/*
- * Copyright (c) 2008, 2009, 2010, 2011 Nicira Networks.
+ * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira, Inc.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
#include <inttypes.h>
#include <stdlib.h>
#include <sys/types.h>
+#include <sys/uio.h>
#include <unistd.h>
#include "coverage.h"
#include "dynamic-string.h"
#include "poll-loop.h"
#include "socket-util.h"
#include "stress.h"
+#include "util.h"
#include "vlog.h"
VLOG_DEFINE_THIS_MODULE(netlink_socket);
* information. Also, at high logging levels we log *all* Netlink messages. */
static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(60, 600);
+static uint32_t nl_sock_allocate_seq(struct nl_sock *, unsigned int n);
static void log_nlmsg(const char *function, int error,
const void *message, size_t size, int protocol);
\f
struct nl_sock
{
int fd;
+ uint32_t next_seq;
uint32_t pid;
int protocol;
struct nl_dump *dump;
+ unsigned int rcvbuf; /* Receive buffer size (SO_RCVBUF). */
};
-static int alloc_pid(uint32_t *);
-static void free_pid(uint32_t);
+/* Compile-time limit on iovecs, so that we can allocate a maximum-size array
+ * of iovecs on the stack. */
+#define MAX_IOVS 128
+
+/* Maximum number of iovecs that may be passed to sendmsg, capped at a
+ * minimum of _XOPEN_IOV_MAX (16) and a maximum of MAX_IOVS.
+ *
+ * Initialized by nl_sock_create(). */
+static int max_iovs;
+
static int nl_sock_cow__(struct nl_sock *);
/* Creates a new netlink socket for the given netlink 'protocol'
{
struct nl_sock *sock;
struct sockaddr_nl local, remote;
+ socklen_t local_size;
+ int rcvbuf;
int retval = 0;
+ if (!max_iovs) {
+ int save_errno = errno;
+ errno = 0;
+
+ max_iovs = sysconf(_SC_UIO_MAXIOV);
+ if (max_iovs < _XOPEN_IOV_MAX) {
+ if (max_iovs == -1 && errno) {
+ VLOG_WARN("sysconf(_SC_UIO_MAXIOV): %s", strerror(errno));
+ }
+ max_iovs = _XOPEN_IOV_MAX;
+ } else if (max_iovs > MAX_IOVS) {
+ max_iovs = MAX_IOVS;
+ }
+
+ errno = save_errno;
+ }
+
*sockp = NULL;
sock = malloc(sizeof *sock);
if (sock == NULL) {
}
sock->protocol = protocol;
sock->dump = NULL;
+ sock->next_seq = 1;
- retval = alloc_pid(&sock->pid);
- if (retval) {
- goto error;
+ rcvbuf = 1024 * 1024;
+ if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUFFORCE,
+ &rcvbuf, sizeof rcvbuf)) {
+ VLOG_WARN_RL(&rl, "setting %d-byte socket receive buffer failed (%s)",
+ rcvbuf, strerror(errno));
}
- /* Bind local address as our selected pid. */
- memset(&local, 0, sizeof local);
- local.nl_family = AF_NETLINK;
- local.nl_pid = sock->pid;
- if (bind(sock->fd, (struct sockaddr *) &local, sizeof local) < 0) {
- VLOG_ERR("bind(%"PRIu32"): %s", sock->pid, strerror(errno));
- goto error_free_pid;
+ retval = get_socket_rcvbuf(sock->fd);
+ if (retval < 0) {
+ retval = -retval;
+ goto error;
}
+ sock->rcvbuf = retval;
- /* Bind remote address as the kernel (pid 0). */
+ /* Connect to kernel (pid 0) as remote address. */
memset(&remote, 0, sizeof remote);
remote.nl_family = AF_NETLINK;
remote.nl_pid = 0;
if (connect(sock->fd, (struct sockaddr *) &remote, sizeof remote) < 0) {
VLOG_ERR("connect(0): %s", strerror(errno));
- goto error_free_pid;
+ goto error;
+ }
+
+ /* Obtain pid assigned by kernel. */
+ local_size = sizeof local;
+ if (getsockname(sock->fd, (struct sockaddr *) &local, &local_size) < 0) {
+ VLOG_ERR("getsockname: %s", strerror(errno));
+ goto error;
+ }
+ if (local_size < sizeof local || local.nl_family != AF_NETLINK) {
+ VLOG_ERR("getsockname returned bad Netlink name");
+ retval = EINVAL;
+ goto error;
}
+ sock->pid = local.nl_pid;
*sockp = sock;
return 0;
-error_free_pid:
- free_pid(sock->pid);
error:
if (retval == 0) {
retval = errno;
sock->dump = NULL;
} else {
close(sock->fd);
- free_pid(sock->pid);
free(sock);
}
}
}
static int
-nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
+nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg,
+ uint32_t nlmsg_seq, bool wait)
{
struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
int error;
nlmsg->nlmsg_len = msg->size;
+ nlmsg->nlmsg_seq = nlmsg_seq;
nlmsg->nlmsg_pid = sock->pid;
do {
int retval;
}
/* Tries to send 'msg', which must contain a Netlink message, to the kernel on
- * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, and
- * nlmsg_pid will be set to 'sock''s pid, before the message is sent.
+ * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
+ * will be set to 'sock''s pid, and nlmsg_seq will be initialized to a fresh
+ * sequence number, before the message is sent.
*
* Returns 0 if successful, otherwise a positive errno value. If
* 'wait' is true, then the send will wait until buffer space is ready;
* otherwise, returns EAGAIN if the 'sock' send buffer is full. */
int
nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
+{
+ return nl_sock_send_seq(sock, msg, nl_sock_allocate_seq(sock, 1), wait);
+}
+
+/* Tries to send 'msg', which must contain a Netlink message, to the kernel on
+ * 'sock'. nlmsg_len in 'msg' will be finalized to match msg->size, nlmsg_pid
+ * will be set to 'sock''s pid, and nlmsg_seq will be initialized to
+ * 'nlmsg_seq', before the message is sent.
+ *
+ * Returns 0 if successful, otherwise a positive errno value. If
+ * 'wait' is true, then the send will wait until buffer space is ready;
+ * otherwise, returns EAGAIN if the 'sock' send buffer is full.
+ *
+ * This function is suitable for sending a reply to a request that was received
+ * with sequence number 'nlmsg_seq'. Otherwise, use nl_sock_send() instead. */
+int
+nl_sock_send_seq(struct nl_sock *sock, const struct ofpbuf *msg,
+ uint32_t nlmsg_seq, bool wait)
{
int error = nl_sock_cow__(sock);
if (error) {
return error;
}
- return nl_sock_send__(sock, msg, wait);
+ return nl_sock_send__(sock, msg, nlmsg_seq, wait);
}
/* This stress option is useful for testing that OVS properly tolerates
5, 1, -1, 100);
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
- * received data will fit in a 2 kB buffer, so we allocate that much space.
- * In case the data is actually bigger than that, we make available enough
- * additional space to allow Netlink messages to be up to 64 kB long (a
- * reasonable figure since that's the maximum length of a Netlink
- * attribute). */
- enum { MAX_SIZE = 65536 };
- enum { HEAD_SIZE = 2048 };
- enum { TAIL_SIZE = MAX_SIZE - HEAD_SIZE };
-
+ /* We can't accurately predict the size of the data to be received. The
+ * caller is supposed to have allocated enough space in 'buf' to handle the
+ * "typical" case. To handle exceptions, we make available enough space in
+ * 'tail' to allow Netlink messages to be up to 64 kB long (a reasonable
+ * figure since that's the maximum length of a Netlink attribute). */
struct nlmsghdr *nlmsghdr;
- uint8_t tail[TAIL_SIZE];
+ uint8_t tail[65536];
struct iovec iov[2];
- struct ofpbuf *buf;
struct msghdr msg;
ssize_t retval;
- *bufp = NULL;
+ assert(buf->allocated >= sizeof *nlmsghdr);
+ ofpbuf_clear(buf);
- buf = ofpbuf_new(HEAD_SIZE);
- iov[0].iov_base = buf->data;
- iov[0].iov_len = HEAD_SIZE;
+ iov[0].iov_base = buf->base;
+ iov[0].iov_len = buf->allocated;
iov[1].iov_base = tail;
- iov[1].iov_len = TAIL_SIZE;
+ iov[1].iov_len = sizeof tail;
memset(&msg, 0, sizeof msg);
msg.msg_iov = iov;
* the kernel tried to send to us. */
COVERAGE_INC(netlink_overflow);
}
- ofpbuf_delete(buf);
return error;
}
if (msg.msg_flags & MSG_TRUNC) {
- VLOG_ERR_RL(&rl, "truncated message (longer than %d bytes)", MAX_SIZE);
- ofpbuf_delete(buf);
+ VLOG_ERR_RL(&rl, "truncated message (longer than %zu bytes)",
+ sizeof tail);
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;
if (retval < sizeof *nlmsghdr
|| nlmsghdr->nlmsg_len < sizeof *nlmsghdr
|| nlmsghdr->nlmsg_len > retval) {
- VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %d)",
- retval, NLMSG_HDRLEN);
- ofpbuf_delete(buf);
+ VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %zu)",
+ retval, sizeof *nlmsghdr);
return EPROTO;
}
if (STRESS(netlink_overflow)) {
- ofpbuf_delete(buf);
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);
COVERAGE_INC(netlink_received);
return 0;
}
-/* Tries to receive a netlink message from the kernel on 'sock'. If
- * successful, stores the received message into '*bufp' and returns 0. The
- * caller is responsible for destroying the message with ofpbuf_delete(). On
- * failure, returns a positive errno value and stores a null pointer into
- * '*bufp'.
+/* Tries to receive a Netlink message from the kernel on 'sock' into 'buf'. If
+ * 'wait' is true, waits for a message to be ready. Otherwise, fails with
+ * EAGAIN if the 'sock' receive buffer is empty.
+ *
+ * The caller must have initialized 'buf' with an allocation of at least
+ * 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.
*
- * If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise,
- * returns EAGAIN if the 'sock' receive buffer is empty. */
+ * 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
-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);
if (error) {
return error;
}
- return nl_sock_recv__(sock, bufp, wait);
+ return nl_sock_recv__(sock, buf, wait);
+}
+
+static void
+nl_sock_record_errors__(struct nl_transaction **transactions, size_t n,
+ int error)
+{
+ size_t i;
+
+ for (i = 0; i < n; i++) {
+ struct nl_transaction *txn = transactions[i];
+
+ txn->error = error;
+ if (txn->reply) {
+ ofpbuf_clear(txn->reply);
+ }
+ }
+}
+
+static int
+nl_sock_transact_multiple__(struct nl_sock *sock,
+ struct nl_transaction **transactions, size_t n,
+ 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 msghdr msg;
+ int error;
+ int i;
+
+ base_seq = nl_sock_allocate_seq(sock, n);
+ *done = 0;
+ for (i = 0; i < n; i++) {
+ struct nl_transaction *txn = transactions[i];
+ struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(txn->request);
+
+ nlmsg->nlmsg_len = txn->request->size;
+ nlmsg->nlmsg_seq = base_seq + i;
+ nlmsg->nlmsg_pid = sock->pid;
+
+ iovs[i].iov_base = txn->request->data;
+ iovs[i].iov_len = txn->request->size;
+ }
+
+ memset(&msg, 0, sizeof msg);
+ msg.msg_iov = iovs;
+ msg.msg_iovlen = n;
+ do {
+ error = sendmsg(sock->fd, &msg, 0) < 0 ? errno : 0;
+ } while (error == EINTR);
+
+ for (i = 0; i < n; i++) {
+ struct nl_transaction *txn = transactions[i];
+
+ log_nlmsg(__func__, error, txn->request->data, txn->request->size,
+ sock->protocol);
+ }
+ if (!error) {
+ COVERAGE_ADD(netlink_sent, n);
+ }
+
+ if (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) {
+ struct nl_transaction *buf_txn, *txn;
+ uint32_t seq;
+
+ /* Find a transaction whose buffer we can use for receiving a reply.
+ * If no such transaction is left, use tmp_txn. */
+ buf_txn = &tmp_txn;
+ for (i = 0; i < n; i++) {
+ if (transactions[i]->reply) {
+ buf_txn = transactions[i];
+ break;
+ }
+ }
+
+ /* Receive a reply. */
+ error = nl_sock_recv__(sock, buf_txn->reply, false);
+ if (error) {
+ if (error == EAGAIN) {
+ nl_sock_record_errors__(transactions, n, 0);
+ *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;
+ }
+ i = seq - base_seq;
+ txn = transactions[i];
+
+ /* Fill in the results for 'txn'. */
+ if (nl_msg_nlmsgerr(buf_txn->reply, &txn->error)) {
+ if (txn->reply) {
+ ofpbuf_clear(txn->reply);
+ }
+ if (txn->error) {
+ VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
+ error, strerror(txn->error));
+ }
+ } else {
+ txn->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;
+ transactions += i + 1;
+ n -= i + 1;
+ base_seq += i + 1;
+ }
+ ofpbuf_uninit(&tmp_reply);
+
+ return error;
+}
+
+/* Sends the 'request' member of the 'n' transactions in 'transactions' on
+ * '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
+ * with a positive errno value. If 'reply' is nonnull, then it will be filled
+ * with the reply if the message receives a detailed reply. In other cases,
+ * i.e. where the request failed or had no reply beyond an indication of
+ * success, 'reply' will be cleared if it is nonnull.
+ *
+ * The caller is responsible for destroying each request and reply, and the
+ * transactions array itself.
+ *
+ * Before sending each message, this function will finalize nlmsg_len in each
+ * '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
+ * reliable delivery and reply semantics on top of bare Netlink. See
+ * nl_sock_transact() for some caveats.
+ */
+void
+nl_sock_transact_multiple(struct nl_sock *sock,
+ struct nl_transaction **transactions, size_t n)
+{
+ int max_batch_count;
+ int error;
+
+ if (!n) {
+ return;
+ }
+
+ error = nl_sock_cow__(sock);
+ if (error) {
+ nl_sock_record_errors__(transactions, n, error);
+ return;
+ }
+
+ /* In theory, every request could have a 64 kB reply. But the default and
+ * maximum socket rcvbuf size with typical Dom0 memory sizes both tend to
+ * be a bit below 128 kB, so that would only allow a single message in a
+ * "batch". So we assume that replies average (at most) 4 kB, which allows
+ * a good deal of batching.
+ *
+ * In practice, most of the requests that we batch either have no reply at
+ * all or a brief reply. */
+ max_batch_count = MAX(sock->rcvbuf / 4096, 1);
+ max_batch_count = MIN(max_batch_count, max_iovs);
+
+ while (n > 0) {
+ size_t count, bytes;
+ size_t done;
+
+ /* Batch up to 'max_batch_count' transactions. But cap it at about a
+ * page of requests total because big skbuffs are expensive to
+ * allocate in the kernel. */
+#if defined(PAGESIZE)
+ enum { MAX_BATCH_BYTES = MAX(1, PAGESIZE - 512) };
+#else
+ enum { MAX_BATCH_BYTES = 4096 - 512 };
+#endif
+ bytes = transactions[0]->request->size;
+ for (count = 1; count < n && count < max_batch_count; count++) {
+ if (bytes + transactions[count]->request->size > MAX_BATCH_BYTES) {
+ break;
+ }
+ bytes += transactions[count]->request->size;
+ }
+
+ error = nl_sock_transact_multiple__(sock, transactions, count, &done);
+ transactions += done;
+ n -= done;
+
+ if (error == ENOBUFS) {
+ VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
+ } else if (error) {
+ VLOG_ERR_RL(&rl, "transaction error (%s)", strerror(error));
+ nl_sock_record_errors__(transactions, n, error);
+ }
+ }
}
/* Sends 'request' to the kernel via 'sock' and waits for a response. If
* on failure '*replyp' is set to NULL. If 'replyp' is null, then the kernel's
* reply, if any, is discarded.
*
- * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will
- * be set to 'sock''s pid, before the message is sent. NLM_F_ACK will be set
- * in nlmsg_flags.
+ * Before the message is sent, nlmsg_len in 'request' will be finalized to
+ * match msg->size, nlmsg_pid will be set to 'sock''s pid, and nlmsg_seq will
+ * be initialized, NLM_F_ACK will be set in nlmsg_flags.
*
* The caller is responsible for destroying 'request'.
*
* needs to be idempotent.
*/
int
-nl_sock_transact(struct nl_sock *sock,
- const struct ofpbuf *request, struct ofpbuf **replyp)
+nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request,
+ struct ofpbuf **replyp)
{
- uint32_t seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
- struct nlmsghdr *nlmsghdr;
- struct ofpbuf *reply;
- int retval;
+ struct nl_transaction *transactionp;
+ struct nl_transaction transaction;
- if (replyp) {
- *replyp = NULL;
- }
+ transaction.request = CONST_CAST(struct ofpbuf *, request);
+ transaction.reply = replyp ? ofpbuf_new(1024) : NULL;
+ transactionp = &transaction;
- /* Ensure that we get a reply even if this message doesn't ordinarily call
- * for one. */
- nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_ACK;
+ nl_sock_transact_multiple(sock, &transactionp, 1);
-send:
- retval = nl_sock_send(sock, request, true);
- if (retval) {
- return retval;
- }
-
-recv:
- retval = nl_sock_recv(sock, &reply, true);
- if (retval) {
- if (retval == ENOBUFS) {
- COVERAGE_INC(netlink_overflow);
- VLOG_DBG_RL(&rl, "receive buffer overflow, resending request");
- goto send;
+ if (replyp) {
+ if (transaction.error) {
+ ofpbuf_delete(transaction.reply);
+ *replyp = NULL;
} else {
- return retval;
+ *replyp = transaction.reply;
}
}
- nlmsghdr = nl_msg_nlmsghdr(reply);
- if (seq != nlmsghdr->nlmsg_seq) {
- VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
- nl_msg_nlmsghdr(reply)->nlmsg_seq, seq);
- ofpbuf_delete(reply);
- goto recv;
- }
- /* If the reply is an error, discard the reply and return the error code.
- *
- * Except: if the reply is just an acknowledgement (error code of 0), and
- * the caller is interested in the reply (replyp != NULL), pass the reply
- * up to the caller. Otherwise the caller will get a return value of 0
- * and null '*replyp', which makes unwary callers likely to segfault. */
- if (nl_msg_nlmsgerr(reply, &retval) && (retval || !replyp)) {
- ofpbuf_delete(reply);
- if (retval) {
- VLOG_DBG_RL(&rl, "received NAK error=%d (%s)",
- retval, strerror(retval));
- }
- return retval != EAGAIN ? retval : EPROTO;
- }
-
- if (replyp) {
- *replyp = reply;
- } else {
- ofpbuf_delete(reply);
- }
- return 0;
+ return transaction.error;
}
/* Drain all the messages currently in 'sock''s receive queue. */
nl_dump_start(struct nl_dump *dump,
struct nl_sock *sock, const struct ofpbuf *request)
{
- struct nlmsghdr *nlmsghdr = nl_msg_nlmsghdr(request);
- nlmsghdr->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
- dump->seq = nlmsghdr->nlmsg_seq;
- dump->buffer = NULL;
+ ofpbuf_init(&dump->buffer, 4096);
if (sock->dump) {
/* 'sock' already has an ongoing dump. Clone the socket because
* Netlink only allows one dump at a time. */
dump->sock = sock;
dump->status = 0;
}
- dump->status = nl_sock_send__(sock, request, true);
+
+ nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK;
+ dump->status = nl_sock_send__(sock, request, nl_sock_allocate_seq(sock, 1),
+ true);
+ dump->seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
}
/* Helper function for nl_dump_next(). */
static int
-nl_dump_recv(struct nl_dump *dump, struct ofpbuf **bufferp)
+nl_dump_recv(struct nl_dump *dump)
{
struct nlmsghdr *nlmsghdr;
- struct ofpbuf *buffer;
int retval;
- retval = nl_sock_recv__(dump->sock, bufferp, true);
+ retval = nl_sock_recv__(dump->sock, &dump->buffer, true);
if (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) {
VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32,
nlmsghdr->nlmsg_seq, dump->seq);
return EAGAIN;
}
- if (nl_msg_nlmsgerr(buffer, &retval)) {
+ if (nl_msg_nlmsgerr(&dump->buffer, &retval)) {
VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
strerror(retval));
return retval && retval != EAGAIN ? retval : EPROTO;
return false;
}
- if (dump->buffer && !dump->buffer->size) {
- ofpbuf_delete(dump->buffer);
- dump->buffer = NULL;
- }
- while (!dump->buffer) {
- int retval = nl_dump_recv(dump, &dump->buffer);
+ while (!dump->buffer.size) {
+ int retval = nl_dump_recv(dump);
if (retval) {
- ofpbuf_delete(dump->buffer);
- dump->buffer = NULL;
+ ofpbuf_clear(&dump->buffer);
if (retval != EAGAIN) {
dump->status = retval;
return false;
}
}
- nlmsghdr = nl_msg_next(dump->buffer, reply);
+ nlmsghdr = nl_msg_next(&dump->buffer, reply);
if (!nlmsghdr) {
VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
dump->status = EPROTO;
nl_sock_destroy(dump->sock);
}
}
- ofpbuf_delete(dump->buffer);
+ ofpbuf_uninit(&dump->buffer);
return dump->status == EOF ? 0 : dump->status;
}
poll_fd_wait(sock->fd, events);
}
-/* Checks whether this socket caused a wakeup in the previous run of the poll
- * loop. */
-short int
-nl_sock_woke(const struct nl_sock *sock)
+/* Returns the underlying fd for 'sock', for use in "poll()"-like operations
+ * that can't use nl_sock_wait().
+ *
+ * It's a little tricky to use the returned fd correctly, because nl_sock does
+ * "copy on write" to allow a single nl_sock to be used for notifications,
+ * transactions, and dumps. If 'sock' is used only for notifications and
+ * transactions (and never for dump) then the usage is safe. */
+int
+nl_sock_fd(const struct nl_sock *sock)
{
- return poll_fd_woke(sock->fd);
+ return sock->fd;
}
/* Returns the PID associated with this socket. */
unsigned int *multicast_group, unsigned int fallback)
{
struct nlattr *family_attrs[ARRAY_SIZE(family_policy)];
- struct ofpbuf all_mcs;
+ const struct nlattr *mc;
struct ofpbuf *reply;
- struct nlattr *mc;
unsigned int left;
int error;
goto exit;
}
- nl_attr_get_nested(family_attrs[CTRL_ATTR_MCAST_GROUPS], &all_mcs);
- NL_ATTR_FOR_EACH (mc, left, all_mcs.data, all_mcs.size) {
+ NL_NESTED_FOR_EACH (mc, left, family_attrs[CTRL_ATTR_MCAST_GROUPS]) {
static const struct nl_policy mc_policy[] = {
[CTRL_ATTR_MCAST_GRP_ID] = {.type = NL_A_U32},
[CTRL_ATTR_MCAST_GRP_NAME] = {.type = NL_A_STRING},
return *number > 0 ? 0 : -*number;
}
\f
-/* Netlink PID.
- *
- * Every Netlink socket must be bound to a unique 32-bit PID. By convention,
- * programs that have a single Netlink socket use their Unix process ID as PID,
- * and programs with multiple Netlink sockets add a unique per-socket
- * identifier in the bits above the Unix process ID.
- *
- * The kernel has Netlink PID 0.
- */
-
-/* Parameters for how many bits in the PID should come from the Unix process ID
- * and how many unique per-socket. */
-#define SOCKET_BITS 10
-#define MAX_SOCKETS (1u << SOCKET_BITS)
-
-#define PROCESS_BITS (32 - SOCKET_BITS)
-#define MAX_PROCESSES (1u << PROCESS_BITS)
-#define PROCESS_MASK ((uint32_t) (MAX_PROCESSES - 1))
-
-/* Bit vector of unused socket identifiers. */
-static uint32_t avail_sockets[ROUND_UP(MAX_SOCKETS, 32)];
-
-/* Allocates and returns a new Netlink PID. */
-static int
-alloc_pid(uint32_t *pid)
+static uint32_t
+nl_sock_allocate_seq(struct nl_sock *sock, unsigned int n)
{
- int i;
+ uint32_t seq = sock->next_seq;
- for (i = 0; i < MAX_SOCKETS; i++) {
- if ((avail_sockets[i / 32] & (1u << (i % 32))) == 0) {
- avail_sockets[i / 32] |= 1u << (i % 32);
- *pid = (getpid() & PROCESS_MASK) | (i << PROCESS_BITS);
- return 0;
- }
+ sock->next_seq += n;
+
+ /* Make it impossible for the next request for sequence numbers to wrap
+ * around to 0. Start over with 1 to avoid ever using a sequence number of
+ * 0, because the kernel uses sequence number 0 for notifications. */
+ if (sock->next_seq >= UINT32_MAX / 2) {
+ sock->next_seq = 1;
}
- VLOG_ERR("netlink pid space exhausted");
- return ENOBUFS;
-}
-/* Makes the specified 'pid' available for reuse. */
-static void
-free_pid(uint32_t pid)
-{
- int sock = pid >> PROCESS_BITS;
- assert(avail_sockets[sock / 32] & (1u << (sock % 32)));
- avail_sockets[sock / 32] &= ~(1u << (sock % 32));
+ return seq;
}
-\f
+
static void
nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
{
if (flags_left) {
ds_put_format(ds, "[OTHER:%"PRIx16"]", flags_left);
}
- ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32"(%d:%d))",
- h->nlmsg_seq, h->nlmsg_pid,
- (int) (h->nlmsg_pid & PROCESS_MASK),
- (int) (h->nlmsg_pid >> PROCESS_BITS));
+ ds_put_format(ds, ", seq=%"PRIx32", pid=%"PRIu32,
+ h->nlmsg_seq, h->nlmsg_pid);
}
static char *
VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg);
free(nlmsg);
}
-
-