summaryrefslogtreecommitdiff
path: root/src/ipcpd/eth/eth.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/ipcpd/eth/eth.c')
-rw-r--r--src/ipcpd/eth/eth.c173
1 files changed, 111 insertions, 62 deletions
diff --git a/src/ipcpd/eth/eth.c b/src/ipcpd/eth/eth.c
index 1dc2b8fe..e47e8689 100644
--- a/src/ipcpd/eth/eth.c
+++ b/src/ipcpd/eth/eth.c
@@ -150,8 +150,9 @@
#define ETH_FRAME_SIZE (ETH_HEADER_SIZE + ETH_MTU_MAX)
#endif
-#define NAME_QUERY_TIMEO 2000 /* ms */
-#define MGMT_TIMEO 100 /* ms */
+#define NAME_QUERY_TIMEO 1900 /* ms total budget */
+#define NAME_QUERY_RETRIES 3 /* retransmits, 4 attempts total */
+#define MGMT_TIMEO 100 /* ms */
#define MGMT_FRAME_SIZE IPCP_ETH_MGMT_FRAME_SIZE
#define ETH_RIB_PATH "eth"
@@ -270,6 +271,7 @@ struct {
size_t n_dlv_f;
size_t n_buf_f;
size_t n_rcv_f;
+ size_t n_snd_f;
size_t kern_rcv;
size_t kern_drp;
} stat;
@@ -426,19 +428,24 @@ static int eth_rib_read(const char * path,
buf[0] = '\0';
if (strcmp(entry, "summary") == 0) {
- int rcvbuf = 0;
- int queued = 0;
-#if defined(HAVE_RAW_SOCKETS) && defined(__linux__)
- struct tpacket_stats tp_stats;
- socklen_t tp_len = sizeof(tp_stats);
-#endif
+ int n;
#if defined(HAVE_RAW_SOCKETS)
+ int rcvbuf = 0;
+ int sndbuf = 0;
+ int queued = 0;
socklen_t optlen = sizeof(rcvbuf);
+# if defined(__linux__)
+ struct tpacket_stats tp_stats;
+ socklen_t tp_len = sizeof(tp_stats);
+# endif
+
getsockopt(eth_data.s_fd, SOL_SOCKET,
SO_RCVBUF, &rcvbuf, &optlen);
+ optlen = sizeof(sndbuf);
+ getsockopt(eth_data.s_fd, SOL_SOCKET,
+ SO_SNDBUF, &sndbuf, &optlen);
ioctl(eth_data.s_fd, FIONREAD, &queued);
-#endif
-#if defined(HAVE_RAW_SOCKETS) && defined(__linux__)
+# if defined(__linux__)
if (getsockopt(eth_data.s_fd, SOL_PACKET,
PACKET_STATISTICS,
&tp_stats, &tp_len) == 0) {
@@ -447,8 +454,9 @@ static int eth_rib_read(const char * path,
FETCH_ADD_RELAXED(&eth_data.stat.kern_drp,
tp_stats.tp_drops);
}
+# endif
#endif
- sprintf(buf,
+ n = sprintf(buf,
"Active flows: %20zu\n"
"Total frames received: %20zu\n"
"Total frames sent: %20zu\n"
@@ -458,10 +466,7 @@ static int eth_rib_read(const char * path,
"Delivery (N+1) failures: %20zu\n"
"Buffer alloc failures: %20zu\n"
"Frame read failures: %20zu\n"
- "Socket rcvbuf (bytes): %20d\n"
- "Socket queued (bytes): %20d\n"
- "Kernel frames received: %20zu\n"
- "Kernel frames dropped: %20zu\n",
+ "Frame send failures: %20zu\n",
LOAD_RELAXED(&eth_data.stat.n_flows),
LOAD_RELAXED(&eth_data.stat.n_rcv),
LOAD_RELAXED(&eth_data.stat.n_snd),
@@ -471,12 +476,22 @@ static int eth_rib_read(const char * path,
LOAD_RELAXED(&eth_data.stat.n_dlv_f),
LOAD_RELAXED(&eth_data.stat.n_buf_f),
LOAD_RELAXED(&eth_data.stat.n_rcv_f),
- rcvbuf,
- queued,
+ LOAD_RELAXED(&eth_data.stat.n_snd_f));
+#if defined(HAVE_RAW_SOCKETS)
+ n += sprintf(buf + n,
+ "Socket rcvbuf (bytes): %20d\n"
+ "Socket sndbuf (bytes): %20d\n"
+ "Socket queued (bytes): %20d\n",
+ rcvbuf, sndbuf, queued);
+# if defined(__linux__)
+ n += sprintf(buf + n,
+ "Kernel frames received: %20zu\n"
+ "Kernel frames dropped: %20zu\n",
LOAD_RELAXED(&eth_data.stat.kern_rcv),
LOAD_RELAXED(&eth_data.stat.kern_drp));
-
- return strlen(buf);
+# endif
+#endif
+ return n;
}
fd = atoi(entry);
@@ -1175,18 +1190,40 @@ static void * eth_ipcp_packet_reader(void * o)
continue;
}
slen = sizeof(src);
+ /* MSG_DONTWAIT: RD_THR>1 race-loser bails with EAGAIN. */
frame_len = recvfrom(eth_data.s_fd, buf,
- ETH_MTU + ETH_HEADER_TOT_SIZE, 0,
+ ETH_MTU + ETH_HEADER_TOT_SIZE,
+ MSG_DONTWAIT,
(struct sockaddr *) &src, &slen);
#endif
- if (frame_len <= 0) {
- log_dbg("Failed to receive frame.");
+ if (frame_len == 0) {
ipcp_spb_release(spb);
+ continue; /* Spurious */
+ }
+
+ if (frame_len < 0) {
+ ipcp_spb_release(spb);
+
+ if (errno == EAGAIN || errno == EWOULDBLOCK)
+ continue;
+
+ log_dbg("Failed to rcv frame: %s.", strerror(errno));
FETCH_ADD_RELAXED(&eth_data.stat.n_rcv_f, 1);
continue;
}
#endif
+#if defined(HAVE_NETMAP)
+ eth_len = hdr.len;
+#elif defined(HAVE_BPF)
+ eth_len = ((struct bpf_hdr *) buf)->bh_caplen;
+#else
+ eth_len = (size_t) frame_len;
+#endif
+ /* Defense in depth: reject before parsing dereferences. */
+ if (eth_len < ETH_HEADER_TOT_SIZE)
+ goto fail_frame;
+
#if defined(HAVE_RAW_SOCKETS)
/* Drop our own egress. */
if (src.sll_pkttype == PACKET_OUTGOING)
@@ -1236,13 +1273,6 @@ static void * eth_ipcp_packet_reader(void * o)
ssap = reverse_bits(e_frame->ssap);
#endif
-#if defined(HAVE_NETMAP)
- eth_len = hdr.len;
-#elif defined(HAVE_BPF)
- eth_len = ((struct bpf_hdr *) buf)->bh_caplen;
-#else
- eth_len = (size_t) frame_len;
-#endif
if (eth_len < ETH_HEADER_TOT_SIZE + (size_t) length)
goto fail_frame;
@@ -1421,6 +1451,7 @@ static void * eth_ipcp_packet_writer(void * o)
FETCH_ADD_RELAXED(
&eth_data.fd_to_ef[fd].stat.p_snd_f,
1);
+ FETCH_ADD_RELAXED(&eth_data.stat.n_snd_f, 1);
} else {
FETCH_ADD_RELAXED(
&eth_data.fd_to_ef[fd].stat.p_snd,
@@ -1776,12 +1807,14 @@ static int eth_init_bpf(struct ifreq * ifr)
return -1;
}
#elif defined(HAVE_RAW_SOCKETS)
+#define SOCKOPT()
static int eth_init_raw_socket(struct ifreq * ifr)
{
int idx;
- int flags;
+ int sndbuf;
+ int rcvbuf;
#if defined(IPCP_ETH_QDISC_BYPASS)
- int qdisc_bypass = 1;
+ int qdisc_bypass = 1;
#endif /* ENABLE_QDISC_BYPASS */
idx = if_nametoindex(ifr->ifr_name);
@@ -1789,6 +1822,7 @@ static int eth_init_raw_socket(struct ifreq * ifr)
log_err("Failed to retrieve interface index.");
return -1;
}
+
memset(&(eth_data.device), 0, sizeof(eth_data.device));
eth_data.device.sll_ifindex = idx;
eth_data.device.sll_family = AF_PACKET;
@@ -1805,17 +1839,6 @@ static int eth_init_raw_socket(struct ifreq * ifr)
goto fail_socket;
}
- flags = fcntl(eth_data.s_fd, F_GETFL, 0);
- if (flags < 0) {
- log_err("Failed to get flags.");
- goto fail_device;
- }
-
- if (fcntl(eth_data.s_fd, F_SETFL, flags | O_NONBLOCK)) {
- log_err("Failed to set socket non-blocking.");
- goto fail_device;
- }
-
#if defined(IPCP_ETH_QDISC_BYPASS)
if (setsockopt(eth_data.s_fd, SOL_PACKET, PACKET_QDISC_BYPASS,
&qdisc_bypass, sizeof(qdisc_bypass))) {
@@ -1823,6 +1846,18 @@ static int eth_init_raw_socket(struct ifreq * ifr)
}
#endif
+ sndbuf = IPCP_ETH_SNDBUF;
+ if (sndbuf > 0 && setsockopt(eth_data.s_fd, SOL_SOCKET, SO_SNDBUF,
+ &sndbuf, sizeof(sndbuf))) {
+ log_info("Failed to set SO_SNDBUF to %d.", sndbuf);
+ }
+
+ rcvbuf = IPCP_ETH_RCVBUF;
+ if (rcvbuf > 0 && setsockopt(eth_data.s_fd, SOL_SOCKET, SO_RCVBUF,
+ &rcvbuf, sizeof(rcvbuf))) {
+ log_info("Failed to set SO_RCVBUF to %d.", rcvbuf);
+ }
+
if (bind(eth_data.s_fd, (struct sockaddr *) &eth_data.device,
sizeof(eth_data.device)) < 0) {
log_err("Failed to bind socket to interface.");
@@ -1999,12 +2034,14 @@ static int eth_ipcp_unreg(const uint8_t * hash)
static int eth_ipcp_query(const uint8_t * hash)
{
uint8_t r_addr[MAC_SIZE];
- struct timespec timeout = TIMESPEC_INIT_MS(NAME_QUERY_TIMEO);
+ struct timespec timeout;
struct dir_query * query;
int ret;
+ int attempt;
uint8_t * buf;
struct mgmt_msg * msg;
size_t len;
+ long per_ms;
if (shim_data_dir_has(eth_data.shim_data, hash))
return 0;
@@ -2024,33 +2061,45 @@ static int eth_ipcp_query(const uint8_t * hash)
memset(r_addr, 0xff, MAC_SIZE);
- query = shim_data_dir_query_create(eth_data.shim_data, hash);
- if (query == NULL) {
- free(buf);
- return -1;
- }
+ per_ms = NAME_QUERY_TIMEO / (NAME_QUERY_RETRIES + 1);
+
+ ret = -1;
+ for (attempt = 0; attempt <= NAME_QUERY_RETRIES; ++attempt) {
+ query = shim_data_dir_query_create(eth_data.shim_data, hash);
+ if (query == NULL) {
+ ret = -1;
+ break;
+ }
- if (eth_ipcp_send_frame(r_addr,
+ if (eth_ipcp_send_frame(r_addr,
#if defined(BUILD_ETH_DIX)
- MGMT_EID,
+ MGMT_EID,
#elif defined(BUILD_ETH_LLC)
- reverse_bits(MGMT_SAP),
- reverse_bits(MGMT_SAP),
+ reverse_bits(MGMT_SAP),
+ reverse_bits(MGMT_SAP),
#endif
- buf, len)) {
- log_err("Failed to send management frame.");
- shim_data_dir_query_destroy(eth_data.shim_data, query);
- free(buf);
- return -1;
- }
+ buf, len)) {
+ log_err("Failed to send management frame.");
+ shim_data_dir_query_destroy(eth_data.shim_data,
+ query);
+ ret = -1;
+ break;
+ }
- FETCH_ADD_RELAXED(&eth_data.stat.n_mgmt_snd, 1);
+ FETCH_ADD_RELAXED(&eth_data.stat.n_mgmt_snd, 1);
- free(buf);
+ timeout.tv_sec = per_ms / 1000;
+ timeout.tv_nsec = (per_ms % 1000) * 1000000L;
+
+ ret = shim_data_dir_query_wait(query, &timeout);
+
+ shim_data_dir_query_destroy(eth_data.shim_data, query);
- ret = shim_data_dir_query_wait(query, &timeout);
+ if (ret != -ETIMEDOUT)
+ break;
+ }
- shim_data_dir_query_destroy(eth_data.shim_data, query);
+ free(buf);
return ret;
}
generated by cgit v1.2.3 (git 2.39.1) at 2026-05-23 19:58:59 +0000