a plain include()
(not include_once()) from several of its own functions (e.g. thold_log()),
keyed off \['base_path']. Once any test in the shared Pest process
exercised one of those call sites, PHP's include-once registry considered
the file already included: a later thold_test_load() (require_once) on the
same resolved path silently no-op'd and never (re)published \
to \, depending on which order the test files happened to run in.

This intermittently broke every test that reads \
(TholdReplaceThresholdTagsTest's THOLDTYPE substitution) as well as tests
several call frames downstream of thold_log()/thold_check_threshold()
(NotificationEmailDeduplicationTest, ThresholdTimeBasedCharacterizationTest),
depending on file execution order.

Add thold_test_load_always()/loadPluginSourceAlways(), which use a plain
require() instead of require_once(), for plugin files that only assign
file-scope variables (no function/class declarations) and are therefore
safe to load more than once. Switch the four test classes that load
includes/arrays.php to the new helper.

* debug: temporary CI diagnostics for notification/restoral failures

* Revert "debug: temporary CI diagnostics for notification/restoral failures"

This reverts commit 21fbd16030.

* debug: temporary CI diagnostics for notification/restoral failures

* test: fix restoral tests for the corrected type-2 restoral operator

Merging develop brought in #796's fix for thold_check_threshold()'s type-2
(time-based) restoral logic (`<` -> `>=` for the warning/alert fail-count
comparisons). This test harness always stubs the log-derived fail count
to 0, and the two restoral tests here relied on the default trigger of 1,
so under the corrected operator (0 >= 1 is false) neither restoral branch
fired anymore and both tests broke.

Override time_fail_trigger/time_warning_fail_trigger to 0 in both tests,
representing a threshold with no minimum streak requirement, so the
harness's stubbed zero count still satisfies the corrected `>=` check and
the tests exercise real restoral behavior again instead of the inverted
bug.

---------

Co-authored-by: Thomas Vincent <thomasvincent@gmail.com>
Co-authored-by: TheWitness <thewitness@cacti.net>
2026-09-21 03:51:56 -07:00