view src/event/quic/bpf/ngx_quic_reuseport_helper.c @ 8946:56dec0d4e5b1 quic

QUIC: avoid excessive buffer allocations in stream output. Previously, when a few bytes were send to a QUIC stream by the application, a 4K buffer was allocated for these bytes. Then a STREAM frame was created and that entire buffer was used as data for that frame. The frame with the buffer were in use up until the frame was acked by client. Meanwhile, when more bytes were send to the stream, more buffers were allocated and assigned as data to newer STREAM frames. In this scenario most buffer memory is unused. Now the unused part of the stream output buffer is available for further stream output while earlier parts of the buffer are waiting to be acked. This is achieved by splitting the output buffer.
author Roman Arutyunyan <arut@nginx.com>
date Fri, 24 Dec 2021 18:13:51 +0300
parents 1a489587e1c8
children a2fbae359828
line wrap: on
line source

#include <errno.h>
#include <linux/string.h>
#include <linux/udp.h>
#include <linux/bpf.h>
/*
 * the bpf_helpers.h is not included into linux-headers, only available
 * with kernel sources in "tools/lib/bpf/bpf_helpers.h" or in libbpf.
 */
#include <bpf/bpf_helpers.h>


#if !defined(SEC)
#define SEC(NAME)  __attribute__((section(NAME), used))
#endif


#if defined(LICENSE_GPL)

/*
 * To see debug:
 *
 *  echo 1 > /sys/kernel/debug/tracing/events/bpf_trace/enable
 *  cat /sys/kernel/debug/tracing/trace_pipe
 *  echo 0 > /sys/kernel/debug/tracing/events/bpf_trace/enable
 */

#define debugmsg(fmt, ...)                                                    \
do {                                                                          \
    char __buf[] = fmt;                                                       \
    bpf_trace_printk(__buf, sizeof(__buf), ##__VA_ARGS__);                    \
} while (0)

#else

#define debugmsg(fmt, ...)

#endif

char _license[] SEC("license") = LICENSE;

/*****************************************************************************/

#define NGX_QUIC_PKT_LONG        0x80  /* header form */
#define NGX_QUIC_SERVER_CID_LEN  20


#define advance_data(nbytes)                                                  \
    offset += nbytes;                                                         \
    if (start + offset > end) {                                               \
        debugmsg("cannot read %ld bytes at offset %ld", nbytes, offset);      \
        goto failed;                                                          \
    }                                                                         \
    data = start + offset - 1;


#define ngx_quic_parse_uint64(p)                                              \
    (((__u64)(p)[0] << 56) |                                                  \
     ((__u64)(p)[1] << 48) |                                                  \
     ((__u64)(p)[2] << 40) |                                                  \
     ((__u64)(p)[3] << 32) |                                                  \
     ((__u64)(p)[4] << 24) |                                                  \
     ((__u64)(p)[5] << 16) |                                                  \
     ((__u64)(p)[6] << 8)  |                                                  \
     ((__u64)(p)[7]))

/*
 * actual map object is created by the "bpf" system call,
 * all pointers to this variable are replaced by the bpf loader
 */
struct bpf_map_def SEC("maps") ngx_quic_sockmap;


SEC(PROGNAME)
int ngx_quic_select_socket_by_dcid(struct sk_reuseport_md *ctx)
{
    int             rc;
    __u64           key;
    size_t          len, offset;
    unsigned char  *start, *end, *data, *dcid;

    start = ctx->data;
    end = (unsigned char *) ctx->data_end;
    offset = 0;

    advance_data(sizeof(struct udphdr)); /* data at UDP header */
    advance_data(1); /* data at QUIC flags */

    if (data[0] & NGX_QUIC_PKT_LONG) {

        advance_data(4); /* data at QUIC version */
        advance_data(1); /* data at DCID len */

        len = data[0];   /* read DCID length */

        if (len < 8) {
            /* it's useless to search for key in such short DCID */
            return SK_PASS;
        }

    } else {
        len = NGX_QUIC_SERVER_CID_LEN;
    }

    dcid = &data[1];
    advance_data(len); /* we expect the packet to have full DCID */

    /* make verifier happy */
    if (dcid + sizeof(__u64) > end) {
        goto failed;
    }

    key = ngx_quic_parse_uint64(dcid);

    rc = bpf_sk_select_reuseport(ctx, &ngx_quic_sockmap, &key, 0);

    switch (rc) {
    case 0:
        debugmsg("nginx quic socket selected by key 0x%llx", key);
        return SK_PASS;

    /* kernel returns positive error numbers, errno.h defines positive */
    case -ENOENT:
        debugmsg("nginx quic default route for key 0x%llx", key);
        /* let the default reuseport logic decide which socket to choose */
        return SK_PASS;

    default:
        debugmsg("nginx quic bpf_sk_select_reuseport err: %d key 0x%llx",
                  rc, key);
        goto failed;
    }

failed:
    /*
     * SK_DROP will generate ICMP, but we may want to process "invalid" packet
     * in userspace quic to investigate further and finally react properly
     * (maybe ignore, maybe send something in response or close connection)
     */
    return SK_PASS;
}