mir/ovs
2
0
Fork 0
mirror of https://github.com/openvswitch/ovs synced 2025-09-01 14:55:18 +00:00
Code Issues Releases Wiki Activity

netlink-socket: Make caller provide message receive buffers.

Browse Source 
Typically an nl_sock client can stack-allocate the buffer for receiving
a Netlink message, which provides a performance boost.

Signed-off-by: Ben Pfaff <blp@nicira.com>
This commit is contained in:
Ben Pfaff
2012-04-09 15:35:29 -07:00
parent 8843668ae5
commit 72d32ac0b3
6 changed files with 210 additions and 172 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
lib/dpif-linux.c
View File

@@ -823,8 +823,12 @@ dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
struct op_auxdata { struct op_auxdata {
struct nl_transaction txn; struct nl_transaction txn;
struct ofpbuf request; struct ofpbuf request;
uint64_t request_stub[1024 / 8]; uint64_t request_stub[1024 / 8];
struct ofpbuf reply;
uint64_t reply_stub[1024 / 8];
} auxes[MAX_OPS]; } auxes[MAX_OPS];
struct nl_transaction *txnsp[MAX_OPS]; struct nl_transaction *txnsp[MAX_OPS];
@@ -843,12 +847,16 @@ dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
aux->request_stub, sizeof aux->request_stub); aux->request_stub, sizeof aux->request_stub);
aux->txn.request = &aux->request; aux->txn.request = &aux->request;
ofpbuf_use_stub(&aux->reply, aux->reply_stub, sizeof aux->reply_stub);
aux->txn.reply = NULL;
switch (op->type) { switch (op->type) {
case DPIF_OP_FLOW_PUT: case DPIF_OP_FLOW_PUT:
put = &op->u.flow_put; put = &op->u.flow_put;
dpif_linux_init_flow_put(dpif_, put, &flow); dpif_linux_init_flow_put(dpif_, put, &flow);
if (put->stats) { if (put->stats) {
flow.nlmsg_flags |= NLM_F_ECHO; flow.nlmsg_flags |= NLM_F_ECHO;
aux->txn.reply = &aux->reply;
} }
dpif_linux_flow_to_ofpbuf(&flow, &aux->request); dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
break; break;
@@ -858,6 +866,7 @@ dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
dpif_linux_init_flow_del(dpif_, del, &flow); dpif_linux_init_flow_del(dpif_, del, &flow);
if (del->stats) { if (del->stats) {
flow.nlmsg_flags |= NLM_F_ECHO; flow.nlmsg_flags |= NLM_F_ECHO;
aux->txn.reply = &aux->reply;
} }
dpif_linux_flow_to_ofpbuf(&flow, &aux->request); dpif_linux_flow_to_ofpbuf(&flow, &aux->request);
break; break;
@@ -879,6 +888,7 @@ dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
nl_sock_transact_multiple(genl_sock, txnsp, n_ops); nl_sock_transact_multiple(genl_sock, txnsp, n_ops);
for (i = 0; i < n_ops; i++) { for (i = 0; i < n_ops; i++) {
struct op_auxdata *aux = &auxes[i];
struct nl_transaction *txn = &auxes[i].txn; struct nl_transaction *txn = &auxes[i].txn;
struct dpif_op *op = ops[i]; struct dpif_op *op = ops[i];
struct dpif_flow_put *put; struct dpif_flow_put *put;
@@ -918,8 +928,8 @@ dpif_linux_operate__(struct dpif *dpif_, struct dpif_op **ops, size_t n_ops)
NOT_REACHED(); NOT_REACHED();
} }
ofpbuf_uninit(txn->request); ofpbuf_uninit(&aux->request);
ofpbuf_delete(txn->reply); ofpbuf_uninit(&aux->reply);
} }
} }
@@ -1121,10 +1131,13 @@ dpif_linux_recv(struct dpif *dpif_, struct dpif_upcall *upcall)
return EAGAIN; return EAGAIN;
} }
error = nl_sock_recv(upcall_sock, &buf, false); buf = ofpbuf_new(2048);
if (error == EAGAIN) { error = nl_sock_recv(upcall_sock, buf, false);
break; if (error) {
} else if (error) { ofpbuf_delete(buf);
if (error == EAGAIN) {
break;
}
return error; return error;
} }

10
lib/netlink-notifier.c
View File

@@ -1,5 +1,5 @@
/* /*
* Copyright (c) 2009, 2010, 2011 Nicira Networks. * Copyright (c) 2009, 2010, 2011, 2012 Nicira Networks.
* *
* Licensed under the Apache License, Version 2.0 (the "License"); * Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License. * you may not use this file except in compliance with the License.
@@ -164,18 +164,20 @@ nln_run(struct nln *nln)
nln->has_run = true; nln->has_run = true;
for (;;) { for (;;) {
struct ofpbuf *buf; uint64_t buf_stub[4096 / 8];
struct ofpbuf buf;
int error; int error;
ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub);
error = nl_sock_recv(nln->notify_sock, &buf, false); error = nl_sock_recv(nln->notify_sock, &buf, false);
if (!error) { if (!error) {
if (nln->parse(buf, nln->change)) { if (nln->parse(&buf, nln->change)) {
nln_report(nln, nln->change); nln_report(nln, nln->change);
} else { } else {
VLOG_WARN_RL(&rl, "received bad netlink message"); VLOG_WARN_RL(&rl, "received bad netlink message");
nln_report(nln, NULL); nln_report(nln, NULL);
} }
ofpbuf_delete(buf); ofpbuf_uninit(&buf);
} else if (error == EAGAIN) { } else if (error == EAGAIN) {
return; return;
} else { } else {

264
lib/netlink-socket.c
View File

@@ -300,32 +300,26 @@ STRESS_OPTION(
5, 1, -1, 100); 5, 1, -1, 100);
static int static int
nl_sock_recv__(struct nl_sock *sock, struct ofpbuf **bufp, bool wait) nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
{ {
/* We can't accurately predict the size of the data to be received. Most /* We can't accurately predict the size of the data to be received. The
* received data will fit in a 2 kB buffer, so we allocate that much space. * caller is supposed to have allocated enough space in 'buf' to handle the
* In case the data is actually bigger than that, we make available enough * "typical" case. To handle exceptions, we make available enough space in
* additional space to allow Netlink messages to be up to 64 kB long (a * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
* reasonable figure since that's the maximum length of a Netlink * figure since that's the maximum length of a Netlink attribute). */
* attribute). */
enum { MAX_SIZE = 65536 };
enum { HEAD_SIZE = 2048 };
enum { TAIL_SIZE = MAX_SIZE - HEAD_SIZE };
struct nlmsghdr *nlmsghdr; struct nlmsghdr *nlmsghdr;
uint8_t tail[TAIL_SIZE]; uint8_t tail[65536];
struct iovec iov[2]; struct iovec iov[2];
struct ofpbuf *buf;
struct msghdr msg; struct msghdr msg;
ssize_t retval; ssize_t retval;
*bufp = NULL; assert(buf->allocated >= sizeof *nlmsghdr);
ofpbuf_clear(buf);
buf = ofpbuf_new(HEAD_SIZE); iov[0].iov_base = buf->base;
iov[0].iov_base = buf->data; iov[0].iov_len = buf->allocated;
iov[0].iov_len = HEAD_SIZE;
iov[1].iov_base = tail; iov[1].iov_base = tail;
iov[1].iov_len = TAIL_SIZE; iov[1].iov_len = sizeof tail;
memset(&msg, 0, sizeof msg); memset(&msg, 0, sizeof msg);
msg.msg_iov = iov; msg.msg_iov = iov;
@@ -342,77 +336,65 @@ nl_sock_recv__(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
* the kernel tried to send to us. */ * the kernel tried to send to us. */
COVERAGE_INC(netlink_overflow); COVERAGE_INC(netlink_overflow);
} }
ofpbuf_delete(buf);
return error; return error;
} }
if (msg.msg_flags & MSG_TRUNC) { if (msg.msg_flags & MSG_TRUNC) {
VLOG_ERR_RL(&rl, "truncated message (longer than %d bytes)", MAX_SIZE); VLOG_ERR_RL(&rl, "truncated message (longer than %zu bytes)",
ofpbuf_delete(buf); sizeof tail);
return E2BIG; return E2BIG;
} }
ofpbuf_put_uninit(buf, MIN(retval, HEAD_SIZE));
if (retval > HEAD_SIZE) {
COVERAGE_INC(netlink_recv_jumbo);
ofpbuf_put(buf, tail, retval - HEAD_SIZE);
}
nlmsghdr = buf->data; nlmsghdr = buf->data;
if (retval < sizeof *nlmsghdr if (retval < sizeof *nlmsghdr
|| nlmsghdr->nlmsg_len < sizeof *nlmsghdr || nlmsghdr->nlmsg_len < sizeof *nlmsghdr
|| nlmsghdr->nlmsg_len > retval) { || nlmsghdr->nlmsg_len > retval) {
VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %d)", VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %zu)",
retval, NLMSG_HDRLEN); retval, sizeof *nlmsghdr);
ofpbuf_delete(buf);
return EPROTO; return EPROTO;
} }
if (STRESS(netlink_overflow)) { if (STRESS(netlink_overflow)) {
ofpbuf_delete(buf);
return ENOBUFS; return ENOBUFS;
} }
*bufp = buf; buf->size = MIN(retval, buf->allocated);
if (retval > buf->allocated) {
COVERAGE_INC(netlink_recv_jumbo);
ofpbuf_put(buf, tail, retval - buf->allocated);
}
log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol); log_nlmsg(__func__, 0, buf->data, buf->size, sock->protocol);
COVERAGE_INC(netlink_received); COVERAGE_INC(netlink_received);
return 0; return 0;
} }
/* Tries to receive a netlink message from the kernel on 'sock'. If /* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
* successful, stores the received message into '*bufp' and returns 0. The * 'wait' is true, waits for a message to be ready. Otherwise, fails with
* caller is responsible for destroying the message with ofpbuf_delete(). On * EAGAIN if the 'sock' receive buffer is empty.
* failure, returns a positive errno value and stores a null pointer into
* '*bufp'.
* *
* If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise, * The caller must have initialized 'buf' with an allocation of at least
* returns EAGAIN if the 'sock' receive buffer is empty. */ * NLMSG_HDRLEN bytes. For best performance, the caller should allocate enough
* space for a "typical" message.
*
* On success, returns 0 and replaces 'buf''s previous content by the received
* message. This function expands 'buf''s allocated memory, as necessary, to
* hold the actual size of the received message.
*
* On failure, returns a positive errno value and clears 'buf' to zero length.
* 'buf' retains its previous memory allocation.
*
* Regardless of success or failure, this function resets 'buf''s headroom to
* 0. */
int int
nl_sock_recv(struct nl_sock *sock, struct ofpbuf **bufp, bool wait) nl_sock_recv(struct nl_sock *sock, struct ofpbuf *buf, bool wait)
{ {
int error = nl_sock_cow__(sock); int error = nl_sock_cow__(sock);
if (error) { if (error) {
return error; return error;
} }
return nl_sock_recv__(sock, bufp, wait); return nl_sock_recv__(sock, buf, wait);
}
static int
find_nl_transaction_by_seq(struct nl_transaction **transactions, size_t n,
uint32_t seq)
{
int i;
for (i = 0; i < n; i++) {
struct nl_transaction *t = transactions[i];
if (seq == nl_msg_nlmsghdr(t->request)->nlmsg_seq) {
return i;
}
}
return -1;
} }
static void static void
@@ -422,8 +404,12 @@ nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
size_t i; size_t i;
for (i = 0; i < n; i++) { for (i = 0; i < n; i++) {
transactions[i]->error = error; struct nl_transaction *txn = transactions[i];
transactions[i]->reply = NULL;
txn->error = error;
if (txn->reply) {
ofpbuf_clear(txn->reply);
}
} }
} }
@@ -432,21 +418,28 @@ nl_sock_transact_multiple__(struct nl_sock *sock,
struct nl_transaction **transactions, size_t n, struct nl_transaction **transactions, size_t n,
size_t *done) size_t *done)
{ {
uint64_t tmp_reply_stub[1024 / 8];
struct nl_transaction tmp_txn;
struct ofpbuf tmp_reply;
uint32_t base_seq;
struct iovec iovs[MAX_IOVS]; struct iovec iovs[MAX_IOVS];
struct msghdr msg; struct msghdr msg;
int error; int error;
int i; int i;
base_seq = nl_sock_allocate_seq(sock, n);
*done = 0; *done = 0;
for (i = 0; i < n; i++) { for (i = 0; i < n; i++) {
struct ofpbuf *request = transactions[i]->request; struct nl_transaction *txn = transactions[i];
struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(request); struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
nlmsg->nlmsg_len = request->size; nlmsg->nlmsg_len = txn->request->size;
nlmsg->nlmsg_seq = base_seq + i;
nlmsg->nlmsg_pid = sock->pid; nlmsg->nlmsg_pid = sock->pid;
iovs[i].iov_base = request->data; iovs[i].iov_base = txn->request->data;
iovs[i].iov_len = request->size; iovs[i].iov_len = txn->request->size;
} }
memset(&msg, 0, sizeof msg); memset(&msg, 0, sizeof msg);
@@ -457,9 +450,9 @@ nl_sock_transact_multiple__(struct nl_sock *sock,
} while (error == EINTR); } while (error == EINTR);
for (i = 0; i < n; i++) { for (i = 0; i < n; i++) {
struct ofpbuf *request = transactions[i]->request; struct nl_transaction *txn = transactions[i];
log_nlmsg(__func__, error, request->data, request->size, log_nlmsg(__func__, error, txn->request->data, txn->request->size,
sock->protocol); sock->protocol);
} }
if (!error) { if (!error) {
@@ -470,62 +463,93 @@ nl_sock_transact_multiple__(struct nl_sock *sock,
return error; return error;
} }
ofpbuf_use_stub(&tmp_reply, tmp_reply_stub, sizeof tmp_reply_stub);
tmp_txn.request = NULL;
tmp_txn.reply = &tmp_reply;
tmp_txn.error = 0;
while (n > 0) { while (n > 0) {
struct ofpbuf *reply; struct nl_transaction *buf_txn, *txn;
uint32_t seq;
error = nl_sock_recv__(sock, &reply, false); /* Find a transaction whose buffer we can use for receiving a reply.
if (error == EAGAIN) { * If no such transaction is left, use tmp_txn. */
nl_sock_record_errors__(transactions, n, 0); buf_txn = &tmp_txn;
*done += n; for (i = 0; i < n; i++) {
return 0; if (transactions[i]->reply) {
} else if (error) { buf_txn = transactions[i];
return error; break;
}
} }
i = find_nl_transaction_by_seq(transactions, n, /* Receive a reply. */
nl_msg_nlmsghdr(reply)->nlmsg_seq); error = nl_sock_recv__(sock, buf_txn->reply, false);
if (i < 0) { if (error) {
VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, if (error == EAGAIN) {
nl_msg_nlmsghdr(reply)->nlmsg_seq); nl_sock_record_errors__(transactions, n, 0);
ofpbuf_delete(reply); *done += n;
error = 0;
}
break;
}
/* Match the reply up with a transaction. */
seq = nl_msg_nlmsghdr(buf_txn->reply)->nlmsg_seq;
if (seq < base_seq || seq >= base_seq + n) {
VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, seq);
continue; continue;
} }
i = seq - base_seq;
txn = transactions[i];
nl_sock_record_errors__(transactions, i, 0); /* Fill in the results for 'txn'. */
if (nl_msg_nlmsgerr(reply, &error)) { if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
transactions[i]->reply = NULL; if (txn->reply) {
transactions[i]->error = error; ofpbuf_clear(txn->reply);
if (error) { }
VLOG_DBG_RL(&rl, "received NAK error=%d (%s)", if (txn->error) {
error, strerror(error)); VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
error, strerror(txn->error));
} }
ofpbuf_delete(reply);
} else { } else {
transactions[i]->reply = reply; txn->error = 0;
transactions[i]->error = 0; if (txn->reply && txn != buf_txn) {
/* Swap buffers. */
struct ofpbuf *reply = buf_txn->reply;
buf_txn->reply = txn->reply;
txn->reply = reply;
}
} }
/* Fill in the results for transactions before 'txn'. (We have to do
* this after the results for 'txn' itself because of the buffer swap
* above.) */
nl_sock_record_errors__(transactions, i, 0);
/* Advance. */
*done += i + 1; *done += i + 1;
transactions += i + 1; transactions += i + 1;
n -= i + 1; n -= i + 1;
base_seq += i + 1;
} }
ofpbuf_uninit(&tmp_reply);
return 0; return error;
} }
/* Sends the 'request' member of the 'n' transactions in 'transactions' to the /* Sends the 'request' member of the 'n' transactions in 'transactions' on
* kernel, in order, and waits for responses to all of them. Fills in the * 'sock', in order, and receives responses to all of them. Fills in the
* 'error' member of each transaction with 0 if it was successful, otherwise * 'error' member of each transaction with 0 if it was successful, otherwise
* with a positive errno value. 'reply' will be NULL on error or if the * with a positive errno value. If 'reply' is nonnull, then it will be filled
* transaction was successful but had no reply beyond an indication of success. * with the reply if the message receives a detailed reply. In other cases,
* For a successful transaction that did have a more detailed reply, 'reply' * i.e. where the request failed or had no reply beyond an indication of
* will be set to the reply message. * success, 'reply' will be cleared if it is nonnull.
* *
* The caller is responsible for destroying each request and reply, and the * The caller is responsible for destroying each request and reply, and the
* transactions array itself. * transactions array itself.
* *
* Before sending each message, this function will finalize nlmsg_len in each * Before sending each message, this function will finalize nlmsg_len in each
* 'request' to match the ofpbuf's size, and set nlmsg_pid to 'sock''s pid. * 'request' to match the ofpbuf's size, set nlmsg_pid to 'sock''s pid, and
* initialize nlmsg_seq.
* *
* Bare Netlink is an unreliable transport protocol. This function layers * Bare Netlink is an unreliable transport protocol. This function layers
* reliable delivery and reply semantics on top of bare Netlink. See * reliable delivery and reply semantics on top of bare Netlink. See
@@ -640,13 +664,20 @@ nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
struct nl_transaction transaction; struct nl_transaction transaction;
transaction.request = (struct ofpbuf *) request; transaction.request = (struct ofpbuf *) request;
transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
transactionp = &transaction; transactionp = &transaction;
nl_sock_transact_multiple(sock, &transactionp, 1); nl_sock_transact_multiple(sock, &transactionp, 1);
if (replyp) { if (replyp) {
*replyp = transaction.reply; if (transaction.error) {
} else { ofpbuf_delete(transaction.reply);
ofpbuf_delete(transaction.reply); *replyp = NULL;
} else {
*replyp = transaction.reply;
}
} }
return transaction.error; return transaction.error;
} }
@@ -724,7 +755,7 @@ void
nl_dump_start(struct nl_dump *dump, nl_dump_start(struct nl_dump *dump,
struct nl_sock *sock, const struct ofpbuf *request) struct nl_sock *sock, const struct ofpbuf *request)
{ {
dump->buffer = NULL; ofpbuf_init(&dump->buffer, 4096);
if (sock->dump) { if (sock->dump) {
/* 'sock' already has an ongoing dump. Clone the socket because /* 'sock' already has an ongoing dump. Clone the socket because
* Netlink only allows one dump at a time. */ * Netlink only allows one dump at a time. */
@@ -745,26 +776,24 @@ nl_dump_start(struct nl_dump *dump,
/* Helper function for nl_dump_next(). */ /* Helper function for nl_dump_next(). */
static int static int
nl_dump_recv(struct nl_dump *dump, struct ofpbuf **bufferp) nl_dump_recv(struct nl_dump *dump)
{ {
struct nlmsghdr *nlmsghdr; struct nlmsghdr *nlmsghdr;
struct ofpbuf *buffer;
int retval; int retval;
retval = nl_sock_recv__(dump->sock, bufferp, true); retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
if (retval) { if (retval) {
return retval == EINTR ? EAGAIN : retval; return retval == EINTR ? EAGAIN : retval;
} }
buffer = *bufferp;
nlmsghdr = nl_msg_nlmsghdr(buffer); nlmsghdr = nl_msg_nlmsghdr(&dump->buffer);
if (dump->seq != nlmsghdr->nlmsg_seq) { if (dump->seq != nlmsghdr->nlmsg_seq) {
VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32, VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
nlmsghdr->nlmsg_seq, dump->seq); nlmsghdr->nlmsg_seq, dump->seq);
return EAGAIN; return EAGAIN;
} }
if (nl_msg_nlmsgerr(buffer, &retval)) { if (nl_msg_nlmsgerr(&dump->buffer, &retval)) {
VLOG_INFO_RL(&rl, "netlink dump request error (%s)", VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
strerror(retval)); strerror(retval));
return retval && retval != EAGAIN ? retval : EPROTO; return retval && retval != EAGAIN ? retval : EPROTO;
@@ -796,15 +825,10 @@ nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
return false; return false;
} }
if (dump->buffer && !dump->buffer->size) { while (!dump->buffer.size) {
ofpbuf_delete(dump->buffer); int retval = nl_dump_recv(dump);
dump->buffer = NULL;
}
while (!dump->buffer) {
int retval = nl_dump_recv(dump, &dump->buffer);
if (retval) { if (retval) {
ofpbuf_delete(dump->buffer); ofpbuf_clear(&dump->buffer);
dump->buffer = NULL;
if (retval != EAGAIN) { if (retval != EAGAIN) {
dump->status = retval; dump->status = retval;
return false; return false;
@@ -812,7 +836,7 @@ nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
} }
} }
nlmsghdr = nl_msg_next(dump->buffer, reply); nlmsghdr = nl_msg_next(&dump->buffer, reply);
if (!nlmsghdr) { if (!nlmsghdr) {
VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment"); VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
dump->status = EPROTO; dump->status = EPROTO;
@@ -847,7 +871,7 @@ nl_dump_done(struct nl_dump *dump)
nl_sock_destroy(dump->sock); nl_sock_destroy(dump->sock);
} }
} }
ofpbuf_delete(dump->buffer); ofpbuf_uninit(&dump->buffer);
return dump->status == EOF ? 0 : dump->status; return dump->status == EOF ? 0 : dump->status;
} }
@@ -1191,5 +1215,3 @@ log_nlmsg(const char *function, int error,
VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg); VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg);
free(nlmsg); free(nlmsg);
} }

20
lib/netlink-socket.h
View File

@@ -1,5 +1,5 @@
/* /*
* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks. * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira Networks.
* *
* Licensed under the Apache License, Version 2.0 (the "License"); * Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License. * you may not use this file except in compliance with the License.
@@ -35,8 +35,8 @@
#include <stdbool.h> #include <stdbool.h>
#include <stddef.h> #include <stddef.h>
#include <stdint.h> #include <stdint.h>
#include "ofpbuf.h"
struct ofpbuf;
struct nl_sock; struct nl_sock;
#ifndef HAVE_NETLINK #ifndef HAVE_NETLINK
@@ -52,9 +52,9 @@ int nl_sock_join_mcgroup(struct nl_sock *, unsigned int multicast_group);
int nl_sock_leave_mcgroup(struct nl_sock *, unsigned int multicast_group); int nl_sock_leave_mcgroup(struct nl_sock *, unsigned int multicast_group);
int nl_sock_send(struct nl_sock *, const struct ofpbuf *, bool wait); int nl_sock_send(struct nl_sock *, const struct ofpbuf *, bool wait);
int nl_sock_recv(struct nl_sock *, struct ofpbuf **, bool wait); int nl_sock_recv(struct nl_sock *, struct ofpbuf *, bool wait);
int nl_sock_transact(struct nl_sock *, const struct ofpbuf *request, int nl_sock_transact(struct nl_sock *, const struct ofpbuf *request,
struct ofpbuf **reply); struct ofpbuf **replyp);
int nl_sock_drain(struct nl_sock *); int nl_sock_drain(struct nl_sock *);
@@ -68,8 +68,14 @@ struct nl_transaction {
/* Filled in by client. */ /* Filled in by client. */
struct ofpbuf *request; /* Request to send. */ struct ofpbuf *request; /* Request to send. */
/* Filled in by nl_sock_transact_batch(). */ /* The client must initialize 'reply' to one of:
struct ofpbuf *reply; /* Reply (NULL if reply was an error code). */ *
* - NULL, if it does not care to examine the reply.
*
* - Otherwise, to an ofpbuf with a memory allocation of at least
* NLMSG_HDRLEN bytes.
*/
struct ofpbuf *reply; /* Reply (empty if reply was an error code). */
int error; /* Positive errno value, 0 if no error. */ int error; /* Positive errno value, 0 if no error. */
}; };
@@ -80,7 +86,7 @@ void nl_sock_transact_multiple(struct nl_sock *,
struct nl_dump { struct nl_dump {
struct nl_sock *sock; /* Socket being dumped. */ struct nl_sock *sock; /* Socket being dumped. */
uint32_t seq; /* Expected nlmsg_seq for replies. */ uint32_t seq; /* Expected nlmsg_seq for replies. */
struct ofpbuf *buffer; /* Receive buffer currently being iterated. */ struct ofpbuf buffer; /* Receive buffer currently being iterated. */
int status; /* 0=OK, EOF=done, or positive errno value. */ int status; /* 0=OK, EOF=done, or positive errno value. */
}; };

14
utilities/nlmon.c
View File

@@ -39,7 +39,9 @@ static const struct nl_policy rtnlgrp_link_policy[] = {
int int
main(int argc OVS_UNUSED, char *argv[]) main(int argc OVS_UNUSED, char *argv[])
{ {
uint64_t buf_stub[4096 / 64];
struct nl_sock *sock; struct nl_sock *sock;
struct ofpbuf buf;
int error; int error;
set_program_name(argv[0]); set_program_name(argv[0]);
@@ -55,9 +57,8 @@ main(int argc OVS_UNUSED, char *argv[])
ovs_fatal(error, "could not join RTNLGRP_LINK multicast group"); ovs_fatal(error, "could not join RTNLGRP_LINK multicast group");
} }
ofpbuf_use_stub(&buf, buf_stub, sizeof buf_stub);
for (;;) { for (;;) {
struct ofpbuf *buf;
error = nl_sock_recv(sock, &buf, false); error = nl_sock_recv(sock, &buf, false);
if (error == EAGAIN) { if (error == EAGAIN) {
/* Nothing to do. */ /* Nothing to do. */
@@ -95,19 +96,17 @@ main(int argc OVS_UNUSED, char *argv[])
struct ifinfomsg *iim; struct ifinfomsg *iim;
int i; int i;
nlh = ofpbuf_at(buf, 0, NLMSG_HDRLEN); nlh = ofpbuf_at(&buf, 0, NLMSG_HDRLEN);
iim = ofpbuf_at(buf, NLMSG_HDRLEN, sizeof *iim); iim = ofpbuf_at(&buf, NLMSG_HDRLEN, sizeof *iim);
if (!iim) { if (!iim) {
ovs_error(0, "received bad rtnl message (no ifinfomsg)"); ovs_error(0, "received bad rtnl message (no ifinfomsg)");
ofpbuf_delete(buf);
continue; continue;
} }
if (!nl_policy_parse(buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg), if (!nl_policy_parse(&buf, NLMSG_HDRLEN + sizeof(struct ifinfomsg),
rtnlgrp_link_policy, rtnlgrp_link_policy,
attrs, ARRAY_SIZE(rtnlgrp_link_policy))) { attrs, ARRAY_SIZE(rtnlgrp_link_policy))) {
ovs_error(0, "received bad rtnl message (policy)"); ovs_error(0, "received bad rtnl message (policy)");
ofpbuf_delete(buf);
continue; continue;
} }
printf("netdev %s changed (%s):\n", printf("netdev %s changed (%s):\n",
@@ -132,7 +131,6 @@ main(int argc OVS_UNUSED, char *argv[])
} }
printf("\tmaster=%"PRIu32" (%s)\n", idx, ifname); printf("\tmaster=%"PRIu32" (%s)\n", idx, ifname);
} }
ofpbuf_delete(buf);
} }
nl_sock_wait(sock, POLLIN); nl_sock_wait(sock, POLLIN);

49
vswitchd/ovs-brcompatd.c
View File

@@ -1,4 +1,4 @@
/* Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks /* Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira Networks
* *
* Licensed under the Apache License, Version 2.0 (the "License"); * Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License. * you may not use this file except in compliance with the License.
@@ -654,56 +654,53 @@ handle_get_ports_cmd(struct ofpbuf *buffer)
return 0; return 0;
} }
static struct ofpbuf * static bool
brc_recv_update__(void) brc_recv_update__(struct ofpbuf *buffer)
{ {
for (;;) { for (;;) {
struct ofpbuf *buffer; int retval = nl_sock_recv(brc_sock, buffer, false);
int retval;
retval = nl_sock_recv(brc_sock, &buffer, false);
switch (retval) { switch (retval) {
case 0: case 0:
if (nl_msg_nlmsgerr(buffer, NULL) if (nl_msg_nlmsgerr(buffer, NULL)
|| nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE) { || nl_msg_nlmsghdr(buffer)->nlmsg_type == NLMSG_DONE) {
break; break;
} }
return buffer; return true;
case ENOBUFS: case ENOBUFS:
break; break;
case EAGAIN: case EAGAIN:
return NULL; return false;
default: default:
VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval)); VLOG_WARN_RL(&rl, "brc_recv_update: %s", strerror(retval));
return NULL; return false;
} }
ofpbuf_delete(buffer);
} }
} }
static void static void
brc_recv_update(void) brc_recv_update(void)
{ {
struct ofpbuf *buffer;
struct genlmsghdr *genlmsghdr; struct genlmsghdr *genlmsghdr;
uint64_t buffer_stub[1024 / 8];
struct ofpbuf buffer;
buffer = brc_recv_update__(); ofpbuf_use_stub(&buffer, buffer_stub, sizeof buffer_stub);
if (!buffer) { if (!brc_recv_update__(&buffer)) {
return; goto error;
} }
genlmsghdr = nl_msg_genlmsghdr(buffer); genlmsghdr = nl_msg_genlmsghdr(&buffer);
if (!genlmsghdr) { if (!genlmsghdr) {
VLOG_WARN_RL(&rl, "received packet too short for generic NetLink"); VLOG_WARN_RL(&rl, "received packet too short for generic NetLink");
goto error; goto error;
} }
if (nl_msg_nlmsghdr(buffer)->nlmsg_type != brc_family) { if (nl_msg_nlmsghdr(&buffer)->nlmsg_type != brc_family) {
VLOG_DBG_RL(&rl, "received type (%"PRIu16") != brcompat family (%d)", VLOG_DBG_RL(&rl, "received type (%"PRIu16") != brcompat family (%d)",
nl_msg_nlmsghdr(buffer)->nlmsg_type, brc_family); nl_msg_nlmsghdr(&buffer)->nlmsg_type, brc_family);
goto error; goto error;
} }
@@ -729,31 +726,31 @@ brc_recv_update(void)
switch (genlmsghdr->cmd) { switch (genlmsghdr->cmd) {
case BRC_GENL_C_DP_ADD: case BRC_GENL_C_DP_ADD:
handle_bridge_cmd(buffer, true); handle_bridge_cmd(&buffer, true);
break; break;
case BRC_GENL_C_DP_DEL: case BRC_GENL_C_DP_DEL:
handle_bridge_cmd(buffer, false); handle_bridge_cmd(&buffer, false);
break; break;
case BRC_GENL_C_PORT_ADD: case BRC_GENL_C_PORT_ADD:
handle_port_cmd(buffer, true); handle_port_cmd(&buffer, true);
break; break;
case BRC_GENL_C_PORT_DEL: case BRC_GENL_C_PORT_DEL:
handle_port_cmd(buffer, false); handle_port_cmd(&buffer, false);
break; break;
case BRC_GENL_C_FDB_QUERY: case BRC_GENL_C_FDB_QUERY:
handle_fdb_query_cmd(buffer); handle_fdb_query_cmd(&buffer);
break; break;
case BRC_GENL_C_GET_BRIDGES: case BRC_GENL_C_GET_BRIDGES:
handle_get_bridges_cmd(buffer); handle_get_bridges_cmd(&buffer);
break; break;
case BRC_GENL_C_GET_PORTS: case BRC_GENL_C_GET_PORTS:
handle_get_ports_cmd(buffer); handle_get_ports_cmd(&buffer);
break; break;
default: default:
@@ -763,7 +760,7 @@ brc_recv_update(void)
} }
error: error:
ofpbuf_delete(buffer); ofpbuf_uninit(&buffer);
} }
static void static void