Files
reverse-engineering-anything/docs/adb-device-analysis.md
T
김준호andt 92f332652d fix(adb): separate unresolvable paths and host capacity in adb spawn failures (#6188)
* fix(adb): name unexecutable adb paths in readiness refusals

When REA_ADB_PATH or the PATH-resolved adb is a directory or a file the
host refuses to execute, the version probe fails at spawn with EACCES or
EPERM. That fell through to the generic "did not report a usable version"
refusal, which named neither the path nor the cause.

Classify those spawn errors as binary_not_executable (provider rejection
executable_missing), name the selected path and errno, and give
remediation for the environment, PATH and host-policy cases. The generic
fallback now names the selected path too.

Fixes #5689

* fix(process): admit synchronous spawn errors as captured failures

Node throws spawn errors other than EACCES, EAGAIN, EMFILE, ENFILE and
ENOENT synchronously from execFile. execFileBytes called execFile inside
its Promise executor, so ENOTDIR or EPERM became a rejection without
capture fields and execFileOutputFailure returned undefined. Consumers
that classify by failure code, such as the adb version probe, could not
see the code, which left the adb EPERM branch unreachable.

Catch synchronous spawn errno errors at the capture boundary and attach
the same empty capture an asynchronous spawn failure carries. Argument
validation errors stay unadmitted.

The adb version probe now reports an ENOTDIR selected path, such as a
REA_ADB_PATH whose parent is a regular file, as a missing binary with
the filesystem error code instead of the generic version refusal.

* fix(process): expose spawn failures from the capture owner

execFileBytes knows when a command never started, but
ExecFileOutputFailure did not say so, so consumers reread raw syscall
and code fields and disagreed. Mark both the synchronous throw and an
asynchronous `spawn <path>` error with `spawnFailed: true`, and set
`path` on the synchronous error to match the asynchronous shape.
Argument validation errors stay unadmitted.

Consumers now use the owner field:

- inspectAdbClient never reports a spawn errno as `exit_code`. ENOENT,
  ENOTDIR, ELOOP and ENAMETOOLONG are `binary_missing`; EACCES and EPERM
  stay `binary_not_executable`; any other spawn errno, such as EINVAL,
  is `binary_not_executable` with a path-naming "could not be started"
  message. Each records `filesystem_error` and `path_source`. EACCES
  remediation also names parent directory search permission, and EPERM
  remediation no longer names host policy as the only cause.
- DarwinProcessRunTokenReader tags every not-started compiler as
  `spawn_failed`, not only ENOENT. EACCES, EPERM and ENOTDIR were
  previously `failed`.
- NativeDmgArtifactReader uses one not-started rule for both
  `processNotStarted` (previously `spawn ` only, so synchronous errors
  were missed) and the ENOENT unavailable/io split (previously any
  `spawn` prefix). Its fake hosts now reject with real execFileOutput
  spawn failures.

* fix(process): classify spawn failures consistently across consumers

Admit synchronous and asynchronous spawn errnos through one helper that
requires an errno and a `spawn` or `spawn <path>` syscall. The
synchronous shape now also gains `spawnargs`, so both shapes match.

Use the owner's `spawnFailed` fact in the remaining consumers:

- inspectAdbClient reports exhausted host capacity (EAGAIN, EMFILE,
  ENFILE, ENOMEM) as a host failure to retry and no longer tells the
  caller to change REA_ADB_PATH. Path-resolution errnos come from a
  predicate the filesystem owner exports.
- runAdbCapture says adb could not be started instead of claiming an
  exit code for a process that never ran.
- The Darwin compiler probe words every spawn_failed outcome as a start
  failure, not a compilation failure.
- Asset catalog inspection treats any assetutil that never started as
  unavailable, not only ENOENT.

* fix(process): report host spawn capacity as a retryable constraint

Spawn errnos that report exhausted host capacity (EAGAIN, EMFILE, ENFILE,
ENOMEM) were classified by adb readiness alone. assetutil reported them as
an unavailable tool, which marked inspect_asset_catalog unavailable in the
session, and adb commands reported them as protocol failures.

Classify them once in the process capture owner and report a retryable
capacity constraint with recovery advice from assetutil and adb commands.

* docs(adb): describe host capacity failures from device operations

* fix(adb): report exhausted host spawn capacity as runtime unavailability

A host that cannot start adb for lack of process, memory or descriptor
capacity is not a configuration problem. Give it its own failure code and
map it to the runtime_unavailable provider rejection instead of
not_configured.

* fix(adb): report probe-time host capacity as a retryable constraint

A device operation whose adb version probe could not start for lack of host
capacity was refused as capability_unavailable without retryable. It now
reports the same retryable capacity constraint as a later adb command.

* fix(adb): keep the selected adb path in probe-time capacity errors

The retryable capacity constraint raised at the version probe dropped the
selected_binary and path_source diagnostics the refusal used to carry.

* refactor(adb): keep the spawn-failure fact to adb consumers

Restore the DMG reader, Darwin compiler probe and assetutil to main. The
capture owner still marks a never-started command, including the spawn
errors Node throws synchronously, and only the adb readiness probe and
adb commands read that fact.

* test(adb): assert spawn-failure facts instead of refusal wording

The new adb tests compared complete refusal sentences and banned an old
phrase. They now check the errno, selected path, path source and the
owner's spawn-failure marker, so wording edits do not break them.

* fix(adb): own one host-capacity remediation and cover the spawn route

The host-capacity constraint and the configuration failure each carried their own remediation sentence. AdbConfiguration now owns the text and both use it, so the constraint's remediation.action equals diagnostics.remediation.

The synchronous spawn-throw test now runs on Node's execFile route and on the owned spawn route adb uses (stopSignal). The provider test covers inspect_adb_client under EMFILE next to list_adb_devices, and the docs say which operations report the retryable capacity constraint and which report runtime_unavailable.

* fix(process): report EMFILE and ENFILE spawn failures without stdio

After an EMFILE or ENFILE spawn failure Node creates no stdio, so the
owned stop-signal route threw a TypeError while reading the streams and
the later spawn error went unhandled. Attach the error listener first and
capture absent streams as empty, so adb reports retryable host capacity.

* fix(adb): keep a completed push when host capacity stops its digest check

A capacity failure starting sha256sum came back as a retryable error even
though adb push had already written the file. Repeating the request with
overwrite=false is then refused because the destination exists.

Return the receipt with an unavailable device digest and name the capacity
failure in its limitations, so the caller is told the transfer is unverified
rather than advised to retry.

* docs(adb): list host capacity as an unverified push case

---------

Co-authored-by: t <t@t>
2026-11-05 15:40:32 +08:00

302 lines
23 KiB
Markdown

# ADB device inspection and acquisition
REA inspects connected Android devices, reads device state, and transfers
files and packages through the caller's `adb` binary. Emulators, USB
devices, and network devices participate alike: REA neither starts nor
manages them and never installs platform tools. Every operation composes a
fixed adb argument vector — caller values occupy positional arguments only
and are quoted when adb forwards them to the device shell.
The adb client may start its local server on first device query; that is adb
client behavior, recorded in the tool result when observed.
`inspect_adb_client` probes only the binary and never contacts the server.
## Supply tools
Use any Android platform-tools `adb` (verified with 34.0.5). Select it
through the environment:
```sh
# Optional: absolute path to the adb binary; without it, adb resolves from PATH.
export REA_ADB_PATH=/absolute/path/to/adb
```
REA never installs the Android SDK, an emulator, or a system image.
Readiness is `rea inspect-adb-client`: it runs `adb version` and reports the
resolved binary, its version, and where it was selected from. A missing
binary reports `capability_unavailable` with the `REA_ADB_PATH` remediation
rather than attempting an installation. A path that cannot resolve to a file
(`ENOENT`, `ENOTDIR`, `ELOOP` or `ENAMETOOLONG`) is reported as missing. A
selected path the host refuses to execute, such as a directory or a file
without execute permission (`EACCES`) or one the host does not permit
(`EPERM`), and any other spawn error report `capability_unavailable` naming
that path. Exhausted host capacity (`EAGAIN`, `EMFILE`, `ENFILE` or `ENOMEM`)
is not a configuration failure. `inspect_adb_client`, and any device
operation whose adb cannot start at its version probe or at a later adb
command, report a retryable `capacity` resource constraint rather than a
capability refusal or an adb protocol failure; the `binary_session` inventory
reports the provider as `runtime_unavailable`. Both say the host could not
start adb and ask for a retry after freeing processes or descriptors, without
blaming `REA_ADB_PATH`. When adb never started, the refusal diagnostics record
the spawn error as `filesystem_error` with `path_source`, never as an
`exit_code`.
The `binary_session` inventory uses the same bounded client-version probe; it
does not query devices, start the adb server, or establish that a requested
serial is reachable. Device state is checked by each device operation.
## Answer an analyst question
```sh
# Device identity
rea inspect-adb-client
rea list-adb-devices
rea inspect-adb-device emulator-5554
rea inspect-adb-display emulator-5554
rea list-adb-features emulator-5554
rea list-adb-services emulator-5554
# Find and acquire an app (the easy-get flow)
rea resolve-adb-packages emulator-5554 whatsapp
rea list-adb-packages emulator-5554 --scope third_party
rea inspect-adb-package emulator-5554 com.whatsapp
rea pull-adb-package emulator-5554 com.whatsapp --output-directory /tmp/acq
rea inspect-android-package /tmp/acq/com.whatsapp/base.apk
# Device state and files
rea list-adb-processes emulator-5554
rea inspect-adb-window emulator-5554
rea read-adb-setting emulator-5554 global adb_enabled
rea read-adb-logcat emulator-5554 --count 200
rea list-adb-directory emulator-5554 /data/data/com.whatsapp
rea pull-adb-file emulator-5554 /data/local/tmp/notes.txt --output-directory /tmp/acq
rea push-adb-file emulator-5554 /tools/frida-server /data/local/tmp/frida-server
rea capture-adb-screen emulator-5554 --output-directory /tmp/shots
rea collect-adb-bugreport emulator-5554 --output-directory /tmp/bugs
# Package lifecycle and dynamic analysis (device-mutating)
rea install-adb-package emulator-5554 /tmp/app.apk --replace
rea start-adb-app emulator-5554 com.example.app
rea read-adb-logcat emulator-5554 --pid 12345
rea start-adb-activity emulator-5554 android.intent.action.VIEW --data-uri https://example.com
rea stop-adb-app emulator-5554 com.example.app
rea uninstall-adb-package emulator-5554 com.example.app
```
Equivalent MCP methods use the snake_case names (`list_adb_devices`,
`pull_adb_package`, `resolve_adb_packages`, …). Device operations take
`serial` exactly as `adb devices` reports it. The pulled APK set feeds the
existing analysis families directly, for example
`rea inspect-android-package /tmp/acq/<package>/base.apk`.
## Operations
| Tool | Kind | What it does |
| ----------------------- | ------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `inspect_adb_client` | read | Resolved binary, version, install path; never contacts the server |
| `list_adb_devices` | read | Serial, state, transport, model metadata, inferred emulator/physical kind with basis |
| `inspect_adb_device` | read | Fixed `getprop` whitelist: build, SDK, patch, fingerprint, product, ABIs, debuggable, `ro.kernel.qemu` |
| `inspect_adb_display` | read | `wm size` and `wm density` as declared |
| `list_adb_features` | read | `pm list features` (hardware/software features, hex GL versions) |
| `list_adb_services` | read | `service list` binder services and interfaces |
| `list_adb_processes` | read | `ps -A`: user, pid, ppid, rss, state, name |
| `inspect_adb_window` | read | `dumpsys window` focus declarations (what is on screen) |
| `read_adb_setting` | read | One `settings get` value; device-side null is reported as such |
| `read_adb_logcat` | read | Bounded `logcat -d -t` dump for one buffer, optional pid filter |
| `list_adb_packages` | read | `pm list packages -f` with scope; base APK device paths |
| `resolve_adb_packages` | read | Case-insensitive package-name substring search with exact-match flag |
| `inspect_adb_package` | read | `dumpsys package` projection: versions, install times, installer, flags, and count of listed requested-permission rows (including Unicode names) |
| `list_adb_directory` | read | NUL-framed immediate children, including hidden entries: kinds, permissions, owner, group, sizes, symlink targets |
| `pull_adb_package` | read | Complete split APK set from `pm path`, each file digested |
| `pull_adb_file` | read | One absolute device path pulled and digested |
| `capture_adb_screen` | read | One `screencap -p` frame (exec-out, no device-side file) with dimensions |
| `collect_adb_bugreport` | read | Device-generated bugreport zip, digested |
| `push_adb_file` | **mutates device** | One local file to a device path (e.g. pushing tools into `/data/local/tmp`); checks for existing targets and refuses them without explicit overwrite (the check is not atomic); compares digests when device `sha256sum` is available |
| `install_adb_package` | **mutates device** | Local APK via `adb install [-r]`; preserves its Success/Failure verdict, records `replace_requested`, and leaves actual replacement `null` because prior package state is not observed |
| `uninstall_adb_package` | **mutates device** | Requests package removal for user 0 via `adb shell pm uninstall --user 0`, without data retention |
| `start_adb_app` | **mutates device** | Resolves the package's launcher activity (`cmd package resolve-activity --brief`) and starts it (`am start -n`) |
| `start_adb_activity` | **mutates device** | Starts an intent (`am start -a` with optional URI, component, and up to 16 typed extras); preserves the activity manager's verdict |
| `stop_adb_app` | **mutates device** | Force-stops every process of a package, ending a dynamic-analysis run |
Screen capture validates the PNG signature, IHDR framing, dimensions, and IHDR
checksum before publishing. It does not validate encoding fields or decode and
validate image contents; that limitation is retained in Evidence.
When `push_adb_file` receives an existing directory, adb places the local file
inside it under the local basename; Evidence reports that effective file path
while retaining the caller's directory path in the input parameters. The
private read-only staging copy preserves the source file's executable bits.
For `inspect_adb_window`, `focus_contexts` retains each display header, its exact
display ID when recognized, and every focus declaration with its original line.
Declarations without a display header are explicitly `unscoped`. A missing
declaration differs from a declaration whose `value` is JSON `null`.
`source_section` identifies the selected dumpsys section; `observed_lines`
preserves its focus declarations and display headers in source order.
`focus_summary` is `single_context` only for one context with no repeated focus
field. The singular `focused_window` and `focused_app` summarize that context;
they do not establish device-wide focus. Multiple contexts or repeated fields
produce `ambiguous` with null singular fields and all declarations retained.
When no focus field was declared, the status is `unavailable`. REA reads the
displays section only when the windows section declares neither focus field.
For `install_adb_package`, `replace_requested` records whether `-r` was sent.
The `replaced` result is `null`: the successful install response does not say
whether the package was installed fresh or replaced, and REA does not query the
device first. The install verdict and host APK digest remain available.
Push and install consume a private snapshot of the selected local input. The
reported size and SHA-256 identify those consumed bytes; `local_path` and
`apk_path` retain the original input path, whose contents may change later.
Snapshot creation needs temporary disk space proportional to the input size.
Failed snapshot cleanup remains owned and is retried before later operations
and during shutdown.
## Interpret the result
- **Package numeric declarations retain failures.** `inspect_adb_package`
admits the first valid `userId` or `appId` as `user_id`: a nonnegative safe
integer, including zero. A null value means no valid declaration was found.
Malformed declarations before or after a valid value remain in
`numeric_issues`: the first ten issues with command, field, reason and original
`raw_line`, plus the exact `total`. Reasons are `invalid_format` or
`unsafe_integer`. Any issue makes package coverage partial; existing section
omissions can also make it partial. `version_code` remains exact source text.
- **Display nulls preserve uncertainty.** `inspect_adb_display` admits positive
safe integers. Absent or malformed optional values are null; `numeric_issues`
distinguishes malformed declarations from absence. It keeps the first ten
issues and exact total, with command, field, original line and reason
(`invalid_format`, `unsafe_integer` or `nonpositive`). The last recognized
labeled declaration wins, even when malformed. A physical-density label
suppresses bare-number fallback even when malformed; without that label, the
first unsigned bare number wins, including zero or an unsafe integer that
leaves the value null. Display coverage is complete only with known physical
density and no numeric issues. Missing or malformed required physical size
produces a protocol failure with `details.partial_observation`: an
`inspect_adb_display` observation from provider `adb`, reason
`physical_size_absent` or `physical_size_malformed`, and a `result` retaining
client, serial, optional display facts, numeric issues, null physical size
and partial coverage.
- **Directory names retain whitespace and newlines.** Symlink targets do too,
including trailing newlines. Records containing invalid UTF-8 remain as
labelled hex in `unparsed_lines` and make directory coverage partial.
- **Inferences carry their basis.** Device `kind` reports
`kind_basis` (serial prefix or USB transport); APK `role` reports
`role_basis: file_name`; `emulator_observed` is derived from the build's
`ro.kernel.qemu` property: `true` only when its value is `"1"`, `false` for
any other reported value, and `null` when the property is absent. The raw
property value remains available in `properties`.
- **Device paths are not host paths.** `base_apk_device_path` values live on
the device filesystem. Only pulls and captures produce host-local paths,
each paired with byte size and SHA-256 digest. A pulled file's host name is
the device path's final POSIX segment. REA preserves the original device
path and filename; a missing final segment or `.`/`..` is refused before
output creation. Windows hosts follow [Windows filename rules](https://learn.microsoft.com/en-us/windows/win32/fileio/naming-a-file),
while POSIX hosts retain valid POSIX filenames such as names with spaces or
colons. A package pull checks reported names against the destination
filesystem before pulling, and refuses names that would share a local path.
- **Package path refusals are diagnosable.** Refusals from `pull_adb_package`
return `unreadable_output` with `retryable: false`, keep the `pm path` output
in `details.captured_output`, and name the refusal in `details.reason`.
`unsafe_device_path` lists refused entries in
`details.diagnostics.rejected_paths` as `{device_path, reason}`;
`duplicate_local_destination` reports the shared `file_name` and both
`device_paths`; `local_destination_unavailable` reports the `device_path`,
`file_name`, and filesystem `error_code`. Diagnostics retain raw device
strings, while the caller-facing message escapes invisible and bidirectional
characters so they cannot hide text or reorder the path.
- **Splits are part of the answer.** `pm path` enumerates the base APK and
every installed split; pulls cover the whole set. System packages whose
APKs keep other names report the `unknown` role rather than guessing.
- **Partial results keep facts.** Failed individual pulls return successful
artifacts with `coverage: "partial"` and per-file `failures`. Listings
and package searches that contain unrecognized lines report the first ten
as `unparsed_lines`, state the exact total in a limitation when more were
seen, and mark partial coverage instead of dropping rows.
- **Mutations are explicit.** `push_adb_file`, `install_adb_package`, and
`uninstall_adb_package` declare `mutatesTarget` and never run implicitly:
pushes check for existing device files before transfer and refuse them
unless `overwrite` is set. A failed or unexpected existence probe stops
transfer even with `overwrite`. A completed `test -e` exit of 1 establishes
absence only with empty stdout and either empty stderr or a recognized,
completed adb-server startup notice sequence. Startup notices are host
observations, not remote-path diagnostics; successful push receipts retain
them in `raw_result.device_path_probes`. Incomplete startup, extra diagnostics,
truncation and cancellation do not establish absence. The check is not atomic, so a different
process can create the path before `adb push` writes it. When available, the device's
`sha256sum` verifies the transfer. A missing command, output without a
recognizable digest, or host capacity that prevents the digest check from
starting reports `device_digest_source: "unavailable"` with an
unverified-transfer limitation rather than a retryable error; installs
preserve the
device's `Failure [...]` reason verbatim.
- **Producer identity stays attached.** Every result embeds the producing
adb client (`binary_path`, `path_source`, `version`, `installed_path`).
## Resource and lifecycle limits
- Caller values reach adb only as positional arguments in fixed vectors;
no tool accepts an arbitrary shell command line, `input`, or `monkey`.
The app/activity lifecycle tools use `am start -W` and `am force-stop`.
- Per-invocation budgets: client/devices 10-20 s, shell reads 20-60 s with
64 KiB-8 MiB capture caps (logcat 4 MiB, screen 32 MiB), transfers and
bugreports 600 s. Output capture overflow reports `resource_constraint`
and names the byte allowance. Deadline expiration, including the
client-version probe, reports `provider_timeout` and names the effective
milliseconds. This release has no runtime override for those deadlines or
capture caps. Command cancellation preserves the caller's exact reason;
a later abort does not replace an already completed command failure.
- Pull and capture outputs are additive: the destination must not already
exist. Re-running a completed pull refuses the existing directory instead
of overwriting evidence.
- Cancellation during a package pull preserves caller-selected output. The
error includes partial observations identifying completed artifacts and the
interrupted local path; remaining APKs were not attempted.
- Screen capture checks cancellation before publishing its exclusive output.
If cancellation arrives after publication, the error retains the committed
file's path, digest and dimensions; the file remains available as evidence.
- Bugreports collect into isolated staging and publish with exclusive creation;
filename collisions select the next numeric suffix. If publication fails, the
completed archive remains in staging. The error's partial observation gives
its recoverable `local_path`, the original `adb_reported_path`, and an unknown
digest status. Later operations and shutdown preserve this retained evidence.
- Failed package acquisition rollback retains its directory identity and retries
cleanup before later work and during shutdown. Cleanup refuses a replacement
directory. Terminal shutdown stops admission and drains active operations.
- The provider holds no long-lived processes; every invocation is one
bounded adb run. The adb server belongs to the caller.
## Related capability and follow-up lanes
UI interaction beyond intent starts (tapping, typing, monkey fuzzing, screen
recording) remains a separate lane: candidates such as
[mobile-device-mcp](https://github.com/srmorete/mobile-device-mcp) install
driver software on the device and pair with the Frida instrumentation lane
(#1359) behind explicit mutation-authority contracts. Network transports
beyond already-connected devices (`adb connect`) are not exposed yet.
## Real verification
`npm run verify:adb` exercises observation and transfer operations against a live device. It
requires `adb` (or `REA_ADB_PATH`) and any attached emulator or phone;
`--serial` selects one when several are attached, `--pull PACKAGE` pulls a
real package and verifies the returned digests against the files on disk.
The optional `--install-apk` lifecycle is unavailable until the selected APK
identity and device cleanup ownership can be established safely.
Record: verified against adb 34.0.5 on Linux against an Android 14
(API 34) x86_64 emulator — all operations above, a 224-package inventory,
287 services, 92 features, a real two-APK split set pulled with byte-exact
digests, a verified push/pull roundtrip, a 1.3 MB screen capture, and the
complete package lifecycle; see `docs/testing.md` for the lane summary.
Intent integer extras use signed decimal text. The ADB argument encoder removes
redundant leading zeros for Android's radix-sensitive `--ei` parser while retained
Evidence keeps the caller's original spelling. Package inspection accepts both
`userId` and Android 16's `appId` spelling for the legacy `user_id` result field.
A completed push and device digest verification are separate observations. A
verification timeout, cancellation, or output budget failure returns the typed
failure with partial Evidence of the completed transfer. An unavailable checksum
command or unrecognized checksum output leaves verification explicitly unknown.