Mbed TLS 2 returns void from mbedtls_sha1_starts/update/finish (the int
versions are the _ret variants), which the system-library build on
Ubuntu (2.28) rejects. crypto/mbedtls_compat.hpp gives one int-returning
spelling for both.
src/crypto/aead.hpp implements the SIP004/SIP007 AEAD ciphers as value
types: AeadEngine holds the libsodium or Mbed TLS GCM state by value,
AeadTcpEncryptor/AeadTcpDecryptor frame the chunk stream with a fixed
reassembly buffer, and aead_udp_seal/open handle packets. HKDF-SHA1 runs
on stack SHA-1 state instead of Mbed TLS's allocating HMAC API.
The bundled Mbed TLS config now enables MBEDTLS_BLOCK_CIPHER_C, which
keeps the AES context inside mbedtls_gcm_context, so GCM key setup no
longer allocates. That alone saves the C programs two allocations per
connection.
test_crypto_aead_cpp checks HKDF against RFC 5869 and crypto_hkdf, runs
TCP streams and UDP packets between the C and C++ implementations in
both directions for all five AEAD methods, and covers tampering,
framing and salt replay. alloc_steady_crypto_aead runs it under the
malloc counter and requires zero allocations across 600 new
connections and 600 packets (tests/alloc_audit/check_steady.py).
ss_strict_warnings() now applies -Wextra per source file so targets
mixing new C++ with converted C modules keep the latter at -Wall, and
Result gains move assignment.
src/event/loop.hpp wraps uv_loop/poll/timer/signal/async handles that
live by value inside their owner and dispatch to an owner member
function chosen at compile time. A Poll owns its socket and carries
read and write interest on one handle, closing the descriptor in
libuv's close callback. CloseGroup collects the asynchronous closes of
an owner's handles and reports completion afterwards, which is how
pooled owners implement AsyncClosable. DeadlineList drives the idle
timeouts of many owners from one timer.
test_event_cpp covers readiness, closing a peer from inside a
callback, recycling a pooled owner only after its handles closed,
repeating timers, deadline ordering and cross-thread wakeups.
ss-redir only builds on Linux, so the macOS conversion checks never
compiled it: getsockname() wants a struct sockaddr *, and memcpy's
void * result needs a cast.
Rename utils, ss_event, udprelay, winsock, local, server, tunnel, redir
and manager from .c to .cpp. Inside the moved files only the
\snippet{doc} file references and header comments change; snippet
references in doc/*.md, tests/test_cli_docs.py, the build files and
CONTRIBUTION.md follow the new names.
Also cast the Windows-only string literal returned by
get_default_conf(), found by the zig x86_64-windows-gnu build.
Every module of src/ except the unbuilt android.c and the vendored
blake3 is now compiled as C++20. The new .cpp files join the no-heap
ratchet baseline.
Cast allocator results and void* callback data explicitly, cast string
literals stored in char * variables, replace compound literals in
ss_event.c with zeroing plus member assignment, and give ss-server's
listener array its MAX_REMOTE_NUM bound instead of a VLA.
Where a goto CLEAN_UP jumps over an initialised declaration (C++
rejects that), split the declaration from its initialisation in place.
The variables were already indeterminate at the label in C, and the
cleanup code does not read them. Checked for every MODULE_* variant of
udprelay.c, plus LIB_ONLY for local.c. Still valid C.
Rename netutils, jconf, stream, aead, aead2022, aead2022_udp, crypto,
acl, plugin, plugin_disabled and resolv from .c to .cpp without changing
their contents, and update source lists, snippet references, test
comments and debian/copyright.
The Doxygen source list and scripts/check_cli_docs.py now accept either
extension for each module, so the remaining programs can move one at a
time. The new .cpp files are added to the no-heap ratchet baseline.
socketserver listens with a backlog of 5. With concurrent connections
through the proxy on a loaded machine (parallel ctest), the backlog
overflows, the kernel resets the origin connection, and the proxy
correctly drops the client, which surfaced as intermittent
alloc_audit_report failures ("getpeername: Invalid argument" in
ss-local, "Connection reset by peer" in ss-server). Reproduced
reliably at 64 concurrent connections; gone with a backlog of 256.
Cast allocator and memchr results explicitly, pass mbedtls_operation_t
to mbedtls_cipher_setkey, and declare the two locals in
ss2022_encrypt_first() before the gotos that jump past them. Drop the
unused RESOLV_MODE global that resolv.c defined by accident through an
anonymous enum declaration. Still valid C.
GCC's C++ -Wall includes -Wsign-compare and -Wclass-memaccess, which C's
-Wall does not, so the converted modules failed -Werror on GCC (Linux,
UCRT64) although nothing in them changed. They keep the warning set they
were written against; code under ss_strict_warnings() re-enables both.
Once the C headers are included from C++ translation units,
bugprone-dynamic-static-initializers reports each extern variable
declaration in them; none is defined in a header, so disable the check.
Mark the two bounded memchr searches in ssurl.cpp, which
bugprone-not-null-terminated-result misreads as string copies.
Cast allocator results and the thread argument explicitly and
value-initialise JOBOBJECT_EXTENDED_LIMIT_INFORMATION with {}. Found by
the zig x86_64-windows-gnu cross build.
Rename base64, json, ipset, process, ppbloom, rule, ssurl and cache from
.c to .cpp without touching their contents, and update the CMake source
lists, debian/copyright and the docs.
Ported code keeps -Wall; -Wextra is now opt-in through
ss_strict_warnings() for code written as C++ (the core tests).
check_no_heap.py gains a ratchet baseline for the converted modules,
which still allocate the way the C code did. A new check_cxx_runtime
test verifies with nm that the programs and libshadowsocks-c.a still
need no C++ runtime library, so C consumers are unaffected.
Wrap the declarations of every project header in extern "C" (after the
system includes, which must keep C++ linkage), so converted modules and
the remaining C files agree on linkage.
In the leaf modules, cast allocator and memchr results explicitly,
rename identifiers called new, replace the getpwnam_r VLA with the
fixed 16 KiB buffer the loop already capped itself at, and make the
rlimit initialiser conversion explicit. All of this is still valid C.
- docker/static: Alpine's libstdc++ headers come with g++, and the C++
tests' static libstdc++ needs an unwinder that compiler-rt lacks; add
g++ and llvm-libunwind.
- FixedBuffer: drop the unused kCapacity (clang-tidy 18 reports it as a
dynamically initialized header static).
- ss-server on Windows: manager_addr was only read in POSIX-only code,
which newer clang rejects (-Wunused-but-set-global); say that the
option is ignored there.
docs/cpp-core.md describes the language profile, the core primitives,
the phase plan and the measured C baseline (77-89 allocations per
connection in ss-server, 27-39 in ss-local). The compatibility contract
now keeps the C11 public header instead of a C11 code base. CLAUDE.md
drops stale libev, submodule and gnu99 references.
alloc_counter is a preloadable malloc counter (glibc, macOS) that writes
a snapshot on SIGUSR2. measure_alloc.py warms ss-local and ss-server up,
snapshots both, relays more connections and reports allocations per
connection. It runs report-only as the alloc_audit_report ctest.
interop.peers() now yields the started processes.
Header-only primitives for the no-heap-after-startup design:
Result/Status, Region, ObjectPool with PoolPtr (std::unique_ptr with a
pool deleter), generation-checked Handle and asynchronous close support,
intrusive RefPtr, FixedBuffer, MonotonicArena, FixedHashMap and FixedLru,
IntrusiveList, and a bounded FixedHeap for third-party allocator hooks.
Unit tests cover each primitive (test_core_*). scripts/check_no_heap.py
rejects heap allocation in C++ sources under src/ and runs as a ctest.
clang-tidy now also analyses C++ sources and the core headers.
Add CXX to the project with C++20 (no extensions) and build C++ sources
with -Wextra -fno-exceptions -fno-rtti -fno-threadsafe-statics on top of
the existing warning flags. Give every toolchain a C++ compiler: the zig
Windows toolchain, the static macOS/Windows CI jobs, the static musl
image (clang++ plus libstdc++ headers) and the RPM spec.
Add `ss-local --printable-salt` / `"printable_salt": true`, disabled by
default. For each outgoing TCP salt it picks a prefix length uniformly
from 6-12 bytes and samples those bytes as printable ASCII (0x20-0x7e)
immediately before key derivation. The salt stays 32 bytes with no
framing or markers, so the wire format is unchanged and any stock
server interoperates. Only chacha20-ietf-poly1305 and aes-256-gcm are
accepted; other ciphers fail at startup. UDP, server responses, and
AEAD-2022 are untouched.
Crypto unit tests cover length distribution, empty first writes, exact
framing overhead, salt-tail preservation, byte-at-a-time decryption,
tampering and replay rejection. CLI tests check local-only exposure and
cipher rejection via flag, config key and bad config type. interop.py
gains --printable-salt and --server-bin for independent stock servers.
docs/printable-salt.md records the design, entropy bound, fingerprint
caveats and route-experiment procedure; docs/encoding-experiments.md
and tests/experiments/ hold the Base85/lowpop85/TLS outer-transport
prototypes. docs/measurements/printable-salt-2026-09-23/ archives the
sanitized results of the aborted field trials.
Co-authored-by: Claude <noreply@anthropic.com>
test_event opens 256 UDP receivers plus a sender on top of the loop's own
descriptors. macOS starts processes with a soft RLIMIT_NOFILE of 256, so
socket() eventually returned -1 and the unchecked descriptor tripped the
bind() assertion when running ctest from a stock terminal (#3058).
Wrap every registered unit test in a small sh launcher that raises the soft
limit to 1024 before exec'ing the binary, and have test_event raise its own
limit as well so it also passes when run directly. Assert on socket()
failures so a limit problem is reported as such instead of as a bind error.
Fixes#3058
Claude-Session: https://claude.ai/code/session_01PSG4XV3jiqitDgdgFSGATx
Document the shadowsocks-libev history and shadowsocks-c rename, and recommend Docker installation with mounted configuration and TCP/UDP port mappings.
Add native AMD64/ARM64 builds of the unprivileged static runtime image, real container relay and shutdown tests, and GHCR publication after both architecture checks pass. All 24 build/test checks pass at reviewed head 3f3f6bf8e01e5614821d1e15fda3ad8f07812902; publication is intentionally skipped on pull requests.
Rename the project to shadowsocks-c while retaining ss-* commands, the embedding ABI, and legacy library/package lookup compatibility.
Replace libev with bundled libuv for IOCP/kqueue event backends, expand asynchronous runtime DNS coverage, and add actionlint/Ruff with strict clang-tidy failure handling. Validate native platforms, static builds, packaging, interoperability, sanitizers, and canonical/legacy consumers.
All 22 hosted checks pass at reviewed head a8493250eb.
Bundle pinned offline dependencies, remove libcork/libipset and submodule requirements, and add portable runtime helpers with relocatable static/shared library installations.
Add Clang static build validation for Linux/musl, macOS and Windows, real TCP/UDP and SIP003 interoperability coverage, and recorded performance tradeoffs. All 21 hosted checks pass at the reviewed PR head 8d504e6218.
crypto_init() initializes the nonce bloom filter on every call, so a test
that builds a crypto_t per case has to tear it down per case as well.
Freeing only once at the end leaked a pair of filters for every crypto_t
after the first, which the valgrind memcheck job caught.
test_crypto.c already frees it alongside each crypto_t; do the same here
by folding it into the local free_crypto() helper.
shadowsocks-rust takes a whole server definition as a URL; libev could
only be given the pieces separately, so importing a server meant taking a
URL apart by hand. ss-local now accepts --server-url and reads the
address, port, cipher, password and SIP003 plugin from it.
Both encodings are handled: the SIP002 form with the userinfo in
base64url, and the older form where the whole body is one base64 blob.
The userinfo is also accepted in the clear, because that is what
shadowsocks-rust's `ssurl` emits for the 2022 ciphers, whose base64
pre-shared keys it percent-encodes instead. IPv6 literals, a trailing
slash, the plugin query and the '#' tag are all covered.
Options given later on the command line still win, since they are applied
as getopt reaches them.
ssurl.c sits with the ss-local sources rather than the shared ones: it
depends on base64.c, which ss-manager does not link.
The tests include URLs copied verbatim from `ssurl --encode` output, so
the parser is pinned against what another implementation really produces
rather than only against my reading of the spec. Verified end to end by
pointing ss-local at a rust-generated URL and proxying through their
ssserver.
A cipher can round-trip perfectly against our own implementation and
still be wrong: header layout, padding rules, and nonce ordering are only
pinned down by talking to another implementation. This drives ss-local
against rust's ssserver and rust's sslocal against ss-server, over a
local origin server so it needs no network access.
Covers the three SIP022 methods plus aes-256-gcm, so a regression in the
legacy path shows up here too. Skips itself when shadowsocks-rust or the
built binaries are absent, so it is safe to run anywhere.
The buffers were all large enough for a uint16, so none of these could
actually overflow, but sprintf gives the compiler and the linters nothing
to check. Passing the destination size makes that guarantee explicit and
survives any later change to the buffers.
parse_udprelay_header writes through a caller-supplied pointer, so it
uses the MAX_PORT_STR_LEN contract its callers already allocate to.
crypto_parse_key sized a stack array from strlen() of the key given on
the command line or in the config file, so a long enough key string
overflowed the stack before any validation ran. Only the first key_len
bytes were ever used, and base64_decode already stops at the size it is
given, so a fixed MAX_KEY_LENGTH buffer is enough. crypto_parse_psk
needs one byte of slack to tell an over-long key from an exact-size one.
The two "here is a valid key instead" error paths were identical; they
are now one helper, which also drops the last input-sized VLA.
Switch the standard from gnu99 to C11 via CMAKE_C_STANDARD rather than a
hand-written -std flag, so CMake picks the right spelling per compiler.
Max Lv
