passt/tcp_splice.c

764 lines
18 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-or-later
/* PASTA - Pack A Subtle Tap Abstraction
* for network namespace/tap device mode
*
* tcp_splice.c - direct namespace forwarding for local connections
*
* Copyright (c) 2020-2022 Red Hat GmbH
* Author: Stefano Brivio <sbrivio@redhat.com>
*/
/**
* DOC: Theory of Operation
*
*
* For local traffic directed to TCP ports configured for direct
* mapping between namespaces, packets are directly translated between
* L4 sockets using a pair of splice() syscalls. These connections are
* tracked by struct tcp_splice_conn entries in the @tc array, using
* these events:
*
* - SPLICE_CONNECT: connection accepted, connecting to target
* - SPLICE_ESTABLISHED: connection to target established
* - OUT_WAIT_0: pipe to accepted socket full, wait for EPOLLOUT
* - OUT_WAIT_1: pipe to target socket full, wait for EPOLLOUT
* - FIN_RCVD_0: FIN (EPOLLRDHUP) seen from accepted socket
* - FIN_RCVD_1: FIN (EPOLLRDHUP) seen from target socket
* - FIN_SENT_0: FIN (write shutdown) sent to accepted socket
* - FIN_SENT_1: FIN (write shutdown) sent to target socket
*
* #syscalls:pasta pipe2|pipe fcntl arm:fcntl64 ppc64:fcntl64 i686:fcntl64
*/
#include <sched.h>
#include <unistd.h>
#include <signal.h>
#include <errno.h>
#include <fcntl.h>
#include <limits.h>
#include <stdint.h>
#include <stdbool.h>
#include <string.h>
#include <time.h>
#include <net/ethernet.h>
#include <netinet/in.h>
#include <netinet/tcp.h>
#include <sys/epoll.h>
#include <sys/types.h>
#include <sys/socket.h>
#include "util.h"
#include "ip.h"
#include "passt.h"
#include "log.h"
#include "tcp_splice.h"
#include "siphash.h"
#include "inany.h"
#include "flow.h"
#include "flow_table.h"
#define MAX_PIPE_SIZE (8UL * 1024 * 1024)
#define TCP_SPLICE_PIPE_POOL_SIZE 32
#define TCP_SPLICE_CONN_PRESSURE 30 /* % of conn_count */
#define TCP_SPLICE_FILE_PRESSURE 30 /* % of c->nofile */
/* Pools for pre-opened sockets (in namespace) */
#define TCP_SOCK_POOL_TSH 16 /* Refill in ns if > x used */
static int ns_sock_pool4 [TCP_SOCK_POOL_SIZE];
static int ns_sock_pool6 [TCP_SOCK_POOL_SIZE];
/* Pool of pre-opened pipes */
static int splice_pipe_pool [TCP_SPLICE_PIPE_POOL_SIZE][2];
#define CONN_HAS(conn, set) (((conn)->events & (set)) == (set))
/* Display strings for connection events */
static const char *tcp_splice_event_str[] __attribute((__unused__)) = {
"SPLICE_CONNECT", "SPLICE_ESTABLISHED", "OUT_WAIT_0", "OUT_WAIT_1",
"FIN_RCVD_0", "FIN_RCVD_1", "FIN_SENT_0", "FIN_SENT_1",
};
/* Display strings for connection flags */
static const char *tcp_splice_flag_str[] __attribute((__unused__)) = {
"SPLICE_V6", "RCVLOWAT_SET_0", "RCVLOWAT_SET_1", "RCVLOWAT_ACT_0",
"RCVLOWAT_ACT_1", "CLOSING",
};
/* Forward declaration */
static int tcp_sock_refill_ns(void *arg);
static int tcp_conn_sock_ns(const struct ctx *c, sa_family_t af);
/**
* conn_at_sidx() - Get spliced TCP connection specific flow at given sidx
* @sidx: Flow and side to retrieve
*
* Return: Spliced TCP connection at @sidx, or NULL of @sidx is invalid.
* Asserts if the flow at @sidx is not FLOW_TCP_SPLICE.
*/
static struct tcp_splice_conn *conn_at_sidx(flow_sidx_t sidx)
{
union flow *flow = flow_at_sidx(sidx);
if (!flow)
return NULL;
ASSERT(flow->f.type == FLOW_TCP_SPLICE);
return &flow->tcp_splice;
}
/**
* tcp_splice_conn_epoll_events() - epoll events masks for given state
* @events: Connection event flags
* @ev: Events to fill in, 0 is accepted socket, 1 is connecting socket
*/
static void tcp_splice_conn_epoll_events(uint16_t events,
struct epoll_event ev[])
{
unsigned sidei;
flow_foreach_sidei(sidei)
ev[sidei].events = 0;
if (events & SPLICE_ESTABLISHED) {
flow_foreach_sidei(sidei) {
if (!(events & FIN_SENT(!sidei)))
ev[sidei].events = EPOLLIN | EPOLLRDHUP;
}
} else if (events & SPLICE_CONNECT) {
ev[1].events = EPOLLOUT;
}
flow_foreach_sidei(sidei)
ev[sidei].events |= (events & OUT_WAIT(sidei)) ? EPOLLOUT : 0;
}
/**
* tcp_splice_epoll_ctl() - Add/modify/delete epoll state from connection events
* @c: Execution context
* @conn: Connection pointer
*
* Return: 0 on success, negative error code on failure (not on deletion)
*/
static int tcp_splice_epoll_ctl(const struct ctx *c,
struct tcp_splice_conn *conn)
{
int m = conn->in_epoll ? EPOLL_CTL_MOD : EPOLL_CTL_ADD;
const union epoll_ref ref[SIDES] = {
{ .type = EPOLL_TYPE_TCP_SPLICE, .fd = conn->s[0],
.flowside = FLOW_SIDX(conn, 0) },
{ .type = EPOLL_TYPE_TCP_SPLICE, .fd = conn->s[1],
.flowside = FLOW_SIDX(conn, 1) }
};
struct epoll_event ev[SIDES] = { { .data.u64 = ref[0].u64 },
{ .data.u64 = ref[1].u64 } };
tcp_splice_conn_epoll_events(conn->events, ev);
if (epoll_ctl(c->epollfd, m, conn->s[0], &ev[0]) ||
epoll_ctl(c->epollfd, m, conn->s[1], &ev[1])) {
int ret = -errno;
flow_err(conn, "ERROR on epoll_ctl(): %s", strerror(errno));
return ret;
}
conn->in_epoll = true;
return 0;
}
/**
* conn_flag_do() - Set/unset given flag, log, update epoll on CLOSING flag
* @c: Execution context
* @conn: Connection pointer
* @flag: Flag to set, or ~flag to unset
*/
static void conn_flag_do(const struct ctx *c, struct tcp_splice_conn *conn,
unsigned long flag)
{
if (flag & (flag - 1)) {
int flag_index = fls(~flag);
if (!(conn->flags & ~flag))
return;
conn->flags &= flag;
if (flag_index >= 0)
flow_dbg(conn, "%s dropped",
tcp_splice_flag_str[flag_index]);
} else {
int flag_index = fls(flag);
if (conn->flags & flag)
return;
conn->flags |= flag;
if (flag_index >= 0)
flow_dbg(conn, "%s", tcp_splice_flag_str[flag_index]);
}
if (flag == CLOSING) {
epoll_ctl(c->epollfd, EPOLL_CTL_DEL, conn->s[0], NULL);
epoll_ctl(c->epollfd, EPOLL_CTL_DEL, conn->s[1], NULL);
}
}
#define conn_flag(c, conn, flag) \
do { \
flow_trace(conn, "flag at %s:%i", __func__, __LINE__); \
conn_flag_do(c, conn, flag); \
} while (0)
/**
* conn_event_do() - Set and log connection events, update epoll state
* @c: Execution context
* @conn: Connection pointer
* @event: Connection event
*/
static void conn_event_do(const struct ctx *c, struct tcp_splice_conn *conn,
unsigned long event)
{
if (event & (event - 1)) {
int flag_index = fls(~event);
if (!(conn->events & ~event))
return;
conn->events &= event;
if (flag_index >= 0)
flow_dbg(conn, "~%s", tcp_splice_event_str[flag_index]);
} else {
int flag_index = fls(event);
if (conn->events & event)
return;
conn->events |= event;
if (flag_index >= 0)
flow_dbg(conn, "%s", tcp_splice_event_str[flag_index]);
}
if (tcp_splice_epoll_ctl(c, conn))
conn_flag(c, conn, CLOSING);
}
#define conn_event(c, conn, event) \
do { \
flow_trace(conn, "event at %s:%i",__func__, __LINE__); \
conn_event_do(c, conn, event); \
} while (0)
/**
* tcp_splice_flow_defer() - Deferred per-flow handling (clean up closed)
* @conn: Connection entry to handle
*
* Return: true if the flow is ready to free, false otherwise
*/
bool tcp_splice_flow_defer(struct tcp_splice_conn *conn)
{
unsigned sidei;
if (!(conn->flags & CLOSING))
return false;
flow_foreach_sidei(sidei) {
/* Flushing might need to block: don't recycle them. */
if (conn->pipe[sidei][0] >= 0) {
close(conn->pipe[sidei][0]);
close(conn->pipe[sidei][1]);
conn->pipe[sidei][0] = conn->pipe[sidei][1] = -1;
}
if (conn->s[sidei] >= 0) {
close(conn->s[sidei]);
conn->s[sidei] = -1;
}
conn->read[sidei] = conn->written[sidei] = 0;
}
conn->events = SPLICE_CLOSED;
conn->flags = 0;
flow_dbg(conn, "CLOSED");
return true;
}
/**
* tcp_splice_connect_finish() - Completion of connect() or call on success
* @c: Execution context
* @conn: Connection pointer
*
* Return: 0 on success, -EIO on failure
*/
static int tcp_splice_connect_finish(const struct ctx *c,
struct tcp_splice_conn *conn)
{
unsigned sidei;
int i = 0;
flow_foreach_sidei(sidei) {
for (; i < TCP_SPLICE_PIPE_POOL_SIZE; i++) {
if (splice_pipe_pool[i][0] >= 0) {
SWAP(conn->pipe[sidei][0],
splice_pipe_pool[i][0]);
SWAP(conn->pipe[sidei][1],
splice_pipe_pool[i][1]);
break;
}
}
if (conn->pipe[sidei][0] < 0) {
if (pipe2(conn->pipe[sidei], O_NONBLOCK | O_CLOEXEC)) {
flow_err(conn, "cannot create %d->%d pipe: %s",
sidei, !sidei, strerror(errno));
conn_flag(c, conn, CLOSING);
return -EIO;
}
if (fcntl(conn->pipe[sidei][0], F_SETPIPE_SZ,
c->tcp.pipe_size) != (int)c->tcp.pipe_size) {
flow_trace(conn,
"cannot set %d->%d pipe size to %zu",
sidei, !sidei, c->tcp.pipe_size);
}
}
}
if (!(conn->events & SPLICE_ESTABLISHED))
conn_event(c, conn, SPLICE_ESTABLISHED);
return 0;
}
/**
* tcp_splice_connect() - Create and connect socket for new spliced connection
* @c: Execution context
* @conn: Connection pointer
*
* Return: 0 for connect() succeeded or in progress, negative value on error
*/
static int tcp_splice_connect(const struct ctx *c, struct tcp_splice_conn *conn)
{
const struct flowside *tgt = &conn->f.side[TGTSIDE];
sa_family_t af = inany_v4(&tgt->eaddr) ? AF_INET : AF_INET6;
uint8_t tgtpif = conn->f.pif[TGTSIDE];
union sockaddr_inany sa;
socklen_t sl;
if (tgtpif == PIF_HOST)
conn->s[1] = tcp_conn_sock(c, af);
else if (tgtpif == PIF_SPLICE)
conn->s[1] = tcp_conn_sock_ns(c, af);
else
ASSERT(0);
if (conn->s[1] < 0)
return -1;
if (setsockopt(conn->s[1], SOL_TCP, TCP_QUICKACK,
&((int){ 1 }), sizeof(int))) {
flow_trace(conn, "failed to set TCP_QUICKACK on socket %i",
conn->s[1]);
}
pif_sockaddr(c, &sa, &sl, tgtpif, &tgt->eaddr, tgt->eport);
if (connect(conn->s[1], &sa.sa, sl)) {
if (errno != EINPROGRESS) {
flow_trace(conn, "Couldn't connect socket for splice: %s",
strerror(errno));
return -errno;
}
conn_event(c, conn, SPLICE_CONNECT);
} else {
conn_event(c, conn, SPLICE_ESTABLISHED);
return tcp_splice_connect_finish(c, conn);
}
return 0;
}
/**
* tcp_conn_sock_ns() - Obtain a connectable socket in the namespace
* @c: Execution context
* @af: Address family (AF_INET or AF_INET6)
*
* Return: Socket fd in the namespace on success, -errno on failure
*/
static int tcp_conn_sock_ns(const struct ctx *c, sa_family_t af)
{
int *p = af == AF_INET6 ? ns_sock_pool6 : ns_sock_pool4;
int s;
if ((s = tcp_conn_pool_sock(p)) >= 0)
return s;
/* If the pool is empty we have to incur the latency of entering the ns.
* Therefore, we might as well refill the whole pool while we're at it.
* This differs from tcp_conn_sock().
*/
NS_CALL(tcp_sock_refill_ns, c);
if ((s = tcp_conn_pool_sock(p)) >= 0)
return s;
err("TCP: No available ns sockets for new connection");
return -1;
}
/**
* tcp_splice_conn_from_sock() - Attempt to init state for a spliced connection
* @c: Execution context
* @flow: flow to initialise
* @s0: Accepted (side 0) socket
* @sa: Peer address of connection
*
* #syscalls:pasta setsockopt
*/
void tcp_splice_conn_from_sock(const struct ctx *c, union flow *flow, int s0)
{
struct tcp_splice_conn *conn = FLOW_SET_TYPE(flow, FLOW_TCP_SPLICE,
tcp_splice);
ASSERT(c->mode == MODE_PASTA);
conn->s[0] = s0;
conn->s[1] = -1;
conn->pipe[0][0] = conn->pipe[0][1] = -1;
conn->pipe[1][0] = conn->pipe[1][1] = -1;
if (setsockopt(s0, SOL_TCP, TCP_QUICKACK, &((int){ 1 }), sizeof(int)))
flow_trace(conn, "failed to set TCP_QUICKACK on %i", s0);
if (tcp_splice_connect(c, conn))
conn_flag(c, conn, CLOSING);
FLOW_ACTIVATE(conn);
}
/**
* tcp_splice_sock_handler() - Handler for socket mapped to spliced connection
* @c: Execution context
* @ref: epoll reference
* @events: epoll events bitmap
*
* #syscalls:pasta splice
*/
void tcp_splice_sock_handler(struct ctx *c, union epoll_ref ref,
uint32_t events)
{
struct tcp_splice_conn *conn = conn_at_sidx(ref.flowside);
unsigned evsidei = ref.flowside.sidei, fromsidei;
uint8_t lowat_set_flag, lowat_act_flag;
int eof, never_read;
ASSERT(conn->f.type == FLOW_TCP_SPLICE);
if (conn->events == SPLICE_CLOSED)
return;
if (events & EPOLLERR) {
int err, rc;
socklen_t sl = sizeof(err);
rc = getsockopt(ref.fd, SOL_SOCKET, SO_ERROR, &err, &sl);
if (rc)
flow_err(conn, "Error retrieving SO_ERROR: %s",
strerror(errno));
else
flow_trace(conn, "Error event on socket: %s",
strerror(err));
goto close;
}
if (conn->events == SPLICE_CONNECT) {
if (!(events & EPOLLOUT))
goto close;
if (tcp_splice_connect_finish(c, conn))
goto close;
}
if (events & EPOLLOUT) {
fromsidei = !evsidei;
conn_event(c, conn, ~OUT_WAIT(evsidei));
} else {
fromsidei = evsidei;
}
if (events & EPOLLRDHUP)
/* For side 0 this is fake, but implied */
conn_event(c, conn, FIN_RCVD(evsidei));
swap:
eof = 0;
never_read = 1;
lowat_set_flag = RCVLOWAT_SET(fromsidei);
lowat_act_flag = RCVLOWAT_ACT(fromsidei);
while (1) {
tcp_splice: splice() all we have to the writing side, not what we just read In tcp_splice_sock_handler(), we try to calculate how much we can move from the pipe to the writing socket: if we just read some bytes, we'll use that amount, but if we haven't, we just try to empty the pipe. However, if we just read something, that doesn't mean that that's all the data we have on the pipe, as it's obvious from this sequence, where: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 98304 from read-side call Flow 0 (TCP connection (spliced)): 33615 from write-side call (passed 98304) Flow 0 (TCP connection (spliced)): -1 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 524288) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 Flow 0 (TCP connection (spliced)): OUT_WAIT_0 we first pile up 98304 - 33615 = 64689 pending bytes, that we read but couldn't write, as the receiver buffer is full, and we set the corresponding OUT_WAIT flag. Then: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 32768 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 32768) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 we splice() 32768 more bytes from our receiving side to the pipe. At some point: pasta: epoll event on connected spliced TCP socket 49 (events: 0x00000004) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:489 Flow 0 (TCP connection (spliced)): ~OUT_WAIT_0 Flow 0 (TCP connection (spliced)): 1320 from read-side call Flow 0 (TCP connection (spliced)): 1320 from write-side call (passed 1320) the receiver is signalling to us that it's ready for more data (EPOLLOUT). We reset the OUT_WAIT flag, read 1320 more bytes from our receiving socket into the pipe, and that's what we write to the receiver, forgetting about the pending 97457 bytes we had, which the receiver might never get (not the same 97547 bytes: we'll actually send 1320 of those). This condition is rather hard to reproduce, and it was observed with Podman pulling container images via HTTPS. In the traces above, the client is side 0 (the initiating peer), and the server is sending the whole data. Instead of splicing from pipe to socket the amount of data we just read, we need to splice all the pending data we piled up until that point. We could do that using 'read' and 'written' counters, but there's actually no need, as the kernel also keeps track of how much data is available on the pipe. So, to make this simple and more robust, just give the whole pipe size as length to splice(). The kernel knows what to do with it. Later in the function, we used 'to_write' for an optimisation meant to reduce wakeups which retries right away to splice() in both directions if we couldn't write to the receiver the whole amount of pending data. Calculate a 'pending' value instead, only if we reach that point. Now that we check for the actual amount of pending data in that optimisation, we need to make sure we don't compare a zero or negative 'written' value: if we met that, it means that the receiver signalled end-of-file, an error, or to try again later. In those three cases, the optimisation doesn't make any sense, so skip it. Reported-by: Ed Santiago <santiago@redhat.com> Reported-by: Paul Holzinger <pholzing@redhat.com> Analysed-by: Paul Holzinger <pholzing@redhat.com> Link: https://github.com/containers/podman/issues/24219 Signed-off-by: Stefano Brivio <sbrivio@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
2024-10-24 09:12:11 +02:00
ssize_t readlen, written, pending;
int more = 0;
retry:
readlen = splice(conn->s[fromsidei], NULL,
conn->pipe[fromsidei][1], NULL,
c->tcp.pipe_size,
SPLICE_F_MOVE | SPLICE_F_NONBLOCK);
flow_trace(conn, "%zi from read-side call", readlen);
if (readlen < 0) {
if (errno == EINTR)
goto retry;
if (errno != EAGAIN)
goto close;
} else if (!readlen) {
eof = 1;
} else {
never_read = 0;
tcp_splice: splice() all we have to the writing side, not what we just read In tcp_splice_sock_handler(), we try to calculate how much we can move from the pipe to the writing socket: if we just read some bytes, we'll use that amount, but if we haven't, we just try to empty the pipe. However, if we just read something, that doesn't mean that that's all the data we have on the pipe, as it's obvious from this sequence, where: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 98304 from read-side call Flow 0 (TCP connection (spliced)): 33615 from write-side call (passed 98304) Flow 0 (TCP connection (spliced)): -1 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 524288) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 Flow 0 (TCP connection (spliced)): OUT_WAIT_0 we first pile up 98304 - 33615 = 64689 pending bytes, that we read but couldn't write, as the receiver buffer is full, and we set the corresponding OUT_WAIT flag. Then: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 32768 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 32768) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 we splice() 32768 more bytes from our receiving side to the pipe. At some point: pasta: epoll event on connected spliced TCP socket 49 (events: 0x00000004) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:489 Flow 0 (TCP connection (spliced)): ~OUT_WAIT_0 Flow 0 (TCP connection (spliced)): 1320 from read-side call Flow 0 (TCP connection (spliced)): 1320 from write-side call (passed 1320) the receiver is signalling to us that it's ready for more data (EPOLLOUT). We reset the OUT_WAIT flag, read 1320 more bytes from our receiving socket into the pipe, and that's what we write to the receiver, forgetting about the pending 97457 bytes we had, which the receiver might never get (not the same 97547 bytes: we'll actually send 1320 of those). This condition is rather hard to reproduce, and it was observed with Podman pulling container images via HTTPS. In the traces above, the client is side 0 (the initiating peer), and the server is sending the whole data. Instead of splicing from pipe to socket the amount of data we just read, we need to splice all the pending data we piled up until that point. We could do that using 'read' and 'written' counters, but there's actually no need, as the kernel also keeps track of how much data is available on the pipe. So, to make this simple and more robust, just give the whole pipe size as length to splice(). The kernel knows what to do with it. Later in the function, we used 'to_write' for an optimisation meant to reduce wakeups which retries right away to splice() in both directions if we couldn't write to the receiver the whole amount of pending data. Calculate a 'pending' value instead, only if we reach that point. Now that we check for the actual amount of pending data in that optimisation, we need to make sure we don't compare a zero or negative 'written' value: if we met that, it means that the receiver signalled end-of-file, an error, or to try again later. In those three cases, the optimisation doesn't make any sense, so skip it. Reported-by: Ed Santiago <santiago@redhat.com> Reported-by: Paul Holzinger <pholzing@redhat.com> Analysed-by: Paul Holzinger <pholzing@redhat.com> Link: https://github.com/containers/podman/issues/24219 Signed-off-by: Stefano Brivio <sbrivio@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
2024-10-24 09:12:11 +02:00
if (readlen >= (long)c->tcp.pipe_size * 90 / 100)
more = SPLICE_F_MORE;
if (conn->flags & lowat_set_flag)
conn_flag(c, conn, lowat_act_flag);
}
eintr:
written = splice(conn->pipe[fromsidei][0], NULL,
tcp_splice: splice() all we have to the writing side, not what we just read In tcp_splice_sock_handler(), we try to calculate how much we can move from the pipe to the writing socket: if we just read some bytes, we'll use that amount, but if we haven't, we just try to empty the pipe. However, if we just read something, that doesn't mean that that's all the data we have on the pipe, as it's obvious from this sequence, where: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 98304 from read-side call Flow 0 (TCP connection (spliced)): 33615 from write-side call (passed 98304) Flow 0 (TCP connection (spliced)): -1 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 524288) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 Flow 0 (TCP connection (spliced)): OUT_WAIT_0 we first pile up 98304 - 33615 = 64689 pending bytes, that we read but couldn't write, as the receiver buffer is full, and we set the corresponding OUT_WAIT flag. Then: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 32768 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 32768) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 we splice() 32768 more bytes from our receiving side to the pipe. At some point: pasta: epoll event on connected spliced TCP socket 49 (events: 0x00000004) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:489 Flow 0 (TCP connection (spliced)): ~OUT_WAIT_0 Flow 0 (TCP connection (spliced)): 1320 from read-side call Flow 0 (TCP connection (spliced)): 1320 from write-side call (passed 1320) the receiver is signalling to us that it's ready for more data (EPOLLOUT). We reset the OUT_WAIT flag, read 1320 more bytes from our receiving socket into the pipe, and that's what we write to the receiver, forgetting about the pending 97457 bytes we had, which the receiver might never get (not the same 97547 bytes: we'll actually send 1320 of those). This condition is rather hard to reproduce, and it was observed with Podman pulling container images via HTTPS. In the traces above, the client is side 0 (the initiating peer), and the server is sending the whole data. Instead of splicing from pipe to socket the amount of data we just read, we need to splice all the pending data we piled up until that point. We could do that using 'read' and 'written' counters, but there's actually no need, as the kernel also keeps track of how much data is available on the pipe. So, to make this simple and more robust, just give the whole pipe size as length to splice(). The kernel knows what to do with it. Later in the function, we used 'to_write' for an optimisation meant to reduce wakeups which retries right away to splice() in both directions if we couldn't write to the receiver the whole amount of pending data. Calculate a 'pending' value instead, only if we reach that point. Now that we check for the actual amount of pending data in that optimisation, we need to make sure we don't compare a zero or negative 'written' value: if we met that, it means that the receiver signalled end-of-file, an error, or to try again later. In those three cases, the optimisation doesn't make any sense, so skip it. Reported-by: Ed Santiago <santiago@redhat.com> Reported-by: Paul Holzinger <pholzing@redhat.com> Analysed-by: Paul Holzinger <pholzing@redhat.com> Link: https://github.com/containers/podman/issues/24219 Signed-off-by: Stefano Brivio <sbrivio@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
2024-10-24 09:12:11 +02:00
conn->s[!fromsidei], NULL, c->tcp.pipe_size,
SPLICE_F_MOVE | more | SPLICE_F_NONBLOCK);
flow_trace(conn, "%zi from write-side call (passed %zi)",
tcp_splice: splice() all we have to the writing side, not what we just read In tcp_splice_sock_handler(), we try to calculate how much we can move from the pipe to the writing socket: if we just read some bytes, we'll use that amount, but if we haven't, we just try to empty the pipe. However, if we just read something, that doesn't mean that that's all the data we have on the pipe, as it's obvious from this sequence, where: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 98304 from read-side call Flow 0 (TCP connection (spliced)): 33615 from write-side call (passed 98304) Flow 0 (TCP connection (spliced)): -1 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 524288) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 Flow 0 (TCP connection (spliced)): OUT_WAIT_0 we first pile up 98304 - 33615 = 64689 pending bytes, that we read but couldn't write, as the receiver buffer is full, and we set the corresponding OUT_WAIT flag. Then: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 32768 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 32768) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 we splice() 32768 more bytes from our receiving side to the pipe. At some point: pasta: epoll event on connected spliced TCP socket 49 (events: 0x00000004) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:489 Flow 0 (TCP connection (spliced)): ~OUT_WAIT_0 Flow 0 (TCP connection (spliced)): 1320 from read-side call Flow 0 (TCP connection (spliced)): 1320 from write-side call (passed 1320) the receiver is signalling to us that it's ready for more data (EPOLLOUT). We reset the OUT_WAIT flag, read 1320 more bytes from our receiving socket into the pipe, and that's what we write to the receiver, forgetting about the pending 97457 bytes we had, which the receiver might never get (not the same 97547 bytes: we'll actually send 1320 of those). This condition is rather hard to reproduce, and it was observed with Podman pulling container images via HTTPS. In the traces above, the client is side 0 (the initiating peer), and the server is sending the whole data. Instead of splicing from pipe to socket the amount of data we just read, we need to splice all the pending data we piled up until that point. We could do that using 'read' and 'written' counters, but there's actually no need, as the kernel also keeps track of how much data is available on the pipe. So, to make this simple and more robust, just give the whole pipe size as length to splice(). The kernel knows what to do with it. Later in the function, we used 'to_write' for an optimisation meant to reduce wakeups which retries right away to splice() in both directions if we couldn't write to the receiver the whole amount of pending data. Calculate a 'pending' value instead, only if we reach that point. Now that we check for the actual amount of pending data in that optimisation, we need to make sure we don't compare a zero or negative 'written' value: if we met that, it means that the receiver signalled end-of-file, an error, or to try again later. In those three cases, the optimisation doesn't make any sense, so skip it. Reported-by: Ed Santiago <santiago@redhat.com> Reported-by: Paul Holzinger <pholzing@redhat.com> Analysed-by: Paul Holzinger <pholzing@redhat.com> Link: https://github.com/containers/podman/issues/24219 Signed-off-by: Stefano Brivio <sbrivio@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
2024-10-24 09:12:11 +02:00
written, c->tcp.pipe_size);
/* Most common case: skip updating counters. */
if (readlen > 0 && readlen == written) {
if (readlen >= (long)c->tcp.pipe_size * 10 / 100)
continue;
if (conn->flags & lowat_set_flag &&
readlen > (long)c->tcp.pipe_size / 10) {
int lowat = c->tcp.pipe_size / 4;
if (setsockopt(conn->s[fromsidei], SOL_SOCKET,
SO_RCVLOWAT,
&lowat, sizeof(lowat))) {
flow_trace(conn,
"Setting SO_RCVLOWAT %i: %s",
lowat, strerror(errno));
} else {
conn_flag(c, conn, lowat_set_flag);
conn_flag(c, conn, lowat_act_flag);
}
}
break;
}
conn->read[fromsidei] += readlen > 0 ? readlen : 0;
conn->written[fromsidei] += written > 0 ? written : 0;
if (written < 0) {
if (errno == EINTR)
goto eintr;
if (errno != EAGAIN)
goto close;
if (conn->read[fromsidei] == conn->written[fromsidei])
break;
conn_event(c, conn, OUT_WAIT(!fromsidei));
break;
}
if (never_read && written == (long)(c->tcp.pipe_size))
goto retry;
tcp_splice: splice() all we have to the writing side, not what we just read In tcp_splice_sock_handler(), we try to calculate how much we can move from the pipe to the writing socket: if we just read some bytes, we'll use that amount, but if we haven't, we just try to empty the pipe. However, if we just read something, that doesn't mean that that's all the data we have on the pipe, as it's obvious from this sequence, where: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 98304 from read-side call Flow 0 (TCP connection (spliced)): 33615 from write-side call (passed 98304) Flow 0 (TCP connection (spliced)): -1 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 524288) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 Flow 0 (TCP connection (spliced)): OUT_WAIT_0 we first pile up 98304 - 33615 = 64689 pending bytes, that we read but couldn't write, as the receiver buffer is full, and we set the corresponding OUT_WAIT flag. Then: pasta: epoll event on connected spliced TCP socket 54 (events: 0x00000001) Flow 0 (TCP connection (spliced)): 32768 from read-side call Flow 0 (TCP connection (spliced)): -1 from write-side call (passed 32768) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:580 we splice() 32768 more bytes from our receiving side to the pipe. At some point: pasta: epoll event on connected spliced TCP socket 49 (events: 0x00000004) Flow 0 (TCP connection (spliced)): event at tcp_splice_sock_handler:489 Flow 0 (TCP connection (spliced)): ~OUT_WAIT_0 Flow 0 (TCP connection (spliced)): 1320 from read-side call Flow 0 (TCP connection (spliced)): 1320 from write-side call (passed 1320) the receiver is signalling to us that it's ready for more data (EPOLLOUT). We reset the OUT_WAIT flag, read 1320 more bytes from our receiving socket into the pipe, and that's what we write to the receiver, forgetting about the pending 97457 bytes we had, which the receiver might never get (not the same 97547 bytes: we'll actually send 1320 of those). This condition is rather hard to reproduce, and it was observed with Podman pulling container images via HTTPS. In the traces above, the client is side 0 (the initiating peer), and the server is sending the whole data. Instead of splicing from pipe to socket the amount of data we just read, we need to splice all the pending data we piled up until that point. We could do that using 'read' and 'written' counters, but there's actually no need, as the kernel also keeps track of how much data is available on the pipe. So, to make this simple and more robust, just give the whole pipe size as length to splice(). The kernel knows what to do with it. Later in the function, we used 'to_write' for an optimisation meant to reduce wakeups which retries right away to splice() in both directions if we couldn't write to the receiver the whole amount of pending data. Calculate a 'pending' value instead, only if we reach that point. Now that we check for the actual amount of pending data in that optimisation, we need to make sure we don't compare a zero or negative 'written' value: if we met that, it means that the receiver signalled end-of-file, an error, or to try again later. In those three cases, the optimisation doesn't make any sense, so skip it. Reported-by: Ed Santiago <santiago@redhat.com> Reported-by: Paul Holzinger <pholzing@redhat.com> Analysed-by: Paul Holzinger <pholzing@redhat.com> Link: https://github.com/containers/podman/issues/24219 Signed-off-by: Stefano Brivio <sbrivio@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
2024-10-24 09:12:11 +02:00
pending = conn->read[fromsidei] - conn->written[fromsidei];
if (!never_read && written > 0 && written < pending)
goto retry;
if (eof)
break;
}
if (conn->read[fromsidei] == conn->written[fromsidei] && eof) {
unsigned sidei;
flow_foreach_sidei(sidei) {
if ((conn->events & FIN_RCVD(sidei)) &&
!(conn->events & FIN_SENT(!sidei))) {
shutdown(conn->s[!sidei], SHUT_WR);
conn_event(c, conn, FIN_SENT(!sidei));
}
}
}
if (CONN_HAS(conn, FIN_SENT(0) | FIN_SENT(1)))
goto close;
if ((events & (EPOLLIN | EPOLLOUT)) == (EPOLLIN | EPOLLOUT)) {
events = EPOLLIN;
fromsidei = !fromsidei;
goto swap;
}
if (events & EPOLLHUP)
goto close;
return;
close:
conn_flag(c, conn, CLOSING);
}
/**
* tcp_set_pipe_size() - Set usable pipe size, probe starting from MAX_PIPE_SIZE
* @c: Execution context
*/
static void tcp_set_pipe_size(struct ctx *c)
{
int probe_pipe[TCP_SPLICE_PIPE_POOL_SIZE][2], i, j;
c->tcp.pipe_size = MAX_PIPE_SIZE;
smaller:
for (i = 0; i < TCP_SPLICE_PIPE_POOL_SIZE; i++) {
if (pipe2(probe_pipe[i], O_CLOEXEC)) {
i++;
break;
}
if (fcntl(probe_pipe[i][0], F_SETPIPE_SZ, c->tcp.pipe_size) < 0)
break;
}
for (j = i - 1; j >= 0; j--) {
close(probe_pipe[j][0]);
close(probe_pipe[j][1]);
}
if (i == TCP_SPLICE_PIPE_POOL_SIZE)
return;
if (!(c->tcp.pipe_size /= 2)) {
c->tcp.pipe_size = MAX_PIPE_SIZE;
return;
}
goto smaller;
}
/**
* tcp_splice_pipe_refill() - Refill pool of pre-opened pipes
* @c: Execution context
*/
static void tcp_splice_pipe_refill(const struct ctx *c)
{
int i;
for (i = 0; i < TCP_SPLICE_PIPE_POOL_SIZE; i++) {
if (splice_pipe_pool[i][0] >= 0)
break;
if (pipe2(splice_pipe_pool[i], O_NONBLOCK | O_CLOEXEC))
continue;
if (fcntl(splice_pipe_pool[i][0], F_SETPIPE_SZ,
c->tcp.pipe_size) != (int)c->tcp.pipe_size) {
trace("TCP (spliced): cannot set pool pipe size to %zu",
c->tcp.pipe_size);
}
}
}
/**
* tcp_sock_refill_ns() - Refill pools of pre-opened sockets in namespace
* @arg: Execution context cast to void *
*
* Return: 0
*/
/* cppcheck-suppress [constParameterCallback, unmatchedSuppression] */
static int tcp_sock_refill_ns(void *arg)
{
const struct ctx *c = (const struct ctx *)arg;
ns_enter(c);
if (c->ifi4) {
int rc = tcp_sock_refill_pool(c, ns_sock_pool4, AF_INET);
if (rc < 0)
warn("TCP: Error refilling IPv4 ns socket pool: %s",
strerror(-rc));
}
if (c->ifi6) {
int rc = tcp_sock_refill_pool(c, ns_sock_pool6, AF_INET6);
if (rc < 0)
warn("TCP: Error refilling IPv6 ns socket pool: %s",
strerror(-rc));
}
return 0;
}
/**
* tcp_splice_refill() - Refill pools of resources needed for splicing
* @c: Execution context
*/
void tcp_splice_refill(const struct ctx *c)
{
if ((c->ifi4 && ns_sock_pool4[TCP_SOCK_POOL_TSH] < 0) ||
(c->ifi6 && ns_sock_pool6[TCP_SOCK_POOL_TSH] < 0))
NS_CALL(tcp_sock_refill_ns, c);
tcp_splice_pipe_refill(c);
}
/**
* tcp_splice_init() - Initialise pipe pool and size
* @c: Execution context
*/
void tcp_splice_init(struct ctx *c)
{
memset(splice_pipe_pool, 0xff, sizeof(splice_pipe_pool));
tcp_set_pipe_size(c);
memset(&ns_sock_pool4, 0xff, sizeof(ns_sock_pool4));
memset(&ns_sock_pool6, 0xff, sizeof(ns_sock_pool6));
NS_CALL(tcp_sock_refill_ns, c);
}
/**
* tcp_splice_timer() - Timer for spliced connections
* @c: Execution context
* @conn: Connection to handle
*/
void tcp_splice_timer(const struct ctx *c, struct tcp_splice_conn *conn)
{
unsigned sidei;
ASSERT(!(conn->flags & CLOSING));
flow_foreach_sidei(sidei) {
if ((conn->flags & RCVLOWAT_SET(sidei)) &&
!(conn->flags & RCVLOWAT_ACT(sidei))) {
if (setsockopt(conn->s[sidei], SOL_SOCKET, SO_RCVLOWAT,
&((int){ 1 }), sizeof(int))) {
flow_trace(conn, "can't set SO_RCVLOWAT on %d",
conn->s[sidei]);
}
conn_flag(c, conn, ~RCVLOWAT_SET(sidei));
}
}
flow_foreach_sidei(sidei)
conn_flag(c, conn, ~RCVLOWAT_ACT(sidei));
}