- Insert .splitter div between .apdu-list and .apdu-detail
- Replace flex:1 / max-height:40% with --detail-height CSS var
- Drag handler with clamp 10%-80%, persisted in localStorage
- Bump 1.22.12, SW cache v43->v44
GsmtapListener.stop() closed the UDP socket permanently, but start()
only set _running=True without recreating it. Second capture session
would silently spin on the dead socket (OSError → None → continue).
- Add _closed flag to GsmtapReceiver, set in close()
- Store bind_port in GsmtapListener for socket recreation
- start() recreates GsmtapReceiver only if previous one was closed
- Uses getattr for backward compat with test fakes
SIGROK-derived PCAPs have relative timestamps (0, 1.37, 15.02) rather
than epoch time. The _REAL_TS_MIN check failed, so close_session() set
ended to wall-clock (same as started → duration 0s). Now close_session
accepts an optional duration_secs param; the import handler passes
last_ts - first_ts so the session gets the correct capture duration.
- Toggle buttons (Tiles/List) in sessions header, preference
persisted in localStorage
- List view: table with Name, Date, Duration, Messages, Mode, Actions
columns; same fields and buttons as tiles, just tabular layout
- Refactored loadSessions() into buildGridHTML/buildListHTML helpers
with shared attachSessionListeners() for both views
- i18n: Tiles, List, Name, Date, Duration, Messages, Mode, Actions
keys added to LANG_RU
- Bump 1.22.6, SW cache v37->v38
- error_names(m) now includes 'Data length mismatch' when
decoded.length_mismatch is set, showing a ⚠ badge + tooltip
in the APDU list row (alongside existing errors like bad FCS)
- Header title: SIM-ME -> UICC
- Bump 1.22.5, SW cache v36->v37
- capture-btn moved to first child of trace-topbar (before search input)
- Button always visible: shows Start/Stop/Connect-to-server as before
- When server capture is disabled: grayed-out Start Capture with
tooltip via existing i18n key 'Capture is disabled ...'
- No more hiding button when viewing saved session or when disabled
- Bump 1.22.4, SW cache v35->v36
- HTTP request logging now off by default (log_message guarded by
RequestHandler.log_requests); --log-requests enables it
- Add session lifecycle logging: POST /api/sessions/import and
DELETE /api/sessions/<id> emit session+name+messages/bytes or
session= lines via the shared _log() helper
- Capture start/stop and periodic status keep logging as before
T=15 in the ATR is a Protocol-and-Parameter-Selection capability,
not a data protocol like T=0/T=1. Previously 'Protocols: T=0, T=15'
was misleading.
- _decode_atr: exclude T=15 from protocols, set pps=True instead
- TCK presence: only-t0 rule now also requires no PPS (T=15 implies TCK)
- PWA ATR detail: 'Protocol: T=0' + 'PPS: supported' rows
- +1 test with the real UICC ATR (3B 9F 95 80 1F C7 ...)
- Bump 1.22.2, SW cache v33->v34
_decode_nai (decode.py) only treated an all-0xFF record as empty.
A record that is a lone identity tag, or a tag with a zero-length
value (e.g. 80 00 + FF erase filler — a standard unused EF_IMPU
record) slipped through to an ascii fallback and rendered as garbage
text (the "Text: �" artifact).
- Treat a lone tag (80 / 81) or tag+zero-length-value (80 00 / 81 00)
as empty.
- Only decode the TLV *value* bytes (UTF-8); never feed the TLV header
or trailing FF filler to the text decoder. Drop the fragile
raw.rstrip(...).decode('ascii','replace') fallback.
Tests: +4 (empty 80 00+FF record, all-FF, real SIP URI, lone tag).
Backend-only; stays on 1.22.0.
Follow-ups from the capture-reliability review:
- All server-side capture log lines are now prefixed with a local
timestamp (YYYY-MM-DD HH:MM:SS) for easier correlation.
- New --log-interval SECONDS CLI flag (default 60) tunes the
"Capture alive" heartbeat interval; passed through to CaptureManager.
Stays on 1.22.0 (backend-only). 369 tests pass.
The GSMTAP capture thread died silently whenever a single non-SIM or
truncated GSMTAP datagram arrived (read_packet() raises ValueError,
uncaught in the loop, killing the daemon thread while UDP kept
flowing) — matching the "stops listening after a sniffer restart"
symptom.
Fixes:
- GsmtapListener.iter_messages() now catches ValueError/struct.error
from read_packet(), drops the offending datagram (counted via a new
`dropped` counter) and keeps listening.
- CaptureManager._capture_loop() guards the DB insert so a transient
error can't kill the thread; a top-level guard logs
"ERROR: capture thread stopped on error …" + traceback and clears
the session so /api/status no longer reports a stuck "active".
- Server now logs capture liveness to stderr: "Capture started", a
periodic "Capture alive: session=N messages=M bytes=B dropped=D"
heartbeat (default every 60 s, configurable via log_interval), and
"Capture stopped" (or "empty session discarded").
Tests: +5 (listener survives non-SIM/short packet, capture loop
survives insert error, logs stop-on-error, heartbeat+stop log).
Stays on 1.22.0 (backend-only).
- decode.py: use the GSMTAP_FLAG_BAD_FCS constant instead of a literal
0x01 in gsmtap_flag_names() (import from ..gsmtap); no behavior change.
- tests: add a DB round-trip test (insert_message with BAD_FCS flag →
get_messages → decoded.errors) covering the end-to-end path, not just
the direct decode_sniff_msg call.
sigrok-iso7816-stream sets GSMTAP_FLAG_BAD_FCS (0x01) in the GSMTAP
header `res` byte on desynced TPDUs (mis-framed exchanges kept for
forensics). Propagate that flag through the whole pipeline so such
packets are explicitly surfaced instead of decoding as garbage:
- gsmtap.py: GSMTAP_FLAG_BAD_FCS const; GsmtapReceiver.read_packet()
now returns (sub_type, data, flags) from the res byte.
- server/capture.py: GsmtapListener forwards the flags to the DB.
- pcap.py: _parse_gsmtap_hdr/_scan_gsmtap/extract_gsmtap and
parse_pcap/parse_pcapng now yield a (ts, type, payload, flags)
tuple, so the flag survives a PCAP/PCAPNG round-trip.
- server/server.py: PCAP import stores the flags.
- server/decode.py: gsmtap_flag_names() maps bit 0x01 to
'bad FCS (desynced)'; merged into decoded errors for tpdu/atr/pps
and rst/vcc, so the frontend ⚠ badge + Errors row flag the entry.
User-visible: desynced TPDUs show a ⚠ "bad FCS (desynced)" marker
in the trace and detail views. Tests: +4 (decode BAD_FCS markers,
PCAP flag round-trip); existing tests updated for the new tuple
arities. Version 1.22.0, SW cache simtrace-analyser-v32.
Support the sigrok-iso7816-stream decoder (FX2 logic-analyzer passive
SIM sniffer, github.com/anttro/sigrok_iso7816_stream) which streams
GSMTAP-SIM on UDP 4729 and adds two custom sub-types for line events.
- gsmtap.py: GSMTAP_SIM_RST_EVENT (0x10) / GSMTAP_SIM_VCC_EVENT (0x11).
- capture.py: gsmtap_msg_type() maps 0x10→'rst', 0x11→'vcc' (shared by
the listener and tests); pcap.py _SUB_TYPE_NAMES gains the same
entries so PCAP re-imports of these captures preserve the events.
- decode.py: decode_line_event() decodes [direction, level, reserved]
into reset asserted/de-asserted or power applied/removed + level;
routed through decode_sniff_msg.
- database.py: reset-assert and power-removal invalidate the per-channel
file selection (card-level transition); de-assert / power-on are
markers only.
- frontend: RST/VCC labels (RESET ASSERTED / VCC OFF, …), a detail-view
branch (Event + Level + raw), and list styling (.apdu-type.rst/.vcc).
README notes the custom sub-type compatibility. Tests: +8 (line-event
vectors, sniff routing, subtype + PCAP round-trip mappings) — 359
total. Version 1.21.0, SW cache simtrace-analyser-v31.
- Pair CLA+INS and P1+P2 into single rows (halves the header rows
per C-APDU): "· CLA, INS 00 — UICC, A4 — SELECT".
- Skip the " — desc" suffix when the parser's desc merely repeats
the hex (generic Data fields showed "6F46 — 6F46").
- Restore the implied-CLA annotation on the C-APDU title row (was
dead code behind an operator-precedence bug).
Version 1.20.2, SW cache simtrace-analyser-v30.
The secured data of an SCP80 packet is a stream of C-APDUs (e.g.
SELECT / UPDATE BINARY sequences for OTA file management). Vendor the
working C-APDU parser from the otaman project (frontend-only, ~10 KB):
PARSE_INS table, describeP1P2, describeInstallDataLv, parseOneApdu and
parseCompactApdus (concatenated streams with implied-CLA handling).
Detail view now shows an "Application message" section for secured
packets:
- unciphered packets (SPI1.b3=0): the on-wire secured data decodes
immediately;
- ciphered packets: decodes from the client-side decrypted payload
once the user supplies the key (before that, ciphertext rows only);
- each C-APDU renders as a numbered row with indented CLA/INS/P1/P2/
Lc/Data/Le + descriptions (nested LV for INSTALL), with implied-CLA
annotation; raw-hex fallback when nothing parses;
- vendored parser uppercases input (otaman's expects uppercase hex).
Tests deferred to the follow-up step (node:test infra, planned).
Version 1.20.0, SW cache simtrace-analyser-v28.
Adds optional decryption of ciphered secured packets, entirely in the
browser — the key never leaves the frontend.
- decode.py: _decode_secured_packet exposes 'cipher_block' (the
contiguous on-wire ciphered octets CNTR..end, ready for CBC) and
raw kic/kid bytes for algorithm selection. Additive only.
- index.html: pure-JS DES/3DES core (FIPS 46-3 bit-array tables,
parity ignored) + CBC; AES-CBC via native SubtleCrypto; both with
zero IV per TS 102 225. decryptSecured() picks the algorithm from
the KIc nibble (02 AES-CBC, 05 3DES-2K, 09 3DES-3K, 01 DES) with
key-length validation.
- UI: ciphered packets show a key-hex input + Decrypt strip below the
detail table; keys are memoized per key-set for the session in an
in-memory OTA_KEYS map — never sent to the server, never persisted.
- Decrypted results cached by cipher_block (survive live-polling
re-renders); plaintext rows rendered as CNTR/PCNTR/RC/Secured data
(decrypted) alongside the ciphertext; wrong-key guard via the
decrypted padding counter (PCNTR > remaining → "invalid padding").
- 3DES implementation validated byte-identical against openssl
des-ede3-cbc (3-key, 2-key ×3 random vectors, single-key).
No CC verification (out of scope). Tests: +2 (cipher_block
contiguity, AES registry guard) — 350 total. Version 1.19.0,
SW cache simtrace-analyser-v27.
User-reported: the decoded CNTR (0x1830A44E41 ≈ 104 billion) cannot
be the real counter — the OTA platform shows ~200 for that card.
Root cause: per TS 102 225 Table 2 note 1, when SPI indicates
ciphering the whole block from CNTR through the secured data
(CNTR, PCNTR, RC/CC/DS, secured data) is ENCRYPTED. Only CPI, CPL,
CHI/CHL, SPI, KIc, KID and TAR travel in the clear. Our decoder
parsed the ciphertext bytes as if they were the real CNTR/PCNTR/RC
and presented them as such — misleading (the true values are only
recoverable after 3DES decryption with the card's keys).
Fix:
- _decode_secured_packet: emit 'ciphered': True when SPI1.b3 is set
(wire hex values kept for reference); docstring documents the
ciphered block.
- _decode_response_packet: accepts ciphered (command SPI2.b5, PoR
ciphering) and flags the response likewise; database context now
carries por_ciphered from the ENVELOPE so the GET RESPONSE PoR
inherits it. Standalone SMS-SUBMIT PoRs carry no SPI — state
unknowable, defaults to False (documented).
- Frontend: CNTR/PCNTR/RC-CC-DS labels get "(encrypted)" when the
respective packet is ciphered (command, response, and ENVELOPE
response renderings).
Tests: +4 TestSecuredCiphering (ciphered flagged, unciphered plain
CNTR e.g. 200, response flag, PoR-via-context). 348 total. Version
1.18.0, SW cache simtrace-analyser-v26.
stk_test1 msg 14036 regression: READ BINARY of EF_SPDI returned
"No decoded data available" — _decode_spdi (added v1.16.0) parsed
the A3 container value as bare 3-byte records (missing the inner
'80' value TLVs) and returned a list of strings, while
_build_summary/_file_summary expects the EF_PLMNwAcT shape
(list of {mcc, mnc} dicts). The resulting TypeError aborted the
whole message decode — this was the source of the 28 "Decode
error ... string indices" messages seen across all sessions.
_decode_spdi now unwraps inner '80' TLVs (bare-records fallback
kept) and always returns {'plmns': [{mcc, mnc}...]}. Replay of all
33 sessions: decode exceptions 28 → 0 (2 remaining undecoded tpdu
are 1-byte truncated capture records, not decodable by nature).
Tests: +3 (standard inner-80 form, bare-records fallback, exact
stk_test1 regression via decode_message). 344 total. Version
1.17.1, SW cache simtrace-analyser-v25.
Next to the relative elapsed time, the detail-view header now shows
the absolute local time (YYYY-MM-DD HH:MM:SS.mmm) computed as
session start + message elapsed offset.
- STATE.sessionStarted populated from /api/capture/start (live
capture), data.session.started (session view), and PCAP imports
(sessions.started already derives from the first packet timestamp
when real).
- Graceful fallback to relative-only when the session start is
unknown; applies to all detail header variants (gap, no-decoded,
state change, Fi/Di, ATR/PPS, APDU).
Version 1.17.0, SW cache simtrace-analyser-v24.
mobileid_activate_setpin_resetpin_20201222 regression: UPDATE BINARY
(plain offset addressing) after an SMS-PP DOWNLOAD / proactive group
showed "Writes: —" in the map view, although selection tracking kept
EF_SMSS selected the whole time.
Root cause: the sel-based branch only emitted result['file'] when
_decode_file_data() produced content — files without a registered
content decoder (EF_SMSS 6F43) silently lost attribution. The first
write only displayed via the frontend's SELECT-note fallback, which
masked the bug until a proactive group split the RW run.
Fix: decouple attribution from content decoding in both the command
branch and the GET RESPONSE branch — when the selected file has no
content decoder (or the offset is non-zero, where content decode is
intentionally skipped), emit {fid, ef} so the map, list, and detail
views always name the file. Content decoders still take priority.
Session 63 now decodes all three UPDATE BINARY ops as ADF/EF_SMSS.
Tests: +4 TestFileAttribution, one expectation updated to the new
contract — 342 total. Version 1.16.1, SW cache v23.
Spec verification against UICC_SPECS.md / TS 102 223 V18.3.0 §8.6
showed the Command Qualifier for SET UP IDLE MODE TEXT is RFU
("this byte is RFU") — no icon self-explanatory bit exists in any
generation (GSM 11.14 predates the command; TS 31.111 does not
override it). The v1.11.0 _idle_text_qualifier decoded bit 1 as
"icon self-explanatory" from an unverified source.
Remove the decoder; the command now uses generic RFU handling
(hidden at '00', raw hex otherwise). Tests updated accordingly.
Version 1.15.1, SW cache v21.
- Pair members (a command ending '61 XX' and its GET RESPONSE, via the
server-side response_for link) get a subtle slate row tint
(--pair-row, themed: rgba(148,163,184,.12) dark / rgba(100,116,139,.15)
light); hover/selected still take precedence.
- Paired GET RESPONSE rows are prefixed with '↳' marking them as a
continuation of the previous command; orphaned GRs stay plain.
- GET RESPONSE label uses cat-gr class: muted color, weight 600 like
other command names.
- Map view untouched (it skips GRs); detail view already shows
"Response for".
Version 1.15.0, SW cache simtrace-analyser-v20.
Self-review of 174df8a/d0f8fb8 found two real bugs plus smaller items:
- _decode_timer_value read semi-octets high-nibble first; TS 23.040
coding (referenced by §8.38) puts the first digit in the LOW nibble
— "09:25:03" decoded as "90:25:03". Regression test added with an
asymmetric vector; the original symmetric vectors masked the bug.
- TIMER MANAGEMENT qualifier used bits b2-b3 instead of the spec's
bits 1-2 (q & 0x03): '01' deactivate was hidden, '02' printed
"deactivate timer", RFU '04' printed "get current timer value".
Decoder fixed and wrong expectations corrected.
- Stale text_note: a compliant 8D Text String arriving after the
non-standard '10' quirk TLV now clears the note.
- Proprietary type format unified to uppercase ("Proprietary (0xF2)").
- GET INKEY: Yes/No bit now suppresses the b1/b2 character-set parts,
matching §8.6 ("character sets ... are disabled").
- Frontend renders string timer_id values verbatim instead of NaN.
Tests: +4 regressions, expectations corrected — 320 total.
Phase 3 of the spec audit: TS 102 223 §8.6 qualifier coverage for every
command that defines a coding (verified against V18.3.0 text):
- Value tables: SET UP CALL ('00-'05), LAUNCH BROWSER, GET READER
STATUS, SET FRAMES, ENCAPSULATED SESSION CONTROL ('00-'04), LSI
COMMAND ('00/'01).
- Bitmask decoders (deviations from default shown; all-default hidden;
pure-RFU falls back to raw hex): SEND SHORT MESSAGE packing bit,
PLAY TONE vibrate, DISPLAY TEXT / DISPLAY MULTIMEDIA MESSAGE
priority + clear-delay, GET INKEY (alphabet/UCS2/Yes-No/immediate-
digit/help), GET INPUT (alphabet/UCS2/hidden/packed/help),
SELECT ITEM (presentation type/soft key/help), SET UP MENU,
TIMER MANAGEMENT (start/deactivate/get value), LANGUAGE NOTIFICATION,
OPEN CHANNEL (link establishment/reconnection/background/DNS with
bearer-dependent note), CLOSE CHANNEL (NAA reuse note), SEND DATA,
DECLARE SERVICE.
- RFU-byte commands (MORE TIME, POLL INTERVAL, SEND SS/USSD, ACTIVATE…)
keep raw-hex fallback.
Tests: +11 TestQualifiers cases — 317 total. Version 1.14.0, SW cache
simtrace-analyser-v17.
TS 102 223 §6.6.13 support for the SET UP IDLE MODE TEXT (0x28)
proactive command:
- Command qualifier: COMMAND_QUALIFIERS entries may now be callables
for bitmask-style qualifiers; _idle_text_qualifier decodes bit 1
as "icon self-explanatory", hides 0x00, falls back to raw hex on
RFU bits.
- Icon Identifier TLV (0x1E) parsed into icon_id (image record
number in EF_IMG, 0 = none); rendered as an "Icon" detail row in
the PWA.
- Zero-length Text String TLV now yields text='' ("remove idle mode
text") instead of being silently skipped.
Tests: +6 TestIdleModeText cases (8-bit/UCS-2 text, icon TLV,
qualifier variants, empty text) — 285 total. SW cache bumped to
simtrace-analyser-v14.
- Add sim-analyser-light.svg (#2563eb, light-theme accent); dark
keeps the existing #3b82f6 variant.
- applyTheme() now swaps the favicon href alongside data-theme and
theme-color, so the icon follows toggle click, saved theme on
load, and OS-preference fallback (all route through applyTheme).
- Precache the new asset in the service worker.
- Add a P2 spec for AUTHENTICATE (auth context bits + specific-reference-data
flag), so P2 is displayed instead of only leaking through cmd.context.
- Fix P1 name to 'No indication' (was the misleading 'Run GSM algorithm').
- Decompose AUTN into SQN(xor)AK / AMF / MAC and AUTS into SQN(xor)AK / MAC-S
per TS 33.102; surface 'specific reference data' key flag.
- SW column now shows green (success 90/92-9F), blue (pending 61/91/6C),
amber (warning 62/63), red (error 64-6F) per ISO 7816-4 / TS 102 221.
- _decode_nai: treat all-FF records as empty instead of emitting
replacement characters; strip FF padding before the ASCII fallback.
- Only update the selected-file state on a successful SELECT (SW1 in
61/62/63/90/91/92/9E/9F); a failed SELECT (e.g. 6A82) leaves the
current file unchanged.
- Track the file structure from the FCP (GET RESPONSE after SELECT).
- Skip file decoding for record ops (READ/UPDATE/SEARCH RECORD) when the
selected file is transparent/DF, marking the result 'stale' instead of
decoding garbage — this catches silently-dropped SELECTs.
- Track the currently selected file (SELECT by FID/path) across the
session and pass it to the decoder; reset on ATR/card change; seed via
a bounded backward scan on paged views.
- Decode READ BINARY/RECORD, UPDATE BINARY/RECORD and (via GET RESPONSE)
SEARCH RECORD bodies using a FILE_DECODERS registry: IMSI, ICCID,
LI/PL, ADN-format (MSISDN/SDN/FDN/...), SMS records (SMSC + TPDU),
UST/SST/IST service tables, PLMN lists (+access tech), EF_DIR,
CBMI/R, ARR, PNN, ECC, SPN, LOCI, ACC, Phase, ISIM IMPI/DOMAIN/IMPU.
- File formats verified against TS 51.011 / TS 102 221 / TS 31.102
(PLMNwAcT is 5-byte entries; EF_SMS = status + SMSC + TPDU; ISIM
identities are 0x80 TLV).
- Offset != 0 READ BINARY/UPDATE BINARY is skipped (fragment reads).
- SEARCH RECORD results (record numbers) decoded in GET RESPONSE.
- GET RESPONSE summary now shows FCP file type/structure/size.
- New 'File data' section in the detail view.
The selected row was white (--bg-input) in light theme, nearly
indistinguishable from the unselected/hover rows. Use a blue-100 tint
(#dbeafe) in light theme only; dark theme unchanged.
Frontend:
- trace empty-state now shows connection status + capture mode (boxed,
colored) in the center, updated live via checkConnection
- capture-mode uses amber warning style (gsmtap/disabled), green for direct
- stopCapture clears currentSessionId
Server:
- gsmtap read_packet catches OSError on socket close (no more EBADF
traceback on stop)
- stop_session joins the loop and discards a session with zero messages;
/api/capture/stop reports discarded
- startup prints server version
Tests: 122 (gsmtap close, empty/non-empty capture lifecycle).
- pcap.py: _parse_tsresol now pads options to a 32-bit boundary
((-length) % 4) and uses 2** for the MSB-set (power-of-2) case, so
nanosecond if_tsresol is read correctly (was defaulting to µs and
scaling timestamps 1000x)
- database/server: set imported session started/ended from the first/last
packet timestamps (falling back to wall-clock for synthetic SPB)
- frontend: formatDuration renders hours
- tests: tsresol (ns after padded option, default µs, ns round-trip) and
session-window/fallback (119 total)
The fetch handler tried to cache every intercepted request, so a
chrome-extension:// request made cache.put() throw 'Request scheme
chrome-extension is unsupported'. Skip non-http(s) and non-GET requests,
and swallow cache-write failures.
First published version. __version__ and pyproject bumped to 1.0.0,
/api/status now returns 'version' so the PWA can check compatibility.
Tests: 107 (status includes version, package version).
Lets a public-hosted PWA reach a local simtrace2-pysniff-server
(http://127.0.0.1:8081); without it the browser blocks the preflight
before any request is sent.
Proactive commands with a body over 127 bytes (e.g. SET UP MENU with a
long item list) use a long-form length (0x81 NN / 0x82 NN NN) in their
D0 wrapper. Previously parse_tlv bailed on those, so the FETCH decoded
as a bare 'FETCH' without its command name or items.
Tests: 105 (long-form 1/2-byte, real SET UP MENU sample).
- pcap.py: parse_pcap (micro/nano, both byte orders) and parse_pcapng
(SHB byte-order, IDB linktype + if_tsresol, EPB/SPB) readers that
unwrap Ethernet/RAW/IPv4/NULL/SLL/SLL2 to UDP and extract GSMTAP-SIM
packets (APDU->tpdu, ATR->atr, PPS->pps), with a heuristic fallback
- database: insert_messages bulk-insert helper
- server: import endpoint reads raw body, auto-detects format, skips
non-GSMTAP packets; 400 'No GSMTAP data found in trace' (no session)
when empty, else creates a 'pcap' session named after the file
Tests: 102 (pcap round-trip, pcapng, no-GSMTAP, endpoint 200/400).
CLA 0x80 is the ETSI/3GPP-defined UICC/USIM class (TS 102 221 Table 10.3),
not ISO 7816-4 'proprietary'. Relabel 0x80 -> ETSI-defined (UICC/USIM),
0xA0 -> ETSI-defined (SIM/GSM), 0x00 -> inter-industry (ISO 7816-4);
drop the now-redundant note field.
The PCAP branch fell back to GSMTAP_SIM_APDU for every non-ATR message,
so PPS/CHANGE/FIDI were written as 'APDU'. Align with the UDP branch and
the server export: only atr→ATR and tpdu→APDU are written.
pcap.py: LINKTYPE_GSMTAP (155) was wrong — 155 is DLT_USER8, so
Wireshark showed 'USER 8 / DLT=155'. Now write Ethernet-encapsulated
PCAP (LINKTYPE 1): each GSMTAP message wrapped in Ethernet + IPv4 +
UDP (port 4729), with correct checksums — matching a real tcpdump
capture of GSMTAP-over-UDP and how Wireshark's gsmtap dissector is keyed.
gsmtap.py: GSMTAP_TYPE_SIM was 0x07 (WiMAX burst) — corrected to 0x04;
GSMTAP_SIM_APDU was 0x02 (PPS req) — corrected to 0x00; added PPS_REQ/
PPS_RSP. These match libosmocore/Wireshark packet-gsmtap.h.
Verified with tshark: PCAP now decodes as GSM SIM (SELECT/STATUS/ATR)
instead of 'USER 8'.
tests: pcap tests assert Ethernet linktype + framing + GSMTAP payload.
pcap.py: build_pcap(packets) — build PCAP bytes in memory (same format
as PcapWriter)
database.py: get_messages_raw() — raw elapsed/type/data without decode
server.py: GET /api/sessions/<id>/pcap generates a GSMTAP-encapsulated
PCAP (ATR/TPDU) with real timestamps and downloads it as
<session-name>.pcap (ASCII filename fallback + RFC 5987 UTF-8 for
spaces/unicode via _content_disposition)
tests: build_pcap + _content_disposition (67 total)
- PLI_QUALIFIERS: TS 102 223 qualifier values (Location Info, IMEI,
NMR, Date/time, Timing Advance, Access Technology, etc.)
- _command_details_name(): appends the qualifier description to
PROVIDE LOCAL INFORMATION (type 0x26) in Command Details
- decode_cat (FETCH) and decode_tr_command (TR) now use it, so both
'cat_command' and 'response_to' carry the qualifier (e.g.
'PROVIDE LOCAL INFORMATION — IMEI')
tests: PLI qualifier cases (TR/FETCH, IMEI/location/date, non-PLI)
decode.py:
- _decode_fcp: FCP/FCI/FMD template — full bit-level file descriptor
(TS 102 221 Table 11.5: file type, structure, shareable, data coding,
record length, num records), file ID (+ KNOWN_FIDS name), DF name,
SFI, life cycle
- _decode_auth: GSM SRES+Kc, or 3G tag DB (RES/CK/IK/KC) / DC (AUTS)
- _decode_tr_result: TERMINAL RESPONSE Result TLV (code + name)
- _decode_response_for(ins, data): GET RESPONSE decodes its body using
the previous command's INS (SELECT→FCP, AUTHENTICATE→auth)
- decode_message wires response decoding for TR and GET RESPONSE
database.py: prev context now carries 'ins' (INS byte) via
_context_from_decoded, used by GET RESPONSE response decoding
tests: FCP, AUTH (GSM/3G), TR result, GET RESPONSE wiring (57 total)
decode.py: split decode_tr_command() out of decode_cat(); TR (0x14)
now stores the echoed proactive command type in 'response_to' instead
of 'cat_command', so it no longer overrides the INS name.
FETCH/ENVELOPE still use cat_command as the primary label.
tests: TR now asserts response_to + no cat_command
decode.py:
- decode_cat: TERMINAL RESPONSE (0x14) now extracts the proactive
command type from its echoed Command Details (tag 81) — no history
lookup needed since TR carries its own type
- decode_message gains prev param; GET RESPONSE (0xC0) resolves
'response_for' from the previous command when its SW1 was 0x61,
else None (orphaned)
- decode_sniff_msg threads prev through
database.py:
- _row_to_message accepts prev; new _decode_rows tracks previous-TPDU
context in sequence; get_messages/get_messages_after seed context
from the message before the window/cursor
tests: TR cat_command + GET RESPONSE context/orphaned cases (49 total)
catarrh