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 | 1 + 1 file changed, 1 insertion(+) diff --git a/Makefile b/Makefile index 6b22408..8bcbbc0 100644 --- a/Makefile +++ b/Makefile @@ -262,6 +262,7 @@ 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..4e56a6c 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 8bcbbc0..9046b0b 100644 --- a/Makefile +++ b/Makefile @@ -45,7 +45,8 @@ 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 4e56a6c..1137f45 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

When we compact the connection tables (both spliced and non-spliced) we need to move entries from one slot to another. That requires some updates in the entries themselves. Add helpers to make all the necessary updates for the spliced and non-spliced cases. This will simplify later cleanups. Signed-off-by: David Gibson --- tcp.c | 16 +++++++++------- tcp_splice.c | 17 ++++++++++++++--- 2 files changed, 23 insertions(+), 10 deletions(-) diff --git a/tcp.c b/tcp.c index b9b15ee..39b6176 100644 --- a/tcp.c +++ b/tcp.c @@ -1291,11 +1291,13 @@ static void tcp_hash_remove(const struct tcp_tap_conn *conn) } /** - * tcp_hash_update() - Update pointer for given connection - * @old: Old connection pointer - * @new: New connection pointer + * tcp_tap_conn_update() - Update tcp_tap_conn when being moved in the table + * @c: Execution context + * @old: Old location of tcp_tap_conn + * @new: New location of tcp_tap_conn */ -static void tcp_hash_update(struct tcp_tap_conn *old, struct tcp_tap_conn *new) +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; @@ -1314,6 +1316,8 @@ static void tcp_hash_update(struct tcp_tap_conn *old, struct tcp_tap_conn *new) debug("TCP: hash table update: old index %li, new index %li, sock %i, " "bucket: %i, old: %p, new: %p", CONN_IDX(old), CONN_IDX(new), new->sock, b, old, new); + + tcp_epoll_ctl(c, new); } /** @@ -1362,9 +1366,7 @@ static void tcp_table_compact(struct ctx *c, struct tcp_tap_conn *hole) memcpy(hole, from, sizeof(*hole)); to = hole; - tcp_hash_update(from, to); - - tcp_epoll_ctl(c, to); + tcp_tap_conn_update(c, from, to); debug("TCP: hash table compaction: old index %li, new index %li, " "sock %i, from: %p, to: %p", diff --git a/tcp_splice.c b/tcp_splice.c index d8be91b..7dcd1cb 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -242,6 +242,19 @@ static void conn_event_do(const struct ctx *c, struct tcp_splice_conn *conn, conn_event_do(c, conn, event); \ } while (0) + +/** + * tcp_splice_conn_update() - Update tcp_splice_conn when being moved in the table + * @c: Execution context + * @new: New location of tcp_splice_conn + */ +static 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 @@ -269,9 +282,7 @@ static void tcp_table_splice_compact(struct ctx *c, debug("TCP (spliced): index %li moved to %li", CONN_IDX(move), CONN_IDX(hole)); - tcp_splice_epoll_ctl(c, hole); - if (tcp_splice_epoll_ctl(c, hole)) - conn_flag(c, hole, CLOSING); + tcp_splice_conn_update(c, hole); } /** -- 2.38.1

These two functions each step through non-spliced and spliced connections respectively and clean up entries for closed connections. To avoid scanning the connection table twice, we merge these into a single function which scans the unified table and performs the appropriate sort of cleanup action on each one. Signed-off-by: David Gibson --- tcp.c | 20 +++++++++++--------- tcp_conn.h | 1 + tcp_splice.c | 24 +----------------------- tcp_splice.h | 1 - 4 files changed, 13 insertions(+), 33 deletions(-) diff --git a/tcp.c b/tcp.c index 01b08c8..598634f 100644 --- a/tcp.c +++ b/tcp.c @@ -1527,21 +1527,23 @@ 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_tap_conn *conn; + union tcp_conn *conn; tcp_l2_flags_buf_flush(c); tcp_l2_data_buf_flush(c); - tcp_splice_defer_handler(c); - - if (c->tcp.conn_count < MIN(max_files, max_conns)) + if ((c->tcp.conn_count < MIN(max_files, max_conns)) && + (c->tcp.splice_conn_count < MIN(max_files / 6, max_conns))) return; - 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) { + if (conn->splice.flags & CLOSING) + tcp_splice_destroy(c, &conn->splice); + } else { + if (conn->tap.events == CLOSED) + tcp_conn_destroy(c, &conn->tap); + } } } diff --git a/tcp_conn.h b/tcp_conn.h index 4295f7d..634e259 100644 --- a/tcp_conn.h +++ b/tcp_conn.h @@ -200,5 +200,6 @@ 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); #endif /* TCP_CONN_H */ diff --git a/tcp_splice.c b/tcp_splice.c index c986a9c..ad2b216 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -111,7 +111,6 @@ static void tcp_splice_conn_epoll_events(uint16_t events, *b |= (events & B_OUT_WAIT) ? EPOLLOUT : 0; } -static void tcp_splice_destroy(struct ctx *c, struct tcp_splice_conn *conn); static int tcp_splice_epoll_ctl(const struct ctx *c, struct tcp_splice_conn *conn); @@ -257,7 +256,7 @@ void tcp_splice_conn_update(struct ctx *c, struct tcp_splice_conn *new) * @c: Execution context * @conn: Connection pointer */ -static void tcp_splice_destroy(struct ctx *c, struct tcp_splice_conn *conn) +void tcp_splice_destroy(struct ctx *c, struct tcp_splice_conn *conn) { if (conn->events & SPLICE_ESTABLISHED) { /* Flushing might need to block: don't recycle them. */ @@ -849,24 +848,3 @@ void tcp_splice_timer(struct ctx *c) tcp_splice_pipe_refill(c); } - -/** - * tcp_splice_defer_handler() - Close connections without timer on file pressure - * @c: Execution context - */ -void tcp_splice_defer_handler(struct ctx *c) -{ - int max_conns = c->tcp.conn_count / 100 * TCP_SPLICE_CONN_PRESSURE; - int max_files = c->nofile / 100 * TCP_SPLICE_FILE_PRESSURE; - struct tcp_splice_conn *conn; - - if (c->tcp.conn_count < MIN(max_files / 6, max_conns)) - return; - - 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 e8c70e9..82e057c 100644 --- a/tcp_splice.h +++ b/tcp_splice.h @@ -10,6 +10,5 @@ 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); -void tcp_splice_defer_handler(struct ctx *c); #endif /* TCP_SPLICE_H */ -- 2.38.1

There is very little common between the tcp_tap_conn and tcp_splice_conn structures. However, both do have an IN_EPOLL flag which has the same meaning in each case, though it's stored in a different location. Simplify things slightly by moving this bit into the common header of the two structures. Signed-off-by: David Gibson --- tcp.c | 10 +++++----- tcp_conn.h | 20 ++++++++++---------- tcp_splice.c | 8 ++++---- 3 files changed, 19 insertions(+), 19 deletions(-) diff --git a/tcp.c b/tcp.c index 08169b6..aac70cd 100644 --- a/tcp.c +++ b/tcp.c @@ -429,8 +429,8 @@ static const char *tcp_state_str[] __attribute((__unused__)) = { }; static const char *tcp_flag_str[] __attribute((__unused__)) = { - "STALLED", "LOCAL", "WND_CLAMPED", "IN_EPOLL", "ACTIVE_CLOSE", - "ACK_TO_TAP_DUE", "ACK_FROM_TAP_DUE", + "STALLED", "LOCAL", "WND_CLAMPED", "ACTIVE_CLOSE", "ACK_TO_TAP_DUE", + "ACK_FROM_TAP_DUE", }; /* Listening sockets, used for automatic port forwarding in pasta mode only */ @@ -660,14 +660,14 @@ static void conn_flag_do(const struct ctx *c, struct tcp_tap_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; + 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) }; struct epoll_event ev = { .data.u64 = ref.u64 }; if (conn->events == CLOSED) { - if (conn->flags & IN_EPOLL) + if (conn->c.in_epoll) epoll_ctl(c->epollfd, EPOLL_CTL_DEL, conn->sock, &ev); if (conn->timer != -1) epoll_ctl(c->epollfd, EPOLL_CTL_DEL, conn->timer, &ev); @@ -679,7 +679,7 @@ static int tcp_epoll_ctl(const struct ctx *c, struct tcp_tap_conn *conn) if (epoll_ctl(c->epollfd, m, conn->sock, &ev)) return -errno; - conn->flags |= IN_EPOLL; /* No need to log this */ + conn->c.in_epoll = true; if (conn->timer != -1) { union epoll_ref ref_t = { .r.proto = IPPROTO_TCP, diff --git a/tcp_conn.h b/tcp_conn.h index 7c450a0..faa63dc 100644 --- a/tcp_conn.h +++ b/tcp_conn.h @@ -14,9 +14,11 @@ /** * struct tcp_conn_common - Common fields for spliced and non-spliced * @spliced: Is this a spliced connection? + * @in_epoll: Is the connection in the epoll set? */ struct tcp_conn_common { bool spliced :1; + bool in_epoll :1; }; extern const char *tcp_common_flag_str[]; @@ -90,10 +92,9 @@ struct tcp_tap_conn { #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) +#define ACTIVE_CLOSE BIT(3) +#define ACK_TO_TAP_DUE BIT(4) +#define ACK_FROM_TAP_DUE BIT(5) unsigned int hash_bucket :TCP_HASH_BUCKET_BITS; @@ -170,12 +171,11 @@ struct tcp_splice_conn { 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) +#define RCVLOWAT_SET_A BIT(1) +#define RCVLOWAT_SET_B BIT(2) +#define RCVLOWAT_ACT_A BIT(3) +#define RCVLOWAT_ACT_B BIT(4) +#define CLOSING BIT(5) uint32_t a_read; uint32_t a_written; diff --git a/tcp_splice.c b/tcp_splice.c index 0ac316d..7a06252 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -83,8 +83,8 @@ static const char *tcp_splice_event_str[] __attribute((__unused__)) = { /* Display strings for connection flags */ static const char *tcp_splice_flag_str[] __attribute((__unused__)) = { - "SPLICE_V6", "SPLICE_IN_EPOLL", "RCVLOWAT_SET_A", "RCVLOWAT_SET_B", - "RCVLOWAT_ACT_A", "RCVLOWAT_ACT_B", "CLOSING", + "SPLICE_V6", "RCVLOWAT_SET_A", "RCVLOWAT_SET_B", "RCVLOWAT_ACT_A", + "RCVLOWAT_ACT_B", "CLOSING", }; /** @@ -164,7 +164,7 @@ 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 & SPLICE_IN_EPOLL) ? EPOLL_CTL_MOD : EPOLL_CTL_ADD; + 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.splice = 1, .r.p.tcp.tcp.index = CONN_IDX(conn), @@ -188,7 +188,7 @@ static int tcp_splice_epoll_ctl(const struct ctx *c, epoll_ctl(c->epollfd, m, conn->b, &ev_b)) goto delete; - conn->flags |= SPLICE_IN_EPOLL; /* No need to log this */ + conn->c.in_epoll = true; return 0; -- 2.38.1

In tcp_sock_handler() we split off to handle spliced sockets before checking anything else. However the first steps of the "new connection" path for each case are the same: allocate a connection entry and accept() the connection. Remove this duplication by making tcp_conn_from_sock() handle both spliced and non-spliced cases, with help from more specific tcp_tap_conn_from_sock and tcp_splice_conn_from_sock functions for the later stages which differ. Signed-off-by: David Gibson --- tcp.c | 68 ++++++++++++++++++++++++++++++++++------------------ tcp_splice.c | 58 +++++++++++++++++++++++--------------------- tcp_splice.h | 4 ++++ 3 files changed, 80 insertions(+), 50 deletions(-) diff --git a/tcp.c b/tcp.c index 72d3b49..e66a82a 100644 --- a/tcp.c +++ b/tcp.c @@ -2753,28 +2753,19 @@ static void tcp_connect_finish(struct ctx *c, struct tcp_tap_conn *conn) } /** - * tcp_conn_from_sock() - Handle new connection request from listening socket + * tcp_tap_conn_from_sock() - Initialize state for non-spliced connection * @c: Execution context * @ref: epoll reference of listening socket + * @conn: connection structure to initialize + * @s: Accepted socket + * @sa: Peer socket address (from accept()) * @now: Current timestamp */ -static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, - const struct timespec *now) +static void tcp_tap_conn_from_sock(struct ctx *c, union epoll_ref ref, + struct tcp_tap_conn *conn, int s, + struct sockaddr *sa, + const struct timespec *now) { - struct sockaddr_storage sa; - struct tcp_tap_conn *conn; - socklen_t sl; - int s; - - if (c->tcp.conn_count >= TCP_MAX_CONNS) - return; - - sl = sizeof(sa); - s = accept4(ref.r.s, (struct sockaddr *)&sa, &sl, SOCK_NONBLOCK); - if (s < 0) - return; - - conn = CONN(c->tcp.conn_count++); conn->c.spliced = false; conn->sock = s; conn->timer = -1; @@ -2784,7 +2775,7 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, if (ref.r.p.tcp.tcp.v6) { struct sockaddr_in6 sa6; - memcpy(&sa6, &sa, sizeof(sa6)); + memcpy(&sa6, sa, sizeof(sa6)); if (IN6_IS_ADDR_LOOPBACK(&sa6.sin6_addr) || IN6_ARE_ADDR_EQUAL(&sa6.sin6_addr, &c->ip6.addr_seen) || @@ -2813,7 +2804,7 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, } else { struct sockaddr_in sa4; - memcpy(&sa4, &sa, sizeof(sa4)); + memcpy(&sa4, sa, sizeof(sa4)); memset(&conn->a.a4.zero, 0, sizeof(conn->a.a4.zero)); memset(&conn->a.a4.one, 0xff, sizeof(conn->a.a4.one)); @@ -2846,6 +2837,37 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, tcp_get_sndbuf(conn); } +/** + * tcp_conn_from_sock() - Handle new connection request from listening socket + * @c: Execution context + * @ref: epoll reference of listening socket + * @now: Current timestamp + */ +static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, + const struct timespec *now) +{ + struct sockaddr_storage sa; + union tcp_conn *conn; + socklen_t sl; + int s; + + if (c->tcp.conn_count >= TCP_MAX_CONNS) + return; + + sl = sizeof(sa); + 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); +} + /** * tcp_timer_handler() - timerfd events: close, send ACK, retransmit, or reset * @c: Execution context @@ -2925,13 +2947,13 @@ void tcp_sock_handler(struct ctx *c, union epoll_ref ref, uint32_t events, return; } - if (ref.r.p.tcp.tcp.splice) { - tcp_sock_handler_splice(c, ref, events); + if (ref.r.p.tcp.tcp.listen) { + tcp_conn_from_sock(c, ref, now); return; } - if (ref.r.p.tcp.tcp.listen) { - tcp_conn_from_sock(c, ref, now); + if (ref.r.p.tcp.tcp.splice) { + tcp_sock_handler_splice(c, ref, events); return; } diff --git a/tcp_splice.c b/tcp_splice.c index 7a06252..7007501 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -501,6 +501,36 @@ static void tcp_splice_dir(struct tcp_splice_conn *conn, int ref_sock, *pipes = *from == conn->a ? conn->pipe_a_b : conn->pipe_b_a; } +/** + * tcp_splice_conn_from_sock() - Initialize state for spliced connection + * @c: Execution context + * @ref: epoll reference of listening socket + * @conn: connection structure to initialize + * @s: Accepted socket + * + * #syscalls:pasta setsockopt + */ +void tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref, + struct tcp_splice_conn *conn, int s) +{ + assert(c->mode == MODE_PASTA); + + 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); +} + /** * tcp_sock_handler_splice() - Handler for socket mapped to spliced connection * @c: Execution context @@ -517,33 +547,7 @@ void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, uint32_t *seq_read, *seq_write; struct tcp_splice_conn *conn; - if (ref.r.p.tcp.tcp.listen) { - int s; - - if (c->tcp.conn_count >= TCP_MAX_CONNS) - return; - - if ((s = accept4(ref.r.s, NULL, NULL, SOCK_NONBLOCK)) < 0) - return; - - if (setsockopt(s, SOL_TCP, TCP_QUICKACK, &((int){ 1 }), - sizeof(int))) { - trace("TCP (spliced): failed to set TCP_QUICKACK on %i", - s); - } - - 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; - - if (tcp_splice_new(c, conn, ref.r.p.tcp.tcp.index, - ref.r.p.tcp.tcp.outbound)) - conn_flag(c, conn, CLOSING); - - return; - } + assert(!ref.r.p.tcp.tcp.listen); conn = CONN(ref.r.p.tcp.tcp.index); diff --git a/tcp_splice.h b/tcp_splice.h index 22024d6..f9462ae 100644 --- a/tcp_splice.h +++ b/tcp_splice.h @@ -6,8 +6,12 @@ #ifndef TCP_SPLICE_H #define TCP_SPLICE_H +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); void tcp_splice_init(struct ctx *c); #endif /* TCP_SPLICE_H */ -- 2.38.1

Currently the epoll reference for tcp sockets includes a bit indicating whether the socket maps to a spliced connection. However, the reference also has the index of the connection structure which also indicates whether it is spliced. We can therefore avoid the splice bit in the epoll_ref by unifying the first part of the non-spliced and spliced handlers where we look up the connection state. Signed-off-by: David Gibson --- tcp.c | 60 +++++++++++++++++++++++++++++----------------------- tcp.h | 2 -- tcp_splice.c | 26 +++++++++-------------- tcp_splice.h | 4 ++-- 4 files changed, 46 insertions(+), 46 deletions(-) diff --git a/tcp.c b/tcp.c index 4065da7..e46330e 100644 --- a/tcp.c +++ b/tcp.c @@ -2851,7 +2851,6 @@ static void tcp_conn_from_sock(struct ctx *c, union epoll_ref ref, 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; @@ -2940,35 +2939,14 @@ static void tcp_timer_handler(struct ctx *c, union epoll_ref ref) } /** - * tcp_sock_handler() - Handle new data from socket, or timerfd event + * tcp_tap_sock_handler() - Handle new data from non-spliced socket * @c: Execution context - * @ref: epoll reference + * @conn: Connection state * @events: epoll events bitmap - * @now: Current timestamp */ -void tcp_sock_handler(struct ctx *c, union epoll_ref ref, uint32_t events, - const struct timespec *now) +static void tcp_tap_sock_handler(struct ctx *c, struct tcp_tap_conn *conn, + uint32_t events) { - struct tcp_tap_conn *conn; - - if (ref.r.p.tcp.tcp.timer) { - tcp_timer_handler(c, ref); - return; - } - - if (ref.r.p.tcp.tcp.listen) { - tcp_conn_from_sock(c, ref, now); - return; - } - - if (ref.r.p.tcp.tcp.splice) { - tcp_sock_handler_splice(c, ref, events); - return; - } - - if (!(conn = conn_at_idx(ref.r.p.tcp.tcp.index))) - return; - if (conn->events == CLOSED) return; @@ -3015,6 +2993,36 @@ void tcp_sock_handler(struct ctx *c, union epoll_ref ref, uint32_t events, } } +/** + * tcp_sock_handler() - Handle new data from socket, or timerfd event + * @c: Execution context + * @ref: epoll reference + * @events: epoll events bitmap + * @now: Current timestamp + */ +void tcp_sock_handler(struct ctx *c, union epoll_ref ref, uint32_t events, + const struct timespec *now) +{ + union tcp_conn *conn; + + if (ref.r.p.tcp.tcp.timer) { + tcp_timer_handler(c, ref); + return; + } + + if (ref.r.p.tcp.tcp.listen) { + tcp_conn_from_sock(c, ref, now); + return; + } + + conn = tc + ref.r.p.tcp.tcp.index; + + if (conn->c.spliced) + tcp_splice_sock_handler(c, &conn->splice, ref.r.s, events); + else + tcp_tap_sock_handler(c, &conn->tap, events); +} + /** * tcp_sock_init4() - Initialise listening sockets for a given IPv4 port * @c: Execution context diff --git a/tcp.h b/tcp.h index f4ed298..a940682 100644 --- a/tcp.h +++ b/tcp.h @@ -32,7 +32,6 @@ 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 - * @splice: Set if descriptor is associated to a spliced connection * @outbound: Listening socket maps to outbound, spliced connection * @v6: Set for IPv6 sockets or connections * @timer: Reference is a timerfd descriptor for connection @@ -42,7 +41,6 @@ void tcp_update_l2_buf(const unsigned char *eth_d, const unsigned char *eth_s, union tcp_epoll_ref { struct { uint32_t listen:1, - splice:1, outbound:1, v6:1, timer:1, diff --git a/tcp_splice.c b/tcp_splice.c index 30d49d4..2852e76 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -166,11 +166,9 @@ 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.splice = 1, .r.p.tcp.tcp.index = CONN_IDX(conn), .r.p.tcp.tcp.v6 = CONN_V6(conn) }; union epoll_ref ref_b = { .r.proto = IPPROTO_TCP, .r.s = conn->b, - .r.p.tcp.tcp.splice = 1, .r.p.tcp.tcp.index = CONN_IDX(conn), .r.p.tcp.tcp.v6 = CONN_V6(conn) }; struct epoll_event ev_a = { .data.u64 = ref_a.u64 }; @@ -549,24 +547,20 @@ bool tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref, } /** - * tcp_sock_handler_splice() - Handler for socket mapped to spliced connection + * tcp_splice_sock_handler() - Handler for socket mapped to spliced connection * @c: Execution context - * @ref: epoll reference + * @conn: Connection state + * @s: Socket fd on which an event has occurred * @events: epoll events bitmap * * #syscalls:pasta splice */ -void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, - uint32_t events) +void tcp_splice_sock_handler(struct ctx *c, struct tcp_splice_conn *conn, + int s, uint32_t events) { uint8_t lowat_set_flag, lowat_act_flag; int from, to, *pipes, eof, never_read; uint32_t *seq_read, *seq_write; - struct tcp_splice_conn *conn; - - assert(!ref.r.p.tcp.tcp.listen); - - conn = CONN(ref.r.p.tcp.tcp.index); if (conn->events == SPLICE_CLOSED) return; @@ -582,25 +576,25 @@ void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, } if (events & EPOLLOUT) { - if (ref.r.s == conn->a) + if (s == conn->a) conn_event(c, conn, ~A_OUT_WAIT); else conn_event(c, conn, ~B_OUT_WAIT); - tcp_splice_dir(conn, ref.r.s, 1, &from, &to, &pipes); + tcp_splice_dir(conn, s, 1, &from, &to, &pipes); } else { - tcp_splice_dir(conn, ref.r.s, 0, &from, &to, &pipes); + tcp_splice_dir(conn, s, 0, &from, &to, &pipes); } if (events & EPOLLRDHUP) { - if (ref.r.s == conn->a) + if (s == conn->a) conn_event(c, conn, A_FIN_RCVD); else conn_event(c, conn, B_FIN_RCVD); } if (events & EPOLLHUP) { - if (ref.r.s == conn->a) + if (s == conn->a) conn_event(c, conn, A_FIN_SENT); /* Fake, but implied */ else conn_event(c, conn, B_FIN_SENT); diff --git a/tcp_splice.h b/tcp_splice.h index 1a915dd..6814ae7 100644 --- a/tcp_splice.h +++ b/tcp_splice.h @@ -8,8 +8,8 @@ struct tcp_splice_conn; -void tcp_sock_handler_splice(struct ctx *c, union epoll_ref ref, - uint32_t events); +void tcp_splice_sock_handler(struct ctx *c, struct tcp_splice_conn *conn, + int s, uint32_t events); bool tcp_splice_conn_from_sock(struct ctx *c, union epoll_ref ref, struct tcp_splice_conn *conn, int s, const struct sockaddr *sa); -- 2.38.1

struct tcp_conn stores an address which could be IPv6 or IPv4 using a union. We can do this without an additional tag by encoding IPv4 addresses as IPv4-mapped IPv6 addresses. This approach is useful wider than the specific place in tcp_conn, so expose a new 'union inany_addr' like this from a new inany.h. Along with that create a number of helper functions to make working with these "inany" addresses easier. Signed-off-by: David Gibson --- Makefile | 6 ++-- inany.h | 68 ++++++++++++++++++++++++++++++++++++++++ tcp.c | 88 +++++++++++++++++++++++++--------------------------- tcp_conn.h | 15 ++------- tcp_splice.c | 1 + 5 files changed, 117 insertions(+), 61 deletions(-) create mode 100644 inany.h diff --git a/Makefile b/Makefile index 9046b0b..ca453aa 100644 --- a/Makefile +++ b/Makefile @@ -44,9 +44,9 @@ SRCS = $(PASST_SRCS) $(QRAP_SRCS) 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_conn.h tcp_splice.h udp.h \ - util.h + inany.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_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/inany.h b/inany.h new file mode 100644 index 0000000..4e53da9 --- /dev/null +++ b/inany.h @@ -0,0 +1,68 @@ +/* SPDX-License-Identifier: AGPL-3.0-or-later + * Copyright Red Hat + * Author: David Gibson + * + * inany.h - Types and helpers for handling addresses which could be + * IPv6 or IPv4 (encoded as IPv4-mapped IPv6 addresses) + */ + +#include + +/** union inany_addr - Represents either an IPv4 or IPv6 address + * @a6: Address as an IPv6 address, may be IPv4-mapped + * @_v4._zero: All zero-bits for an IPv4 address + * @_v4._one: All one-bits for an IPv4 address + * @_v4.a4: If @a6 is an IPv4 mapped address, this is the raw IPv4 address + * + * Fields starting with _ shouldn't be accessed except via helpers. + */ +union inany_addr { + struct in6_addr a6; + struct { + uint8_t _zero[10]; + uint8_t _one[2]; + struct in_addr a4; + } _v4mapped; +}; + +/** inany_v4 - Extract IPv4 address, if present, from IPv[46] address + * @addr: IPv4 or IPv6 address + * + * Return: IPv4 address if @addr is IPv4, NULL otherwise + */ +static inline const struct in_addr *inany_v4(const union inany_addr *addr) +{ + if (!IN6_IS_ADDR_V4MAPPED(&addr->a6)) + return NULL; + return &addr->_v4mapped.a4; +} + +/** inany_equals - Compare two IPv[46] addresses + * @a, @b: IPv[46] addresses + * + * Return: true if @a and @b are the same address + */ +static inline bool inany_equals(const union inany_addr *a, + const union inany_addr *b) +{ + return IN6_ARE_ADDR_EQUAL(&a->a6, &b->a6); +} + +/** inany_from_af - Set IPv[46] address from IPv4 or IPv6 address + * @aa: Pointer to store IPv[46] address + * @af: Address family of @addr + * @addr: struct in_addr (IPv4) or struct in6_addr (IPv6) + */ +static inline void inany_from_af(union inany_addr *aa, int af, const void *addr) +{ + if (af == AF_INET6) { + aa->a6 = *((struct in6_addr *)addr); + } else if (af == AF_INET) { + memset(&aa->_v4mapped._zero, 0, sizeof(aa->_v4mapped._zero)); + memset(&aa->_v4mapped._one, 0xff, sizeof(aa->_v4mapped._one)); + aa->_v4mapped.a4 = *((struct in_addr *)addr); + } else { + /* Not valid to call with other address families */ + assert(0); + } +} diff --git a/tcp.c b/tcp.c index 7686766..4040198 100644 --- a/tcp.c +++ b/tcp.c @@ -301,6 +301,7 @@ #include "conf.h" #include "tcp_splice.h" #include "log.h" +#include "inany.h" #include "tcp_conn.h" @@ -404,7 +405,7 @@ struct tcp6_l2_head { /* For MSS6 macro: keep in sync with tcp6_l2_buf_t */ #define OPT_SACK 5 #define OPT_TS 8 -#define CONN_V4(conn) IN6_IS_ADDR_V4MAPPED(&conn->a.a6) +#define CONN_V4(conn) (!!inany_v4(&(conn)->addr)) #define CONN_V6(conn) (!CONN_V4(conn)) #define CONN_IS_CLOSING(conn) \ ((conn->events & ESTABLISHED) && \ @@ -438,7 +439,7 @@ static int tcp_sock_init_ext [NUM_PORTS][IP_VERSIONS]; static int tcp_sock_ns [NUM_PORTS][IP_VERSIONS]; /* Table of destinations with very low RTT (assumed to be local), LRU */ -static struct in6_addr low_rtt_dst[LOW_RTT_TABLE_SIZE]; +static union inany_addr low_rtt_dst[LOW_RTT_TABLE_SIZE]; /* Static buffers */ @@ -861,7 +862,7 @@ static int tcp_rtt_dst_low(const struct tcp_tap_conn *conn) int i; for (i = 0; i < LOW_RTT_TABLE_SIZE; i++) - if (IN6_ARE_ADDR_EQUAL(&conn->a.a6, low_rtt_dst + i)) + if (inany_equals(&conn->addr, low_rtt_dst + i)) return 1; return 0; @@ -883,7 +884,7 @@ static void tcp_rtt_dst_check(const struct tcp_tap_conn *conn, return; for (i = 0; i < LOW_RTT_TABLE_SIZE; i++) { - if (IN6_ARE_ADDR_EQUAL(&conn->a.a6, low_rtt_dst + i)) + if (inany_equals(&conn->addr, low_rtt_dst + i)) return; if (hole == -1 && IN6_IS_ADDR_UNSPECIFIED(low_rtt_dst + i)) hole = i; @@ -895,10 +896,10 @@ static void tcp_rtt_dst_check(const struct tcp_tap_conn *conn, if (hole == -1) return; - memcpy(low_rtt_dst + hole++, &conn->a.a6, sizeof(conn->a.a6)); + low_rtt_dst[hole++] = conn->addr; if (hole == LOW_RTT_TABLE_SIZE) hole = 0; - memcpy(low_rtt_dst + hole, &in6addr_any, sizeof(conn->a.a6)); + inany_from_af(low_rtt_dst + hole, AF_INET6, &in6addr_any); #else (void)conn; (void)tinfo; @@ -1187,13 +1188,14 @@ 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) && - !memcmp(&conn->a.a4.a, addr, sizeof(conn->a.a4.a)) && + 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->a.a6, addr) && + IN6_ARE_ADDR_EQUAL(&conn->addr.a6, addr) && conn->tap_port == tap_port && conn->sock_port == sock_port) return 1; @@ -1253,11 +1255,13 @@ 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) { - if (CONN_V6(conn)) - return tcp_hash(c, AF_INET6, &conn->a.a6, + 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_INET, &conn->a.a4.a, + return tcp_hash(c, AF_INET6, &conn->addr.a6, conn->tap_port, conn->sock_port); } @@ -1582,6 +1586,7 @@ static size_t tcp_l2_buf_fill_headers(const struct ctx *c, void *p, size_t plen, const uint16_t *check, uint32_t seq) { + const struct in_addr *a4 = inany_v4(&conn->addr); size_t ip_len, eth_len; #define SET_TCP_HEADER_COMMON_V4_V6(b, conn, seq) \ @@ -1599,13 +1604,33 @@ do { \ } \ } while (0) - if (CONN_V6(conn)) { + if (a4) { + struct tcp4_l2_buf_t *b = (struct tcp4_l2_buf_t *)p; + + ip_len = plen + sizeof(struct iphdr) + sizeof(struct tcphdr); + b->iph.tot_len = htons(ip_len); + b->iph.saddr = a4->s_addr; + b->iph.daddr = c->ip4.addr_seen.s_addr; + + if (check) + b->iph.check = *check; + else + tcp_update_check_ip4(b); + + SET_TCP_HEADER_COMMON_V4_V6(b, conn, seq); + + tcp_update_check_tcp4(b); + + eth_len = ip_len + sizeof(struct ethhdr); + if (c->mode == MODE_PASST) + b->vnet_len = htonl(eth_len); + } else { struct tcp6_l2_buf_t *b = (struct tcp6_l2_buf_t *)p; ip_len = plen + sizeof(struct ipv6hdr) + sizeof(struct tcphdr); b->ip6h.payload_len = htons(plen + sizeof(struct tcphdr)); - b->ip6h.saddr = conn->a.a6; + b->ip6h.saddr = conn->addr.a6; if (IN6_IS_ADDR_LINKLOCAL(&b->ip6h.saddr)) b->ip6h.daddr = c->ip6.addr_ll_seen; else @@ -1621,26 +1646,6 @@ do { \ b->ip6h.flow_lbl[1] = (conn->sock >> 8) & 0xff; b->ip6h.flow_lbl[2] = (conn->sock >> 0) & 0xff; - eth_len = ip_len + sizeof(struct ethhdr); - if (c->mode == MODE_PASST) - b->vnet_len = htonl(eth_len); - } else { - struct tcp4_l2_buf_t *b = (struct tcp4_l2_buf_t *)p; - - ip_len = plen + sizeof(struct iphdr) + sizeof(struct tcphdr); - b->iph.tot_len = htons(ip_len); - b->iph.saddr = conn->a.a4.a.s_addr; - b->iph.daddr = c->ip4.addr_seen.s_addr; - - if (check) - b->iph.check = *check; - else - tcp_update_check_ip4(b); - - SET_TCP_HEADER_COMMON_V4_V6(b, conn, seq); - - tcp_update_check_tcp4(b); - eth_len = ip_len + sizeof(struct ethhdr); if (c->mode == MODE_PASST) b->vnet_len = htonl(eth_len); @@ -2144,18 +2149,14 @@ static void tcp_conn_from_tap(struct ctx *c, int af, const void *addr, if (!(conn->wnd_from_tap = (htons(th->window) >> conn->ws_from_tap))) conn->wnd_from_tap = 1; + inany_from_af(&conn->addr, af, addr); + if (af == AF_INET) { sa = (struct sockaddr *)&addr4; sl = sizeof(addr4); - - memset(&conn->a.a4.zero, 0, sizeof(conn->a.a4.zero)); - memset(&conn->a.a4.one, 0xff, sizeof(conn->a.a4.one)); - memcpy(&conn->a.a4.a, addr, sizeof(conn->a.a4.a)); } else { sa = (struct sockaddr *)&addr6; sl = sizeof(addr6); - - memcpy(&conn->a.a6, addr, sizeof(conn->a.a6)); } conn->sock_port = ntohs(th->dest); @@ -2808,7 +2809,7 @@ static void tcp_tap_conn_from_sock(struct ctx *c, union epoll_ref ref, memcpy(&sa6.sin6_addr, src, sizeof(*src)); } - memcpy(&conn->a.a6, &sa6.sin6_addr, sizeof(conn->a.a6)); + inany_from_af(&conn->addr, AF_INET6, &sa6.sin6_addr); conn->sock_port = ntohs(sa6.sin6_port); conn->tap_port = ref.r.p.tcp.tcp.index; @@ -2824,15 +2825,12 @@ static void tcp_tap_conn_from_sock(struct ctx *c, union epoll_ref ref, memcpy(&sa4, sa, sizeof(sa4)); - memset(&conn->a.a4.zero, 0, sizeof(conn->a.a4.zero)); - memset(&conn->a.a4.one, 0xff, sizeof(conn->a.a4.one)); - if (IN4_IS_ADDR_LOOPBACK(&sa4.sin_addr) || IN4_IS_ADDR_UNSPECIFIED(&sa4.sin_addr) || IN4_ARE_ADDR_EQUAL(&sa4.sin_addr, &c->ip4.addr_seen)) sa4.sin_addr = c->ip4.gw; - conn->a.a4.a = sa4.sin_addr; + inany_from_af(&conn->addr, AF_INET, &sa4.sin_addr); conn->sock_port = ntohs(sa4.sin_port); conn->tap_port = ref.r.p.tcp.tcp.index; diff --git a/tcp_conn.h b/tcp_conn.h index 4bffe9a..bf50e1c 100644 --- a/tcp_conn.h +++ b/tcp_conn.h @@ -35,10 +35,7 @@ extern const char *tcp_common_flag_str[]; * @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 + * @addr: Remote address (IPv4 or IPv6) * @tap_port: Guest-facing tap port * @sock_port: Remote, socket-facing port * @wnd_from_tap: Last window size from tap, unscaled (as received) @@ -108,15 +105,7 @@ struct tcp_tap_conn { 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; - + union inany_addr addr; in_port_t tap_port; in_port_t sock_port; diff --git a/tcp_splice.c b/tcp_splice.c index 2852e76..30ab0eb 100644 --- a/tcp_splice.c +++ b/tcp_splice.c @@ -52,6 +52,7 @@ #include "passt.h" #include "log.h" #include "tcp_splice.h" +#include "inany.h" #include "tcp_conn.h" -- 2.38.1

tcp_hash_insert() takes an address to control which hash bucket the connection will go into. However, an inany_addr representation of that address is already stored in struct tcp_conn. Now that we've made the hashing of IPv4 and IPv4-mapped IPv6 addresses equivalent, we can simplify tcp_hash_insert() to use the address in struct tcp_conn, rather than taking it as an extra parameter. Signed-off-by: David Gibson --- tcp.c | 17 +++++------------ 1 file changed, 5 insertions(+), 12 deletions(-) diff --git a/tcp.c b/tcp.c index 56da864..ca7f295 100644 --- a/tcp.c +++ b/tcp.c @@ -1248,17 +1248,12 @@ static unsigned int tcp_conn_hash(const struct ctx *c, * tcp_hash_insert() - Insert connection into hash table, chain link * @c: Execution context * @conn: Connection pointer - * @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_tap_conn *conn, - int af, const void *addr) +static void tcp_hash_insert(const struct ctx *c, struct tcp_tap_conn *conn) { - union inany_addr aany; int b; - inany_from_af(&aany, af, addr); - b = tcp_hash(c, &aany, conn->tap_port, conn->sock_port); + b = tcp_hash(c, &conn->addr, conn->tap_port, conn->sock_port); conn->next_index = tc_hash[b] ? CONN_IDX(tc_hash[b]) : -1; tc_hash[b] = conn; @@ -2153,7 +2148,7 @@ static void tcp_conn_from_tap(struct ctx *c, int af, const void *addr, conn->seq_to_tap = tcp_seq_init(c, af, addr, th->dest, th->source, now); conn->seq_ack_from_tap = conn->seq_to_tap + 1; - tcp_hash_insert(c, conn, af, addr); + tcp_hash_insert(c, conn); if (!bind(s, sa, sl)) { tcp_rst(c, conn); /* Nobody is listening then */ @@ -2802,8 +2797,6 @@ static void tcp_tap_conn_from_sock(struct ctx *c, union epoll_ref ref, conn->sock_port, conn->tap_port, now); - - tcp_hash_insert(c, conn, AF_INET6, &sa6.sin6_addr); } else { struct sockaddr_in sa4; @@ -2823,10 +2816,10 @@ static void tcp_tap_conn_from_sock(struct ctx *c, union epoll_ref ref, conn->sock_port, conn->tap_port, now); - - tcp_hash_insert(c, conn, AF_INET, &sa4.sin_addr); } + tcp_hash_insert(c, conn); + conn->seq_ack_from_tap = conn->seq_to_tap + 1; conn->wnd_from_tap = WINDOW_DEFAULT; -- 2.38.1

tcp_seq_init() has separate paths for IPv4 and IPv6 addresses, which means we will calculate different sequence numbers for IPv4 and equivalent IPv4-mapped IPv6 addresses. Change it to treat these the same by always converting the input address into an inany_addr representation and use that to calculate the sequence number. Signed-off-by: David Gibson --- siphash.c | 1 + tcp.c | 46 ++++++++++++++++++---------------------------- 2 files changed, 19 insertions(+), 28 deletions(-) diff --git a/siphash.c b/siphash.c index 516a508..811918b 100644 --- a/siphash.c +++ b/siphash.c @@ -123,6 +123,7 @@ uint64_t siphash_8b(const uint8_t *in, const uint64_t *k) * * Return: 32 bits obtained by XORing the two halves of the 64-bit hash output */ +/* cppcheck-suppress unusedFunction */ uint32_t siphash_12b(const uint8_t *in, const uint64_t *k) { uint32_t *in32 = (uint32_t *)in; diff --git a/tcp.c b/tcp.c index f566060..d4e9838 100644 --- a/tcp.c +++ b/tcp.c @@ -1942,37 +1942,27 @@ 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) { + union inany_addr aany; + struct { + union inany_addr src; + in_port_t srcport; + union inany_addr dst; + in_port_t dstport; + } __attribute__((__packed__)) in = { + .srcport = srcport, + .dstport = dstport, + }; uint32_t ns, seq = 0; - if (af == AF_INET) { - struct { - struct in_addr src; - in_port_t srcport; - struct in_addr dst; - in_port_t dstport; - } __attribute__((__packed__)) in = { - .src = *(struct in_addr *)addr, - .srcport = srcport, - .dst = c->ip4.addr, - .dstport = dstport, - }; - - seq = siphash_12b((uint8_t *)&in, c->tcp.hash_secret); - } else if (af == AF_INET6) { - struct { - struct in6_addr src; - in_port_t srcport; - struct in6_addr dst; - in_port_t dstport; - } __attribute__((__packed__)) in = { - .src = *(struct in6_addr *)addr, - .srcport = srcport, - .dst = c->ip6.addr, - .dstport = dstport, - }; + inany_from_af(&aany, af, addr); + in.src = aany; + if (af == AF_INET) + inany_from_af(&aany, AF_INET, &c->ip4.addr); + else + inany_from_af(&aany, AF_INET6, &c->ip6.addr); + in.dst = aany; - seq = siphash_36b((uint8_t *)&in, c->tcp.hash_secret); - } + seq = siphash_36b((uint8_t *)&in, c->tcp.hash_secret); ns = now->tv_sec * 1E9; ns += now->tv_nsec >> 5; /* 32ns ticks, overflows 32 bits every 137s */ -- 2.38.1

It looks like tcp_seq_init() is supposed to advance the sequence number by one every 32ns. However we only right shift the ns part of the timespec not the seconds part, meaning that we'll advance by an extra 32 steps on each second. I don't know if that's exploitable in any way, but it doesn't appear to be the intent, nor what RFC 6528 suggests. In addition, we convert from seconds to nanoseconds with a multiplication by '1E9'. In C '1E9' is a floating point constant, forcing a conversion to floating point and back for what should be an integer calculation (confirmed with objdump and Makefile default compiler flags). Spell out 1000000000 in full to avoid that. Signed-off-by: David Gibson --- tcp.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tcp.c b/tcp.c index 7156246..0513b3b 100644 --- a/tcp.c +++ b/tcp.c @@ -1957,8 +1957,8 @@ static void tcp_seq_init(const struct ctx *c, struct tcp_tap_conn *conn, seq = siphash_36b((uint8_t *)&in, c->tcp.hash_secret); - ns = now->tv_sec * 1E9; - ns += now->tv_nsec >> 5; /* 32ns ticks, overflows 32 bits every 137s */ + /* 32ns ticks, overflows 32 bits every 137s */ + ns = (now->tv_sec * 1000000000 + now->tv_nsec) >> 5; conn->seq_to_tap = seq + ns; } -- 2.38.1

passt usually doesn't NAT, but it does do so for the remapping of the gateway address to refer to the host. Currently we perform this NAT with slightly different rules on both IPv4 addresses and IPv6 addresses, but not on IPv4-mapped IPv6 addresses. This means we won't correctly handle the case of an IPv4 connection over an IPv6 socket, which is possible on Linux (and probably other platforms). Refactor tcp_conn_from_sock() to perform the NAT after converting either address family into an inany_addr, so IPv4 and and IPv4-mapped addresses have the same representation. With two new helpers this lets us remove the IPv4 and IPv6 specific paths from tcp_conn_from_sock(). Signed-off-by: David Gibson --- inany.h | 30 ++++++++++++++++++++++++++-- tcp.c | 62 ++++++++++++++++++++++++--------------------------------- 2 files changed, 54 insertions(+), 38 deletions(-) diff --git a/inany.h b/inany.h index 4e53da9..a677aa7 100644 --- a/inany.h +++ b/inany.h @@ -30,11 +30,11 @@ union inany_addr { * * Return: IPv4 address if @addr is IPv4, NULL otherwise */ -static inline const struct in_addr *inany_v4(const union inany_addr *addr) +static inline struct in_addr *inany_v4(const union inany_addr *addr) { if (!IN6_IS_ADDR_V4MAPPED(&addr->a6)) return NULL; - return &addr->_v4mapped.a4; + return (struct in_addr *)&addr->_v4mapped.a4; } /** inany_equals - Compare two IPv[46] addresses @@ -66,3 +66,29 @@ static inline void inany_from_af(union inany_addr *aa, int af, const void *addr) assert(0); } } + +/** inany_from_sockaddr - Extract IPv[46] address and port number from sockaddr + * @a: Pointer to store IPv[46] address + * @port: Pointer to store port number, host order + * @addr: struct sockaddr_in (IPv4) or struct sockaddr_in6 (IPv6) + */ +static inline void inany_from_sockaddr(union inany_addr *aa, in_port_t *port, + const void *sa_) +{ + const struct sockaddr *sa = (const struct sockaddr *)sa_; + + if (sa->sa_family == AF_INET6) { + struct sockaddr_in6 *sa6 = (struct sockaddr_in6 *)sa; + + inany_from_af(aa, AF_INET6, &sa6->sin6_addr); + *port = ntohs(sa6->sin6_port); + } else if (sa->sa_family == AF_INET) { + struct sockaddr_in *sa4 = (struct sockaddr_in *)sa; + + inany_from_af(aa, AF_INET, &sa4->sin_addr); + *port = ntohs(sa4->sin_port); + } else { + /* Not valid to call with other address families */ + assert(0); + } +} diff --git a/tcp.c b/tcp.c index b05ed6c..fca5df4 100644 --- a/tcp.c +++ b/tcp.c @@ -2724,6 +2724,29 @@ static void tcp_connect_finish(struct ctx *c, struct tcp_tap_conn *conn) conn_flag(c, conn, ACK_FROM_TAP_DUE); } +static void tcp_snat_inbound(const struct ctx *c, union inany_addr *addr) +{ + struct in_addr *addr4 = inany_v4(addr); + + if (addr4) { + if (IN4_IS_ADDR_LOOPBACK(addr4) || + IN4_IS_ADDR_UNSPECIFIED(addr4) || + IN4_ARE_ADDR_EQUAL(addr4, &c->ip4.addr_seen)) + *addr4 = c->ip4.gw; + } else { + struct in6_addr *addr6 = &addr->a6; + + if (IN6_IS_ADDR_LOOPBACK(addr6) || + IN6_ARE_ADDR_EQUAL(addr6, &c->ip6.addr_seen) || + IN6_ARE_ADDR_EQUAL(addr6, &c->ip6.addr)) { + if (IN6_IS_ADDR_LINKLOCAL(&c->ip6.gw)) + *addr6 = c->ip6.gw; + else + *addr6 = c->ip6.addr_ll; + } + } +} + /** * tcp_tap_conn_from_sock() - Initialize state for non-spliced connection * @c: Execution context @@ -2744,43 +2767,10 @@ 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 (sa->sa_family == AF_INET6) { - struct sockaddr_in6 sa6; - - memcpy(&sa6, sa, sizeof(sa6)); - - if (IN6_IS_ADDR_LOOPBACK(&sa6.sin6_addr) || - IN6_ARE_ADDR_EQUAL(&sa6.sin6_addr, &c->ip6.addr_seen) || - IN6_ARE_ADDR_EQUAL(&sa6.sin6_addr, &c->ip6.addr)) { - struct in6_addr *src; + inany_from_sockaddr(&conn->addr, &conn->sock_port, sa); + conn->tap_port = ref.r.p.tcp.tcp.index; - if (IN6_IS_ADDR_LINKLOCAL(&c->ip6.gw)) - src = &c->ip6.gw; - else - src = &c->ip6.addr_ll; - - memcpy(&sa6.sin6_addr, src, sizeof(*src)); - } - - inany_from_af(&conn->addr, AF_INET6, &sa6.sin6_addr); - - conn->sock_port = ntohs(sa6.sin6_port); - conn->tap_port = ref.r.p.tcp.tcp.index; - } else { - struct sockaddr_in sa4; - - memcpy(&sa4, sa, sizeof(sa4)); - - if (IN4_IS_ADDR_LOOPBACK(&sa4.sin_addr) || - IN4_IS_ADDR_UNSPECIFIED(&sa4.sin_addr) || - IN4_ARE_ADDR_EQUAL(&sa4.sin_addr, &c->ip4.addr_seen)) - sa4.sin_addr = c->ip4.gw; - - inany_from_af(&conn->addr, AF_INET, &sa4.sin_addr); - - conn->sock_port = ntohs(sa4.sin_port); - conn->tap_port = ref.r.p.tcp.tcp.index; - } + tcp_snat_inbound(c, &conn->addr); tcp_seq_init(c, conn, now); tcp_hash_insert(c, conn); -- 2.38.1

Previous cleanups mean that tcp_sock_init4() and tcp_sock_init6() are almost identical, and the remaining differences can be easily parameterized. Combine both into a single tcp_sock_init_af() function. Signed-off-by: David Gibson --- tcp.c | 50 +++++++++++++++----------------------------------- 1 file changed, 15 insertions(+), 35 deletions(-) diff --git a/tcp.c b/tcp.c index fca5df4..616b9d0 100644 --- a/tcp.c +++ b/tcp.c @@ -2973,52 +2973,32 @@ 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 + * tcp_sock_init_af() - Initialise listening socket for a given af and port * @c: Execution context + * @af: Address family to listen on + * @port: Port, host order * @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 + * + * Return: fd for the new listening socket, or -1 on failure */ -static void tcp_sock_init4(const struct ctx *c, const struct in_addr *addr, - const char *ifname, in_port_t port) +static int tcp_sock_init_af(const struct ctx *c, int af, in_port_t port, + const struct in_addr *addr, const char *ifname) { in_port_t idx = port + c->tcp.fwd_in.delta[port]; union tcp_epoll_ref tref = { .tcp.listen = 1, .tcp.index = idx }; int s; - 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, IPPROTO_TCP, addr, ifname, port, tref.u32); if (c->tcp.fwd_in.mode == FWD_AUTO) - tcp_sock_init_ext[port][V4] = s; -} + tcp_sock_init_ext[port][(af == AF_INET) ? V4 : V6] = s; -/** - * tcp_sock_init6() - Initialise listening sockets for a given IPv6 port - * @c: Execution context - * @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, - const struct in6_addr *addr, const char *ifname, - in_port_t port) -{ - in_port_t idx = port + c->tcp.fwd_in.delta[port]; - 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); - if (s >= 0) - tcp_sock_set_bufsize(c, s); - else - s = -1; + if (s < 0) + return -1; - if (c->tcp.fwd_in.mode == FWD_AUTO) - tcp_sock_init_ext[port][V6] = s; + tcp_sock_set_bufsize(c, s); + return s; } /** @@ -3033,9 +3013,9 @@ 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, addr, ifname, port); + tcp_sock_init_af(c, AF_INET, port, addr, ifname); if ((af == AF_INET6 || af == AF_UNSPEC) && c->ifi6) - tcp_sock_init6(c, addr, ifname, port); + tcp_sock_init_af(c, AF_INET6, port, addr, ifname); } /** -- 2.38.1

According to its doc comments, sock_l4() returns -1 on error. It does, except in one case where it returns -EIO. Fix this inconsistency to match the docs and always return -1. Signed-off-by: David Gibson --- util.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/util.c b/util.c index fc629ed..e2222b8 100644 --- a/util.c +++ b/util.c @@ -141,7 +141,7 @@ int sock_l4(const struct ctx *c, int af, uint8_t proto, if (fd > SOCKET_MAX) { close(fd); - return -EIO; + return -1; } ref.r.s = fd; -- 2.38.1