MFU-C AUTH and iCLASS READCHECK skipped the next record before checking
it was the tag's reply. With an unanswered command, the reader's next
frame was swallowed: a retried 1A 00 lost the whole MFU-C auth, and a
CHECK right after READCHECK was taken as the epurse, hiding the lookup.
Now the record is only consumed when it is a tag frame of the expected
shape.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
.data shared the bootphase2 output section with the RAM-executed code, so
the PM5 bootrom linked with "LOAD segment with RWX permissions" (CMake
silenced it with --no-warn-rwx-segments). Put .data in .bootphase2_data
with its own PT_LOAD, placed right after .bootphase2 in RAM and flash so
the existing copy loop still covers it; asserts guard the contiguity.
Same image size on RDV4/PM5/PM5+BWM; code identical bar .ramfunc moving
ahead of .data, data bytes unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
common_arm/Makefile.hal already includes it, so a direct "make -C bootrom"
appended the standalone define twice and stopped with "You must choose only
one Standalone mode!: LF_SAMYRUN LF_SAMYRUN". Matches armsrc/Makefile.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
v is unsigned long; PRIu64 is "llu" on macOS, so clang fails the build
with the PrintAndLogEx format attribute.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The 4-byte packing loop read up to 3 bytes beyond src when length was not
a multiple of 4 (ASan heap-buffer-overflow in data test_ss8). Pack byte by
byte into the zeroed backing buffer instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
save_buffer8/32/S32 allocate a backing buffer that the self tests never
released. Found by cppcheck 2.22 follow-up, confirmed with LeakSanitizer
(696 bytes in 3 allocations per run of the three tests).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
size_t printed with %lu/%ld/PRIu64 and uint64_t with %lu are correct on
64-bit Linux/macOS only; with the new format attribute they fail -Werror
builds on 32-bit hosts (RPi, 32-bit Windows), and %lu for uint64_t also on
64-bit Windows. unsigned long printed with PRIu64 likewise.
Found by a 32-bit (arm-none-eabi) syntax-only pass over the client.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add a printf format attribute to PrintAndLogEx so gcc/clang flag
argument/specifier mismatches (previous commit fixed the 34 existing ones).
On mingw use __MINGW_PRINTF_FORMAT, matching the GNU printf selected by
_ISOC99_SOURCE rather than ms_printf.
PrintAndLogEx(x, "") is the established way to print an empty line, so
-Wno-format-zero-length is added to every build that compiles callers:
client Makefile, client CMake, experimental_lib, and cliparser (Makefile
and cmake).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Found by building with a printf format attribute on PrintAndLogEx.
User visible:
- hw version: GENERIC device/firmware name was passed as an unused arg
and never printed
- hf saflok selftest: PRIx32 lacked '%', printed a literal 'x'
- hf mfp: key index value was never printed
- hf waveshare: output filename was never printed
Also: size_t/int64_t/uint64_t specifiers (%zu, PRId64, PRIX64), int
macros printed with %zu, ' ' flag on %u, unused arg in flash_prepare,
and non-literal strings passed as the format.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The save-state self tests and add_more_candidates() dereferenced calloc
results without a NULL check. Follow each file's existing OOM handling.
Found by cppcheck 2.22.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The 16/32-bit parity fallbacks folded x but indexed g_odd_byte_parity[256]
with the unmasked value, reading out of bounds. Only reachable with a
compiler that does not define __GNUC__. Found by cppcheck 2.22.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
%llu reads a 64-bit argument for a size_t, which is only correct on LP64
hosts; 32-bit builds (e.g. RPi Zero, 32-bit Windows) print garbage.
Found by cppcheck 2.22.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The decrypted/selected ADF strings were strdup'd and uppercased for the
compare, but never freed on either return path and used without NULL check.
Use strcasecmp on the originals instead. Found by cppcheck 2.22.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
set_my_executable_path() checked my_executable_path instead of the freshly
allocated my_executable_directory, so a failed calloc led to strncpy on NULL.
Found by cppcheck 2.22.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Start and End columns were unsigned differences to the first frame,
so a frame stamped before it or after the timer restarted showed as
about 4294967296, eg -281 in traces/lf_hitag_crypto_dump.trace and
-29708 in traces/hf_14b_cryptorf_select.trace. Same as the frame delay
time in d11643c8f.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Most `xx list` aliases pass "<protocol> -c", so adding -c yourself, eg
`hf 14a list -c`, gave "extraneous option -c". Allow up to two.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Both lived in extract_uid, so a 14a anticollision after an iCLASS SELECT
overwrote the CSN and its CHECK gave no lookup line. Also take CL2 only
after a CL1 with the cascade tag, and CL3 after such a CL2, instead of
printing a UID made of whatever the buffer held.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
extractChall_ev2() only took a 21 byte tag reply, ie wrapped framing
without CID, so a native EV2 First / Non First auth, or one with a CID,
printed no challenge. Locate it like the AES branch does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
loadFile_safeEx, loadFile_TXTsafe, loadFileEML_safe and loadFileMCT_safe
freed *pdata on a short read or a failed realloc but left the caller's
pointer set, so a caller keeping it freed it again. Same as
loadFileDICTIONARY_safe_ex already does.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The -r and --frame delay was an unsigned difference, so a frame starting
before the end of the previous one, or after the timer restarted, eg
between the sessions of traces/hf_14b_xerox_info.trace or the frames of
traces/lf_HitagS256_dump.trace, showed as about 4294967296 instead of
-83024 or -117.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
With -r the Data header said "' denotes short bytes", but short bytes
are marked (n), and in both modes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It returned PM3_SUCCESS when the trace download or writing the file
failed, so scripts carried on as if the trace was saved.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CmdTraceListAlias() built "-t <protocol> <args>" in a 128 byte buffer,
so a longer command line was cut silently, eg `hf mf list -f` with a
long dictionary path tried to load a truncated file name and fell back
to the default keys. Allocate the buffer to fit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The trace buffer was cleared before the load, so a missing or empty
file lost the trace already loaded. Loading straight into gs_trace also
left it dangling when loadFile_safe() freed it on a short read, to be
freed again by the next clear or load. Load into a local buffer and
swap it in on success.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A PM5 BigBuf is several hundred KiB, so its traces outgrow the 16 bit
trace length and record positions. The device already reports the
length as 32 bit in download_done_t.extra, but download_trace_ex() kept
it in a uint16_t and cut it modulo 65536, as `trace load` did for files.
Make the client side trace length and every record position 32 bit, and
drop the 64 KiB refusal from d36d821c5.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The trace buffer length and all record positions are 16 bit. `trace
load` and the spiffs import (ImportTraceBuffer took a uint16_t, its
caller passes a uint32_t) cut a longer length modulo 65536, eg a 70005
byte file loaded as 4469 bytes. Report it and keep the buffer empty.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Same order as the native branch. No change in output, the reply is a
tag frame the next pass ignores anyway.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
They were matched on the low nibble of the first byte only. A 14a CL2
SELECT, 95 70 UID BCC CRC, is 9 bytes like a CHECK and was printed as
its MACs, and a MIFARE AUTH B, 61 .., was taken as a SELECT and ate the
frame after it.
A SELECT now needs CSN[8] and a reply with a valid iCLASS CRC, which is
only skipped when it matched. A CHECK carries no CRC, so a frame with a
valid CRC_A is not one.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
They are static and were never cleared, so an iCLASS CHECK in a trace
without a SELECT gave an `hf iclass lookup` line with the CSN and epurse
of a trace extracted earlier.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Both ignored the result of download_trace(). After a timeout they went
on with whatever the client side trace buffer held, ie an older trace,
under a fresh "Recorded activity" line. Offline they printed the failed
upload error and then listed the buffer anyway.
Download only when a device is present and return its error, offline
uses the trace buffer like `trace save` does. The "removing parameter
-1" hint is only given when -1 was.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Any reader frame with the top two bits clear was searched for an auth
command at its first two INF bytes. A MIFARE Ultralight READBLOCK(10),
30 0A CRC, was reported as a native DESFire auth with the CRC byte as
key number and page data as challenge (traces/hf_14a_mfuc.trace), and
so was a native SelectApplication of an AID starting with 0A.
Require an I-block PCB, take the command after CLA 90 only when the
frame is wrapped, and the key number before the CRC. long_jmp now
follows the framing instead of guessing it from the frame length.
The switch loses one indent level, see git diff -w.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The ISO auth branch took a tag reply longer than 20 bytes as 3TDEA. A
native 3TDEA reply is 20 bytes, so it was shown as 2TDEA with half the
challenge. Use what is left after the framing, SW and CRC instead.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The OP card manager SELECT built its sAPDU_t from a static AID, so
clang emitted a constant template with an absolute pointer at offset 5
of the PACKED struct. Apple ld rejects it:
ld: pointer not aligned in 'l___const.gp_info.op'+0x5 (cmdhfgp.o)
Keep the AID on the stack, like the other sAPDU_t users.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The tag challenge of the DES, ISO and AES auth branches was read from
frame + 1. In native framing that is the AF status byte, and with a CID
it is the CID byte, so the challenge came out shifted by one, eg
traces/hf_mfdes_sniff.trace gave 000675359294E7CD instead of
0675359294E7CDA1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The MFDES AES auth branch printed 8 of the 16 RndB bytes and 16 of the
32 reader bytes, and accepted a 9 byte tag reply.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The retry check estimated the next attempt from the longest one so far.
That does not bound it:
- get_pulse_length() waits up to 4 bit periods
- one find_double_listen_window() can take about 3.2 s with no signal
- a login + read attempt can take over 6 s
- setup before tstart was not counted
A slow attempt after fast failures could end after the client gave up,
leaving the late reply queued for the next command.
read and info now set a deadline after setup, 1 s before the client
timeout. The listen window, ACK and word decode loops stop at the
deadline, as do the retries. With no deadline set (write, brute, sim,
...) they behave as before. The client waits 5 s for read, as it
already did for info / dump.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Makefile.defs hardening sets -D_FORTIFY_SOURCE=3. Under clang, glibc's
stdio2.h then defines fprintf as a macro, so the stdout hook in
deps/reveng/config.h redefined it and -Werror stopped the build in
bmpbit.c, cli.c, model.c, poly.c, preset.c and src/cmdcrc.c. Undefine
fprintf, fputs and puts before hooking them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
brew's "openssl" alias follows the newest major (now openssl@4), so
`brew --prefix openssl` returns a prefix that isn't installed on setups
with only openssl@3, and the build fails with evp.h / des.h not found.
Probe the candidate kegs and use the first that actually has the headers.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
hf gp info also tries the OpenPlatform card manager AID
A000000003000000 before falling back to the default applet. That
fallback now warns and is titled "Default applet" instead of
"Security Domain".
hf gp aids returns PM3_ESOFT when the card accepts any SELECT, instead
of PM3_SUCCESS.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>