X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetlink-socket.c;h=713049aab0ed2367b472eb646830002700d7b8ce;hb=e0edde6fee279cdbbf3c179f5f50adaf0c7c7f1e;hp=7f496261a45634f2cb200f05bd770e6ff36a8466;hpb=8522ba0996907af86eb63afd4c69e9fd6bb1178a;p=sliver-openvswitch.git diff --git a/lib/netlink-socket.c b/lib/netlink-socket.c index 7f496261a..713049aab 100644 --- a/lib/netlink-socket.c +++ b/lib/netlink-socket.c @@ -1,5 +1,5 @@ /* - * 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. @@ -54,6 +54,7 @@ COVERAGE_DEFINE(netlink_sent); * 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); @@ -62,6 +63,7 @@ static void log_nlmsg(const char *function, int error, struct nl_sock { int fd; + uint32_t next_seq; uint32_t pid; int protocol; struct nl_dump *dump; @@ -78,8 +80,6 @@ struct nl_sock * Initialized by nl_sock_create(). */ static int max_iovs; -static int alloc_pid(uint32_t *); -static void free_pid(uint32_t); static int nl_sock_cow__(struct nl_sock *); /* Creates a new netlink socket for the given netlink 'protocol' @@ -90,6 +90,8 @@ nl_sock_create(int protocol, struct nl_sock **sockp) { struct nl_sock *sock; struct sockaddr_nl local, remote; + socklen_t local_size; + int rcvbuf; int retval = 0; if (!max_iovs) { @@ -122,6 +124,14 @@ nl_sock_create(int protocol, struct nl_sock **sockp) } sock->protocol = protocol; sock->dump = NULL; + sock->next_seq = 1; + + 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)); + } retval = get_socket_rcvbuf(sock->fd); if (retval < 0) { @@ -130,34 +140,31 @@ nl_sock_create(int protocol, struct nl_sock **sockp) } sock->rcvbuf = retval; - retval = alloc_pid(&sock->pid); - if (retval) { - goto error; - } - - /* 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; - } - - /* 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; @@ -190,7 +197,6 @@ nl_sock_destroy(struct nl_sock *sock) sock->dump = NULL; } else { close(sock->fd); - free_pid(sock->pid); free(sock); } } @@ -253,6 +259,7 @@ nl_sock_send__(struct nl_sock *sock, const struct ofpbuf *msg, bool wait) int error; nlmsg->nlmsg_len = msg->size; + nlmsg->nlmsg_seq = nl_sock_allocate_seq(sock, 1); nlmsg->nlmsg_pid = sock->pid; do { int retval; @@ -293,32 +300,26 @@ STRESS_OPTION( 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; @@ -335,77 +336,65 @@ nl_sock_recv__(struct nl_sock *sock, struct ofpbuf **bufp, bool wait) * 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); -} - -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; + return nl_sock_recv__(sock, buf, wait); } static void @@ -415,8 +404,12 @@ nl_sock_record_errors__(struct nl_transaction **transactions, size_t n, size_t i; for (i = 0; i < n; i++) { - transactions[i]->error = error; - transactions[i]->reply = NULL; + struct nl_transaction *txn = transactions[i]; + + txn->error = error; + if (txn->reply) { + ofpbuf_clear(txn->reply); + } } } @@ -425,26 +418,28 @@ 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 ofpbuf *request = transactions[i]->request; - struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(request); + struct nl_transaction *txn = transactions[i]; + 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; - if (i == n - 1) { - /* Ensure that we get a reply even if the final request doesn't - * ordinarily call for one. */ - nlmsg->nlmsg_flags |= NLM_F_ACK; - } - iovs[i].iov_base = request->data; - iovs[i].iov_len = request->size; + iovs[i].iov_base = txn->request->data; + iovs[i].iov_len = txn->request->size; } memset(&msg, 0, sizeof msg); @@ -455,9 +450,9 @@ nl_sock_transact_multiple__(struct nl_sock *sock, } while (error == EINTR); 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); } if (!error) { @@ -468,59 +463,93 @@ nl_sock_transact_multiple__(struct nl_sock *sock, 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 ofpbuf *reply; + 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; + } + } - error = nl_sock_recv__(sock, &reply, true); + /* Receive a reply. */ + error = nl_sock_recv__(sock, buf_txn->reply, false); if (error) { - return error; + if (error == EAGAIN) { + nl_sock_record_errors__(transactions, n, 0); + *done += n; + error = 0; + } + break; } - i = find_nl_transaction_by_seq(transactions, n, - nl_msg_nlmsghdr(reply)->nlmsg_seq); - if (i < 0) { - VLOG_DBG_RL(&rl, "ignoring unexpected seq %#"PRIx32, - nl_msg_nlmsghdr(reply)->nlmsg_seq); - ofpbuf_delete(reply); + /* 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]; - nl_sock_record_errors__(transactions, i, 0); - if (nl_msg_nlmsgerr(reply, &error)) { - transactions[i]->reply = NULL; - transactions[i]->error = error; - if (error) { + /* 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(error)); + error, strerror(txn->error)); } - ofpbuf_delete(reply); } else { - transactions[i]->reply = reply; - transactions[i]->error = 0; + 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 0; + return error; } -/* Sends the 'request' member of the 'n' transactions in 'transactions' to the - * kernel, in order, and waits for responses to all of them. Fills in the +/* 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. 'reply' will be NULL on error or if the - * transaction was successful but had no reply beyond an indication of success. - * For a successful transaction that did have a more detailed reply, 'reply' - * will be set to the reply message. + * 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, and set nlmsg_pid to 'sock''s pid. - * NLM_F_ACK will be added to some requests' nlmsg_flags. + * '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 @@ -595,9 +624,9 @@ nl_sock_transact_multiple(struct nl_sock *sock, * 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'. * @@ -635,13 +664,20 @@ nl_sock_transact(struct nl_sock *sock, const struct ofpbuf *request, struct nl_transaction transaction; transaction.request = (struct ofpbuf *) request; + transaction.reply = replyp ? ofpbuf_new(1024) : NULL; transactionp = &transaction; + nl_sock_transact_multiple(sock, &transactionp, 1); + if (replyp) { - *replyp = transaction.reply; - } else { - ofpbuf_delete(transaction.reply); + if (transaction.error) { + ofpbuf_delete(transaction.reply); + *replyp = NULL; + } else { + *replyp = transaction.reply; + } } + return transaction.error; } @@ -719,10 +755,7 @@ void 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. */ @@ -735,31 +768,32 @@ nl_dump_start(struct nl_dump *dump, dump->sock = sock; dump->status = 0; } + + nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK; dump->status = nl_sock_send__(sock, request, 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; @@ -791,15 +825,10 @@ nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply) 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; @@ -807,7 +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) { VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment"); dump->status = EPROTO; @@ -842,7 +871,7 @@ nl_dump_done(struct nl_dump *dump) nl_sock_destroy(dump->sock); } } - ofpbuf_delete(dump->buffer); + ofpbuf_uninit(&dump->buffer); return dump->status == EOF ? 0 : dump->status; } @@ -1055,54 +1084,23 @@ nl_lookup_genl_family(const char *name, int *number) return *number > 0 ? 0 : -*number; } -/* 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; } - + static void nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds) { @@ -1151,10 +1149,8 @@ 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 * @@ -1219,5 +1215,3 @@ log_nlmsg(const char *function, int error, VLOG_DBG_RL(&rl, "%s (%s): %s", function, strerror(error), nlmsg); free(nlmsg); } - -