freebsd-dev/sys/net/debugnet.c

1072 lines
26 KiB
C
Raw Normal View History

Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2019 Isilon Systems, LLC.
* Copyright (c) 2005-2014 Sandvine Incorporated. All rights reserved.
* Copyright (c) 2000 Darrell Anderson
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_ddb.h"
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
#include "opt_inet.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/endian.h>
#include <sys/errno.h>
#include <sys/eventhandler.h>
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
#include <sys/socket.h>
#include <sys/sysctl.h>
#ifdef DDB
#include <ddb/ddb.h>
#include <ddb/db_lex.h>
#endif
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
#include <net/ethernet.h>
#include <net/if.h>
#include <net/if_arp.h>
#include <net/if_dl.h>
#include <net/if_types.h>
#include <net/if_var.h>
#include <netinet/in.h>
#include <netinet/in_systm.h>
#include <netinet/in_var.h>
#include <netinet/ip.h>
#include <netinet/ip_var.h>
#include <netinet/ip_options.h>
#include <netinet/udp.h>
#include <netinet/udp_var.h>
#include <machine/in_cksum.h>
#include <machine/pcb.h>
#include <net/debugnet.h>
#define DEBUGNET_INTERNAL
#include <net/debugnet_int.h>
FEATURE(debugnet, "Debugnet support");
SYSCTL_NODE(_net, OID_AUTO, debugnet, CTLFLAG_RD, NULL,
"debugnet parameters");
unsigned debugnet_debug;
SYSCTL_UINT(_net_debugnet, OID_AUTO, debug, CTLFLAG_RWTUN,
&debugnet_debug, 0,
"Debug message verbosity (0: off; 1: on; 2: verbose)");
int debugnet_npolls = 2000;
SYSCTL_INT(_net_debugnet, OID_AUTO, npolls, CTLFLAG_RWTUN,
&debugnet_npolls, 0,
"Number of times to poll before assuming packet loss (0.5ms per poll)");
int debugnet_nretries = 10;
SYSCTL_INT(_net_debugnet, OID_AUTO, nretries, CTLFLAG_RWTUN,
&debugnet_nretries, 0,
"Number of retransmit attempts before giving up");
static bool g_debugnet_pcb_inuse;
static struct debugnet_pcb g_dnet_pcb;
/*
* Simple accessors for opaque PCB.
*/
const unsigned char *
debugnet_get_gw_mac(const struct debugnet_pcb *pcb)
{
MPASS(g_debugnet_pcb_inuse && pcb == &g_dnet_pcb &&
pcb->dp_state >= DN_STATE_HAVE_GW_MAC);
return (pcb->dp_gw_mac.octet);
}
/*
* Start of network primitives, beginning with output primitives.
*/
/*
* Handles creation of the ethernet header, then places outgoing packets into
* the tx buffer for the NIC
*
* Parameters:
* m The mbuf containing the packet to be sent (will be freed by
* this function or the NIC driver)
* ifp The interface to send on
* dst The destination ethernet address (source address will be looked
* up using ifp)
* etype The ETHERTYPE_* value for the protocol that is being sent
*
* Returns:
* int see errno.h, 0 for success
*/
int
debugnet_ether_output(struct mbuf *m, struct ifnet *ifp, struct ether_addr dst,
u_short etype)
{
struct ether_header *eh;
if (((ifp->if_flags & (IFF_MONITOR | IFF_UP)) != IFF_UP) ||
(ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING) {
if_printf(ifp, "%s: interface isn't up\n", __func__);
m_freem(m);
return (ENETDOWN);
}
/* Fill in the ethernet header. */
M_PREPEND(m, ETHER_HDR_LEN, M_NOWAIT);
if (m == NULL) {
printf("%s: out of mbufs\n", __func__);
return (ENOBUFS);
}
eh = mtod(m, struct ether_header *);
memcpy(eh->ether_shost, IF_LLADDR(ifp), ETHER_ADDR_LEN);
memcpy(eh->ether_dhost, dst.octet, ETHER_ADDR_LEN);
eh->ether_type = htons(etype);
return (ifp->if_debugnet_methods->dn_transmit(ifp, m));
}
/*
* Unreliable transmission of an mbuf chain to the debugnet server
* Note: can't handle fragmentation; fails if the packet is larger than
* ifp->if_mtu after adding the UDP/IP headers
*
* Parameters:
* pcb The debugnet context block
* m mbuf chain
*
* Returns:
* int see errno.h, 0 for success
*/
static int
debugnet_udp_output(struct debugnet_pcb *pcb, struct mbuf *m)
{
struct udphdr *udp;
MPASS(pcb->dp_state >= DN_STATE_HAVE_GW_MAC);
M_PREPEND(m, sizeof(*udp), M_NOWAIT);
if (m == NULL) {
printf("%s: out of mbufs\n", __func__);
return (ENOBUFS);
}
udp = mtod(m, void *);
udp->uh_ulen = htons(m->m_pkthdr.len);
/* Use this src port so that the server can connect() the socket */
udp->uh_sport = htons(pcb->dp_client_port);
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
udp->uh_dport = htons(pcb->dp_server_port);
/* Computed later (protocol-dependent). */
udp->uh_sum = 0;
return (debugnet_ip_output(pcb, m));
}
int
debugnet_ack_output(struct debugnet_pcb *pcb, uint32_t seqno /* net endian */)
{
struct debugnet_ack *dn_ack;
struct mbuf *m;
DNETDEBUG("Acking with seqno %u\n", ntohl(seqno));
m = m_gethdr(M_NOWAIT, MT_DATA);
if (m == NULL) {
printf("%s: Out of mbufs\n", __func__);
return (ENOBUFS);
}
m->m_len = sizeof(*dn_ack);
m->m_pkthdr.len = sizeof(*dn_ack);
MH_ALIGN(m, sizeof(*dn_ack));
dn_ack = mtod(m, void *);
dn_ack->da_seqno = seqno;
return (debugnet_udp_output(pcb, m));
}
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
/*
* Dummy free function for debugnet clusters.
*/
static void
debugnet_mbuf_free(struct mbuf *m __unused)
{
}
/*
* Construct and reliably send a debugnet packet. May fail from a resource
* shortage or extreme number of unacknowledged retransmissions. Wait for
* an acknowledgement before returning. Splits packets into chunks small
* enough to be sent without fragmentation (looks up the interface MTU)
*
* Parameters:
* type debugnet packet type (HERALD, FINISHED, ...)
* data data
* datalen data size (bytes)
* auxdata optional auxiliary information
*
* Returns:
* int see errno.h, 0 for success
*/
int
debugnet_send(struct debugnet_pcb *pcb, uint32_t type, const void *data,
uint32_t datalen, const struct debugnet_proto_aux *auxdata)
{
struct debugnet_msg_hdr *dn_msg_hdr;
struct mbuf *m, *m2;
uint64_t want_acks;
uint32_t i, pktlen, sent_so_far;
int retries, polls, error;
if (pcb->dp_state == DN_STATE_REMOTE_CLOSED)
return (ECONNRESET);
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
want_acks = 0;
pcb->dp_rcvd_acks = 0;
retries = 0;
retransmit:
/* Chunks can be too big to fit in packets. */
for (i = sent_so_far = 0; sent_so_far < datalen ||
(i == 0 && datalen == 0); i++) {
pktlen = datalen - sent_so_far;
/* Bound: the interface MTU (assume no IP options). */
pktlen = min(pktlen, pcb->dp_ifp->if_mtu -
sizeof(struct udpiphdr) - sizeof(struct debugnet_msg_hdr));
/*
* Check if it is retransmitting and this has been ACKed
* already.
*/
if ((pcb->dp_rcvd_acks & (1 << i)) != 0) {
sent_so_far += pktlen;
continue;
}
/*
* Get and fill a header mbuf, then chain data as an extended
* mbuf.
*/
m = m_gethdr(M_NOWAIT, MT_DATA);
if (m == NULL) {
printf("%s: Out of mbufs\n", __func__);
return (ENOBUFS);
}
m->m_len = sizeof(struct debugnet_msg_hdr);
m->m_pkthdr.len = sizeof(struct debugnet_msg_hdr);
MH_ALIGN(m, sizeof(struct debugnet_msg_hdr));
dn_msg_hdr = mtod(m, struct debugnet_msg_hdr *);
dn_msg_hdr->mh_seqno = htonl(pcb->dp_seqno + i);
dn_msg_hdr->mh_type = htonl(type);
dn_msg_hdr->mh_len = htonl(pktlen);
if (auxdata != NULL) {
dn_msg_hdr->mh_offset =
htobe64(auxdata->dp_offset_start + sent_so_far);
dn_msg_hdr->mh_aux2 = htobe32(auxdata->dp_aux2);
} else {
dn_msg_hdr->mh_offset = htobe64(sent_so_far);
dn_msg_hdr->mh_aux2 = 0;
}
if (pktlen != 0) {
m2 = m_get(M_NOWAIT, MT_DATA);
if (m2 == NULL) {
m_freem(m);
printf("%s: Out of mbufs\n", __func__);
return (ENOBUFS);
}
MEXTADD(m2, __DECONST(char *, data) + sent_so_far,
pktlen, debugnet_mbuf_free, NULL, NULL, 0,
EXT_DISPOSABLE);
m2->m_len = pktlen;
m_cat(m, m2);
m->m_pkthdr.len += pktlen;
}
error = debugnet_udp_output(pcb, m);
if (error != 0)
return (error);
/* Note that we're waiting for this packet in the bitfield. */
want_acks |= (1 << i);
sent_so_far += pktlen;
}
if (i >= DEBUGNET_MAX_IN_FLIGHT)
printf("Warning: Sent more than %d packets (%d). "
"Acknowledgements will fail unless the size of "
"rcvd_acks/want_acks is increased.\n",
DEBUGNET_MAX_IN_FLIGHT, i);
/*
* Wait for acks. A *real* window would speed things up considerably.
*/
polls = 0;
while (pcb->dp_rcvd_acks != want_acks) {
if (polls++ > debugnet_npolls) {
if (retries++ > debugnet_nretries)
return (ETIMEDOUT);
printf(". ");
goto retransmit;
}
debugnet_network_poll(pcb);
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
DELAY(500);
if (pcb->dp_state == DN_STATE_REMOTE_CLOSED)
return (ECONNRESET);
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
}
pcb->dp_seqno += i;
return (0);
}
/*
* Network input primitives.
*/
/*
* Just introspect the header enough to fire off a seqno ack and validate
* length fits.
*/
static void
debugnet_handle_rx_msg(struct debugnet_pcb *pcb, struct mbuf **mb)
{
const struct debugnet_msg_hdr *dnh;
struct mbuf *m;
int error;
m = *mb;
if (m->m_pkthdr.len < sizeof(*dnh)) {
DNETDEBUG("ignoring small debugnet_msg packet\n");
return;
}
/* Get ND header. */
if (m->m_len < sizeof(*dnh)) {
m = m_pullup(m, sizeof(*dnh));
*mb = m;
if (m == NULL) {
DNETDEBUG("m_pullup failed\n");
return;
}
}
dnh = mtod(m, const void *);
if (ntohl(dnh->mh_len) + sizeof(*dnh) > m->m_pkthdr.len) {
DNETDEBUG("Dropping short packet.\n");
return;
}
/*
* If the issue is transient (ENOBUFS), sender should resend. If
* non-transient (like driver objecting to rx -> tx from the same
* thread), not much else we can do.
*/
error = debugnet_ack_output(pcb, dnh->mh_seqno);
if (error != 0)
return;
if (ntohl(dnh->mh_type) == DEBUGNET_FINISHED) {
printf("Remote shut down the connection on us!\n");
pcb->dp_state = DN_STATE_REMOTE_CLOSED;
/*
* Continue through to the user handler so they are signalled
* not to wait for further rx.
*/
}
pcb->dp_rx_handler(pcb, mb);
}
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
static void
debugnet_handle_ack(struct debugnet_pcb *pcb, struct mbuf **mb, uint16_t sport)
{
const struct debugnet_ack *dn_ack;
struct mbuf *m;
uint32_t rcv_ackno;
m = *mb;
/* Get Ack. */
if (m->m_len < sizeof(*dn_ack)) {
m = m_pullup(m, sizeof(*dn_ack));
*mb = m;
if (m == NULL) {
DNETDEBUG("m_pullup failed\n");
return;
}
}
dn_ack = mtod(m, const void *);
/* Debugnet processing. */
/*
* Packet is meant for us. Extract the ack sequence number and the
* port number if necessary.
*/
rcv_ackno = ntohl(dn_ack->da_seqno);
if (pcb->dp_state < DN_STATE_GOT_HERALD_PORT) {
pcb->dp_server_port = sport;
pcb->dp_state = DN_STATE_GOT_HERALD_PORT;
}
if (rcv_ackno >= pcb->dp_seqno + DEBUGNET_MAX_IN_FLIGHT)
printf("%s: ACK %u too far in future!\n", __func__, rcv_ackno);
else if (rcv_ackno >= pcb->dp_seqno) {
/* We're interested in this ack. Record it. */
pcb->dp_rcvd_acks |= 1 << (rcv_ackno - pcb->dp_seqno);
}
}
void
debugnet_handle_udp(struct debugnet_pcb *pcb, struct mbuf **mb)
{
const struct udphdr *udp;
struct mbuf *m;
uint16_t sport, ulen;
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
/* UDP processing. */
m = *mb;
if (m->m_pkthdr.len < sizeof(*udp)) {
DNETDEBUG("ignoring small UDP packet\n");
return;
}
/* Get UDP headers. */
if (m->m_len < sizeof(*udp)) {
m = m_pullup(m, sizeof(*udp));
*mb = m;
if (m == NULL) {
DNETDEBUG("m_pullup failed\n");
return;
}
}
udp = mtod(m, const void *);
/* We expect to receive UDP packets on the configured client port. */
if (ntohs(udp->uh_dport) != pcb->dp_client_port) {
DNETDEBUG("not on the expected port.\n");
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
return;
}
/* Check that ulen does not exceed actual size of data. */
ulen = ntohs(udp->uh_ulen);
if (m->m_pkthdr.len < ulen) {
DNETDEBUG("ignoring runt UDP packet\n");
return;
}
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
sport = ntohs(udp->uh_sport);
m_adj(m, sizeof(*udp));
ulen -= sizeof(*udp);
if (ulen == sizeof(struct debugnet_ack)) {
debugnet_handle_ack(pcb, mb, sport);
return;
}
if (pcb->dp_rx_handler == NULL) {
if (ulen < sizeof(struct debugnet_ack))
DNETDEBUG("ignoring small ACK packet\n");
else
DNETDEBUG("ignoring unexpected non-ACK packet on "
"half-duplex connection.\n");
return;
}
debugnet_handle_rx_msg(pcb, mb);
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
}
/*
* Handler for incoming packets directly from the network adapter
* Identifies the packet type (IP or ARP) and passes it along to one of the
* helper functions debugnet_handle_ip or debugnet_handle_arp.
*
* It needs to partially replicate the behaviour of ether_input() and
* ether_demux().
*
* Parameters:
* ifp the interface the packet came from
* m an mbuf containing the packet received
*/
static void
debugnet_pkt_in(struct ifnet *ifp, struct mbuf *m)
{
struct ifreq ifr;
struct ether_header *eh;
u_short etype;
/* Ethernet processing. */
if ((m->m_flags & M_PKTHDR) == 0) {
DNETDEBUG_IF(ifp, "discard frame without packet header\n");
goto done;
}
if (m->m_len < ETHER_HDR_LEN) {
DNETDEBUG_IF(ifp,
"discard frame without leading eth header (len %u pktlen %u)\n",
m->m_len, m->m_pkthdr.len);
goto done;
}
if ((m->m_flags & M_HASFCS) != 0) {
m_adj(m, -ETHER_CRC_LEN);
m->m_flags &= ~M_HASFCS;
}
eh = mtod(m, struct ether_header *);
etype = ntohs(eh->ether_type);
if ((m->m_flags & M_VLANTAG) != 0 || etype == ETHERTYPE_VLAN) {
DNETDEBUG_IF(ifp, "ignoring vlan packets\n");
goto done;
}
if (if_gethwaddr(ifp, &ifr) != 0) {
DNETDEBUG_IF(ifp, "failed to get hw addr for interface\n");
goto done;
}
if (memcmp(ifr.ifr_addr.sa_data, eh->ether_dhost,
ETHER_ADDR_LEN) != 0 &&
(etype != ETHERTYPE_ARP || !ETHER_IS_BROADCAST(eh->ether_dhost))) {
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
DNETDEBUG_IF(ifp,
"discard frame with incorrect destination addr\n");
goto done;
}
MPASS(g_debugnet_pcb_inuse);
/* Done ethernet processing. Strip off the ethernet header. */
m_adj(m, ETHER_HDR_LEN);
switch (etype) {
case ETHERTYPE_ARP:
debugnet_handle_arp(&g_dnet_pcb, &m);
break;
case ETHERTYPE_IP:
debugnet_handle_ip(&g_dnet_pcb, &m);
break;
default:
DNETDEBUG_IF(ifp, "dropping unknown ethertype %hu\n", etype);
break;
}
done:
if (m != NULL)
m_freem(m);
}
/*
* Network polling primitive.
*
* Instead of assuming that most of the network stack is sane, we just poll the
* driver directly for packets.
*/
void
debugnet_network_poll(struct debugnet_pcb *pcb)
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
{
struct ifnet *ifp;
ifp = pcb->dp_ifp;
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
ifp->if_debugnet_methods->dn_poll(ifp, 1000);
}
/*
* Start of consumer API surface.
*/
void
debugnet_free(struct debugnet_pcb *pcb)
{
struct ifnet *ifp;
MPASS(g_debugnet_pcb_inuse);
MPASS(pcb == &g_dnet_pcb);
ifp = pcb->dp_ifp;
if (ifp != NULL) {
if (pcb->dp_drv_input != NULL)
ifp->if_input = pcb->dp_drv_input;
if (pcb->dp_event_started)
ifp->if_debugnet_methods->dn_event(ifp, DEBUGNET_END);
}
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
debugnet_mbuf_finish();
g_debugnet_pcb_inuse = false;
memset(&g_dnet_pcb, 0xfd, sizeof(g_dnet_pcb));
}
int
debugnet_connect(const struct debugnet_conn_params *dcp,
struct debugnet_pcb **pcb_out)
{
struct debugnet_proto_aux herald_auxdata;
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
struct debugnet_pcb *pcb;
struct ifnet *ifp;
int error;
if (g_debugnet_pcb_inuse) {
printf("%s: Only one connection at a time.\n", __func__);
return (EBUSY);
}
pcb = &g_dnet_pcb;
*pcb = (struct debugnet_pcb) {
.dp_state = DN_STATE_INIT,
.dp_client = dcp->dc_client,
.dp_server = dcp->dc_server,
.dp_gateway = dcp->dc_gateway,
.dp_server_port = dcp->dc_herald_port, /* Initially */
.dp_client_port = dcp->dc_client_port,
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
.dp_seqno = 1,
.dp_ifp = dcp->dc_ifp,
.dp_rx_handler = dcp->dc_rx_handler,
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
};
/* Switch to the debugnet mbuf zones. */
debugnet_mbuf_start();
/* At least one needed parameter is missing; infer it. */
if (pcb->dp_client == INADDR_ANY || pcb->dp_gateway == INADDR_ANY ||
pcb->dp_ifp == NULL) {
struct sockaddr_in dest_sin, *gw_sin, *local_sin;
struct rtentry *dest_rt;
struct ifnet *rt_ifp;
memset(&dest_sin, 0, sizeof(dest_sin));
dest_sin = (struct sockaddr_in) {
.sin_len = sizeof(dest_sin),
.sin_family = AF_INET,
.sin_addr.s_addr = pcb->dp_server,
};
CURVNET_SET(vnet0);
dest_rt = rtalloc1((struct sockaddr *)&dest_sin, 0,
RTF_RNH_LOCKED);
CURVNET_RESTORE();
if (dest_rt == NULL) {
printf("%s: Could not get route for that server.\n",
__func__);
error = ENOENT;
goto cleanup;
}
if (dest_rt->rt_gateway->sa_family == AF_INET)
gw_sin = (struct sockaddr_in *)dest_rt->rt_gateway;
else {
if (dest_rt->rt_gateway->sa_family == AF_LINK)
DNETDEBUG("Destination address is on link.\n");
gw_sin = NULL;
}
MPASS(dest_rt->rt_ifa->ifa_addr->sa_family == AF_INET);
local_sin = (struct sockaddr_in *)dest_rt->rt_ifa->ifa_addr;
rt_ifp = dest_rt->rt_ifp;
if (pcb->dp_client == INADDR_ANY)
pcb->dp_client = local_sin->sin_addr.s_addr;
if (pcb->dp_gateway == INADDR_ANY && gw_sin != NULL)
pcb->dp_gateway = gw_sin->sin_addr.s_addr;
if (pcb->dp_ifp == NULL)
pcb->dp_ifp = rt_ifp;
RTFREE_LOCKED(dest_rt);
}
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
ifp = pcb->dp_ifp;
if (debugnet_debug > 0) {
char serbuf[INET_ADDRSTRLEN], clibuf[INET_ADDRSTRLEN],
gwbuf[INET_ADDRSTRLEN];
inet_ntop(AF_INET, &pcb->dp_server, serbuf, sizeof(serbuf));
inet_ntop(AF_INET, &pcb->dp_client, clibuf, sizeof(clibuf));
if (pcb->dp_gateway != INADDR_ANY)
inet_ntop(AF_INET, &pcb->dp_gateway, gwbuf, sizeof(gwbuf));
DNETDEBUG("Connecting to %s:%d%s%s from %s:%d on %s\n",
serbuf, pcb->dp_server_port,
(pcb->dp_gateway == INADDR_ANY) ? "" : " via ",
(pcb->dp_gateway == INADDR_ANY) ? "" : gwbuf,
clibuf, pcb->dp_client_port, if_name(ifp));
}
/* Validate iface is online and supported. */
if (!DEBUGNET_SUPPORTED_NIC(ifp)) {
printf("%s: interface '%s' does not support debugnet\n",
__func__, if_name(ifp));
error = ENODEV;
goto cleanup;
}
if ((if_getflags(ifp) & IFF_UP) == 0) {
printf("%s: interface '%s' link is down\n", __func__,
if_name(ifp));
error = ENXIO;
goto cleanup;
}
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
ifp->if_debugnet_methods->dn_event(ifp, DEBUGNET_START);
pcb->dp_event_started = true;
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
/*
* We maintain the invariant that g_debugnet_pcb_inuse is always true
* while the debugnet ifp's if_input is overridden with
* debugnet_pkt_in.
*/
g_debugnet_pcb_inuse = true;
/* Make the card use *our* receive callback. */
pcb->dp_drv_input = ifp->if_input;
ifp->if_input = debugnet_pkt_in;
printf("%s: searching for %s MAC...\n", __func__,
(dcp->dc_gateway == INADDR_ANY) ? "server" : "gateway");
error = debugnet_arp_gw(pcb);
if (error != 0) {
printf("%s: failed to locate MAC address\n", __func__);
goto cleanup;
}
MPASS(pcb->dp_state == DN_STATE_HAVE_GW_MAC);
herald_auxdata = (struct debugnet_proto_aux) {
.dp_offset_start = dcp->dc_herald_offset,
.dp_aux2 = dcp->dc_herald_aux2,
};
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
error = debugnet_send(pcb, DEBUGNET_HERALD, dcp->dc_herald_data,
dcp->dc_herald_datalen, &herald_auxdata);
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
if (error != 0) {
printf("%s: failed to herald debugnet server\n", __func__);
goto cleanup;
}
*pcb_out = pcb;
return (0);
cleanup:
debugnet_free(pcb);
return (error);
}
/*
* Pre-allocated dump-time mbuf tracking.
*
* We just track the high water mark we've ever seen and allocate appropriately
* for that iface/mtu combo.
*/
static struct {
int nmbuf;
int ncl;
int clsize;
} dn_hwm;
static struct mtx dn_hwm_lk;
MTX_SYSINIT(debugnet_hwm_lock, &dn_hwm_lk, "Debugnet HWM lock", MTX_DEF);
static void
dn_maybe_reinit_mbufs(int nmbuf, int ncl, int clsize)
{
bool any;
any = false;
mtx_lock(&dn_hwm_lk);
if (nmbuf > dn_hwm.nmbuf) {
any = true;
dn_hwm.nmbuf = nmbuf;
} else
nmbuf = dn_hwm.nmbuf;
if (ncl > dn_hwm.ncl) {
any = true;
dn_hwm.ncl = ncl;
} else
ncl = dn_hwm.ncl;
if (clsize > dn_hwm.clsize) {
any = true;
dn_hwm.clsize = clsize;
} else
clsize = dn_hwm.clsize;
mtx_unlock(&dn_hwm_lk);
if (any)
debugnet_mbuf_reinit(nmbuf, ncl, clsize);
}
void
debugnet_any_ifnet_update(struct ifnet *ifp)
{
int clsize, nmbuf, ncl, nrxr;
if (!DEBUGNET_SUPPORTED_NIC(ifp))
return;
ifp->if_debugnet_methods->dn_init(ifp, &nrxr, &ncl, &clsize);
KASSERT(nrxr > 0, ("invalid receive ring count %d", nrxr));
/*
* We need two headers per message on the transmit side. Multiply by
* four to give us some breathing room.
*/
nmbuf = ncl * (4 + nrxr);
ncl *= nrxr;
/*
* Bandaid for drivers that (incorrectly) advertise LinkUp before their
* dn_init method is available.
*/
if (nmbuf == 0 || ncl == 0 || clsize == 0) {
printf("%s: Bad dn_init result from %s (ifp %p), ignoring.\n",
__func__, if_name(ifp), ifp);
return;
}
Split out a more generic debugnet(4) from netdump(4) Debugnet is a simplistic and specialized panic- or debug-time reliable datagram transport. It can drive a single connection at a time and is currently unidirectional (debug/panic machine transmit to remote server only). It is mostly a verbatim code lift from netdump(4). Netdump(4) remains the only consumer (until the rest of this patch series lands). The INET-specific logic has been extracted somewhat more thoroughly than previously in netdump(4), into debugnet_inet.c. UDP-layer logic and up, as much as possible as is protocol-independent, remains in debugnet.c. The separation is not perfect and future improvement is welcome. Supporting INET6 is a long-term goal. Much of the diff is "gratuitous" renaming from 'netdump_' or 'nd_' to 'debugnet_' or 'dn_' -- sorry. I thought keeping the netdump name on the generic module would be more confusing than the refactoring. The only functional change here is the mbuf allocation / tracking. Instead of initiating solely on netdump-configured interface(s) at dumpon(8) configuration time, we watch for any debugnet-enabled NIC for link activation and query it for mbuf parameters at that time. If they exceed the existing high-water mark allocation, we re-allocate and track the new high-water mark. Otherwise, we leave the pre-panic mbuf allocation alone. In a future patch in this series, this will allow initiating netdump from panic ddb(4) without pre-panic configuration. No other functional change intended. Reviewed by: markj (earlier version) Some discussion with: emaste, jhb Objection from: marius Differential Revision: https://reviews.freebsd.org/D21421
2019-10-17 16:23:03 +00:00
dn_maybe_reinit_mbufs(nmbuf, ncl, clsize);
}
/*
* Unfortunately, the ifnet_arrival_event eventhandler hook is mostly useless
* for us because drivers tend to if_attach before invoking DEBUGNET_SET().
*
* On the other hand, hooking DEBUGNET_SET() itself may still be too early,
* because the driver is still in attach. Since we cannot use down interfaces,
* maybe hooking ifnet_event:IFNET_EVENT_UP is sufficient? ... Nope, at least
* with vtnet and dhcpclient that event just never occurs.
*
* So that's how I've landed on the lower level ifnet_link_event.
*/
static void
dn_ifnet_event(void *arg __unused, struct ifnet *ifp, int link_state)
{
if (link_state == LINK_STATE_UP)
debugnet_any_ifnet_update(ifp);
}
static eventhandler_tag dn_attach_cookie;
static void
dn_evh_init(void *ctx __unused)
{
dn_attach_cookie = EVENTHANDLER_REGISTER(ifnet_link_event,
dn_ifnet_event, NULL, EVENTHANDLER_PRI_ANY);
}
SYSINIT(dn_evh_init, SI_SUB_EVENTHANDLER + 1, SI_ORDER_ANY, dn_evh_init, NULL);
/*
* DDB parsing helpers for debugnet(4) consumers.
*/
#ifdef DDB
struct my_inet_opt {
bool has_opt;
const char *printname;
in_addr_t *result;
};
static int
dn_parse_optarg_ipv4(struct my_inet_opt *opt)
{
in_addr_t tmp;
unsigned octet;
int t;
tmp = 0;
for (octet = 0; octet < 4; octet++) {
t = db_read_token_flags(DRT_WSPACE | DRT_DECIMAL);
if (t != tNUMBER) {
db_printf("%s:%s: octet %u expected number; found %d\n",
__func__, opt->printname, octet, t);
return (EINVAL);
}
/*
* db_lex lexes '-' distinctly from the number itself, but
* let's document that invariant.
*/
MPASS(db_tok_number >= 0);
if (db_tok_number > UINT8_MAX) {
db_printf("%s:%s: octet %u out of range: %jd\n", __func__,
opt->printname, octet, (intmax_t)db_tok_number);
return (EDOM);
}
/* Constructed host-endian and converted to network later. */
tmp = (tmp << 8) | db_tok_number;
if (octet < 3) {
t = db_read_token_flags(DRT_WSPACE);
if (t != tDOT) {
db_printf("%s:%s: octet %u expected '.'; found"
" %d\n", __func__, opt->printname, octet,
t);
return (EINVAL);
}
}
}
*opt->result = htonl(tmp);
opt->has_opt = true;
return (0);
}
int
debugnet_parse_ddb_cmd(const char *cmd, struct debugnet_ddb_config *result)
{
struct ifnet *ifp;
int t, error;
bool want_ifp;
char ch;
struct my_inet_opt opt_client = {
.printname = "client",
.result = &result->dd_client,
},
opt_server = {
.printname = "server",
.result = &result->dd_server,
},
opt_gateway = {
.printname = "gateway",
.result = &result->dd_gateway,
},
*cur_inet_opt;
ifp = NULL;
memset(result, 0, sizeof(*result));
/*
* command [space] [-] [opt] [[space] [optarg]] ...
*
* db_command has already lexed 'command' for us.
*/
t = db_read_token_flags(DRT_WSPACE);
if (t == tWSPACE)
t = db_read_token_flags(DRT_WSPACE);
while (t != tEOL) {
if (t != tMINUS) {
db_printf("%s: Bad syntax; expected '-', got %d\n",
cmd, t);
goto usage;
}
t = db_read_token_flags(DRT_WSPACE);
if (t != tIDENT) {
db_printf("%s: Bad syntax; expected tIDENT, got %d\n",
cmd, t);
goto usage;
}
if (strlen(db_tok_string) > 1) {
db_printf("%s: Bad syntax; expected single option "
"flag, got '%s'\n", cmd, db_tok_string);
goto usage;
}
want_ifp = false;
cur_inet_opt = NULL;
switch ((ch = db_tok_string[0])) {
default:
DNETDEBUG("Unexpected: '%c'\n", ch);
/* FALLTHROUGH */
case 'h':
goto usage;
case 'c':
cur_inet_opt = &opt_client;
break;
case 'g':
cur_inet_opt = &opt_gateway;
break;
case 's':
cur_inet_opt = &opt_server;
break;
case 'i':
want_ifp = true;
break;
}
t = db_read_token_flags(DRT_WSPACE);
if (t != tWSPACE) {
db_printf("%s: Bad syntax; expected space after "
"flag %c, got %d\n", cmd, ch, t);
goto usage;
}
if (want_ifp) {
t = db_read_token_flags(DRT_WSPACE);
if (t != tIDENT) {
db_printf("%s: Expected interface but got %d\n",
cmd, t);
goto usage;
}
CURVNET_SET(vnet0);
/*
* We *don't* take a ref here because the only current
* consumer, db_netdump_cmd, does not need it. It
* (somewhat redundantly) extracts the if_name(),
* re-lookups the ifp, and takes its own reference.
*/
ifp = ifunit(db_tok_string);
CURVNET_RESTORE();
if (ifp == NULL) {
db_printf("Could not locate interface %s\n",
db_tok_string);
goto cleanup;
}
} else {
MPASS(cur_inet_opt != NULL);
/* Assume IPv4 for now. */
error = dn_parse_optarg_ipv4(cur_inet_opt);
if (error != 0)
goto cleanup;
}
/* Skip (mandatory) whitespace after option, if not EOL. */
t = db_read_token_flags(DRT_WSPACE);
if (t == tEOL)
break;
if (t != tWSPACE) {
db_printf("%s: Bad syntax; expected space after "
"flag %c option; got %d\n", cmd, ch, t);
goto usage;
}
t = db_read_token_flags(DRT_WSPACE);
}
if (!opt_server.has_opt) {
db_printf("%s: need a destination server address\n", cmd);
goto usage;
}
result->dd_has_client = opt_client.has_opt;
result->dd_has_gateway = opt_gateway.has_opt;
result->dd_ifp = ifp;
/* We parsed the full line to tEOL already, or bailed with an error. */
return (0);
usage:
db_printf("Usage: %s -s <server> [-g <gateway> -c <localip> "
"-i <interface>]\n", cmd);
error = EINVAL;
/* FALLTHROUGH */
cleanup:
db_skip_to_eol();
return (error);
}
#endif /* DDB */