X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetlink-socket.c;h=8c8d16be766fa8fd35c4c62ccee2bf353b3f2540;hb=79f108b14e7944ddc4669e9c03fc34b40a3a2288;hp=23c05c5e812d8040086faa61904edf4fa88a2f23;hpb=3442636d01d2a73a557952ad9140de07418c28c2;p=sliver-openvswitch.git diff --git a/lib/netlink-socket.c b/lib/netlink-socket.c index 23c05c5e8..8c8d16be7 100644 --- a/lib/netlink-socket.c +++ b/lib/netlink-socket.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013 Nicira, Inc. + * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -31,6 +31,7 @@ #include "ofpbuf.h" #include "ovs-thread.h" #include "poll-loop.h" +#include "seq.h" #include "socket-util.h" #include "util.h" #include "vlog.h" @@ -40,7 +41,6 @@ VLOG_DEFINE_THIS_MODULE(netlink_socket); COVERAGE_DEFINE(netlink_overflow); COVERAGE_DEFINE(netlink_received); COVERAGE_DEFINE(netlink_recv_jumbo); -COVERAGE_DEFINE(netlink_send); COVERAGE_DEFINE(netlink_sent); /* Linux header file confusion causes this to be undefined. */ @@ -338,7 +338,7 @@ nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait) } if (msg.msg_flags & MSG_TRUNC) { - VLOG_ERR_RL(&rl, "truncated message (longer than %zu bytes)", + VLOG_ERR_RL(&rl, "truncated message (longer than %"PRIuSIZE" bytes)", sizeof tail); return E2BIG; } @@ -347,7 +347,7 @@ nl_sock_recv__(struct nl_sock *sock, struct ofpbuf *buf, bool wait) if (retval < sizeof *nlmsghdr || nlmsghdr->nlmsg_len < sizeof *nlmsghdr || nlmsghdr->nlmsg_len > retval) { - VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %zu)", + VLOG_ERR_RL(&rl, "received invalid nlmsg (%"PRIuSIZE"d bytes < %"PRIuSIZE")", retval, sizeof *nlmsghdr); return EPROTO; } @@ -676,105 +676,125 @@ nl_sock_drain(struct nl_sock *sock) * Netlink socket created with the given 'protocol', and initializes 'dump' to * reflect the state of the operation. * - * nlmsg_len in 'msg' will be finalized to match msg->size, and nlmsg_pid will - * be set to the Netlink socket's pid, before the message is sent. NLM_F_DUMP - * and NLM_F_ACK will be set in nlmsg_flags. + * 'request' must contain a Netlink message. Before sending the message, + * nlmsg_len will be finalized to match request->size, and nlmsg_pid will be + * set to the Netlink socket's pid. NLM_F_DUMP and NLM_F_ACK will be set in + * nlmsg_flags. * * The design of this Netlink socket library ensures that the dump is reliable. * - * This function provides no status indication. An error status for the entire - * dump operation is provided when it is completed by calling nl_dump_done(). + * This function provides no status indication. nl_dump_done() provides an + * error status for the entire dump operation. * - * The caller is responsible for destroying 'request'. + * The caller must eventually destroy 'request'. */ void nl_dump_start(struct nl_dump *dump, int protocol, const struct ofpbuf *request) { - ofpbuf_init(&dump->buffer, 4096); - dump->status = nl_pool_alloc(protocol, &dump->sock); - if (dump->status) { + int status = nl_pool_alloc(protocol, &dump->sock); + + if (status) { return; } nl_msg_nlmsghdr(request)->nlmsg_flags |= NLM_F_DUMP | NLM_F_ACK; - dump->status = nl_sock_send__(dump->sock, request, - nl_sock_allocate_seq(dump->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 nlmsghdr *nlmsghdr; - int retval; - - retval = nl_sock_recv__(dump->sock, &dump->buffer, true); - if (retval) { - return retval == EINTR ? EAGAIN : retval; - } - - 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(&dump->buffer, &retval)) { - VLOG_INFO_RL(&rl, "netlink dump request error (%s)", - ovs_strerror(retval)); - return retval && retval != EAGAIN ? retval : EPROTO; - } - - return 0; + status = nl_sock_send__(dump->sock, request, + nl_sock_allocate_seq(dump->sock, 1), true); + atomic_init(&dump->status, status << 1); + dump->nl_seq = nl_msg_nlmsghdr(request)->nlmsg_seq; + dump->status_seq = seq_create(); } -/* Attempts to retrieve another reply from 'dump', which must have been - * initialized with nl_dump_start(). +/* Attempts to retrieve another reply from 'dump' into 'buffer'. 'dump' must + * have been initialized with nl_dump_start(), and 'buffer' must have been + * initialized. 'buffer' should be at least NL_DUMP_BUFSIZE bytes long. * * If successful, returns true and points 'reply->data' and 'reply->size' to - * the message that was retrieved. The caller must not modify 'reply' (because - * it points into the middle of a larger buffer). + * the message that was retrieved. The caller must not modify 'reply' (because + * it points within 'buffer', which will be used by future calls to this + * function). * * On failure, returns false and sets 'reply->data' to NULL and 'reply->size' * to 0. Failure might indicate an actual error or merely the end of replies. * An error status for the entire dump operation is provided when it is * completed by calling nl_dump_done(). + * + * Multiple threads may call this function, passing the same nl_dump, however + * each must provide independent buffers. This function may cache multiple + * replies in the buffer, and these will be processed before more replies are + * fetched. When this function returns false, other threads may continue to + * process replies in their buffers, but they will not fetch more replies. */ bool -nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply) +nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply, struct ofpbuf *buffer) { struct nlmsghdr *nlmsghdr; + int error = 0; reply->data = NULL; reply->size = 0; - if (dump->status) { - return false; - } - while (!dump->buffer.size) { - int retval = nl_dump_recv(dump); + /* If 'buffer' is empty, fetch another batch of nlmsgs. */ + while (!buffer->size) { + unsigned int status; + int retval, seq; + + seq = seq_read(dump->status_seq); + atomic_read(&dump->status, &status); + if (status) { + return false; + } + + retval = nl_sock_recv__(dump->sock, buffer, false); if (retval) { - ofpbuf_clear(&dump->buffer); - if (retval != EAGAIN) { - dump->status = retval; - return false; + ofpbuf_clear(buffer); + if (retval == EAGAIN) { + nl_sock_wait(dump->sock, POLLIN); + seq_wait(dump->status_seq, seq); + poll_block(); + continue; + } else { + error = retval; + goto exit; } } + + nlmsghdr = nl_msg_nlmsghdr(buffer); + if (dump->nl_seq != nlmsghdr->nlmsg_seq) { + VLOG_DBG_RL(&rl, "ignoring seq %#"PRIx32" != expected %#"PRIx32, + nlmsghdr->nlmsg_seq, dump->nl_seq); + ofpbuf_clear(buffer); + continue; + } + + if (nl_msg_nlmsgerr(buffer, &retval) && retval) { + VLOG_INFO_RL(&rl, "netlink dump request error (%s)", + ovs_strerror(retval)); + error = retval == EAGAIN ? EPROTO : retval; + ofpbuf_clear(buffer); + goto exit; + } } - nlmsghdr = nl_msg_next(&dump->buffer, reply); + /* Fetch the next nlmsg in the current batch. */ + nlmsghdr = nl_msg_next(buffer, reply); if (!nlmsghdr) { VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment"); - dump->status = EPROTO; - return false; + error = EPROTO; } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) { - dump->status = EOF; - return false; + error = EOF; } - return true; +exit: + if (error == EOF) { + unsigned int old; + atomic_or(&dump->status, 1, &old); + seq_change(dump->status_seq); + } else if (error) { + atomic_store(&dump->status, error << 1); + seq_change(dump->status_seq); + } + return !error; } /* Completes Netlink dump operation 'dump', which must have been initialized @@ -783,19 +803,28 @@ nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply) int nl_dump_done(struct nl_dump *dump) { + int status; + /* Drain any remaining messages that the client didn't read. Otherwise the * kernel will continue to queue them up and waste buffer space. * * XXX We could just destroy and discard the socket in this case. */ - while (!dump->status) { - struct ofpbuf reply; - if (!nl_dump_next(dump, &reply)) { - ovs_assert(dump->status); + atomic_read(&dump->status, &status); + if (!status) { + uint64_t tmp_reply_stub[NL_DUMP_BUFSIZE / 8]; + struct ofpbuf reply, buf; + + ofpbuf_use_stub(&buf, tmp_reply_stub, sizeof tmp_reply_stub); + while (nl_dump_next(dump, &reply, &buf)) { + /* Nothing to do. */ } + atomic_read(&dump->status, &status); + ovs_assert(status); + ofpbuf_uninit(&buf); } nl_pool_release(dump->sock); - ofpbuf_uninit(&dump->buffer); - return dump->status == EOF ? 0 : dump->status; + seq_destroy(dump->status_seq); + return status >> 1; } /* Causes poll_block() to wake up when any of the specified 'events' (which is