clang-tools 15.0.0 appears to have added a new warning that will always complain about assignments in if statements, which we use in a number of places in passt/pasta. Encountered on Fedora 37 with clang-tools-extra-15.0.0-3.fc37.x86_64. Suppress the new warning so that we can compile and test. Signed-off-by: David Gibson --- Makefile | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/Makefile b/Makefile index 1dc2df5..cc4f014 100644 --- a/Makefile +++ b/Makefile @@ -259,11 +259,16 @@ docs: README.md # - readability-identifier-length # Complains about any identifier <3 characters, reasonable for # globals, pointlessly verbose for locals and parameters. +# +# - bugprone-assignment-in-if-condition +# Dubious value over the compiler's built-in warning. Would +# increase verbosity. clang-tidy: $(SRCS) $(HEADERS) clang-tidy -checks=*,-modernize-*,\ -clang-analyzer-valist.Uninitialized,\ -cppcoreguidelines-init-variables,\ + -bugprone-assignment-in-if-condition,\ -bugprone-macro-parentheses,\ -google-readability-braces-around-statements,\ -hicpp-braces-around-statements,\ -- 2.38.1

This obvious include was omitted, which means that declarations in the header weren't checked against definitions in the .c file. This shows up an old declaration for a function that is now static, and a duplicate #define. Signed-off-by: David Gibson --- tcp_splice.c | 2 +- tcp_splice.h | 3 --- 2 files changed, 1 insertion(+), 4 deletions(-) diff --git a/tcp_splice.c b/tcp_splice.c index 99c5fa7..3c5c111 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -49,9 +49,9 @@ #include "util.h" #include "passt.h" #include "log.h" +#include "tcp_splice.h" #define MAX_PIPE_SIZE (8UL * 1024 * 1024) -#define TCP_SPLICE_MAX_CONNS (128 * 1024) #define TCP_SPLICE_PIPE_POOL_SIZE 16 #define TCP_SPLICE_CONN_PRESSURE 30 /* % of splice_conn_count */ #define TCP_SPLICE_FILE_PRESSURE 30 /* % of c->nofile */ diff --git a/tcp_splice.h b/tcp_splice.h index 63ffc68..2c4bff3 100644 --- a/tcp_splice.h +++ b/tcp_splice.h @@ -8,11 +8,8 @@ #define TCP_SPLICE_MAX_CONNS (128 * 1024) -struct tcp_splice_conn; - void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, uint32_t events); -void tcp_splice_destroy(struct ctx *c, struct tcp_splice_conn *conn); void tcp_splice_init(struct ctx *c); void tcp_splice_timer(struct ctx *c); void tcp_splice_defer_handler(struct ctx *c); -- 2.38.1

The macro CONN_OR_NULL() is used to look up connections by index with bounds checking. Replace it with an inline function, which means: - Better type checking - No danger of multiple evaluation of an @index with side effects Also add a helper to perform the reverse translation: from connection pointer to index. Introduce a macro for this which will make later cleanups easier and safer. Signed-off-by: David Gibson --- tcp.c | 83 ++++++++++++++++++++++++++++++++--------------------------- 1 file changed, 45 insertions(+), 38 deletions(-) diff --git a/tcp.c b/tcp.c index d043123..34d7d45 100644 --- a/tcp.c +++ b/tcp.c @@ -518,14 +518,6 @@ struct tcp_conn { (conn->events & (SOCK_FIN_RCVD | TAP_FIN_RCVD))) #define CONN_HAS(conn, set) ((conn->events & (set)) == (set)) -#define CONN(index) (tc + (index)) - -/* We probably don't want to use gcc statement expressions (for portability), so - * use this only after well-defined sequence points (no pre-/post-increments). - */ -#define CONN_OR_NULL(index) \ - (((int)(index) >= 0 && (index) < TCP_MAX_CONNS) ? (tc + (index)) : NULL) - static const char *tcp_event_str[] __attribute((__unused__)) = { "SOCK_ACCEPTED", "TAP_SYN_RCVD", "ESTABLISHED", "TAP_SYN_ACK_SENT", @@ -705,6 +697,21 @@ static size_t tcp6_l2_flags_buf_bytes; /* TCP connections */ static struct tcp_conn tc[TCP_MAX_CONNS]; +#define CONN(index) (tc + (index)) +#define CONN_IDX(conn) ((conn) - tc) + +/** conn_at_idx() - Find a connection by index, if present + * @index: Index of connection to lookup + * + * Return: pointer to connection, or NULL if @index is out of bounds + */ +static inline struct tcp_conn *conn_at_idx(int index) +{ + if ((index < 0) || (index >= TCP_MAX_CONNS)) + return NULL; + return CONN(index); +} + /* Table for lookup from remote address, local port, remote port */ static struct tcp_conn *tc_hash[TCP_HASH_TABLE_SIZE]; @@ -761,7 +768,7 @@ static int tcp_epoll_ctl(const struct ctx *c, struct tcp_conn *conn) { int m = (conn->flags & IN_EPOLL) ? EPOLL_CTL_MOD : EPOLL_CTL_ADD; union epoll_ref ref = { .r.proto = IPPROTO_TCP, .r.s = conn->sock, - .r.p.tcp.tcp.index = conn - tc, + .r.p.tcp.tcp.index = CONN_IDX(conn), .r.p.tcp.tcp.v6 = CONN_V6(conn) }; struct epoll_event ev = { .data.u64 = ref.u64 }; @@ -784,7 +791,7 @@ static int tcp_epoll_ctl(const struct ctx *c, struct tcp_conn *conn) union epoll_ref ref_t = { .r.proto = IPPROTO_TCP, .r.s = conn->sock, .r.p.tcp.tcp.timer = 1, - .r.p.tcp.tcp.index = conn - tc }; + .r.p.tcp.tcp.index = CONN_IDX(conn) }; struct epoll_event ev_t = { .data.u64 = ref_t.u64, .events = EPOLLIN | EPOLLET }; @@ -813,7 +820,7 @@ static void tcp_timer_ctl(const struct ctx *c, struct tcp_conn *conn) union epoll_ref ref = { .r.proto = IPPROTO_TCP, .r.s = conn->sock, .r.p.tcp.tcp.timer = 1, - .r.p.tcp.tcp.index = conn - tc }; + .r.p.tcp.tcp.index = CONN_IDX(conn) }; struct epoll_event ev = { .data.u64 = ref.u64, .events = EPOLLIN | EPOLLET }; int fd; @@ -846,7 +853,7 @@ static void tcp_timer_ctl(const struct ctx *c, struct tcp_conn *conn) it.it_value.tv_sec = ACT_TIMEOUT; } - debug("TCP: index %li, timer expires in %lu.%03lus", conn - tc, + debug("TCP: index %li, timer expires in %lu.%03lus", CONN_IDX(conn), it.it_value.tv_sec, it.it_value.tv_nsec / 1000 / 1000); timerfd_settime(conn->timer, 0, &it, NULL); @@ -867,7 +874,7 @@ static void conn_flag_do(const struct ctx *c, struct tcp_conn *conn, conn->flags &= flag; if (fls(~flag) >= 0) { - debug("TCP: index %li: %s dropped", conn - tc, + debug("TCP: index %li: %s dropped", CONN_IDX(conn), tcp_flag_str[fls(~flag)]); } } else { @@ -876,7 +883,7 @@ static void conn_flag_do(const struct ctx *c, struct tcp_conn *conn, conn->flags |= flag; if (fls(flag) >= 0) { - debug("TCP: index %li: %s", conn - tc, + debug("TCP: index %li: %s", CONN_IDX(conn), tcp_flag_str[fls(flag)]); } } @@ -926,12 +933,12 @@ static void conn_event_do(const struct ctx *c, struct tcp_conn *conn, new += 5; if (prev != new) { - debug("TCP: index %li, %s: %s -> %s", conn - tc, + debug("TCP: index %li, %s: %s -> %s", CONN_IDX(conn), num == -1 ? "CLOSED" : tcp_event_str[num], prev == -1 ? "CLOSED" : tcp_state_str[prev], (new == -1 || num == -1) ? "CLOSED" : tcp_state_str[new]); } else { - debug("TCP: index %li, %s", conn - tc, + debug("TCP: index %li, %s", CONN_IDX(conn), num == -1 ? "CLOSED" : tcp_event_str[num]); } @@ -1355,12 +1362,12 @@ static void tcp_hash_insert(const struct ctx *c, struct tcp_conn *conn, int b; b = tcp_hash(c, af, addr, conn->tap_port, conn->sock_port); - conn->next_index = tc_hash[b] ? tc_hash[b] - tc : -1; + conn->next_index = tc_hash[b] ? CONN_IDX(tc_hash[b]) : -1; tc_hash[b] = conn; conn->hash_bucket = b; debug("TCP: hash table insert: index %li, sock %i, bucket: %i, next: " - "%p", conn - tc, conn->sock, b, CONN_OR_NULL(conn->next_index)); + "%p", CONN_IDX(conn), conn->sock, b, conn_at_idx(conn->next_index)); } /** @@ -1373,19 +1380,19 @@ static void tcp_hash_remove(const struct tcp_conn *conn) int b = conn->hash_bucket; for (entry = tc_hash[b]; entry; - prev = entry, entry = CONN_OR_NULL(entry->next_index)) { + prev = entry, entry = conn_at_idx(entry->next_index)) { if (entry == conn) { if (prev) prev->next_index = conn->next_index; else - tc_hash[b] = CONN_OR_NULL(conn->next_index); + tc_hash[b] = conn_at_idx(conn->next_index); break; } } debug("TCP: hash table remove: index %li, sock %i, bucket: %i, new: %p", - conn - tc, conn->sock, b, - prev ? CONN_OR_NULL(prev->next_index) : tc_hash[b]); + CONN_IDX(conn), conn->sock, b, + prev ? conn_at_idx(prev->next_index) : tc_hash[b]); } /** @@ -1399,10 +1406,10 @@ static void tcp_hash_update(struct tcp_conn *old, struct tcp_conn *new) int b = old->hash_bucket; for (entry = tc_hash[b]; entry; - prev = entry, entry = CONN_OR_NULL(entry->next_index)) { + prev = entry, entry = conn_at_idx(entry->next_index)) { if (entry == old) { if (prev) - prev->next_index = new - tc; + prev->next_index = CONN_IDX(new); else tc_hash[b] = new; break; @@ -1411,7 +1418,7 @@ static void tcp_hash_update(struct tcp_conn *old, struct tcp_conn *new) debug("TCP: hash table update: old index %li, new index %li, sock %i, " "bucket: %i, old: %p, new: %p", - old - tc, new - tc, new->sock, b, old, new); + CONN_IDX(old), CONN_IDX(new), new->sock, b, old, new); } /** @@ -1431,7 +1438,7 @@ static struct tcp_conn *tcp_hash_lookup(const struct ctx *c, int af, int b = tcp_hash(c, af, addr, tap_port, sock_port); struct tcp_conn *conn; - for (conn = tc_hash[b]; conn; conn = CONN_OR_NULL(conn->next_index)) { + for (conn = tc_hash[b]; conn; conn = conn_at_idx(conn->next_index)) { if (tcp_hash_match(conn, af, addr, tap_port, sock_port)) return conn; } @@ -1448,9 +1455,9 @@ static void tcp_table_compact(struct ctx *c, struct tcp_conn *hole) { struct tcp_conn *from, *to; - if ((hole - tc) == --c->tcp.conn_count) { + if (CONN_IDX(hole) == --c->tcp.conn_count) { debug("TCP: hash table compaction: maximum index was %li (%p)", - hole - tc, hole); + CONN_IDX(hole), hole); memset(hole, 0, sizeof(*hole)); return; } @@ -1465,7 +1472,7 @@ static void tcp_table_compact(struct ctx *c, struct tcp_conn *hole) debug("TCP: hash table compaction: old index %li, new index %li, " "sock %i, from: %p, to: %p", - from - tc, to - tc, from->sock, from, to); + CONN_IDX(from), CONN_IDX(to), from->sock, from, to); memset(from, 0, sizeof(*from)); } @@ -1488,7 +1495,7 @@ static void tcp_conn_destroy(struct ctx *c, struct tcp_conn *conn) static void tcp_rst_do(struct ctx *c, struct tcp_conn *conn); #define tcp_rst(c, conn) \ do { \ - debug("TCP: index %li, reset at %s:%i", conn - tc, \ + debug("TCP: index %li, reset at %s:%i", CONN_IDX(conn), \ __func__, __LINE__); \ tcp_rst_do(c, conn); \ } while (0) @@ -2734,7 +2741,7 @@ int tcp_tap_handler(struct ctx *c, int af, const void *addr, return 1; } - trace("TCP: packet length %lu from tap for index %lu", len, conn - tc); + trace("TCP: packet length %lu from tap for index %lu", len, CONN_IDX(conn)); if (th->rst) { conn_event(c, conn, CLOSED); @@ -2942,7 +2949,7 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, */ static void tcp_timer_handler(struct ctx *c, union epoll_ref ref) { - struct tcp_conn *conn = CONN_OR_NULL(ref.r.p.tcp.tcp.index); + struct tcp_conn *conn = conn_at_idx(ref.r.p.tcp.tcp.index); struct itimerspec check_armed = { { 0 }, { 0 } }; if (!conn) @@ -2961,17 +2968,17 @@ static void tcp_timer_handler(struct ctx *c, union epoll_ref ref) conn_flag(c, conn, ~ACK_TO_TAP_DUE); } else if (conn->flags & ACK_FROM_TAP_DUE) { if (!(conn->events & ESTABLISHED)) { - debug("TCP: index %li, handshake timeout", conn - tc); + debug("TCP: index %li, handshake timeout", CONN_IDX(conn)); tcp_rst(c, conn); } else if (CONN_HAS(conn, SOCK_FIN_SENT | TAP_FIN_ACKED)) { - debug("TCP: index %li, FIN timeout", conn - tc); + debug("TCP: index %li, FIN timeout", CONN_IDX(conn)); tcp_rst(c, conn); } else if (conn->retrans == TCP_MAX_RETRANS) { debug("TCP: index %li, retransmissions count exceeded", - conn - tc); + CONN_IDX(conn)); tcp_rst(c, conn); } else { - debug("TCP: index %li, ACK timeout, retry", conn - tc); + debug("TCP: index %li, ACK timeout, retry", CONN_IDX(conn)); conn->retrans++; conn->seq_to_tap = conn->seq_ack_from_tap; tcp_data_from_sock(c, conn); @@ -2989,7 +2996,7 @@ static void tcp_timer_handler(struct ctx *c, union epoll_ref ref) */ timerfd_settime(conn->timer, 0, &new, &old); if (old.it_value.tv_sec == ACT_TIMEOUT) { - debug("TCP: index %li, activity timeout", conn - tc); + debug("TCP: index %li, activity timeout", CONN_IDX(conn)); tcp_rst(c, conn); } } @@ -3022,7 +3029,7 @@ void tcp_sock_handler(struct ctx *c, union epoll_ref ref, uint32_t events, return; } - if (!(conn = CONN_OR_NULL(ref.r.p.tcp.tcp.index))) + if (!(conn = conn_at_idx(ref.r.p.tcp.tcp.index))) return; if (conn->events == CLOSED) -- 2.38.1

Currently spliced and non-spliced connections use completely independent tracking structures. We want to unify these, so as a preliminary step move the definitions for both variants into a new tcp_conn.h header, shared by tcp.c and tcp_splice.c. This requires renaming some #defines with the same name but different meanings between the two cases. In the process we correct some places that are slightly out of sync between the comments and the code for various event bit names. Signed-off-by: David Gibson --- Makefile | 3 +- tcp.c | 206 +++++++++++++-------------------------------------- tcp_conn.h | 168 +++++++++++++++++++++++++++++++++++++++++ tcp_splice.c | 93 +++++++---------------- 4 files changed, 245 insertions(+), 225 deletions(-) create mode 100644 tcp_conn.h diff --git a/Makefile b/Makefile index cc4f014..a9b4f79 100644 --- a/Makefile +++ b/Makefile @@ -47,7 +47,8 @@ MANPAGES = passt.1 pasta.1 qrap.1 PASST_HEADERS = arch.h arp.h checksum.h conf.h dhcp.h dhcpv6.h icmp.h \ isolation.h lineread.h log.h ndp.h netlink.h packet.h passt.h pasta.h \ - pcap.h port_fwd.h siphash.h tap.h tcp.h tcp_splice.h udp.h util.h + pcap.h port_fwd.h siphash.h tap.h tcp.h tcp_conn.h tcp_splice.h udp.h \ + util.h HEADERS = $(PASST_HEADERS) seccomp.h # On gcc 11 and 12, with -O2 and -flto, tcp_hash() and siphash_20b(), if diff --git a/tcp.c b/tcp.c index 34d7d45..189041c 100644 --- a/tcp.c +++ b/tcp.c @@ -98,7 +98,7 @@ * Connection tracking and storage * ------------------------------- * - * Connections are tracked by the @tc array of struct tcp_conn, containing + * Connections are tracked by the @tc array of struct tcp_tap_conn, containing * addresses, ports, TCP states and parameters. This is statically allocated and * indexed by an arbitrary connection number. The array is compacted whenever a * connection is closed, by remapping the highest connection index in use to the @@ -301,6 +301,8 @@ #include "tcp_splice.h" #include "log.h" +#include "tcp_conn.h" + #define TCP_FRAMES_MEM 128 #define TCP_FRAMES \ (c->mode == MODE_PASST ? TCP_FRAMES_MEM : 1) @@ -308,7 +310,6 @@ #define TCP_FILE_PRESSURE 30 /* % of c->nofile */ #define TCP_CONN_PRESSURE 30 /* % of c->tcp.conn_count */ -#define TCP_HASH_BUCKET_BITS (TCP_CONN_INDEX_BITS + 1) #define TCP_HASH_TABLE_LOAD 70 /* % */ #define TCP_HASH_TABLE_SIZE (TCP_MAX_CONNS * 100 / \ TCP_HASH_TABLE_LOAD) @@ -402,117 +403,8 @@ struct tcp6_l2_head { /* For MSS6 macro: keep in sync with tcp6_l2_buf_t */ #define OPT_SACK 5 #define OPT_TS 8 -/** - * struct tcp_conn - Descriptor for a TCP connection (not spliced) - * @next_index: Connection index of next item in hash chain, -1 for none - * @tap_mss: MSS advertised by tap/guest, rounded to 2 ^ TCP_MSS_BITS - * @sock: Socket descriptor number - * @events: Connection events, implying connection states - * @timer: timerfd descriptor for timeout events - * @flags: Connection flags representing internal attributes - * @hash_bucket: Bucket index in connection lookup hash table - * @retrans: Number of retransmissions occurred due to ACK_TIMEOUT - * @ws_from_tap: Window scaling factor advertised from tap/guest - * @ws_to_tap: Window scaling factor advertised to tap/guest - * @sndbuf: Sending buffer in kernel, rounded to 2 ^ SNDBUF_BITS - * @seq_dup_ack_approx: Last duplicate ACK number sent to tap - * @a.a6: IPv6 remote address, can be IPv4-mapped - * @a.a4.zero: Zero prefix for IPv4-mapped, see RFC 6890, Table 20 - * @a.a4.one: Ones prefix for IPv4-mapped - * @a.a4.a: IPv4 address - * @tap_port: Guest-facing tap port - * @sock_port: Remote, socket-facing port - * @wnd_from_tap: Last window size from tap, unscaled (as received) - * @wnd_to_tap: Sending window advertised to tap, unscaled (as sent) - * @seq_to_tap: Next sequence for packets to tap - * @seq_ack_from_tap: Last ACK number received from tap - * @seq_from_tap: Next sequence for packets from tap (not actually sent) - * @seq_ack_to_tap: Last ACK number sent to tap - * @seq_init_from_tap: Initial sequence number from tap - */ -struct tcp_conn { - int next_index :TCP_CONN_INDEX_BITS + 2; - -#define TCP_RETRANS_BITS 3 - unsigned int retrans :TCP_RETRANS_BITS; -#define TCP_MAX_RETRANS ((1U << TCP_RETRANS_BITS) - 1) - -#define TCP_WS_BITS 4 /* RFC 7323 */ -#define TCP_WS_MAX 14 - unsigned int ws_from_tap :TCP_WS_BITS; - unsigned int ws_to_tap :TCP_WS_BITS; - - - int sock :SOCKET_REF_BITS; - - uint8_t events; -#define CLOSED 0 -#define SOCK_ACCEPTED BIT(0) /* implies SYN sent to tap */ -#define TAP_SYN_RCVD BIT(1) /* implies socket connecting */ -#define TAP_SYN_ACK_SENT BIT( 3) /* implies socket connected */ -#define ESTABLISHED BIT(2) -#define SOCK_FIN_RCVD BIT( 3) -#define SOCK_FIN_SENT BIT( 4) -#define TAP_FIN_RCVD BIT( 5) -#define TAP_FIN_SENT BIT( 6) -#define TAP_FIN_ACKED BIT( 7) - -#define CONN_STATE_BITS /* Setting these clears other flags */ \ - (SOCK_ACCEPTED | TAP_SYN_RCVD | ESTABLISHED) - - - int timer :SOCKET_REF_BITS; - - uint8_t flags; -#define STALLED BIT(0) -#define LOCAL BIT(1) -#define WND_CLAMPED BIT(2) -#define IN_EPOLL BIT(3) -#define ACTIVE_CLOSE BIT(4) -#define ACK_TO_TAP_DUE BIT(5) -#define ACK_FROM_TAP_DUE BIT(6) - - - unsigned int hash_bucket :TCP_HASH_BUCKET_BITS; - -#define TCP_MSS_BITS 14 - unsigned int tap_mss :TCP_MSS_BITS; -#define MSS_SET(conn, mss) (conn->tap_mss = (mss >> (16 - TCP_MSS_BITS))) -#define MSS_GET(conn) (conn->tap_mss << (16 - TCP_MSS_BITS)) - - -#define SNDBUF_BITS 24 - unsigned int sndbuf :SNDBUF_BITS; -#define SNDBUF_SET(conn, bytes) (conn->sndbuf = ((bytes) >> (32 - SNDBUF_BITS))) -#define SNDBUF_GET(conn) (conn->sndbuf << (32 - SNDBUF_BITS)) - - uint8_t seq_dup_ack_approx; - - - union { - struct in6_addr a6; - struct { - uint8_t zero[10]; - uint8_t one[2]; - struct in_addr a; - } a4; - } a; #define CONN_V4(conn) IN6_IS_ADDR_V4MAPPED(&conn->a.a6) #define CONN_V6(conn) (!CONN_V4(conn)) - - in_port_t tap_port; - in_port_t sock_port; - - uint16_t wnd_from_tap; - uint16_t wnd_to_tap; - - uint32_t seq_to_tap; - uint32_t seq_ack_from_tap; - uint32_t seq_from_tap; - uint32_t seq_ack_to_tap; - uint32_t seq_init_from_tap; -}; - #define CONN_IS_CLOSING(conn) \ ((conn->events & ESTABLISHED) && \ (conn->events & (SOCK_FIN_RCVD | TAP_FIN_RCVD))) @@ -695,7 +587,7 @@ static unsigned int tcp6_l2_flags_buf_used; static size_t tcp6_l2_flags_buf_bytes; /* TCP connections */ -static struct tcp_conn tc[TCP_MAX_CONNS]; +static struct tcp_tap_conn tc[TCP_MAX_CONNS]; #define CONN(index) (tc + (index)) #define CONN_IDX(conn) ((conn) - tc) @@ -705,7 +597,7 @@ static struct tcp_conn tc[TCP_MAX_CONNS]; * * Return: pointer to connection, or NULL if @index is out of bounds */ -static inline struct tcp_conn *conn_at_idx(int index) +static inline struct tcp_tap_conn *conn_at_idx(int index) { if ((index < 0) || (index >= TCP_MAX_CONNS)) return NULL; @@ -713,7 +605,7 @@ static inline struct tcp_conn *conn_at_idx(int index) } /* Table for lookup from remote address, local port, remote port */ -static struct tcp_conn *tc_hash[TCP_HASH_TABLE_SIZE]; +static struct tcp_tap_conn *tc_hash[TCP_HASH_TABLE_SIZE]; /* Pools for pre-opened sockets */ int init_sock_pool4 [TCP_SOCK_POOL_SIZE]; @@ -749,7 +641,7 @@ static uint32_t tcp_conn_epoll_events(uint8_t events, uint8_t conn_flags) return EPOLLRDHUP; } -static void conn_flag_do(const struct ctx *c, struct tcp_conn *conn, +static void conn_flag_do(const struct ctx *c, struct tcp_tap_conn *conn, unsigned long flag); #define conn_flag(c, conn, flag) \ do { \ @@ -764,7 +656,7 @@ static void conn_flag_do(const struct ctx *c, struct tcp_conn *conn, * * Return: 0 on success, negative error code on failure (not on deletion) */ -static int tcp_epoll_ctl(const struct ctx *c, struct tcp_conn *conn) +static int tcp_epoll_ctl(const struct ctx *c, struct tcp_tap_conn *conn) { int m = (conn->flags & IN_EPOLL) ? EPOLL_CTL_MOD : EPOLL_CTL_ADD; union epoll_ref ref = { .r.proto = IPPROTO_TCP, .r.s = conn->sock, @@ -809,7 +701,7 @@ static int tcp_epoll_ctl(const struct ctx *c, struct tcp_conn *conn) * * #syscalls timerfd_create timerfd_settime */ -static void tcp_timer_ctl(const struct ctx *c, struct tcp_conn *conn) +static void tcp_timer_ctl(const struct ctx *c, struct tcp_tap_conn *conn) { struct itimerspec it = { { 0 }, { 0 } }; @@ -865,7 +757,7 @@ static void tcp_timer_ctl(const struct ctx *c, struct tcp_conn *conn) * @conn: Connection pointer * @flag: Flag to set, or ~flag to unset */ -static void conn_flag_do(const struct ctx *c, struct tcp_conn *conn, +static void conn_flag_do(const struct ctx *c, struct tcp_tap_conn *conn, unsigned long flag) { if (flag & (flag - 1)) { @@ -903,7 +795,7 @@ static void conn_flag_do(const struct ctx *c, struct tcp_conn *conn, * @conn: Connection pointer * @event: Connection event */ -static void conn_event_do(const struct ctx *c, struct tcp_conn *conn, +static void conn_event_do(const struct ctx *c, struct tcp_tap_conn *conn, unsigned long event) { int prev, new, num = fls(event); @@ -963,7 +855,7 @@ static void conn_event_do(const struct ctx *c, struct tcp_conn *conn, * * Return: 1 if destination is in low RTT table, 0 otherwise */ -static int tcp_rtt_dst_low(const struct tcp_conn *conn) +static int tcp_rtt_dst_low(const struct tcp_tap_conn *conn) { int i; @@ -979,7 +871,7 @@ static int tcp_rtt_dst_low(const struct tcp_conn *conn) * @conn: Connection pointer * @tinfo: Pointer to struct tcp_info for socket */ -static void tcp_rtt_dst_check(const struct tcp_conn *conn, +static void tcp_rtt_dst_check(const struct tcp_tap_conn *conn, const struct tcp_info *tinfo) { #ifdef HAS_MIN_RTT @@ -1016,7 +908,7 @@ static void tcp_rtt_dst_check(const struct tcp_conn *conn, * tcp_get_sndbuf() - Get, scale SO_SNDBUF between thresholds (1 to 0.5 usage) * @conn: Connection pointer */ -static void tcp_get_sndbuf(struct tcp_conn *conn) +static void tcp_get_sndbuf(struct tcp_tap_conn *conn) { int s = conn->sock, sndbuf; socklen_t sl; @@ -1290,7 +1182,8 @@ static int tcp_opt_get(const char *opts, size_t len, uint8_t type_find, * * Return: 1 on match, 0 otherwise */ -static int tcp_hash_match(const struct tcp_conn *conn, int af, const void *addr, +static int tcp_hash_match(const struct tcp_tap_conn *conn, + int af, const void *addr, in_port_t tap_port, in_port_t sock_port) { if (af == AF_INET && CONN_V4(conn) && @@ -1356,7 +1249,7 @@ static unsigned int tcp_hash(const struct ctx *c, int af, const void *addr, * @af: Address family, AF_INET or AF_INET6 * @addr: Remote address, pointer to in_addr or in6_addr */ -static void tcp_hash_insert(const struct ctx *c, struct tcp_conn *conn, +static void tcp_hash_insert(const struct ctx *c, struct tcp_tap_conn *conn, int af, const void *addr) { int b; @@ -1374,9 +1267,9 @@ static void tcp_hash_insert(const struct ctx *c, struct tcp_conn *conn, * tcp_hash_remove() - Drop connection from hash table, chain unlink * @conn: Connection pointer */ -static void tcp_hash_remove(const struct tcp_conn *conn) +static void tcp_hash_remove(const struct tcp_tap_conn *conn) { - struct tcp_conn *entry, *prev = NULL; + struct tcp_tap_conn *entry, *prev = NULL; int b = conn->hash_bucket; for (entry = tc_hash[b]; entry; @@ -1400,9 +1293,9 @@ static void tcp_hash_remove(const struct tcp_conn *conn) * @old: Old connection pointer * @new: New connection pointer */ -static void tcp_hash_update(struct tcp_conn *old, struct tcp_conn *new) +static void tcp_hash_update(struct tcp_tap_conn *old, struct tcp_tap_conn *new) { - struct tcp_conn *entry, *prev = NULL; + struct tcp_tap_conn *entry, *prev = NULL; int b = old->hash_bucket; for (entry = tc_hash[b]; entry; @@ -1431,12 +1324,13 @@ static void tcp_hash_update(struct tcp_conn *old, struct tcp_conn *new) * * Return: connection pointer, if found, -ENOENT otherwise */ -static struct tcp_conn *tcp_hash_lookup(const struct ctx *c, int af, - const void *addr, - in_port_t tap_port, in_port_t sock_port) +static struct tcp_tap_conn *tcp_hash_lookup(const struct ctx *c, + int af, const void *addr, + in_port_t tap_port, + in_port_t sock_port) { int b = tcp_hash(c, af, addr, tap_port, sock_port); - struct tcp_conn *conn; + struct tcp_tap_conn *conn; for (conn = tc_hash[b]; conn; conn = conn_at_idx(conn->next_index)) { if (tcp_hash_match(conn, af, addr, tap_port, sock_port)) @@ -1451,9 +1345,9 @@ static struct tcp_conn *tcp_hash_lookup(const struct ctx *c, int af, * @c: Execution context * @hole: Pointer to recently closed connection */ -static void tcp_table_compact(struct ctx *c, struct tcp_conn *hole) +static void tcp_table_compact(struct ctx *c, struct tcp_tap_conn *hole) { - struct tcp_conn *from, *to; + struct tcp_tap_conn *from, *to; if (CONN_IDX(hole) == --c->tcp.conn_count) { debug("TCP: hash table compaction: maximum index was %li (%p)", @@ -1482,7 +1376,7 @@ static void tcp_table_compact(struct ctx *c, struct tcp_conn *hole) * @c: Execution context * @conn: Connection pointer */ -static void tcp_conn_destroy(struct ctx *c, struct tcp_conn *conn) +static void tcp_conn_destroy(struct ctx *c, struct tcp_tap_conn *conn) { close(conn->sock); if (conn->timer != -1) @@ -1492,7 +1386,7 @@ static void tcp_conn_destroy(struct ctx *c, struct tcp_conn *conn) tcp_table_compact(c, conn); } -static void tcp_rst_do(struct ctx *c, struct tcp_conn *conn); +static void tcp_rst_do(struct ctx *c, struct tcp_tap_conn *conn); #define tcp_rst(c, conn) \ do { \ debug("TCP: index %li, reset at %s:%i", CONN_IDX(conn), \ @@ -1627,7 +1521,7 @@ void tcp_defer_handler(struct ctx *c) { int max_conns = c->tcp.conn_count / 100 * TCP_CONN_PRESSURE; int max_files = c->nofile / 100 * TCP_FILE_PRESSURE; - struct tcp_conn *conn; + struct tcp_tap_conn *conn; tcp_l2_flags_buf_flush(c); tcp_l2_data_buf_flush(c); @@ -1656,7 +1550,7 @@ void tcp_defer_handler(struct ctx *c) * Return: 802.3 length, host order */ static size_t tcp_l2_buf_fill_headers(const struct ctx *c, - const struct tcp_conn *conn, + const struct tcp_tap_conn *conn, void *p, size_t plen, const uint16_t *check, uint32_t seq) { @@ -1738,7 +1632,7 @@ do { \ * * Return: 1 if sequence or window were updated, 0 otherwise */ -static int tcp_update_seqack_wnd(const struct ctx *c, struct tcp_conn *conn, +static int tcp_update_seqack_wnd(const struct ctx *c, struct tcp_tap_conn *conn, int force_seq, struct tcp_info *tinfo) { uint32_t prev_wnd_to_tap = conn->wnd_to_tap << conn->ws_to_tap; @@ -1824,7 +1718,7 @@ out: * * Return: negative error code on connection reset, 0 otherwise */ -static int tcp_send_flag(struct ctx *c, struct tcp_conn *conn, int flags) +static int tcp_send_flag(struct ctx *c, struct tcp_tap_conn *conn, int flags) { uint32_t prev_ack_to_tap = conn->seq_ack_to_tap; uint32_t prev_wnd_to_tap = conn->wnd_to_tap; @@ -1971,7 +1865,7 @@ static int tcp_send_flag(struct ctx *c, struct tcp_conn *conn, int flags) * @c: Execution context * @conn: Connection pointer */ -static void tcp_rst_do(struct ctx *c, struct tcp_conn *conn) +static void tcp_rst_do(struct ctx *c, struct tcp_tap_conn *conn) { if (conn->events == CLOSED) return; @@ -1986,7 +1880,7 @@ static void tcp_rst_do(struct ctx *c, struct tcp_conn *conn) * @opts: Pointer to start of TCP options * @optlen: Bytes in options: caller MUST ensure available length */ -static void tcp_get_tap_ws(struct tcp_conn *conn, +static void tcp_get_tap_ws(struct tcp_tap_conn *conn, const char *opts, size_t optlen) { int ws = tcp_opt_get(opts, optlen, OPT_WS, NULL, NULL); @@ -2003,7 +1897,7 @@ static void tcp_get_tap_ws(struct tcp_conn *conn, * @conn: Connection pointer * @window: Window value, host order, unscaled */ -static void tcp_clamp_window(const struct ctx *c, struct tcp_conn *conn, +static void tcp_clamp_window(const struct ctx *c, struct tcp_tap_conn *conn, unsigned wnd) { uint32_t prev_scaled = conn->wnd_from_tap << conn->ws_from_tap; @@ -2125,7 +2019,7 @@ static int tcp_conn_new_sock(const struct ctx *c, sa_family_t af) * Return: clamped MSS value */ static uint16_t tcp_conn_tap_mss(const struct ctx *c, - const struct tcp_conn *conn, + const struct tcp_tap_conn *conn, const char *opts, size_t optlen) { unsigned int mss; @@ -2172,7 +2066,7 @@ static void tcp_conn_from_tap(struct ctx *c, int af, const void *addr, .sin6_addr = *(struct in6_addr *)addr, }; const struct sockaddr *sa; - struct tcp_conn *conn; + struct tcp_tap_conn *conn; socklen_t sl; int s, mss; @@ -2280,7 +2174,7 @@ static void tcp_conn_from_tap(struct ctx *c, int af, const void *addr, * * Return: 0 on success, negative error code from recv() on failure */ -static int tcp_sock_consume(struct tcp_conn *conn, uint32_t ack_seq) +static int tcp_sock_consume(struct tcp_tap_conn *conn, uint32_t ack_seq) { /* Simply ignore out-of-order ACKs: we already consumed the data we * needed from the buffer, and we won't rewind back to a lower ACK @@ -2307,7 +2201,7 @@ static int tcp_sock_consume(struct tcp_conn *conn, uint32_t ack_seq) * @seq: Sequence number to be sent * @now: Current timestamp */ -static void tcp_data_to_tap(struct ctx *c, struct tcp_conn *conn, +static void tcp_data_to_tap(struct ctx *c, struct tcp_tap_conn *conn, ssize_t plen, int no_csum, uint32_t seq) { struct iovec *iov; @@ -2344,7 +2238,7 @@ static void tcp_data_to_tap(struct ctx *c, struct tcp_conn *conn, * * #syscalls recvmsg */ -static int tcp_data_from_sock(struct ctx *c, struct tcp_conn *conn) +static int tcp_data_from_sock(struct ctx *c, struct tcp_tap_conn *conn) { uint32_t wnd_scaled = conn->wnd_from_tap << conn->ws_from_tap; int fill_bufs, send_bufs = 0, last_len, iov_rem = 0; @@ -2475,7 +2369,7 @@ zero_len: * * #syscalls sendmsg */ -static void tcp_data_from_tap(struct ctx *c, struct tcp_conn *conn, +static void tcp_data_from_tap(struct ctx *c, struct tcp_tap_conn *conn, const struct pool *p) { int i, iov_i, ack = 0, fin = 0, retr = 0, keep = -1, partial_send = 0; @@ -2675,7 +2569,7 @@ out: * @opts: Pointer to start of options * @optlen: Bytes in options: caller MUST ensure available length */ -static void tcp_conn_from_sock_finish(struct ctx *c, struct tcp_conn *conn, +static void tcp_conn_from_sock_finish(struct ctx *c, struct tcp_tap_conn *conn, const struct tcphdr *th, const char *opts, size_t optlen) { @@ -2714,7 +2608,7 @@ static void tcp_conn_from_sock_finish(struct ctx *c, struct tcp_conn *conn, int tcp_tap_handler(struct ctx *c, int af, const void *addr, const struct pool *p, const struct timespec *now) { - struct tcp_conn *conn; + struct tcp_tap_conn *conn; size_t optlen, len; struct tcphdr *th; int ack_due = 0; @@ -2829,7 +2723,7 @@ int tcp_tap_handler(struct ctx *c, int af, const void *addr, * @c: Execution context * @conn: Connection pointer */ -static void tcp_connect_finish(struct ctx *c, struct tcp_conn *conn) +static void tcp_connect_finish(struct ctx *c, struct tcp_tap_conn *conn) { socklen_t sl; int so; @@ -2857,7 +2751,7 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, const struct timespec *now) { struct sockaddr_storage sa; - struct tcp_conn *conn; + struct tcp_tap_conn *conn; socklen_t sl; int s; @@ -2949,7 +2843,7 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, */ static void tcp_timer_handler(struct ctx *c, union epoll_ref ref) { - struct tcp_conn *conn = conn_at_idx(ref.r.p.tcp.tcp.index); + struct tcp_tap_conn *conn = conn_at_idx(ref.r.p.tcp.tcp.index); struct itimerspec check_armed = { { 0 }, { 0 } }; if (!conn) @@ -3012,7 +2906,7 @@ static void tcp_timer_handler(struct ctx *c, union epoll_ref ref) void tcp_sock_handler(struct ctx *c, union epoll_ref ref, uint32_t events, const struct timespec *now) { - struct tcp_conn *conn; + struct tcp_tap_conn *conn; if (ref.r.p.tcp.tcp.timer) { tcp_timer_handler(c, ref); @@ -3510,7 +3404,7 @@ static int tcp_port_rebind(void *arg) void tcp_timer(struct ctx *c, const struct timespec *ts) { struct tcp_sock_refill_arg refill_arg = { c, 0 }; - struct tcp_conn *conn; + struct tcp_tap_conn *conn; (void)ts; diff --git a/tcp_conn.h b/tcp_conn.h new file mode 100644 index 0000000..db4c2d9 --- /dev/null +++ b/tcp_conn.h @@ -0,0 +1,168 @@ +/* SPDX-License-Identifier: AGPL-3.0-or-later + * Copyright Red Hat + * Author: Stefano Brivio + * Author: David Gibson + * + * TCP connection tracking data structures, used by tcp.c and + * tcp_splice.c. Shouldn't be included in non-TCP code. + */ +#ifndef TCP_CONN_H +#define TCP_CONN_H + +#define TCP_HASH_BUCKET_BITS (TCP_CONN_INDEX_BITS + 1) + +/** + * struct tcp_tap_conn - Descriptor for a TCP connection (not spliced) + * @next_index: Connection index of next item in hash chain, -1 for none + * @tap_mss: MSS advertised by tap/guest, rounded to 2 ^ TCP_MSS_BITS + * @sock: Socket descriptor number + * @events: Connection events, implying connection states + * @timer: timerfd descriptor for timeout events + * @flags: Connection flags representing internal attributes + * @hash_bucket: Bucket index in connection lookup hash table + * @retrans: Number of retransmissions occurred due to ACK_TIMEOUT + * @ws_from_tap: Window scaling factor advertised from tap/guest + * @ws_to_tap: Window scaling factor advertised to tap/guest + * @sndbuf: Sending buffer in kernel, rounded to 2 ^ SNDBUF_BITS + * @seq_dup_ack_approx: Last duplicate ACK number sent to tap + * @a.a6: IPv6 remote address, can be IPv4-mapped + * @a.a4.zero: Zero prefix for IPv4-mapped, see RFC 6890, Table 20 + * @a.a4.one: Ones prefix for IPv4-mapped + * @a.a4.a: IPv4 address + * @tap_port: Guest-facing tap port + * @sock_port: Remote, socket-facing port + * @wnd_from_tap: Last window size from tap, unscaled (as received) + * @wnd_to_tap: Sending window advertised to tap, unscaled (as sent) + * @seq_to_tap: Next sequence for packets to tap + * @seq_ack_from_tap: Last ACK number received from tap + * @seq_from_tap: Next sequence for packets from tap (not actually sent) + * @seq_ack_to_tap: Last ACK number sent to tap + * @seq_init_from_tap: Initial sequence number from tap + */ +struct tcp_tap_conn { + int next_index :TCP_CONN_INDEX_BITS + 2; + +#define TCP_RETRANS_BITS 3 + unsigned int retrans :TCP_RETRANS_BITS; +#define TCP_MAX_RETRANS ((1U << TCP_RETRANS_BITS) - 1) + +#define TCP_WS_BITS 4 /* RFC 7323 */ +#define TCP_WS_MAX 14 + unsigned int ws_from_tap :TCP_WS_BITS; + unsigned int ws_to_tap :TCP_WS_BITS; + + + int sock :SOCKET_REF_BITS; + + uint8_t events; +#define CLOSED 0 +#define SOCK_ACCEPTED BIT(0) /* implies SYN sent to tap */ +#define TAP_SYN_RCVD BIT(1) /* implies socket connecting */ +#define TAP_SYN_ACK_SENT BIT( 3) /* implies socket connected */ +#define ESTABLISHED BIT(2) +#define SOCK_FIN_RCVD BIT( 3) +#define SOCK_FIN_SENT BIT( 4) +#define TAP_FIN_RCVD BIT( 5) +#define TAP_FIN_SENT BIT( 6) +#define TAP_FIN_ACKED BIT( 7) + +#define CONN_STATE_BITS /* Setting these clears other flags */ \ + (SOCK_ACCEPTED | TAP_SYN_RCVD | ESTABLISHED) + + + int timer :SOCKET_REF_BITS; + + uint8_t flags; +#define STALLED BIT(0) +#define LOCAL BIT(1) +#define WND_CLAMPED BIT(2) +#define IN_EPOLL BIT(3) +#define ACTIVE_CLOSE BIT(4) +#define ACK_TO_TAP_DUE BIT(5) +#define ACK_FROM_TAP_DUE BIT(6) + + + unsigned int hash_bucket :TCP_HASH_BUCKET_BITS; + +#define TCP_MSS_BITS 14 + unsigned int tap_mss :TCP_MSS_BITS; +#define MSS_SET(conn, mss) (conn->tap_mss = (mss >> (16 - TCP_MSS_BITS))) +#define MSS_GET(conn) (conn->tap_mss << (16 - TCP_MSS_BITS)) + + +#define SNDBUF_BITS 24 + unsigned int sndbuf :SNDBUF_BITS; +#define SNDBUF_SET(conn, bytes) (conn->sndbuf = ((bytes) >> (32 - SNDBUF_BITS))) +#define SNDBUF_GET(conn) (conn->sndbuf << (32 - SNDBUF_BITS)) + + uint8_t seq_dup_ack_approx; + + + union { + struct in6_addr a6; + struct { + uint8_t zero[10]; + uint8_t one[2]; + struct in_addr a; + } a4; + } a; + + in_port_t tap_port; + in_port_t sock_port; + + uint16_t wnd_from_tap; + uint16_t wnd_to_tap; + + uint32_t seq_to_tap; + uint32_t seq_ack_from_tap; + uint32_t seq_from_tap; + uint32_t seq_ack_to_tap; + uint32_t seq_init_from_tap; +}; + +/** + * struct tcp_splice_conn - Descriptor for a spliced TCP connection + * @a: File descriptor number of socket for accepted connection + * @pipe_a_b: Pipe ends for splice() from @a to @b + * @b: File descriptor number of peer connected socket + * @pipe_b_a: Pipe ends for splice() from @b to @a + * @events: Events observed/actions performed on connection + * @flags: Connection flags (attributes, not events) + * @a_read: Bytes read from @a (not fully written to @b in one shot) + * @a_written: Bytes written to @a (not fully written from one @b read) + * @b_read: Bytes read from @b (not fully written to @a in one shot) + * @b_written: Bytes written to @b (not fully written from one @a read) +*/ +struct tcp_splice_conn { + int a; + int pipe_a_b[2]; + int b; + int pipe_b_a[2]; + + uint8_t events; +#define SPLICE_CLOSED 0 +#define SPLICE_CONNECT BIT(0) +#define SPLICE_ESTABLISHED BIT(1) +#define A_OUT_WAIT BIT(2) +#define B_OUT_WAIT BIT(3) +#define A_FIN_RCVD BIT(4) +#define B_FIN_RCVD BIT(5) +#define A_FIN_SENT BIT(6) +#define B_FIN_SENT BIT(7) + + uint8_t flags; +#define SPLICE_V6 BIT(0) +#define SPLICE_IN_EPOLL BIT(1) +#define RCVLOWAT_SET_A BIT(2) +#define RCVLOWAT_SET_B BIT(3) +#define RCVLOWAT_ACT_A BIT(4) +#define RCVLOWAT_ACT_B BIT(5) +#define CLOSING BIT(6) + + uint32_t a_read; + uint32_t a_written; + uint32_t b_read; + uint32_t b_written; +}; + +#endif /* TCP_CONN_H */ diff --git a/tcp_splice.c b/tcp_splice.c index 4cc4ad2..cbfab01 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -21,12 +21,12 @@ * * - SPLICE_CONNECT: connection accepted, connecting to target * - SPLICE_ESTABLISHED: connection to target established - * - SPLICE_A_OUT_WAIT: pipe to accepted socket full, wait for EPOLLOUT - * - SPLICE_B_OUT_WAIT: pipe to target socket full, wait for EPOLLOUT - * - SPLICE_A_FIN_RCVD: FIN (EPOLLRDHUP) seen from accepted socket - * - SPLICE_B_FIN_RCVD: FIN (EPOLLRDHUP) seen from target socket - * - SPLICE_A_FIN_RCVD: FIN (write shutdown) sent to accepted socket - * - SPLICE_B_FIN_RCVD: FIN (write shutdown) sent to target socket + * - A_OUT_WAIT: pipe to accepted socket full, wait for EPOLLOUT + * - B_OUT_WAIT: pipe to target socket full, wait for EPOLLOUT + * - A_FIN_RCVD: FIN (EPOLLRDHUP) seen from accepted socket + * - B_FIN_RCVD: FIN (EPOLLRDHUP) seen from target socket + * - A_FIN_RCVD: FIN (write shutdown) sent to accepted socket + * - B_FIN_RCVD: FIN (write shutdown) sent to target socket * * #syscalls:pasta pipe2|pipe fcntl armv6l:fcntl64 armv7l:fcntl64 ppc64:fcntl64 */ @@ -52,6 +52,8 @@ #include "log.h" #include "tcp_splice.h" +#include "tcp_conn.h" + #define MAX_PIPE_SIZE (8UL * 1024 * 1024) #define TCP_SPLICE_PIPE_POOL_SIZE 16 #define TCP_SPLICE_CONN_PRESSURE 30 /* % of splice_conn_count */ @@ -66,52 +68,7 @@ extern int ns_sock_pool6 [TCP_SOCK_POOL_SIZE]; /* Pool of pre-opened pipes */ static int splice_pipe_pool [TCP_SPLICE_PIPE_POOL_SIZE][2][2]; -/** - * struct tcp_splice_conn - Descriptor for a spliced TCP connection - * @a: File descriptor number of socket for accepted connection - * @pipe_a_b: Pipe ends for splice() from @a to @b - * @b: File descriptor number of peer connected socket - * @pipe_b_a: Pipe ends for splice() from @b to @a - * @events: Events observed/actions performed on connection - * @flags: Connection flags (attributes, not events) - * @a_read: Bytes read from @a (not fully written to @b in one shot) - * @a_written: Bytes written to @a (not fully written from one @b read) - * @b_read: Bytes read from @b (not fully written to @a in one shot) - * @b_written: Bytes written to @b (not fully written from one @a read) -*/ -struct tcp_splice_conn { - int a; - int pipe_a_b[2]; - int b; - int pipe_b_a[2]; - - uint8_t events; -#define CLOSED 0 -#define CONNECT BIT(0) -#define ESTABLISHED BIT(1) -#define A_OUT_WAIT BIT(2) -#define B_OUT_WAIT BIT(3) -#define A_FIN_RCVD BIT(4) -#define B_FIN_RCVD BIT(5) -#define A_FIN_SENT BIT(6) -#define B_FIN_SENT BIT(7) - - uint8_t flags; -#define SOCK_V6 BIT(0) -#define IN_EPOLL BIT(1) -#define RCVLOWAT_SET_A BIT(2) -#define RCVLOWAT_SET_B BIT(3) -#define RCVLOWAT_ACT_A BIT(4) -#define RCVLOWAT_ACT_B BIT(5) -#define CLOSING BIT(6) - - uint32_t a_read; - uint32_t a_written; - uint32_t b_read; - uint32_t b_written; -}; - -#define CONN_V6(x) (x->flags & SOCK_V6) +#define CONN_V6(x) (x->flags & SPLICE_V6) #define CONN_V4(x) (!CONN_V6(x)) #define CONN_HAS(conn, set) ((conn->events & (set)) == (set)) #define CONN(index) (tc_splice + (index)) @@ -122,13 +79,13 @@ static struct tcp_splice_conn tc_splice[TCP_SPLICE_MAX_CONNS]; /* Display strings for connection events */ static const char *tcp_splice_event_str[] __attribute((__unused__)) = { - "CONNECT", "ESTABLISHED", "A_OUT_WAIT", "B_OUT_WAIT", + "SPLICE_CONNECT", "SPLICE_ESTABLISHED", "A_OUT_WAIT", "B_OUT_WAIT", "A_FIN_RCVD", "B_FIN_RCVD", "A_FIN_SENT", "B_FIN_SENT", }; /* Display strings for connection flags */ static const char *tcp_splice_flag_str[] __attribute((__unused__)) = { - "SOCK_V6", "IN_EPOLL", "RCVLOWAT_SET_A", "RCVLOWAT_SET_B", + "SPLICE_V6", "SPLICE_IN_EPOLL", "RCVLOWAT_SET_A", "RCVLOWAT_SET_B", "RCVLOWAT_ACT_A", "RCVLOWAT_ACT_B", "CLOSING", }; @@ -143,12 +100,12 @@ static void tcp_splice_conn_epoll_events(uint16_t events, { *a = *b = 0; - if (events & ESTABLISHED) { + if (events & SPLICE_ESTABLISHED) { if (!(events & B_FIN_SENT)) *a = EPOLLIN | EPOLLRDHUP; if (!(events & A_FIN_SENT)) *b = EPOLLIN | EPOLLRDHUP; - } else if (events & CONNECT) { + } else if (events & SPLICE_CONNECT) { *b = EPOLLOUT; } @@ -210,7 +167,7 @@ static void conn_flag_do(const struct ctx *c, struct tcp_splice_conn *conn, static int tcp_splice_epoll_ctl(const struct ctx *c, struct tcp_splice_conn *conn) { - int m = (conn->flags & IN_EPOLL) ? EPOLL_CTL_MOD : EPOLL_CTL_ADD; + int m = (conn->flags & SPLICE_IN_EPOLL) ? EPOLL_CTL_MOD : EPOLL_CTL_ADD; union epoll_ref ref_a = { .r.proto = IPPROTO_TCP, .r.s = conn->a, .r.p.tcp.tcp.splice = 1, .r.p.tcp.tcp.index = CONN_IDX(conn), @@ -234,7 +191,7 @@ static int tcp_splice_epoll_ctl(const struct ctx *c, epoll_ctl(c->epollfd, m, conn->b, &ev_b)) goto delete; - conn->flags |= IN_EPOLL; /* No need to log this */ + conn->flags |= SPLICE_IN_EPOLL; /* No need to log this */ return 0; @@ -323,7 +280,7 @@ static void tcp_table_splice_compact(struct ctx *c, */ static void tcp_splice_destroy(struct ctx *c, struct tcp_splice_conn *conn) { - if (conn->events & ESTABLISHED) { + if (conn->events & SPLICE_ESTABLISHED) { /* Flushing might need to block: don't recycle them. */ if (conn->pipe_a_b[0] != -1) { close(conn->pipe_a_b[0]); @@ -337,7 +294,7 @@ static void tcp_splice_destroy(struct ctx *c, struct tcp_splice_conn *conn) } } - if (conn->events & CONNECT) { + if (conn->events & SPLICE_CONNECT) { close(conn->b); conn->b = -1; } @@ -346,7 +303,7 @@ static void tcp_splice_destroy(struct ctx *c, struct tcp_splice_conn *conn) conn->a = -1; conn->a_read = conn->a_written = conn->b_read = conn->b_written = 0; - conn->events = CLOSED; + conn->events = SPLICE_CLOSED; conn->flags = 0; debug("TCP (spliced): index %li, CLOSED", CONN_IDX(conn)); @@ -397,8 +354,8 @@ static int tcp_splice_connect_finish(const struct ctx *c, } } - if (!(conn->events & ESTABLISHED)) - conn_event(c, conn, ESTABLISHED); + if (!(conn->events & SPLICE_ESTABLISHED)) + conn_event(c, conn, SPLICE_ESTABLISHED); return 0; } @@ -466,9 +423,9 @@ static int tcp_splice_connect(const struct ctx *c, struct tcp_splice_conn *conn, close(sock_conn); return ret; } - conn_event(c, conn, CONNECT); + conn_event(c, conn, SPLICE_CONNECT); } else { - conn_event(c, conn, ESTABLISHED); + conn_event(c, conn, SPLICE_ESTABLISHED); return tcp_splice_connect_finish(c, conn); } @@ -598,7 +555,7 @@ void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, conn = CONN(c->tcp.splice_conn_count++); conn->a = s; - conn->flags = ref.r.p.tcp.tcp.v6 ? SOCK_V6 : 0; + conn->flags = ref.r.p.tcp.tcp.v6 ? SPLICE_V6 : 0; if (tcp_splice_new(c, conn, ref.r.p.tcp.tcp.index, ref.r.p.tcp.tcp.outbound)) @@ -609,13 +566,13 @@ void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, conn = CONN(ref.r.p.tcp.tcp.index); - if (conn->events == CLOSED) + if (conn->events == SPLICE_CLOSED) return; if (events & EPOLLERR) goto close; - if (conn->events == CONNECT) { + if (conn->events == SPLICE_CONNECT) { if (!(events & EPOLLOUT)) goto close; if (tcp_splice_connect_finish(c, conn)) -- 2.38.1

Currently spliced and non-spliced connections are stored in completely separate tables, so there are completely independent limits on the number of spliced and non-spliced connections. This is a bit counter-intuitive. More importantly, the fact that the tables are separate prevents us from unifying some other logic between the two cases. So, merge these two tables into one, using the 'c.spliced' common field to distinguish between them when necessary. For now we keep a common limit of 128k connections, whether they're spliced or non-spliced, which means we save memory overall. If necessary we could increase this to a 256k or higher total, which would cost memory but give some more flexibility. For now, the code paths which need to step through all extant connections are still separate for the two cases, just skipping over entries which aren't for them. We'll improve that in later patches. Signed-off-by: David Gibson --- tcp.c | 46 ++++++++++++++++++++---------------- tcp.h | 2 +- tcp_conn.h | 6 +++++ tcp_splice.c | 66 ++++++++++++++-------------------------------------- tcp_splice.h | 2 -- 5 files changed, 51 insertions(+), 71 deletions(-) diff --git a/tcp.c b/tcp.c index 90eb840..61700d6 100644 --- a/tcp.c +++ b/tcp.c @@ -98,11 +98,11 @@ * Connection tracking and storage * ------------------------------- * - * Connections are tracked by the @tc array of struct tcp_tap_conn, containing - * addresses, ports, TCP states and parameters. This is statically allocated and - * indexed by an arbitrary connection number. The array is compacted whenever a - * connection is closed, by remapping the highest connection index in use to the - * one freed up. + * Connections are tracked by struct tcp_tap_conn entries in the @tc + * array, containing addresses, ports, TCP states and parameters. This + * is statically allocated and indexed by an arbitrary connection + * number. The array is compacted whenever a connection is closed, by + * remapping the highest connection index in use to the one freed up. * * References used for the epoll interface report the connection index used for * the @tc array. @@ -588,10 +588,10 @@ static unsigned int tcp6_l2_flags_buf_used; static size_t tcp6_l2_flags_buf_bytes; /* TCP connections */ -static struct tcp_tap_conn tc[TCP_MAX_CONNS]; +union tcp_conn tc[TCP_MAX_CONNS]; -#define CONN(index) (tc + (index)) -#define CONN_IDX(conn) ((conn) - tc) +#define CONN(index) (&tc[(index)].tap) +#define CONN_IDX(conn) ((union tcp_conn *)(conn) - tc) /** conn_at_idx() - Find a connection by index, if present * @index: Index of connection to lookup @@ -1351,26 +1351,28 @@ static struct tcp_tap_conn *tcp_hash_lookup(const struct ctx *c, * @c: Execution context * @hole: Pointer to recently closed connection */ -static void tcp_table_compact(struct ctx *c, struct tcp_tap_conn *hole) +void tcp_table_compact(struct ctx *c, union tcp_conn *hole) { - struct tcp_tap_conn *from, *to; + union tcp_conn *from; if (CONN_IDX(hole) == --c->tcp.conn_count) { - debug("TCP: hash table compaction: maximum index was %li (%p)", + debug("TCP: table compaction: maximum index was %li (%p)", CONN_IDX(hole), hole); memset(hole, 0, sizeof(*hole)); return; } - from = CONN(c->tcp.conn_count); + from = tc + c->tcp.conn_count; memcpy(hole, from, sizeof(*hole)); - to = hole; - tcp_tap_conn_update(c, from, to); + if (from->c.spliced) + tcp_splice_conn_update(c, &hole->splice); + else + tcp_tap_conn_update(c, &from->tap, &hole->tap); - debug("TCP: hash table compaction: old index %li, new index %li, " - "sock %i, from: %p, to: %p", - CONN_IDX(from), CONN_IDX(to), from->sock, from, to); + debug("TCP: table compaction (spliced=%d): old index %li, new index %li, " + "from: %p, to: %p", + from->c.spliced, CONN_IDX(from), CONN_IDX(hole), from, hole); memset(from, 0, sizeof(*from)); } @@ -1387,7 +1389,7 @@ static void tcp_conn_destroy(struct ctx *c, struct tcp_tap_conn *conn) close(conn->timer); tcp_hash_remove(conn); - tcp_table_compact(c, conn); + tcp_table_compact(c, (union tcp_conn *)conn); } static void tcp_rst_do(struct ctx *c, struct tcp_tap_conn *conn); @@ -1535,7 +1537,9 @@ void tcp_defer_handler(struct ctx *c) if (c->tcp.conn_count < MIN(max_files, max_conns)) return; - for (conn = CONN(c->tcp.conn_count - 1); conn >= tc; conn--) { + for (conn = CONN(c->tcp.conn_count - 1); conn >= CONN(0); conn--) { + if (conn->c.spliced) + continue; if (conn->events == CLOSED) tcp_conn_destroy(c, conn); } @@ -3433,7 +3437,9 @@ void tcp_timer(struct ctx *c, const struct timespec *ts) } } - for (conn = CONN(c->tcp.conn_count - 1); conn >= tc; conn--) { + for (conn = CONN(c->tcp.conn_count - 1); conn >= CONN(0); conn--) { + if (conn->c.spliced) + continue; if (conn->events == CLOSED) tcp_conn_destroy(c, conn); } diff --git a/tcp.h b/tcp.h index bba0f38..49738ef 100644 --- a/tcp.h +++ b/tcp.h @@ -54,7 +54,7 @@ union tcp_epoll_ref { /** * struct tcp_ctx - Execution context for TCP routines * @hash_secret: 128-bit secret for hash functions, ISN and hash table - * @conn_count: Count of connections (not spliced) in connection table + * @conn_count: Count of total connections in connection table * @splice_conn_count: Count of spliced connections in connection table * @port_to_tap: Ports bound host-side, packets to tap or spliced * @fwd_in: Port forwarding configuration for inbound packets diff --git a/tcp_conn.h b/tcp_conn.h index 39d104a..4295f7d 100644 --- a/tcp_conn.h +++ b/tcp_conn.h @@ -195,4 +195,10 @@ union tcp_conn { struct tcp_splice_conn splice; }; +/* TCP connections */ +extern union tcp_conn tc[]; + +void tcp_splice_conn_update(struct ctx *c, struct tcp_splice_conn *new); +void tcp_table_compact(struct ctx *c, union tcp_conn *hole); + #endif /* TCP_CONN_H */ diff --git a/tcp_splice.c b/tcp_splice.c index 7dcd1cb..c986a9c 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -16,7 +16,7 @@ * For local traffic directed to TCP ports configured for direct * mapping between namespaces, packets are directly translated between * L4 sockets using a pair of splice() syscalls. These connections are - * tracked in the @tc_splice array of struct tcp_splice_conn, using + * tracked by struct tcp_splice_conn entries in the @tc array, using * these events: * * - SPLICE_CONNECT: connection accepted, connecting to target @@ -57,7 +57,7 @@ #define MAX_PIPE_SIZE (8UL * 1024 * 1024) #define TCP_SPLICE_PIPE_POOL_SIZE 16 -#define TCP_SPLICE_CONN_PRESSURE 30 /* % of splice_conn_count */ +#define TCP_SPLICE_CONN_PRESSURE 30 /* % of conn_count */ #define TCP_SPLICE_FILE_PRESSURE 30 /* % of c->nofile */ /* From tcp.c */ @@ -72,11 +72,8 @@ static int splice_pipe_pool [TCP_SPLICE_PIPE_POOL_SIZE][2][2]; #define CONN_V6(x) (x->flags & SPLICE_V6) #define CONN_V4(x) (!CONN_V6(x)) #define CONN_HAS(conn, set) ((conn->events & (set)) == (set)) -#define CONN(index) (tc_splice + (index)) -#define CONN_IDX(conn) ((conn) - tc_splice) - -/* Spliced connections */ -static struct tcp_splice_conn tc_splice[TCP_SPLICE_MAX_CONNS]; +#define CONN(index) (&tc[(index)].splice) +#define CONN_IDX(conn) ((union tcp_conn *)(conn) - tc) /* Display strings for connection events */ static const char *tcp_splice_event_str[] __attribute((__unused__)) = { @@ -248,43 +245,13 @@ static void conn_event_do(const struct ctx *c, struct tcp_splice_conn *conn, * @c: Execution context * @new: New location of tcp_splice_conn */ -static void tcp_splice_conn_update(struct ctx *c, struct tcp_splice_conn *new) +void tcp_splice_conn_update(struct ctx *c, struct tcp_splice_conn *new) { tcp_splice_epoll_ctl(c, new); if (tcp_splice_epoll_ctl(c, new)) conn_flag(c, new, CLOSING); } -/** - * tcp_table_splice_compact - Compact spliced connection table - * @c: Execution context - * @hole: Pointer to recently closed connection - */ -static void tcp_table_splice_compact(struct ctx *c, - struct tcp_splice_conn *hole) -{ - struct tcp_splice_conn *move; - - if (CONN_IDX(hole) == --c->tcp.splice_conn_count) { - debug("TCP (spliced): index %li (max) removed", CONN_IDX(hole)); - return; - } - - move = CONN(c->tcp.splice_conn_count); - - memcpy(hole, move, sizeof(*hole)); - - move->a = move->b = -1; - move->a_read = move->a_written = move->b_read = move->b_written = 0; - move->pipe_a_b[0] = move->pipe_a_b[1] = -1; - move->pipe_b_a[0] = move->pipe_b_a[1] = -1; - move->flags = move->events = 0; - - debug("TCP (spliced): index %li moved to %li", - CONN_IDX(move), CONN_IDX(hole)); - tcp_splice_conn_update(c, hole); -} - /** * tcp_splice_destroy() - Close spliced connection and pipes, clear * @c: Execution context @@ -319,7 +286,8 @@ static void tcp_splice_destroy(struct ctx *c, struct tcp_splice_conn *conn) conn->flags = 0; debug("TCP (spliced): index %li, CLOSED", CONN_IDX(conn)); - tcp_table_splice_compact(c, conn); + c->tcp.splice_conn_count--; + tcp_table_compact(c, (union tcp_conn *)conn); } /** @@ -553,7 +521,7 @@ void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, if (ref.r.p.tcp.tcp.listen) { int s; - if (c->tcp.splice_conn_count >= TCP_SPLICE_MAX_CONNS) + if (c->tcp.conn_count >= TCP_MAX_CONNS) return; if ((s = accept4(ref.r.s, NULL, NULL, SOCK_NONBLOCK)) < 0) @@ -565,8 +533,9 @@ void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, s); } - conn = CONN(c->tcp.splice_conn_count++); + conn = CONN(c->tcp.conn_count++); conn->c.spliced = true; + c->tcp.splice_conn_count++; conn->a = s; conn->flags = ref.r.p.tcp.tcp.v6 ? SPLICE_V6 : 0; @@ -845,9 +814,10 @@ void tcp_splice_timer(struct ctx *c) { struct tcp_splice_conn *conn; - for (conn = CONN(c->tcp.splice_conn_count - 1); - conn >= tc_splice; - conn--) { + for (conn = CONN(c->tcp.conn_count - 1); conn >= CONN(0); conn--) { + if (!conn->c.spliced) + continue; + if (conn->flags & CLOSING) { tcp_splice_destroy(c, conn); return; @@ -890,12 +860,12 @@ void tcp_splice_defer_handler(struct ctx *c) int max_files = c->nofile / 100 * TCP_SPLICE_FILE_PRESSURE; struct tcp_splice_conn *conn; - if (c->tcp.splice_conn_count < MIN(max_files / 6, max_conns)) + if (c->tcp.conn_count < MIN(max_files / 6, max_conns)) return; - for (conn = CONN(c->tcp.splice_conn_count - 1); - conn >= tc_splice; - conn--) { + for (conn = CONN(c->tcp.conn_count - 1); conn >= CONN(0); conn--) { + if (!conn->c.spliced) + continue; if (conn->flags & CLOSING) tcp_splice_destroy(c, conn); } diff --git a/tcp_splice.h b/tcp_splice.h index 2c4bff3..e8c70e9 100644 --- a/tcp_splice.h +++ b/tcp_splice.h @@ -6,8 +6,6 @@ #ifndef TCP_SPLICE_H #define TCP_SPLICE_H -#define TCP_SPLICE_MAX_CONNS (128 * 1024) - void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, uint32_t events); void tcp_splice_init(struct ctx *c); -- 2.38.1

These two functions scan all the non-splced and spliced connections respectively and perform timed updates on them. Avoid scanning the now unified table twice, by having tcp_timer scan it once calling the relevant per-connection function for each one. Signed-off-by: David Gibson --- tcp.c | 18 ++++++++--------- tcp_conn.h | 3 +++ tcp_splice.c | 57 +++++++++++++++++++++++----------------------------- tcp_splice.h | 1 - 4 files changed, 37 insertions(+), 42 deletions(-) diff --git a/tcp.c b/tcp.c index a9cc811..24cd10b 100644 --- a/tcp.c +++ b/tcp.c @@ -3283,8 +3283,6 @@ int tcp_init(struct ctx *c) refill_arg.ns = 1; NS_CALL(tcp_sock_refill, &refill_arg); - - tcp_splice_timer(c); } return 0; @@ -3416,7 +3414,7 @@ static int tcp_port_rebind(void *arg) void tcp_timer(struct ctx *c, const struct timespec *ts) { struct tcp_sock_refill_arg refill_arg = { c, 0 }; - struct tcp_tap_conn *conn; + union tcp_conn *conn; (void)ts; @@ -3439,11 +3437,13 @@ void tcp_timer(struct ctx *c, const struct timespec *ts) } } - for (conn = CONN(c->tcp.conn_count - 1); conn >= CONN(0); conn--) { - if (conn->c.spliced) - continue; - if (conn->events == CLOSED) - tcp_conn_destroy(c, conn); + for (conn = tc + c->tcp.conn_count - 1; conn >= tc; conn--) { + if (conn->c.spliced) { + tcp_splice_timer(c, &conn->splice); + } else { + if (conn->tap.events == CLOSED) + tcp_conn_destroy(c, &conn->tap); + } } tcp_sock_refill(&refill_arg); @@ -3453,6 +3453,6 @@ void tcp_timer(struct ctx *c, const struct timespec *ts) (c->ifi6 && ns_sock_pool6[TCP_SOCK_POOL_TSH] < 0)) NS_CALL(tcp_sock_refill, &refill_arg); - tcp_splice_timer(c); + tcp_splice_pipe_refill(c); } } diff --git a/tcp_conn.h b/tcp_conn.h index 634e259..7c450a0 100644 --- a/tcp_conn.h +++ b/tcp_conn.h @@ -201,5 +201,8 @@ extern union tcp_conn tc[]; void tcp_splice_conn_update(struct ctx *c, struct tcp_splice_conn *new); void tcp_table_compact(struct ctx *c, union tcp_conn *hole); void tcp_splice_destroy(struct ctx *c, struct tcp_splice_conn *conn); +void tcp_splice_timer(struct ctx *c, struct tcp_splice_conn *conn); +void tcp_splice_pipe_refill(const struct ctx *c); + #endif /* TCP_CONN_H */ diff --git a/tcp_splice.c b/tcp_splice.c index ad2b216..0ac316d 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -766,7 +766,7 @@ smaller: * tcp_splice_pipe_refill() - Refill pool of pre-opened pipes * @c: Execution context */ -static void tcp_splice_pipe_refill(const struct ctx *c) +void tcp_splice_pipe_refill(const struct ctx *c) { int i; @@ -803,48 +803,41 @@ void tcp_splice_init(struct ctx *c) { memset(splice_pipe_pool, 0xff, sizeof(splice_pipe_pool)); tcp_set_pipe_size(c); + tcp_splice_pipe_refill(c); } /** * tcp_splice_timer() - Timer for spliced connections * @c: Execution context + * @conn: Spliced connection */ -void tcp_splice_timer(struct ctx *c) +void tcp_splice_timer(struct ctx *c, struct tcp_splice_conn *conn) { - struct tcp_splice_conn *conn; - - for (conn = CONN(c->tcp.conn_count - 1); conn >= CONN(0); conn--) { - if (!conn->c.spliced) - continue; - - if (conn->flags & CLOSING) { - tcp_splice_destroy(c, conn); - return; - } + if (conn->flags & CLOSING) { + tcp_splice_destroy(c, conn); + return; + } - if ( (conn->flags & RCVLOWAT_SET_A) && - !(conn->flags & RCVLOWAT_ACT_A)) { - if (setsockopt(conn->a, SOL_SOCKET, SO_RCVLOWAT, - &((int){ 1 }), sizeof(int))) { - trace("TCP (spliced): can't set SO_RCVLOWAT on " - "%i", conn->a); - } - conn_flag(c, conn, ~RCVLOWAT_SET_A); + if ( (conn->flags & RCVLOWAT_SET_A) && + !(conn->flags & RCVLOWAT_ACT_A)) { + if (setsockopt(conn->a, SOL_SOCKET, SO_RCVLOWAT, + &((int){ 1 }), sizeof(int))) { + trace("TCP (spliced): can't set SO_RCVLOWAT on " + "%i", conn->a); } + conn_flag(c, conn, ~RCVLOWAT_SET_A); + } - if ( (conn->flags & RCVLOWAT_SET_B) && - !(conn->flags & RCVLOWAT_ACT_B)) { - if (setsockopt(conn->b, SOL_SOCKET, SO_RCVLOWAT, - &((int){ 1 }), sizeof(int))) { - trace("TCP (spliced): can't set SO_RCVLOWAT on " - "%i", conn->b); - } - conn_flag(c, conn, ~RCVLOWAT_SET_B); + if ( (conn->flags & RCVLOWAT_SET_B) && + !(conn->flags & RCVLOWAT_ACT_B)) { + if (setsockopt(conn->b, SOL_SOCKET, SO_RCVLOWAT, + &((int){ 1 }), sizeof(int))) { + trace("TCP (spliced): can't set SO_RCVLOWAT on " + "%i", conn->b); } - - conn_flag(c, conn, ~RCVLOWAT_ACT_A); - conn_flag(c, conn, ~RCVLOWAT_ACT_B); + conn_flag(c, conn, ~RCVLOWAT_SET_B); } - tcp_splice_pipe_refill(c); + conn_flag(c, conn, ~RCVLOWAT_ACT_A); + conn_flag(c, conn, ~RCVLOWAT_ACT_B); } diff --git a/tcp_splice.h b/tcp_splice.h index 82e057c..22024d6 100644 --- a/tcp_splice.h +++ b/tcp_splice.h @@ -9,6 +9,5 @@ void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, uint32_t events); void tcp_splice_init(struct ctx *c); -void tcp_splice_timer(struct ctx *c); #endif /* TCP_SPLICE_H */ -- 2.38.1

tcp_sock_init*() can create either sockets listening on the host, or in the pasta network namespace (with @ns==1). There are, however, a number of differences in how these two cases work in practice though. "ns" sockets are only used in pasta mode, and they always lead to spliced connections only. The functions are also only ever called in "ns" mode with a NULL address and interface name, and it doesn't really make sense for them to be called any other way. Later changes will introduce further differences in behaviour between these two cases, so it makes more sense to use separate functions for creating the ns listening sockets than the regular external/host listening sockets. --- conf.c | 6 +-- tcp.c | 132 ++++++++++++++++++++++++++++++++++++++------------------- tcp.h | 4 +- 3 files changed, 93 insertions(+), 49 deletions(-) diff --git a/conf.c b/conf.c index b07d661..4721c97 100644 --- a/conf.c +++ b/conf.c @@ -209,7 +209,7 @@ static int conf_ports(const struct ctx *c, char optname, const char *optarg, for (i = 0; i < PORT_EPHEMERAL_MIN; i++) { if (optname == 't') - tcp_sock_init(c, 0, AF_UNSPEC, NULL, NULL, i); + tcp_sock_init(c, AF_UNSPEC, NULL, NULL, i); else if (optname == 'u') udp_sock_init(c, 0, AF_UNSPEC, NULL, NULL, i); } @@ -287,7 +287,7 @@ static int conf_ports(const struct ctx *c, char optname, const char *optarg, bitmap_set(fwd->map, i); if (optname == 't') - tcp_sock_init(c, 0, af, addr, ifname, i); + tcp_sock_init(c, af, addr, ifname, i); else if (optname == 'u') udp_sock_init(c, 0, af, addr, ifname, i); } @@ -333,7 +333,7 @@ static int conf_ports(const struct ctx *c, char optname, const char *optarg, fwd->delta[i] = mapped_range.first - orig_range.first; if (optname == 't') - tcp_sock_init(c, 0, af, addr, ifname, i); + tcp_sock_init(c, af, addr, ifname, i); else if (optname == 'u') udp_sock_init(c, 0, af, addr, ifname, i); } diff --git a/tcp.c b/tcp.c index 35fca31..306f928 100644 --- a/tcp.c +++ b/tcp.c @@ -2987,15 +2987,15 @@ void tcp_sock_handler(struct ctx *c, union epoll_ref ref, uint32_t events, /** * tcp_sock_init4() - Initialise listening sockets for a given IPv4 port * @c: Execution context - * @ns: In pasta mode, if set, bind with loopback address in namespace * @addr: Pointer to address for binding, NULL if not configured * @ifname: Name of interface to bind to, NULL if not configured * @port: Port, host order */ -static void tcp_sock_init4(const struct ctx *c, int ns, const struct in_addr *addr, +static void tcp_sock_init4(const struct ctx *c, const struct in_addr *addr, const char *ifname, in_port_t port) { - union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.outbound = ns }; + in_port_t idx = port + c->tcp.fwd_in.delta[port]; + union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.index = idx }; bool spliced = false, tap = true; int s; @@ -3006,14 +3006,9 @@ static void tcp_sock_init4(const struct ctx *c, int ns, const struct in_addr *ad if (!addr) addr = &c->ip4.addr; - tap = !ns && !IN4_IS_ADDR_LOOPBACK(addr); + tap = !IN4_IS_ADDR_LOOPBACK(addr); } - if (ns) - tref.tcp.index = (in_port_t)(port + c->tcp.fwd_out.delta[port]); - else - tref.tcp.index = (in_port_t)(port + c->tcp.fwd_in.delta[port]); - if (tap) { s = sock_l4(c, AF_INET, IPPROTO_TCP, addr, ifname, port, tref.u32); @@ -3039,29 +3034,25 @@ static void tcp_sock_init4(const struct ctx *c, int ns, const struct in_addr *ad else s = -1; - if (c->tcp.fwd_out.mode == FWD_AUTO) { - if (ns) - tcp_sock_ns[port][V4] = s; - else - tcp_sock_init_lo[port][V4] = s; - } + if (c->tcp.fwd_out.mode == FWD_AUTO) + tcp_sock_init_lo[port][V4] = s; } } /** * tcp_sock_init6() - Initialise listening sockets for a given IPv6 port * @c: Execution context - * @ns: In pasta mode, if set, bind with loopback address in namespace * @addr: Pointer to address for binding, NULL if not configured * @ifname: Name of interface to bind to, NULL if not configured * @port: Port, host order */ -static void tcp_sock_init6(const struct ctx *c, int ns, +static void tcp_sock_init6(const struct ctx *c, const struct in6_addr *addr, const char *ifname, in_port_t port) { - union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.outbound = ns, - .tcp.v6 = 1 }; + in_port_t idx = port + c->tcp.fwd_in.delta[port]; + union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.v6 = 1, + .tcp.index = idx }; bool spliced = false, tap = true; int s; @@ -3073,14 +3064,9 @@ static void tcp_sock_init6(const struct ctx *c, int ns, if (!addr) addr = &c->ip6.addr; - tap = !ns && !IN6_IS_ADDR_LOOPBACK(addr); + tap = !IN6_IS_ADDR_LOOPBACK(addr); } - if (ns) - tref.tcp.index = (in_port_t)(port + c->tcp.fwd_out.delta[port]); - else - tref.tcp.index = (in_port_t)(port + c->tcp.fwd_in.delta[port]); - if (tap) { s = sock_l4(c, AF_INET6, IPPROTO_TCP, addr, ifname, port, tref.u32); @@ -3105,40 +3091,99 @@ static void tcp_sock_init6(const struct ctx *c, int ns, else s = -1; - if (c->tcp.fwd_out.mode == FWD_AUTO) { - if (ns) - tcp_sock_ns[port][V6] = s; - else - tcp_sock_init_lo[port][V6] = s; - } + if (c->tcp.fwd_out.mode == FWD_AUTO) + tcp_sock_init_lo[port][V6] = s; } } /** - * tcp_sock_init() - Initialise listening sockets for a given port + * tcp_sock_init() - Create listening sockets for a given host ("inbound") port * @c: Execution context - * @ns: In pasta mode, if set, bind with loopback address in namespace * @af: Address family to select a specific IP version, or AF_UNSPEC * @addr: Pointer to address for binding, NULL if not configured * @ifname: Name of interface to bind to, NULL if not configured * @port: Port, host order */ -void tcp_sock_init(const struct ctx *c, int ns, sa_family_t af, - const void *addr, const char *ifname, in_port_t port) +void tcp_sock_init(const struct ctx *c, sa_family_t af, const void *addr, + const char *ifname, in_port_t port) { if ((af == AF_INET || af == AF_UNSPEC) && c->ifi4) - tcp_sock_init4(c, ns, addr, ifname, port); + tcp_sock_init4(c, addr, ifname, port); if ((af == AF_INET6 || af == AF_UNSPEC) && c->ifi6) - tcp_sock_init6(c, ns, addr, ifname, port); + tcp_sock_init6(c, addr, ifname, port); +} + +/** + * tcp_ns_sock_init4() - Init socket to listen for outbound IPv4 connections + * @c: Execution context + * @port: Port, host order + */ +static void tcp_ns_sock_init4(const struct ctx *c, in_port_t port) +{ + in_port_t idx = port + c->tcp.fwd_out.delta[port]; + union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.outbound = 1, + .tcp.splice = 1, .tcp.index = idx }; + struct in_addr loopback = { htonl(INADDR_LOOPBACK) }; + int s; + + assert(c->mode == MODE_PASTA); + + s = sock_l4(c, AF_INET, IPPROTO_TCP, &loopback, NULL, port, tref.u32); + if (s >= 0) + tcp_sock_set_bufsize(c, s); + else + s = -1; + + if (c->tcp.fwd_out.mode == FWD_AUTO) + tcp_sock_ns[port][V4] = s; } /** - * tcp_sock_init_ns() - Bind sockets in namespace for outbound connections + * tcp_ns_sock_init6() - Init socket to listen for outbound IPv6 connections + * @c: Execution context + * @port: Port, host order + */ +static void tcp_ns_sock_init6(const struct ctx *c, in_port_t port) +{ + in_port_t idx = port + c->tcp.fwd_out.delta[port]; + union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.outbound = 1, + .tcp.splice = 1, .tcp.v6 = 1, + .tcp.index = idx }; + int s; + + assert(c->mode == MODE_PASTA); + + s = sock_l4(c, AF_INET6, IPPROTO_TCP, &in6addr_loopback, NULL, port, + tref.u32); + if (s >= 0) + tcp_sock_set_bufsize(c, s); + else + s = -1; + + if (c->tcp.fwd_out.mode == FWD_AUTO) + tcp_sock_ns[port][V6] = s; +} + +/** + * tcp_ns_sock_init() - Init socket to listen for spliced outbound connections + * @c: Execution context + * @port: Port, host order + */ +void tcp_ns_sock_init(const struct ctx *c, in_port_t port) +{ + if (c->ifi4) + tcp_ns_sock_init4(c, port); + if (c->ifi6) + tcp_ns_sock_init6(c, port); +} + +/** + * tcp_ns_socks_init() - Bind sockets in namespace for outbound connections * @arg: Execution context * * Return: 0 */ -static int tcp_sock_init_ns(void *arg) +static int tcp_ns_socks_init(void *arg) { struct ctx *c = (struct ctx *)arg; unsigned port; @@ -3149,7 +3194,7 @@ static int tcp_sock_init_ns(void *arg) if (!bitmap_isset(c->tcp.fwd_out.map, port)) continue; - tcp_sock_init(c, 1, AF_UNSPEC, NULL, NULL, port); + tcp_ns_sock_init(c, port); } return 0; @@ -3279,7 +3324,7 @@ int tcp_init(struct ctx *c) if (c->mode == MODE_PASTA) { tcp_splice_init(c); - NS_CALL(tcp_sock_init_ns, c); + NS_CALL(tcp_ns_socks_init, c); refill_arg.ns = 1; NS_CALL(tcp_sock_refill, &refill_arg); @@ -3364,8 +3409,7 @@ static int tcp_port_rebind(void *arg) if ((a->c->ifi4 && tcp_sock_ns[port][V4] == -1) || (a->c->ifi6 && tcp_sock_ns[port][V6] == -1)) - tcp_sock_init(a->c, 1, AF_UNSPEC, NULL, NULL, - port); + tcp_ns_sock_init(a->c, port); } } else { for (port = 0; port < NUM_PORTS; port++) { @@ -3398,7 +3442,7 @@ static int tcp_port_rebind(void *arg) if ((a->c->ifi4 && tcp_sock_init_ext[port][V4] == -1) || (a->c->ifi6 && tcp_sock_init_ext[port][V6] == -1)) - tcp_sock_init(a->c, 0, AF_UNSPEC, NULL, NULL, + tcp_sock_init(a->c, AF_UNSPEC, NULL, NULL, port); } } diff --git a/tcp.h b/tcp.h index 49738ef..f4ed298 100644 --- a/tcp.h +++ b/tcp.h @@ -19,8 +19,8 @@ void tcp_sock_handler(struct ctx *c, union epoll_ref ref, uint32_t events, const struct timespec *now); int tcp_tap_handler(struct ctx *c, int af, const void *addr, const struct pool *p, const struct timespec *now); -void tcp_sock_init(const struct ctx *c, int ns, sa_family_t af, - const void *addr, const char *ifname, in_port_t port); +void tcp_sock_init(const struct ctx *c, sa_family_t af, const void *addr, + const char *ifname, in_port_t port); int tcp_init(struct ctx *c); void tcp_timer(struct ctx *c, const struct timespec *ts); void tcp_defer_handler(struct ctx *c); -- 2.38.1

In pasta mode, tcp_sock_init[46]() create separate sockets to listen for spliced connections (these are bound to localhost) and non-spliced connections (these are bound to the host address). This introduces a subtle behavioural difference between pasta and passt: by default, pasta will listen only on a single host address, whereas passt will listen on all addresses (0.0.0.0 or ::). This also prevents us using some additional optimizations that only work with the unspecified (0.0.0.0 or ::) address. However, it turns out we don't need to do this. We can splice a connection if and only if it originates from the loopback address. Currently we ensure this by having the "spliced" listening sockets listening only on loopback. Instead, defer the decision about whether to splice a connection until after accept(), by checking if the connection was made from the loopback address. Signed-off-by: David Gibson --- tcp.c | 127 +++++++++++++-------------------------------------- tcp_splice.c | 26 +++++++++-- tcp_splice.h | 5 +- 3 files changed, 56 insertions(+), 102 deletions(-) diff --git a/tcp.c b/tcp.c index a53c70d..7df3c09 100644 --- a/tcp.c +++ b/tcp.c @@ -434,7 +434,6 @@ static const char *tcp_flag_str[] __attribute((__unused__)) = { }; /* Listening sockets, used for automatic port forwarding in pasta mode only */ -static int tcp_sock_init_lo [NUM_PORTS][IP_VERSIONS]; static int tcp_sock_init_ext [NUM_PORTS][IP_VERSIONS]; static int tcp_sock_ns [NUM_PORTS][IP_VERSIONS]; @@ -2851,21 +2850,31 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, socklen_t sl; int s; + assert(ref.r.p.tcp.tcp.listen); + assert(!ref.r.p.tcp.tcp.splice); + if (c->tcp.conn_count >= TCP_MAX_CONNS) return; sl = sizeof(sa); + /* FIXME: Workaround clang-tidy not realizing that accept4() + * writes the socket address. See + * https://github.com/llvm/llvm-project/issues/58992 + */ + memset(&sa, 0, sizeof(struct sockaddr_in6)); s = accept4(ref.r.s, (struct sockaddr *)&sa, &sl, SOCK_NONBLOCK); if (s < 0) return; conn = tc + c->tcp.conn_count++; - if (ref.r.p.tcp.tcp.splice) - tcp_splice_conn_from_sock(c, ref, &conn->splice, s); - else - tcp_tap_conn_from_sock(c, ref, &conn->tap, s, - (struct sockaddr *)&sa, now); + if (c->mode == MODE_PASTA && + tcp_splice_conn_from_sock(c, ref, &conn->splice, + s, (struct sockaddr *)&sa)) + return; + + tcp_tap_conn_from_sock(c, ref, &conn->tap, s, + (struct sockaddr *)&sa, now); } /** @@ -3018,47 +3027,16 @@ static void tcp_sock_init4(const struct ctx *c, const struct in_addr *addr, { in_port_t idx = port + c->tcp.fwd_in.delta[port]; union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.index = idx }; - bool spliced = false, tap = true; int s; - if (c->mode == MODE_PASTA) { - spliced = !addr || IN4_IS_ADDR_UNSPECIFIED(addr) || - IN4_IS_ADDR_LOOPBACK(addr); - - if (!addr) - addr = &c->ip4.addr; - - tap = !IN4_IS_ADDR_LOOPBACK(addr); - } - - if (tap) { - s = sock_l4(c, AF_INET, IPPROTO_TCP, addr, ifname, port, - tref.u32); - if (s >= 0) - tcp_sock_set_bufsize(c, s); - else - s = -1; - - if (c->tcp.fwd_in.mode == FWD_AUTO) - tcp_sock_init_ext[port][V4] = s; - } - - if (spliced) { - struct in_addr loopback = { htonl(INADDR_LOOPBACK) }; - tref.tcp.splice = 1; - - addr = &loopback; - - s = sock_l4(c, AF_INET, IPPROTO_TCP, addr, ifname, port, - tref.u32); - if (s >= 0) - tcp_sock_set_bufsize(c, s); - else - s = -1; + s = sock_l4(c, AF_INET, IPPROTO_TCP, addr, ifname, port, tref.u32); + if (s >= 0) + tcp_sock_set_bufsize(c, s); + else + s = -1; - if (c->tcp.fwd_out.mode == FWD_AUTO) - tcp_sock_init_lo[port][V4] = s; - } + if (c->tcp.fwd_in.mode == FWD_AUTO) + tcp_sock_init_ext[port][V4] = s; } /** @@ -3075,47 +3053,16 @@ static void tcp_sock_init6(const struct ctx *c, in_port_t idx = port + c->tcp.fwd_in.delta[port]; union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.v6 = 1, .tcp.index = idx }; - bool spliced = false, tap = true; int s; - if (c->mode == MODE_PASTA) { - spliced = !addr || - IN6_IS_ADDR_UNSPECIFIED(addr) || - IN6_IS_ADDR_LOOPBACK(addr); - - if (!addr) - addr = &c->ip6.addr; - - tap = !IN6_IS_ADDR_LOOPBACK(addr); - } - - if (tap) { - s = sock_l4(c, AF_INET6, IPPROTO_TCP, addr, ifname, port, - tref.u32); - if (s >= 0) - tcp_sock_set_bufsize(c, s); - else - s = -1; - - if (c->tcp.fwd_in.mode == FWD_AUTO) - tcp_sock_init_ext[port][V6] = s; - } - - if (spliced) { - tref.tcp.splice = 1; - - addr = &in6addr_loopback; - - s = sock_l4(c, AF_INET6, IPPROTO_TCP, addr, ifname, port, - tref.u32); - if (s >= 0) - tcp_sock_set_bufsize(c, s); - else - s = -1; + s = sock_l4(c, AF_INET6, IPPROTO_TCP, addr, ifname, port, tref.u32); + if (s >= 0) + tcp_sock_set_bufsize(c, s); + else + s = -1; - if (c->tcp.fwd_out.mode == FWD_AUTO) - tcp_sock_init_lo[port][V6] = s; - } + if (c->tcp.fwd_in.mode == FWD_AUTO) + tcp_sock_init_ext[port][V6] = s; } /** @@ -3144,7 +3091,7 @@ static void tcp_ns_sock_init4(const struct ctx *c, in_port_t port) { in_port_t idx = port + c->tcp.fwd_out.delta[port]; union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.outbound = 1, - .tcp.splice = 1, .tcp.index = idx }; + .tcp.index = idx }; struct in_addr loopback = { htonl(INADDR_LOOPBACK) }; int s; @@ -3169,8 +3116,7 @@ static void tcp_ns_sock_init6(const struct ctx *c, in_port_t port) { in_port_t idx = port + c->tcp.fwd_out.delta[port]; union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.outbound = 1, - .tcp.splice = 1, .tcp.v6 = 1, - .tcp.index = idx }; + .tcp.v6 = 1, .tcp.index = idx }; int s; assert(c->mode == MODE_PASTA); @@ -3337,7 +3283,6 @@ int tcp_init(struct ctx *c) memset(init_sock_pool6, 0xff, sizeof(init_sock_pool6)); memset(ns_sock_pool4, 0xff, sizeof(ns_sock_pool4)); memset(ns_sock_pool6, 0xff, sizeof(ns_sock_pool6)); - memset(tcp_sock_init_lo, 0xff, sizeof(tcp_sock_init_lo)); memset(tcp_sock_init_ext, 0xff, sizeof(tcp_sock_init_ext)); memset(tcp_sock_ns, 0xff, sizeof(tcp_sock_ns)); @@ -3445,16 +3390,6 @@ static int tcp_port_rebind(void *arg) close(tcp_sock_init_ext[port][V6]); tcp_sock_init_ext[port][V6] = -1; } - - if (tcp_sock_init_lo[port][V4] >= 0) { - close(tcp_sock_init_lo[port][V4]); - tcp_sock_init_lo[port][V4] = -1; - } - - if (tcp_sock_init_lo[port][V6] >= 0) { - close(tcp_sock_init_lo[port][V6]); - tcp_sock_init_lo[port][V6] = -1; - } continue; } diff --git a/tcp_splice.c b/tcp_splice.c index 2fd88ec..356d194 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -502,30 +502,48 @@ static void tcp_splice_dir(struct tcp_splice_conn *conn, int ref_sock, } /** - * tcp_splice_conn_from_sock() - Initialize state for spliced connection + * tcp_splice_conn_from_sock() - Attempt to init state for a spliced connection * @c: Execution context * @ref: epoll reference of listening socket * @conn: connection structure to initialize * @s: Accepted socket + * @sa: Peer address of connection * + * Return: true if able to create a spliced connection, false otherwise * #syscalls:pasta setsockopt */ -void tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref, - struct tcp_splice_conn *conn, int s) +bool tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref, + struct tcp_splice_conn *conn, int s, + const struct sockaddr *sa) { assert(c->mode == MODE_PASTA); + if (ref.r.p.tcp.tcp.v6) { + const struct sockaddr_in6 *sa6; + + sa6 = (const struct sockaddr_in6 *)sa; + if (!IN6_IS_ADDR_LOOPBACK(&sa6->sin6_addr)) + return false; + conn->flags = SPLICE_V6; + } else { + const struct sockaddr_in *sa4 = (const struct sockaddr_in *)sa; + if (!IN4_IS_ADDR_LOOPBACK(&sa4->sin_addr)) + return false; + conn->flags = 0; + } + if (setsockopt(s, SOL_TCP, TCP_QUICKACK, &((int){ 1 }), sizeof(int))) trace("TCP (spliced): failed to set TCP_QUICKACK on %i", s); conn->c.spliced = true; c->tcp.splice_conn_count++; conn->a = s; - conn->flags = ref.r.p.tcp.tcp.v6 ? SPLICE_V6 : 0; if (tcp_splice_new(c, conn, ref.r.p.tcp.tcp.index, ref.r.p.tcp.tcp.outbound)) conn_flag(c, conn, CLOSING); + + return true; } /** diff --git a/tcp_splice.h b/tcp_splice.h index f9462ae..1a915dd 100644 --- a/tcp_splice.h +++ b/tcp_splice.h @@ -10,8 +10,9 @@ struct tcp_splice_conn; void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, uint32_t events); -void tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref, - struct tcp_splice_conn *conn, int s); +bool tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref, + struct tcp_splice_conn *conn, int s, + const struct sockaddr *sa); void tcp_splice_init(struct ctx *c); #endif /* TCP_SPLICE_H */ -- 2.38.1

Currently when we insert a connection into the hash table, we store its bucket number so we can find it when removing entries. However, we can recompute the hash value from other contents of the structure so we don't need to store it. This brings the size of tcp_tap_conn down to 64 bytes again, which means it will fit in a single cacheline on common machines. This change also removes a non-obvious constraint that the hash table have less than twice TCP_MAX_CONNS buckets, because of the way TCP_HASH_BUCKET_BITS was constructed. Signed-off-by: David Gibson --- tcp.c | 29 ++++++++++++++++++++++++----- tcp_conn.h | 5 ----- 2 files changed, 24 insertions(+), 10 deletions(-) diff --git a/tcp.c b/tcp.c index b11ce6e..e9803aa 100644 --- a/tcp.c +++ b/tcp.c @@ -1243,6 +1243,24 @@ static unsigned int tcp_hash(const struct ctx *c, int af, const void *addr, return (unsigned int)(b % TCP_HASH_TABLE_SIZE); } +/** + * tcp_conn_hash() - Calculate hash bucket of an existing connection + * @c: Execution context + * @conn: Connection + * + * Return: hash value, already modulo size of the hash table + */ +static unsigned int tcp_conn_hash(const struct ctx *c, + const struct tcp_tap_conn *conn) +{ + if (CONN_V6(conn)) + return tcp_hash(c, AF_INET6, &conn->a.a6, + conn->tap_port, conn->sock_port); + else + return tcp_hash(c, AF_INET, &conn->a.a4.a, + conn->tap_port, conn->sock_port); +} + /** * tcp_hash_insert() - Insert connection into hash table, chain link * @c: Execution context @@ -1258,7 +1276,6 @@ static void tcp_hash_insert(const struct ctx *c, struct tcp_tap_conn *conn, b = tcp_hash(c, af, addr, conn->tap_port, conn->sock_port); conn->next_index = tc_hash[b] ? CONN_IDX(tc_hash[b]) : -1; tc_hash[b] = conn; - conn->hash_bucket = b; debug("TCP: hash table insert: index %li, sock %i, bucket: %i, next: " "%p", CONN_IDX(conn), conn->sock, b, conn_at_idx(conn->next_index)); @@ -1266,12 +1283,14 @@ static void tcp_hash_insert(const struct ctx *c, struct tcp_tap_conn *conn, /** * tcp_hash_remove() - Drop connection from hash table, chain unlink + * @c: Execution context * @conn: Connection pointer */ -static void tcp_hash_remove(const struct tcp_tap_conn *conn) +static void tcp_hash_remove(const struct ctx *c, + const struct tcp_tap_conn *conn) { struct tcp_tap_conn *entry, *prev = NULL; - int b = conn->hash_bucket; + int b = tcp_conn_hash(c, conn); for (entry = tc_hash[b]; entry; prev = entry, entry = conn_at_idx(entry->next_index)) { @@ -1299,7 +1318,7 @@ static void tcp_tap_conn_update(struct ctx *c, struct tcp_tap_conn *old, struct tcp_tap_conn *new) { struct tcp_tap_conn *entry, *prev = NULL; - int b = old->hash_bucket; + int b = tcp_conn_hash(c, old); for (entry = tc_hash[b]; entry; prev = entry, entry = conn_at_idx(entry->next_index)) { @@ -1387,7 +1406,7 @@ static void tcp_conn_destroy(struct ctx *c, struct tcp_tap_conn *conn) if (conn->timer != -1) close(conn->timer); - tcp_hash_remove(conn); + tcp_hash_remove(c, conn); tcp_table_compact(c, (union tcp_conn *)conn); } diff --git a/tcp_conn.h b/tcp_conn.h index faa63dc..4bffe9a 100644 --- a/tcp_conn.h +++ b/tcp_conn.h @@ -9,8 +9,6 @@ #ifndef TCP_CONN_H #define TCP_CONN_H -#define TCP_HASH_BUCKET_BITS (TCP_CONN_INDEX_BITS + 1) - /** * struct tcp_conn_common - Common fields for spliced and non-spliced * @spliced: Is this a spliced connection? @@ -32,7 +30,6 @@ extern const char *tcp_common_flag_str[]; * @events: Connection events, implying connection states * @timer: timerfd descriptor for timeout events * @flags: Connection flags representing internal attributes - * @hash_bucket: Bucket index in connection lookup hash table * @retrans: Number of retransmissions occurred due to ACK_TIMEOUT * @ws_from_tap: Window scaling factor advertised from tap/guest * @ws_to_tap: Window scaling factor advertised to tap/guest @@ -97,8 +94,6 @@ struct tcp_tap_conn { #define ACK_FROM_TAP_DUE BIT(5) - unsigned int hash_bucket :TCP_HASH_BUCKET_BITS; - #define TCP_MSS_BITS 14 unsigned int tap_mss :TCP_MSS_BITS; #define MSS_SET(conn, mss) (conn->tap_mss = (mss >> (16 - TCP_MSS_BITS))) -- 2.38.1

In the tcp_conn structure, we represent the address with an inany_addr which could be an IPv4 or IPv6 address. However, we have different paths which will calculate different hashes for IPv4 and equivalent IPv4-mapped IPv6 addresses. This will cause problems for some future changes. Make the hash function work the same for these two cases, by taking an inany_addr directly. Since this represents IPv4 and IPv4-mapped IPv6 addresses the same way, it will trivially hash the same for both cases. Callers are changed to construct an inany_addr from whatever they have. Some of that will be elided in later changes. Signed-off-by: David Gibson --- siphash.c | 1 + tcp.c | 52 ++++++++++++++++++---------------------------------- 2 files changed, 19 insertions(+), 34 deletions(-) diff --git a/siphash.c b/siphash.c index 37a6d73..516a508 100644 --- a/siphash.c +++ b/siphash.c @@ -104,6 +104,7 @@ * * Return: the 64-bit hash output */ +/* cppcheck-suppress unusedFunction */ uint64_t siphash_8b(const uint8_t *in, const uint64_t *k) { PREAMBLE(8); diff --git a/tcp.c b/tcp.c index d1008ae..92dd422 100644 --- a/tcp.c +++ b/tcp.c @@ -1205,8 +1205,7 @@ static int tcp_hash_match(const struct tcp_tap_conn *conn, /** * tcp_hash() - Calculate hash value for connection given address and ports * @c: Execution context - * @af: Address family, AF_INET or AF_INET6 - * @addr: Remote address, pointer to in_addr or in6_addr + * @addr: Remote address * @tap_port: tap-facing port * @sock_port: Socket-facing port * @@ -1215,32 +1214,19 @@ static int tcp_hash_match(const struct tcp_tap_conn *conn, #if TCP_HASH_NOINLINE __attribute__((__noinline__)) /* See comment in Makefile */ #endif -static unsigned int tcp_hash(const struct ctx *c, int af, const void *addr, +static unsigned int tcp_hash(const struct ctx *c, const union inany_addr *addr, in_port_t tap_port, in_port_t sock_port) { + struct { + union inany_addr addr; + in_port_t tap_port; + in_port_t sock_port; + } __attribute__((__packed__)) in = { + *addr, tap_port, sock_port + }; uint64_t b = 0; - if (af == AF_INET) { - struct { - struct in_addr addr; - in_port_t tap_port; - in_port_t sock_port; - } __attribute__((__packed__)) in = { - *(struct in_addr *)addr, tap_port, sock_port, - }; - - b = siphash_8b((uint8_t *)&in, c->tcp.hash_secret); - } else if (af == AF_INET6) { - struct { - struct in6_addr addr; - in_port_t tap_port; - in_port_t sock_port; - } __attribute__((__packed__)) in = { - *(struct in6_addr *)addr, tap_port, sock_port, - }; - - b = siphash_20b((uint8_t *)&in, c->tcp.hash_secret); - } + b = siphash_20b((uint8_t *)&in, c->tcp.hash_secret); return (unsigned int)(b % TCP_HASH_TABLE_SIZE); } @@ -1255,14 +1241,7 @@ static unsigned int tcp_hash(const struct ctx *c, int af, const void *addr, static unsigned int tcp_conn_hash(const struct ctx *c, const struct tcp_tap_conn *conn) { - const struct in_addr *a4 = inany_v4(&conn->addr); - - if (a4) - return tcp_hash(c, AF_INET, a4, - conn->tap_port, conn->sock_port); - else - return tcp_hash(c, AF_INET6, &conn->addr.a6, - conn->tap_port, conn->sock_port); + return tcp_hash(c, &conn->addr, conn->tap_port, conn->sock_port); } /** @@ -1275,9 +1254,11 @@ static unsigned int tcp_conn_hash(const struct ctx *c, static void tcp_hash_insert(const struct ctx *c, struct tcp_tap_conn *conn, int af, const void *addr) { + union inany_addr aany; int b; - b = tcp_hash(c, af, addr, conn->tap_port, conn->sock_port); + inany_from_af(&aany, af, addr); + b = tcp_hash(c, &aany, conn->tap_port, conn->sock_port); conn->next_index = tc_hash[b] ? CONN_IDX(tc_hash[b]) : -1; tc_hash[b] = conn; @@ -1357,9 +1338,12 @@ static struct tcp_tap_conn *tcp_hash_lookup(const struct ctx *c, in_port_t tap_port, in_port_t sock_port) { - int b = tcp_hash(c, af, addr, tap_port, sock_port); + union inany_addr aany; struct tcp_tap_conn *conn; + int b; + inany_from_af(&aany, af, addr); + b = tcp_hash(c, &aany, tap_port, sock_port); for (conn = tc_hash[b]; conn; conn = conn_at_idx(conn->next_index)) { if (tcp_hash_match(conn, af, addr, tap_port, sock_port)) return conn; -- 2.38.1

tcp_hash_match() can take either an IPv4 (struct in_addr) or IPv6 (struct in6_addr) address. It has two different paths for each of those cases. However, its only caller has already constructed an equivalent inany representation of the address, so we can have tcp_hash_match take that directly and use a simpler comparison with the inany_equals() helper. Signed-off-by: David Gibson --- tcp.c | 16 ++++------------ 1 file changed, 4 insertions(+), 12 deletions(-) diff --git a/tcp.c b/tcp.c index 0ce7944..8ebd90a 100644 --- a/tcp.c +++ b/tcp.c @@ -1177,25 +1177,17 @@ static int tcp_opt_get(const char *opts, size_t len, uint8_t type_find, /** * tcp_hash_match() - Check if a connection entry matches address and ports * @conn: Connection entry to match against - * @af: Address family, AF_INET or AF_INET6 - * @addr: Remote address, pointer to in_addr or in6_addr + * @addr: Remote address * @tap_port: tap-facing port * @sock_port: Socket-facing port * * Return: 1 on match, 0 otherwise */ static int tcp_hash_match(const struct tcp_tap_conn *conn, - int af, const void *addr, + const union inany_addr *addr, in_port_t tap_port, in_port_t sock_port) { - const struct in_addr *a4 = inany_v4(&conn->addr); - - if (af == AF_INET && a4 && !memcmp(a4, addr, sizeof(*a4)) && - conn->tap_port == tap_port && conn->sock_port == sock_port) - return 1; - - if (af == AF_INET6 && - IN6_ARE_ADDR_EQUAL(&conn->addr.a6, addr) && + if (inany_equals(&conn->addr, addr) && conn->tap_port == tap_port && conn->sock_port == sock_port) return 1; @@ -1340,7 +1332,7 @@ static struct tcp_tap_conn *tcp_hash_lookup(const struct ctx *c, inany_from_af(&aany, af, addr); b = tcp_hash(c, &aany, tap_port, sock_port); for (conn = tc_hash[b]; conn; conn = conn_at_idx(conn->next_index)) { - if (tcp_hash_match(conn, af, addr, tap_port, sock_port)) + if (tcp_hash_match(conn, &aany, tap_port, sock_port)) return conn; } -- 2.38.1

tcp_seq_init() takes a number of parameters for the connection, but at every call site, these are already populated in the tcp_conn structure. Likewise we always store the result into the @seq_to_tap field. Use this to simplify tcp_seq_init(). Signed-off-by: David Gibson --- tcp.c | 36 ++++++++++-------------------------- 1 file changed, 10 insertions(+), 26 deletions(-) diff --git a/tcp.c b/tcp.c index ac64b81..dabcbc4 100644 --- a/tcp.c +++ b/tcp.c @@ -1930,17 +1930,11 @@ static void tcp_clamp_window(const struct ctx *c, struct tcp_tap_conn *conn, /** * tcp_seq_init() - Calculate initial sequence number according to RFC 6528 * @c: Execution context - * @af: Address family, AF_INET or AF_INET6 - * @addr: Remote address, pointer to in_addr or in6_addr - * @dstport: Destination port, connection-wise, network order - * @srcport: Source port, connection-wise, network order + * @conn: TCP connection, with addr, sock_port and tap_port populated * @now: Current timestamp - * - * Return: initial TCP sequence */ -static uint32_t tcp_seq_init(const struct ctx *c, int af, const void *addr, - in_port_t dstport, in_port_t srcport, - const struct timespec *now) +static void tcp_seq_init(const struct ctx *c, struct tcp_tap_conn *conn, + const struct timespec *now) { union inany_addr aany; struct { @@ -1949,14 +1943,13 @@ static uint32_t tcp_seq_init(const struct ctx *c, int af, const void *addr, union inany_addr dst; in_port_t dstport; } __attribute__((__packed__)) in = { - .srcport = srcport, - .dstport = dstport, + .src = conn->addr, + .srcport = conn->tap_port, + .dstport = conn->sock_port, }; uint32_t ns, seq = 0; - inany_from_af(&aany, af, addr); - in.src = aany; - if (af == AF_INET) + if (CONN_V4(conn)) inany_from_af(&aany, AF_INET, &c->ip4.addr); else inany_from_af(&aany, AF_INET6, &c->ip6.addr); @@ -1967,7 +1960,7 @@ static uint32_t tcp_seq_init(const struct ctx *c, int af, const void *addr, ns = now->tv_sec * 1E9; ns += now->tv_nsec >> 5; /* 32ns ticks, overflows 32 bits every 137s */ - return seq + ns; + conn->seq_to_tap = seq + ns; } /** @@ -2127,7 +2120,7 @@ static void tcp_conn_from_tap(struct ctx *c, int af, const void *addr, conn->seq_from_tap = conn->seq_init_from_tap + 1; conn->seq_ack_to_tap = conn->seq_from_tap; - conn->seq_to_tap = tcp_seq_init(c, af, addr, th->dest, th->source, now); + tcp_seq_init(c, conn, now); conn->seq_ack_from_tap = conn->seq_to_tap + 1; tcp_hash_insert(c, conn); @@ -2774,11 +2767,6 @@ static void tcp_tap_conn_from_sock(struct ctx *c, union epoll_ref ref, conn->sock_port = ntohs(sa6.sin6_port); conn->tap_port = ref.r.p.tcp.tcp.index; - - conn->seq_to_tap = tcp_seq_init(c, AF_INET6, &sa6.sin6_addr, - conn->sock_port, - conn->tap_port, - now); } else { struct sockaddr_in sa4; @@ -2793,13 +2781,9 @@ static void tcp_tap_conn_from_sock(struct ctx *c, union epoll_ref ref, conn->sock_port = ntohs(sa4.sin_port); conn->tap_port = ref.r.p.tcp.tcp.index; - - conn->seq_to_tap = tcp_seq_init(c, AF_INET, &sa4.sin_addr, - conn->sock_port, - conn->tap_port, - now); } + tcp_seq_init(c, conn, now); tcp_hash_insert(c, conn); conn->seq_ack_from_tap = conn->seq_to_tap + 1; -- 2.38.1

This bit in the TCP specific epoll reference indicates whether the connection is IPv6 or IPv4. However the sites which refer to it are already calling accept() which (optionally) returns an address for the remote end of the connection. We can use the sa_family field in that address to determine the connection type independent of the epoll reference. This does have a cost: for the spliced case, it means we now need to get that address from accept() which introduces an extran copy_to_user(). However, in future we want to allow handling IPv4 connectons through IPv6 sockets, which means we won't be able to determine the IP version at the time we create the listening socket and epoll reference. So, at some point we'll have to pay this cost anyway. Signed-off-by: David Gibson --- tcp.c | 10 ++++------ tcp.h | 2 -- tcp_splice.c | 8 +++----- 3 files changed, 7 insertions(+), 13 deletions(-) diff --git a/tcp.c b/tcp.c index 559e271..7d5ac6c 100644 --- a/tcp.c +++ b/tcp.c @@ -662,8 +662,7 @@ static int tcp_epoll_ctl(const struct ctx *c, struct tcp_tap_conn *conn) { int m = conn->c.in_epoll ? EPOLL_CTL_MOD : EPOLL_CTL_ADD; union epoll_ref ref = { .r.proto = IPPROTO_TCP, .r.s = conn->sock, - .r.p.tcp.tcp.index = CONN_IDX(conn), - .r.p.tcp.tcp.v6 = CONN_V6(conn) }; + .r.p.tcp.tcp.index = CONN_IDX(conn) }; struct epoll_event ev = { .data.u64 = ref.u64 }; if (conn->events == CLOSED) { @@ -2745,7 +2744,7 @@ static void tcp_tap_conn_from_sock(struct ctx *c, union epoll_ref ref, conn->ws_to_tap = conn->ws_from_tap = 0; conn_event(c, conn, SOCK_ACCEPTED); - if (ref.r.p.tcp.tcp.v6) { + if (sa->sa_family == AF_INET6) { struct sockaddr_in6 sa6; memcpy(&sa6, sa, sizeof(sa6)); @@ -3019,8 +3018,7 @@ static void tcp_sock_init6(const struct ctx *c, in_port_t port) { in_port_t idx = port + c->tcp.fwd_in.delta[port]; - union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.v6 = 1, - .tcp.index = idx }; + union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.index = idx }; int s; s = sock_l4(c, AF_INET6, IPPROTO_TCP, addr, ifname, port, tref.u32); @@ -3084,7 +3082,7 @@ static void tcp_ns_sock_init6(const struct ctx *c, in_port_t port) { in_port_t idx = port + c->tcp.fwd_out.delta[port]; union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.outbound = 1, - .tcp.v6 = 1, .tcp.index = idx }; + .tcp.index = idx }; int s; assert(c->mode == MODE_PASTA); diff --git a/tcp.h b/tcp.h index a940682..739b451 100644 --- a/tcp.h +++ b/tcp.h @@ -33,7 +33,6 @@ void tcp_update_l2_buf(const unsigned char *eth_d, const unsigned char *eth_s, * union tcp_epoll_ref - epoll reference portion for TCP connections * @listen: Set if this file descriptor is a listening socket * @outbound: Listening socket maps to outbound, spliced connection - * @v6: Set for IPv6 sockets or connections * @timer: Reference is a timerfd descriptor for connection * @index: Index of connection in table, or port for bound sockets * @u32: Opaque u32 value of reference @@ -42,7 +41,6 @@ union tcp_epoll_ref { struct { uint32_t listen:1, outbound:1, - v6:1, timer:1, index:20; } tcp; diff --git a/tcp_splice.c b/tcp_splice.c index 2be9b77..1bbd085 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -167,11 +167,9 @@ static int tcp_splice_epoll_ctl(const struct ctx *c, { int m = conn->c.in_epoll ? EPOLL_CTL_MOD : EPOLL_CTL_ADD; union epoll_ref ref_a = { .r.proto = IPPROTO_TCP, .r.s = conn->a, - .r.p.tcp.tcp.index = CONN_IDX(conn), - .r.p.tcp.tcp.v6 = CONN_V6(conn) }; + .r.p.tcp.tcp.index = CONN_IDX(conn) }; union epoll_ref ref_b = { .r.proto = IPPROTO_TCP, .r.s = conn->b, - .r.p.tcp.tcp.index = CONN_IDX(conn), - .r.p.tcp.tcp.v6 = CONN_V6(conn) }; + .r.p.tcp.tcp.index = CONN_IDX(conn) }; struct epoll_event ev_a = { .data.u64 = ref_a.u64 }; struct epoll_event ev_b = { .data.u64 = ref_b.u64 }; uint32_t events_a, events_b; @@ -517,7 +515,7 @@ bool tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref, { assert(c->mode == MODE_PASTA); - if (ref.r.p.tcp.tcp.v6) { + if (sa->sa_family == AF_INET6) { const struct sockaddr_in6 *sa6; sa6 = (const struct sockaddr_in6 *)sa; -- 2.38.1

For non-spliced connections we now treat IPv4-mapped IPv6 addresses the same as the corresponding IPv4 addresses. However currently we won't splice a connection from ::ffff:127.0.0.1 the way we would one from 127.0.0.1. Correct this so that we can splice connections from IPv4 localhost that have been received on an IPv6 dual stack socket. Signed-off-by: David Gibson --- tcp_splice.c | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-) diff --git a/tcp_splice.c b/tcp_splice.c index 1bbd085..e2f0ce1 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -513,20 +513,23 @@ bool tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref, struct tcp_splice_conn *conn, int s, const struct sockaddr *sa) { + const struct in_addr *a4; + union inany_addr aany; + in_port_t port; + assert(c->mode == MODE_PASTA); - if (sa->sa_family == AF_INET6) { - const struct sockaddr_in6 *sa6; + inany_from_sockaddr(&aany, &port, sa); + a4 = inany_v4(&aany); - sa6 = (const struct sockaddr_in6 *)sa; - if (!IN6_IS_ADDR_LOOPBACK(&sa6->sin6_addr)) + if (a4) { + if (!IN4_IS_ADDR_LOOPBACK(a4)) return false; - conn->flags = SPLICE_V6; + conn->flags = 0; } else { - const struct sockaddr_in *sa4 = (const struct sockaddr_in *)sa; - if (!IN4_IS_ADDR_LOOPBACK(&sa4->sin_addr)) + if (!IN6_IS_ADDR_LOOPBACK(&aany.a6)) return false; - conn->flags = 0; + conn->flags = SPLICE_V6; } if (setsockopt(s, SOL_TCP, TCP_QUICKACK, &((int){ 1 }), sizeof(int))) -- 2.38.1

Currently, when instructed to open an IPv6 socket, sock_l4() explicitly sets the IPV6_V6ONLY socket option so that the socket will only respond to IPv6 connections. Linux (and probably other platforms) allow "dual stack" sockets: IPv6 sockets which can also accept IPv4 connections. Extend sock_l4() to be able to make such sockets, by passing AF_UNSPEC as the address family and no bind address (binding to a specific address would defeat the purpose). We add a Makefile define 'DUAL_STACK_SOCKETS' to indicate availability of this feature on the target platform. Signed-off-by: David Gibson --- Makefile | 5 +++++ util.c | 17 +++++++++++++++-- 2 files changed, 20 insertions(+), 2 deletions(-) diff --git a/Makefile b/Makefile index ce159e5..5a0f96a 100644 --- a/Makefile +++ b/Makefile @@ -11,6 +11,10 @@ VERSION ?= $(shell git describe --tags HEAD 2>/dev/null || echo "unknown\ version") +# Does the target platform allow IPv4 connections to be handled via +# the IPv6 socket API? (Linux does) +DUAL_STACK_SOCKETS := 1 + RLIMIT_STACK_VAL := $(shell /bin/sh -c 'ulimit -s') ifeq ($(RLIMIT_STACK_VAL),unlimited) RLIMIT_STACK_VAL := 1024 @@ -36,6 +40,7 @@ FLAGS += -DPASST_AUDIT_ARCH=AUDIT_ARCH_$(AUDIT_ARCH) FLAGS += -DRLIMIT_STACK_VAL=$(RLIMIT_STACK_VAL) FLAGS += -DARCH=\"$(TARGET_ARCH)\" FLAGS += -DVERSION=\"$(VERSION)\" +FLAGS += -DDUAL_STACK_SOCKETS=$(DUAL_STACK_SOCKETS) PASST_SRCS = arch.c arp.c checksum.c conf.c dhcp.c dhcpv6.c icmp.c igmp.c \ isolation.c lineread.c log.c mld.c ndp.c netlink.c packet.c passt.c \ diff --git a/util.c b/util.c index 2125a67..59c6641 100644 --- a/util.c +++ b/util.c @@ -22,6 +22,8 @@ #include #include #include +#include +#include #include "util.h" #include "passt.h" @@ -112,6 +114,7 @@ int sock_l4(const struct ctx *c, int af, uint8_t proto, 0, IN6ADDR_ANY_INIT, 0, }; const struct sockaddr *sa; + bool dual_stack = false; struct epoll_event ev; int fd, sl, y = 1; @@ -119,6 +122,13 @@ int sock_l4(const struct ctx *c, int af, uint8_t proto, proto != IPPROTO_ICMP && proto != IPPROTO_ICMPV6) return -1; /* Not implemented. */ + if (af == AF_UNSPEC) { + if (!DUAL_STACK_SOCKETS || bind_addr) + return -1; + dual_stack = true; + af = AF_INET6; + } + if (proto == IPPROTO_TCP) fd = socket(af, SOCK_STREAM | SOCK_NONBLOCK, proto); else @@ -158,8 +168,11 @@ int sock_l4(const struct ctx *c, int af, uint8_t proto, sa = (const struct sockaddr *)&addr6; sl = sizeof(addr6); - if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &y, sizeof(y))) - debug("Failed to set IPV6_V6ONLY on socket %i", fd); + if (!dual_stack) + if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, + &y, sizeof(y))) + debug("Failed to set IPV6_V6ONLY on socket %i", + fd); } if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &y, sizeof(y))) -- 2.38.1

Platforms like Linux allow IPv6 sockets to listen for IPv4 connections as well as native IPv6 connections. By doing this we halve the number of listening sockets we need for TCP (assuming passt/pasta is listening on the same ports for IPv4 and IPv6). When forwarding many ports (e.g. -t all) this can significantly reduce the amount of kernel memory that passt consumes. When forwarding all TCP and UDP ports for both IPv4 and IPv6 (-t all -u all), this reduces kernel memory usage from ~677MiB to ~487MiB (kernel version 6.0.8 on Fedora 37, x86_64). Signed-off-by: David Gibson --- tcp.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-) diff --git a/tcp.c b/tcp.c index c199863..8044617 100644 --- a/tcp.c +++ b/tcp.c @@ -2996,8 +2996,12 @@ static int tcp_sock_init_af(const struct ctx *c, int af, in_port_t port, s = sock_l4(c, af, IPPROTO_TCP, addr, ifname, port, tref.u32); - if (c->tcp.fwd_in.mode == FWD_AUTO) - tcp_sock_init_ext[port][(af == AF_INET) ? V4 : V6] = s; + if (c->tcp.fwd_in.mode == FWD_AUTO) { + if (af == AF_INET || af == AF_UNSPEC) + tcp_sock_init_ext[port][V4] = s; + if (af == AF_INET6 || af == AF_UNSPEC) + tcp_sock_init_ext[port][V6] = s; + } if (s < 0) return -1; @@ -3017,6 +3021,12 @@ static int tcp_sock_init_af(const struct ctx *c, int af, in_port_t port, void tcp_sock_init(const struct ctx *c, sa_family_t af, const void *addr, const char *ifname, in_port_t port) { + if (af == AF_UNSPEC && c->ifi4 && c->ifi6) + /* Attempt to get a dual stack socket */ + if (tcp_sock_init_af(c, AF_UNSPEC, port, addr, ifname) >= 0) + return; + + /* Otherwise create a socket per IP version */ if ((af == AF_INET || af == AF_UNSPEC) && c->ifi4) tcp_sock_init_af(c, AF_INET, port, addr, ifname); if ((af == AF_INET6 || af == AF_UNSPEC) && c->ifi6) -- 2.38.1