The valid_path_strict() regex previously allowed space characters,
permitting values like '/bin/sh -c id' to pass validation. When passed
to CodeIgniter's popen() sink, the shell interprets the space as an
argument separator, enabling arbitrary command execution.
Remove space from the allowed character set. Legitimate sendmail paths
(/usr/sbin/sendmail, C:\wamp64\bin\sendmail\sendmail.exe) do not
contain spaces.
Update existing tests and add regression cases for shell -c payloads.
---------
Co-authored-by: objecttothis <17935339+objecttothis@users.noreply.github.com>
Adds DatabaseBootstrapTrait and OsposConfigFixtureTrait to replace
duplicated per-file setUp()/tearDown() bootstrap and OSPOS config mock
boilerplate across test classes, bringing them in line with the
ItemsControllerTest gold-standard shape.
Fixes the real root cause of tabular_helperTest's suite-order-dependent
failure: dropAllForeignKeyConstraints()/foreignKeyExists()/
dropColumnIfExists() temporarily blank the shared connection's DBPrefix
via overridePrefix() but never restored it if the query between the two
calls threw (e.g. migration_helperTest's mocked-failure tests) -
permanently corrupting prefixed table lookups for the rest of the
process. Wrapped each in try/finally. Also widens executeScript() to
accept an explicit connection so migration SQL can't silently run
against the wrong DB group.
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
These landed on master by accident via a local branch that was
mistakenly created tracking OpensourcePOS/master directly, rather than
pushed as its own branch. Reverting to restore master to its prior
state; the reverted work will come back through a proper PR instead.
This reverts commit 00bf4cb8af.
This reverts commit 37cc1b15e7.
Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
Adds DatabaseBootstrapTrait and OsposConfigFixtureTrait to replace
duplicated per-file setUp()/tearDown() bootstrap and OSPOS config mock
boilerplate across test classes, bringing them in line with the
ItemsControllerTest gold-standard shape.
Fixes the real root cause of tabular_helperTest's suite-order-dependent
failure: dropAllForeignKeyConstraints()/foreignKeyExists()/
dropColumnIfExists() temporarily blank the shared connection's DBPrefix
via overridePrefix() but never restored it if the query between the two
calls threw (e.g. migration_helperTest's mocked-failure tests) -
permanently corrupting prefixed table lookups for the rest of the
process. Wrapped each in try/finally. Also widens executeScript() to
accept an explicit connection so migration SQL can't silently run
against the wrong DB group.
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
* fix(security): report unwritable .env.lock, make throttle limits configurable
envFileIsWritable() previously checked is_writable(.env) (the file), which passes in Docker even when the mutex file is root-owned by a prior env:provision run. It now checks the real write path: the directory (to create .env.tmp.* + .env.lock) and any existing .env.lock must be writable, so the app throws the clear 'run env:provision' error instead of crashing on 'Unable to open .env.lock'.
The Throttle filter now reads throttle.capacity / throttle.seconds from env (default 5 per 60s); capacity <= 0 disables throttling, so operators serving sequential HTTP clients (e.g. Zabbix) are not caught by the lockout.
* fix(security): address CodeRabbit review findings
- Throttle: validate throttle.capacity as an integer before treating a non-positive value as 'disabled', so a non-numeric value (e.g. 'five') falls back to the default instead of silently bypassing the lockout. Apply the same validation to throttle.seconds.
- envFileIsWritable(): also reject an existing non-writable .env on Windows (where rename() cannot replace a read-only destination); keep the check Windows-only since POSIX rename() replaces a read-only dest when the directory is writable.
- Tests: restore the prior throttle.capacity env state in ThrottleTest (capture/restore instead of delete); add an invalid-capacity fallback case; skip the not-writable fixtures when running as root (where is_writable() is bypassed).
* fix(migration): guard ConvertToCI4 key-write branches with envFileIsWritable()
The migration's 'no key' and 'CI3 key' branches called rotateEncryptionKey()/rotateEncryptionKeyTransaction() directly, bypassing the envFileIsWritable() guard that checkEncryption() uses. On a fresh Docker/Compose install where the web runtime cannot write /app/.env.lock, this produced a raw 'fopen(/app/.env.lock): Permission denied' error instead of the actionable 'run php spark env:provision' message.
A valid CI4 key now short-circuits to checkEncryption() (no write); every write branch is gated on envFileIsWritable() first.
* fix: read provisioned encryption.key so config sees it
env:provision persists the key as 'encryption.key' in .env (matching the
throttle path and .env.example), but Config\Encryption only read the
ENCRYPTION_KEY env var. On a fresh Docker instance the provisioned key was
invisible to config('Encryption')->key, so the app believed no key existed
and tried to write one -- hitting the .env.lock permission wall.
Read encryption.key (via $_SERVER/$_ENV/getenv) first, then fall back to
ENCRYPTION_KEY for Docker '-e' usage. Mirrors checkThrottleEncryption().
* fix: cascade encryption.key lookup past empty-string sources
* refactor: extract Encryption::resolveKey() and test it without global env mutation
* fix(security): decode fallback-selected encryption key like BaseConfig
A key picked in the constructor fallback (notably ENCRYPTION_KEY, which
BaseConfig never inspects) was assigned verbatim, bypassing the
hex2bin:/base64: decode the parent applies to `encryption.key`. Route the
selected key through a parseKey() helper mirroring BaseConfig's
parseEncryptionKey() so prefixed values decrypt consistently, and add a
pure regression test.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* chore: remove advisory ID from comments and tighten verbose comments
Per review: treat GHSA advisory IDs like secrets (drop from code) and
replace the multi-paragraph comments with concise one-liners that keep
the non-obvious "why". No logic changes.
---------
Co-authored-by: objecttothis <17935339+objecttothis@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
fix(validation): allow Windows sendmail paths, tighten shell metachar exclusions
Broaden PLAIN_FILESYSTEM_PATH_STRICT to accept real-world sendmail formats
while blocking command injection characters not needed in valid paths.
- OSPOSRules.php: allow space, colon, backslash for Windows paths
(e.g. C:\wamp64\...) and trailing args (-t -i); still excludes
ampersand, backtick, subshell, redirect, and cmd.exe metacharacters
- OSPOSRulesTest.php: add cases for Windows paths, trailing args, and
injection payloads
- Remove 7 ConfigTest assertions that expected metacharacter rejection;
add acceptance test for sendmail path with trailing args
i18n(lang): expand mailpath_invalid message across all locales
- Fill previously empty mailpath_invalid keys across all locales
- Update existing translations (de-CH, de-DE, es-ES, es-MX, fr, nl-BE,
nl-NL) to reflect newly allowed characters; nl locales corrected from
English loanwords to proper Dutch terms
- Add missing key to ckb/Config.php
docs: remove security advisory IDs from public-facing files
- AGENTS.md: extend no-advisory-ID rule to documentation and URLs
- INSTALL.md: drop GHSA reference and advisory link from Host Header
Injection guidance; rationale and fix instructions remain intact
Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
Signed-off-by: objecttothis <17935339+objecttothis@users.noreply.github.com>
- Introduce `valid_path_strict` rule in `OSPOSRules` to enforce stricter path validation, preventing security issues like injection attempts with newline or special characters.
- Update mail configuration validation in `Config` controller to use the new rule for the `mailpath` field.
- Add unit tests in `OSPOSRulesTest` to cover edge cases for `valid_path_strict`.
Signed-off-by: objecttothis <17935339+objecttothis@users.noreply.github.com>
* fix(auth): validate gcaptcha before password to prevent bypass
Move gcaptcha check before credential validation so a valid captcha
is required prior to any login attempt. Previously, password auth
ran first, allowing timing-based enumeration without captcha.
Signed-off-by: Travis Garrison <travis@chiraqbookstore.com>
* test(auth): add regression tests for gcaptcha validation order in OSPOSRules
Guards fix from 5dea748b0: gcaptcha must be validated before
Employee::login() is attempted to prevent auth bypass.
Change gcaptcha_check visibility to protected to allow testing.
Signed-off-by: Travis Garrison <travis@chiraqbookstore.com>
---------
Signed-off-by: Travis Garrison <travis@chiraqbookstore.com>
Co-authored-by: Travis Garrison <travis@chiraqbookstore.com>
* Add fallback for allowedHostnames environment variable
- In some cases allowedHostnames is set in env but not loaded at the time of check, yet available in other sources. This adds fallback checks.
- Add UnitTest
Signed-off-by: objec <objecttothis@gmail.com>
* Improve the fallback logic for allowedHostnames environment variable
Signed-off-by: objec <objecttothis@gmail.com>
---------
Signed-off-by: objec <objecttothis@gmail.com>
Allow configuring allowed hostnames via ALLOWED_HOSTNAMES environment
variable as an alternative to app.allowedHostnames in .env file. This
is more convenient for Docker/Compose deployments where environment
variables are set directly in compose files.
The ALLOWED_HOSTNAMES variable takes precedence over app.allowedHostnames
if both are set, allowing deployment-specific overrides.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Ollama <ollama@steganos.dev>
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* Fix: Improve allowedHostnames .env configuration and fail-fast in production
Addresses GitHub issue #4480: .env app.allowedHostnames does not work as intended
## Problem
- CodeIgniter 4 cannot override array properties from .env
- Setting app.allowedHostnames.0, app.allowedHostnames.1 did NOT populate the array
- Application always fell back to 'localhost' silently in production
- Host header injection protection was effectively disabled
## Solution
1. Support comma-separated .env values: app.allowedHostnames = 'domain1.com,domain2.com'
2. Fail explicitly in production if not configured (throws RuntimeException)
3. Allow localhost fallback in development/testing with ERROR-level logging
4. Update documentation with clear setup instructions
## Changes
- app/Config/App.php: Parse comma-separated .env values, fail in production
- .env.example: Update format documentation
- INSTALL.md: Add prominent security section
- tests/Config/AppTest.php: Comprehensive tests for new behavior
Fixes#4480
Related: GHSA-jchf-7hr6-h4f3
---------
Co-authored-by: Ollama <ollama@steganos.dev>
Security: Prevent Host Header Injection attacks by validating HTTP_HOST
against a whitelist of allowed hostnames before constructing the baseURL.
Changes:
- Add getValidHost() method to validate HTTP_HOST against allowedHostnames
- If allowedHostnames is empty, log warning and fall back to 'localhost'
- If host not in whitelist, log warning and use first allowed hostname
- Update .env.example with allowedHostnames documentation
- Add security configuration section to INSTALL.md
- Add unit tests for host validation
This addresses the security advisory where the application constructed
baseURL from the attacker-controllable HTTP_HOST header, allowing:
- Login form phishing via manipulated form actions
- Cache poisoning via poisoned asset URLs
Fixes GHSA-jchf-7hr6-h4f3
