* Add a SysdiagnoseInfo module to check-sysdiagnose
check-sysdiagnose had no module of its own: it prepared the archive for
plugin modules and refused to run without one. SysdiagnoseInfo is the
first built-in module. It writes sysdiagnose_info.json with details
about the device and the archive: product type and model, iOS version
and build, serial number, IMEI, MEID and UDID from remotectl_dumpstate.txt
and the mobile activation request, the Apple account name and email from
the App Store daemon database, and the archive's original file name and
creation time from sysdiagnose.log. The build is checked against the
known iOS versions the way BackupInfo does.
The App Store database is copied out of the archive together with its
-wal and -shm sidecars before it is opened, so rows still in the
write-ahead log are read.
With a built-in module the command's list is never empty, so the "no
custom modules" error and its test go. The module joins
IOS_CHECK_IOCS_MODULES like every other module that writes a results
file.
* Note that newer sysdiagnoses lack the App Store daemon database
* Keep refusing check-sysdiagnose runs without a custom module
* Warn instead of refusing when no forensic sysdiagnose module is loaded
MVTSettings.initialise() constructs the settings once with load_env=False
and writes the result to config.yaml, so that values taken from MVT_*
environment variables are never persisted, then constructs them again
with the environment applied.
Since #716 settings_customise_sources() has added env_settings
unconditionally, making load_env dead code. Any MVT_* variable set for a
single run, including MVT_IOS_BACKUP_PASSWORD, MVT_ANDROID_BACKUP_PASSWORD
and MVT_VT_API_KEY, was written in plaintext to config.yaml and read back
on every later run.
Only add env_settings when load_env is true, and add a regression test.
Fixes#915
dumpsys prints a blank line after every namespace block and after a
change history, and the generation registry and any vendor dumps follow
the last block before the section trailer. A record ran until the next
`_id:` line, heading or trailer, so the last row of the section
absorbed those dumps into whichever field came last.
A blank line now closes the record being read; lines that follow it
without an `_id:` are skipped until the next row or heading. The fixture
carries a generation registry after its last block, modelled on the
AOSP dump, and the last row is pinned to exactly its own fields.
A dumpsys settings row can end in `tag:` and, on some vendor builds, in
`isValuePreservedInRestore:` or a bare `notPreservedInRestore` token.
The parser only peeled `default:` and `defaultSystemSet:` off the end
of a record, so on a row without a default those tokens stayed inside
the value, and on a row with one they landed in `defaultSystemSet`. A
value of `0 tag:null` is not the safe value `0`, so a setting at its
safe value was reported as dangerous; these are the false positives
measured in #912.
The metadata keys are ordered fields like the rest of the record, so
`_split_fields` reads them now, which also replaces the separate
handling of the default. The bare token has no `key:` shape and is
printed last, so it is stripped first and recorded as
`isValuePreservedInRestore: false`.
The bugreport fixture gains the three row shapes from #912: a restore
flag after `defaultSystemSet:`, and a `tag:` or a bare token directly
after the value. Two of them sit on dangerous settings at their safe
value, so the unchanged alert count of one is the false-positive check.
The coverage comment never posted: the per-file table with a link on
every file and missing line range exceeds GitHub's 65536-character
comment limit, so PRs got only a badge. Every matrix job also raced to
post the same comment, and the action was unpinned at @main.
Post from one job only, limit the table to files changed in the PR,
drop per-line links, and pin the action. Write the full coverage table
to the job summary as well, which also works for PRs from forks where
the token is read-only.
Set UV_PYTHON from the matrix. Without it, .python-version pins 3.10
and `uv run` rebuilt the venv with 3.10 after `uv sync --python X`, so
all five matrix jobs were testing on Python 3.10.
The `dumpsys settings` parser matched a single regex per line, which
truncated every value that spans more than one line and, when
`defaultSystemSet:` did not fall on the first line, left the trailing
`default:` metadata inside the value. It also keyed results by setting
name within a namespace, so a name recorded twice kept only the last row
and a row without a `pkg:` field was dropped entirely.
Replace it with a line loop that accumulates one record at a time and
splits the `key:value` fields once the whole record has been read.
Results become a list of records carrying the fields dumpsys prints:
namespace, user, _id, name, value, pkg, default and defaultSystemSet,
plus the per-setting change history. History timestamps are printed
without a year, so they are resolved against the "ending at:" time of
the section and serialized into the timeline. This shows which package
changed a security-relevant setting, and when.
The androidqf settings module shares this artifact, so it now emits the
same record shape.
The package version now comes from the latest v* git tag, so a release
is a tag and nothing else. A scheduled workflow tags main on the first
of the month when something was merged since the last release, creates
the GitHub release with generated notes, and publishes to PyPI and the
container registry. Pushing a v* tag by hand goes through the same path.
Each console script imports its CLI module before Click can answer a
shell completion request, and the completion scripts run the program on
every keystroke. Importing mvt.ios.cli or mvt.android.cli took ~230 ms,
of which building the command tree needed almost nothing: cli_plugins
imported one constant from module_loader, which pulled in MVTModule, the
indicators, the pydantic settings, requests and rich; the command
implementations pulled in the same, and the iOS CLI imported iOSbackup
(pycryptodome) for decrypt-backup.
The two platform CLI modules now only build the command tree: each
command imports what it runs when it is invoked. cli_plugins owns the
custom command prefix instead of importing it from module_loader, and
exec_or_profile() loads the settings when it runs.
Completion of mvt-ios and mvt-android drops from ~260 ms to ~85 ms per
keystroke on a clean install, and every command starts that much
sooner. A test fails as soon as a CLI module imports the module
machinery again.
A module is part of check-iocs for a platform when it declares the
check-iocs pair, as before. It is now also part of check-iocs when it
overrides check_indicators() and supports at least one command of that
platform.
The rule lives in module_supports_command(). CmdCheckIOCS already uses
that function to pick its modules, so --list-modules and --module follow
it too. Built-in modules are unaffected. check-iocs takes them from
IOS_CHECK_IOCS_MODULES and ANDROID_CHECK_IOCS_MODULES.
A custom module which replaces a built-in module is covered by the same
rule. If it subclasses the module it replaces, it inherits its
check_indicators() and takes over the re-check of the results file. The
"Replacing a built-in module" section of the plugin documentation is
updated to say that.
A custom module which extends a built-in one ran alongside it, and both
wrote to the same results file when they shared a slug, with the run
order deciding the surviving content.
Custom modules can now name the module class they supersede in a
`replaces` attribute. When both are available to a command, the named
module is dropped from the run and the substitution is logged with the
origin of the replacement, so it is recorded in command.log. Only the
replacements which are applied are reported: a declaration from a
disabled module is ignored, so that replacing a module cannot silently
disable it, and modules which replace each other in a cycle all keep
running and replace nothing, with a warning naming every one of them.
Dependencies are remapped along with the modules themselves: a module
depending on a replaced class is ordered against, and receives the
results of, the module which took its place. Replacing a module which
others depend on therefore does not make that dependency unavailable.
The remapping applies wherever dependencies are read, so a replacement
which cannot run is skipped like any other module with an unavailable
dependency, and takes the modules depending on the module it replaced
with it. Those warnings name the dependency the author declared as well
as the module which replaces it.
A replacement does not have to keep the class name of the module it
replaces, so `--module` now falls back to the name of a replaced module
and runs its replacement. A name which matches no module at all stops
the run with a warning instead of silently analyzing nothing, as does a
selection left with nothing to run once skipped modules are dropped.
Sharing a slug outside a replacement stays possible and is now reported.
Two modules writing to the same results file is a forensic-integrity
problem rather than an error, so both still run and a warning names them,
where each came from, and the file the later one overwrites. Taking over
the slug of a replaced module is not reported, because that module is no
longer part of the run.
* Add the mvt.plugin import surface
mvt.plugin re-exports the names a plugin needs from MVT under one import
path. It holds the module base classes and Command, the alert and result
types, the database errors a module raises, the timestamp converters, the
plugin settings API, MVT's settings, get_plugin_logger() and MVT_VERSION.
The names it exports are kept working on a best-effort basis. Changes to
them are announced in the release notes. Anything else in mvt can still be
imported, and may change between releases without notice.
get_plugin_logger(__name__) returns a logger under mvt.ext for plugin code
outside a module class. Its records then reach the console and the
command.log file of a run. A file loaded with --load-module or
--load-command is named after the file.
* Document how to write MVT plugins
The custom modules page now leads with plugin packages. Loading module
files with --load-module and MVT_CUSTOM_MODULES moves to a section on
developing a module locally.
A new "Writing a module" section shows a module which subclasses
IOSExtraction. It lists each base class, the command pair it serves and the
helpers it provides. "Depending on a built-in module" says to import a
built-in class from its family package.
"Importing from MVT" says what mvt.plugin exports and what importing from
it means. The custom commands page shows a Command subclass which lists its
own modules. The sysdiagnose and plugin configuration pages import from
mvt.plugin.
* Share the check-iocs module lists between the CLI and the code
check-iocs re-checks the results a previous run stored, so its module
list is every module of the platform that could have written one. Each
platform's CLI composed that list inline, concatenating the families by
hand, so the list existed only inside the click callback: anything else
needing to know what check-iocs runs had to build its own copy, and the
two could drift apart without a test noticing.
Give each platform a command_modules.py holding the one list, and have
its CLI assign it. The modules check-iocs runs are unchanged, and a
test pins each list to the families it is composed of.
* Pin that check-iocs re-checks the results of custom modules
check-iocs takes its custom modules from load_custom_modules() like every
check-* command and matches result files to modules by slug, so a plugin
module's stored results are re-checked whenever it declares the
check-iocs pair of its platform; nothing asserted it.
* Find the console log handler by type when changing verbosity
set_verbose_logging() adjusted the first handler on the "mvt" logger,
whichever handler that happened to be. Anything else attaching a
handler to that logger - an embedding application, a plugin, a test
harness - had it mistaken for the console and raised or lowered behind
its back, and the same slot can hold the file handler a command
attaches to its output folder, whose level a --verbose flag should
never decide: command.log records the whole run either way.
Walk the handlers instead and adjust only MVT's own console handler,
found by its MVTLogHandler type. Every other handler on the logger is
left alone. Behaviour is otherwise unchanged.
* Add a --verbose option to the mvt, mvt-ios and mvt-android commands
Verbosity was an option of each module-running command, so
"mvt-ios --verbose check-backup" was a usage error, a plugin command had to
define a flag of its own, and there was no way to get debug output from mvt at
all.
The option now sits on the three commands themselves and sets the level of
MVT's console handler for the run, through set_verbose_logging(), before any
command runs. Plugin commands registered on any of the three CLIs get the
option for free and need none of their own.
The per-command --verbose of the check-* commands is kept for backward
compatibility. It only ever raises the level, so the CLI's choice is never
undone by a command's default, and its help text says it is kept for
compatibility. It is to be removed in a later release.
* Add plugin update checking
Report available updates to the installed MVT plugin packages in the
startup banner, for plugins installed from a package index and for
plugins installed directly from a repository. Repository installs pinned
to a commit or a tag are never reported as outdated.
MVT only prints the command which upgrades a plugin. Installing the
update stays a deliberate choice of the analyst. The check runs at most
once every twelve hours, and in between prints the findings of the
latest check which still apply to what is installed.
Nothing about the check can interrupt a running command: the parts of
the suggested command come from package metadata and are quoted for the
shell, the repository query refuses to prompt for credentials and never
passes metadata as a git option, and a corrupt or stale cache is
discarded rather than trusted.
* Add a plugins command to list installed plugins and check updates
Add a "plugins" command to the platform-neutral mvt command. "plugins
list" shows every installed plugin package with its version, where it was
installed from, how many forensic modules it contributes and which
commands it adds. "plugins check-updates" checks for updates immediately,
without waiting for the automatic check, and prints the command which
upgrades a plugin instead of installing anything.
It lives on mvt only. The packages it lists extend mvt-ios and mvt-android
too, but auditing them is not the job of a command which analyses one
platform, and the two platform CLIs should not carry commands which are
not about an acquisition.
The command is registered as a built-in, before any external command, so
that an installed package cannot replace this audit surface.
* Add a platform-neutral mvt command
Several MVT commands have nothing to do with the acquisition of one
platform, yet they were reachable only through mvt-ios and mvt-android.
Asking which version is installed or downloading the public indicators
meant picking one of the two platform commands arbitrarily, and each of
those tasks had to be written, documented and maintained twice.
Add a third console script, mvt, hosting the commands which belong to no
platform: version and download-iocs for now, with completion following
in a later commit.
Commands installed in the new mvt.cli_plugins entry-point group are
registered on mvt, and on mvt only, so that a command package chooses
the CLI each of its commands is added to: mvt.ios.cli_plugins for
mvt-ios, mvt.android.cli_plugins for mvt-android and mvt.cli_plugins for
mvt. A command wanted on both platform CLIs is registered in both
platform groups; no group adds a command to every CLI. The
MVT_CUSTOM_COMMANDS variable loads command files and folders into mvt
the way the platform variables already do for mvt-ios and mvt-android.
Run on its own, mvt prints the banner and its help instead of a usage
error. The help text reminds that the forensic analysis of an
acquisition runs through mvt-ios and mvt-android, so that the command
which knows nothing about acquisitions says where they are analysed.
version and download-iocs stay on mvt-ios and mvt-android for now, so
that no documented invocation stops working. They are to be dropped from
the platform CLIs in a later release, once mvt has been available long
enough for the change to be announced.
Unlike mvt-ios and mvt-android, which point at their subpackages, the
console script points at mvt.cli:main and the mvt package re-exports
nothing of it. Importing mvt has to stay cheap and free of side effects:
it is the package plugins import from, and pulling in Click, the CLI and
everything the commands import merely because something imported mvt
would work against that.
While here, give the version command of both platform CLIs the context
settings every other command already has, so that "mvt-ios version -h"
prints its help instead of failing on an unknown option.
* Reduce indent for MVT CLI header
* Move shell completion to the mvt command and generate one script for every MVT command
Setting up shell completion had nothing to do with the acquisition of one
platform, yet it was a command of mvt-ios and mvt-android, each
generating the script of the program it ran under only. Completion now
leaves the platform CLIs for mvt, which emits one script covering mvt,
mvt-ios and mvt-android, installed as a single file loaded when the shell
starts. Click names the completion function of each program after the
program, so the three scripts concatenate without colliding, and fish
takes the file in conf.d rather than one named after a single command.
With one script there is nothing left for the platform commands to
generate, so their completion command goes, and with it the banner
suppression which was keyed on the command name: mvt-ios and mvt-android
now print the banner for every command they have.
The long help line says which commands the completion covers, so a
short_help keeps "mvt --help" from truncating it.
* Point the README and the command docs at plugin packages
Loading a command by path is for local development; a package is how a
command is distributed. The README's summary of what extends MVT now
names plugin packages and nothing else, and the section on loading a
command file says what it is for: keeping a command being written
loadable without reinstalling its package after every change.
register_cli_plugins() ran while mvt.ios and mvt.android were being
imported, so importing any part of MVT executed the entry points of every
installed command package. That made plugin loading depend on import
order: a plugin importing from MVT while MVT was still initializing got an
ImportError and was quietly demoted to a broken command, and the same
plugin worked when MVT happened to be imported first.
Move the call into a main() function in each CLI module and point the
console scripts at it, so registration happens once when the program
starts and importing MVT no longer runs third-party code. For packagers:
mvt.ios:cli and mvt.android:cli stay importable, but a wrapper invoking
cli() directly no longer registers the installed plugin commands and
should call main() instead.
A module declaring a dependency its command does not provide made
_ordered_modules() give up on the whole run, so a single wrong declaration
in a module scoped to several commands turned a forensic analysis into
zero executed modules with one warning to explain it.
Drop only the modules that cannot run: the one with the unavailable
dependency, and anything depending on it. Each gets its own warning naming
the module missing a dependency and the dependency it is missing, and the
remaining modules run in the same stable topological order as before. A
cycle in the dependency graph is still a programming error and still stops
the run.
* Add namespaced plugin configuration support
Plugin packages need somewhere to keep their own settings, but MVT
rewrites its config.yaml with only the fields it knows about, so any
foreign section is dropped. Add MVTPluginSettings, a pydantic-settings
base class that gives each plugin its own file under the MVT config
folder and its own MVT_PLUGIN_<NAME>_ environment variable namespace.
Settings resolve from constructor arguments, then the environment, then
the plugin file, then the field defaults. Saving skips the values the
environment currently supplies, so credentials passed as environment
variables are not copied to disk, and writes through a private temporary
file so a settings file is never partially written or briefly readable
by other users.
* Add per-plugin data folders
Plugins had no sanctioned place to keep the data they persist, so the
plugin configuration documentation suggested a CACHE_FOLDER setting
defaulting to ~/.cache/example-plugin. That is a Linux convention which
is wrong on macOS, nothing expands or creates it, and it turns a path
into a setting a user can be asked to configure.
Add plugin_data_folder(), which returns the folder a plugin should use
for caches, downloaded artifacts, synchronization state or anything
else it writes to disk, and creates it if it is missing. The plugin
name is validated before anything is created, so a name holding a path
separator raises an error and leaves no folder behind, and calling the
function again returns the same folder with its contents untouched.
The folder sits under plugin-data rather than under the plugins folder
which holds the settings files. On macOS the configuration folder and
the data folder are the same directory, so reusing the plugins name
would leave each plugin's data folder in among the settings files.
Both the plugin-data folder and the folder of each plugin are created
with 0700 permissions. MVT is a forensic tool, and what a plugin keeps
there, such as API responses or sample metadata, is private by default.
The path is resolved on every call, as the configuration folder already
is, so it follows the current environment rather than whatever it was
when MVT was imported.
The documentation now points plugins at the helper, and the example
settings class carries a plain integer setting in place of its cache
folder.
* Derive the data folder of a plugin from its settings class
A plugin with a settings class already names itself in `plugin_name`;
passing the name again to plugin_data_folder() repeats it and can drift.
Add a `data_folder()` class method on MVTPluginSettings which returns
plugin_data_folder() for the class's validated plugin name (works on the
class and on an instance); plugin_data_folder(name) stays as the function
underneath for plugins without a settings class.
---------
Co-authored-by: Donncha Ó Cearbhaill <google@donncha.is>
The Development section holds more than one page and will grow further,
so give it its own directory. development.md becomes development/index.md,
which keeps its published URL, and the custom CLI command page moves
alongside it.
Update the nav, the cross-link between the two pages, and the README
link to the custom command documentation, which is now published at
/development/custom_commands/.
Keys are matched as bare prefixes, so `Caused by:` inside an abort message
reaches the `Cause` key, fails the key comparison and raises — and the
per-line loop turns that into an error that drops the entire text
tombstone, stack trace included.
A key mismatch means "this line is not that key", not "this file is
broken": decline the line and let the remaining keys have their turn. A
line with no colon is declined the same way instead of raising on the
unpack. The same trap has a second form in the field, HiSilicon/Huawei
tombstones printing `code around pc:` against the `code` key.
from_ab() raises InvalidAndroidBackup instead of exiting when it runs as a
sub-command, which check-androidqf catches to skip the backup modules. The
two password branches still called sys.exit(1) unconditionally, and since
run_backup_cmd() runs inside finish(), that ended the parent run before the
intrusion-logs command and before the timeline, alerts, urls, info and run
manifest were stored — leaving an output directory that looks complete but
has no alerts.json.
Also drop "as backup.ab is malformed" from the skip warning: it covers a
missing or wrong password too.
Donncha Ó Cearbhaill