#include <unistd.h>
#include "coverage.h"
#include "dynamic-string.h"
+#include "hash.h"
+#include "hmap.h"
#include "netlink.h"
#include "netlink-protocol.h"
#include "ofpbuf.h"
COVERAGE_DEFINE(netlink_overflow);
COVERAGE_DEFINE(netlink_received);
-COVERAGE_DEFINE(netlink_recv_retry);
+COVERAGE_DEFINE(netlink_recv_jumbo);
COVERAGE_DEFINE(netlink_send);
COVERAGE_DEFINE(netlink_sent);
}
sock->protocol = protocol;
sock->any_groups = false;
+ sock->dump = NULL;
retval = alloc_pid(&sock->pid);
if (retval) {
return nl_sock_send__(sock, msg, wait);
}
-/* Tries to send the 'n_iov' chunks of data in 'iov' to the kernel on 'sock' as
- * a single Netlink message. (The message must be fully formed and not require
- * finalization of its nlmsg_len or nlmsg_pid fields.)
- *
- * 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_sendv(struct nl_sock *sock, const struct iovec iov[], size_t n_iov,
- bool wait)
-{
- struct msghdr msg;
- int error;
-
- COVERAGE_INC(netlink_send);
- memset(&msg, 0, sizeof msg);
- msg.msg_iov = (struct iovec *) iov;
- msg.msg_iovlen = n_iov;
- do {
- int retval;
- retval = sendmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
- error = retval < 0 ? errno : 0;
- } while (error == EINTR);
- if (error != EAGAIN) {
- log_nlmsg(__func__, error, iov[0].iov_base, iov[0].iov_len,
- sock->protocol);
- if (!error) {
- COVERAGE_INC(netlink_sent);
- }
- }
- return error;
-}
-
/* This stress option is useful for testing that OVS properly tolerates
* -ENOBUFS on NetLink sockets. Such errors are unavoidable because they can
* occur if the kernel cannot temporarily allocate enough GFP_ATOMIC memory to
static int
nl_sock_recv__(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
{
- uint8_t tmp;
- ssize_t bufsize = 2048;
- ssize_t nbytes, nbytes2;
- struct ofpbuf *buf;
+ /* 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 };
+
struct nlmsghdr *nlmsghdr;
- struct iovec iov;
- struct msghdr msg = {
- .msg_name = NULL,
- .msg_namelen = 0,
- .msg_iov = &iov,
- .msg_iovlen = 1,
- .msg_control = NULL,
- .msg_controllen = 0,
- .msg_flags = 0
- };
+ uint8_t tail[TAIL_SIZE];
+ struct iovec iov[2];
+ struct ofpbuf *buf;
+ struct msghdr msg;
+ ssize_t retval;
- buf = ofpbuf_new(bufsize);
*bufp = NULL;
-try_again:
- /* Attempt to read the message. We don't know the size of the data
- * yet, so we take a guess at 2048. If we're wrong, we keep trying
- * and doubling the buffer size each time.
- */
- nlmsghdr = ofpbuf_put_uninit(buf, bufsize);
- iov.iov_base = nlmsghdr;
- iov.iov_len = bufsize;
+ buf = ofpbuf_new(HEAD_SIZE);
+ iov[0].iov_base = buf->data;
+ iov[0].iov_len = HEAD_SIZE;
+ iov[1].iov_base = tail;
+ iov[1].iov_len = TAIL_SIZE;
+
+ memset(&msg, 0, sizeof msg);
+ msg.msg_iov = iov;
+ msg.msg_iovlen = 2;
+
do {
- nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | MSG_PEEK);
- } while (nbytes < 0 && errno == EINTR);
- if (nbytes < 0) {
+ retval = recvmsg(sock->fd, &msg, wait ? 0 : MSG_DONTWAIT);
+ } while (retval < 0 && errno == EINTR);
+
+ if (retval < 0) {
+ int error = errno;
+ if (error == ENOBUFS) {
+ /* Socket receive buffer overflow dropped one or more messages that
+ * the kernel tried to send to us. */
+ COVERAGE_INC(netlink_overflow);
+ }
ofpbuf_delete(buf);
- return errno;
+ return error;
}
+
if (msg.msg_flags & MSG_TRUNC) {
- COVERAGE_INC(netlink_recv_retry);
- bufsize *= 2;
- ofpbuf_reinit(buf, bufsize);
- goto try_again;
+ VLOG_ERR_RL(&rl, "truncated message (longer than %d bytes)", MAX_SIZE);
+ ofpbuf_delete(buf);
+ return E2BIG;
}
- buf->size = nbytes;
- /* We successfully read the message, so recv again to clear the queue */
- iov.iov_base = &tmp;
- iov.iov_len = 1;
- do {
- nbytes2 = recvmsg(sock->fd, &msg, MSG_DONTWAIT);
- } while (nbytes2 < 0 && errno == EINTR);
- if (nbytes2 < 0) {
- if (errno == ENOBUFS) {
- /* The kernel is notifying us that a message it tried to send to us
- * was dropped. We have to pass this along to the caller in case
- * it wants to retry a request. So kill the buffer, which we can
- * re-read next time. */
- COVERAGE_INC(netlink_overflow);
- ofpbuf_delete(buf);
- return ENOBUFS;
- } else {
- VLOG_ERR_RL(&rl, "failed to remove nlmsg from socket: %s\n",
- strerror(errno));
- }
+ ofpbuf_put_uninit(buf, MIN(retval, HEAD_SIZE));
+ if (retval > HEAD_SIZE) {
+ COVERAGE_INC(netlink_recv_jumbo);
+ ofpbuf_put(buf, tail, retval - HEAD_SIZE);
}
- if (nbytes < sizeof *nlmsghdr
+
+ nlmsghdr = buf->data;
+ if (retval < sizeof *nlmsghdr
|| nlmsghdr->nlmsg_len < sizeof *nlmsghdr
- || nlmsghdr->nlmsg_len > nbytes) {
+ || nlmsghdr->nlmsg_len > retval) {
VLOG_ERR_RL(&rl, "received invalid nlmsg (%zd bytes < %d)",
- bufsize, NLMSG_HDRLEN);
+ retval, NLMSG_HDRLEN);
ofpbuf_delete(buf);
return EPROTO;
}
\f
/* Miscellaneous. */
+struct genl_family {
+ struct hmap_node hmap_node;
+ uint16_t id;
+ char *name;
+};
+
+static struct hmap genl_families = HMAP_INITIALIZER(&genl_families);
+
static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
[CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
};
+static struct genl_family *
+find_genl_family_by_id(uint16_t id)
+{
+ struct genl_family *family;
+
+ HMAP_FOR_EACH_IN_BUCKET (family, hmap_node, hash_int(id, 0),
+ &genl_families) {
+ if (family->id == id) {
+ return family;
+ }
+ }
+ return NULL;
+}
+
+static void
+define_genl_family(uint16_t id, const char *name)
+{
+ struct genl_family *family = find_genl_family_by_id(id);
+
+ if (family) {
+ if (!strcmp(family->name, name)) {
+ return;
+ }
+ free(family->name);
+ } else {
+ family = xmalloc(sizeof *family);
+ family->id = id;
+ hmap_insert(&genl_families, &family->hmap_node, hash_int(id, 0));
+ }
+ family->name = xstrdup(name);
+}
+
+static const char *
+genl_family_to_name(uint16_t id)
+{
+ if (id == GENL_ID_CTRL) {
+ return "control";
+ } else {
+ struct genl_family *family = find_genl_family_by_id(id);
+ return family ? family->name : "unknown";
+ }
+}
+
static int do_lookup_genl_family(const char *name)
{
struct nl_sock *sock;
retval = nl_attr_get_u16(attrs[CTRL_ATTR_FAMILY_ID]);
if (retval == 0) {
retval = -EPROTO;
+ } else {
+ define_genl_family(retval, name);
}
nl_sock_destroy(sock);
ofpbuf_delete(reply);
+
return retval;
}
}
\f
static void
-nlmsghdr_to_string(const struct nlmsghdr *h, struct ds *ds)
+nlmsghdr_to_string(const struct nlmsghdr *h, int protocol, struct ds *ds)
{
struct nlmsg_flag {
unsigned int bits;
ds_put_cstr(ds, "(overrun)");
} else if (h->nlmsg_type < NLMSG_MIN_TYPE) {
ds_put_cstr(ds, "(reserved)");
+ } else if (protocol == NETLINK_GENERIC) {
+ ds_put_format(ds, "(%s)", genl_family_to_name(h->nlmsg_type));
} else {
ds_put_cstr(ds, "(family-defined)");
}
struct ds ds = DS_EMPTY_INITIALIZER;
const struct nlmsghdr *h = ofpbuf_at(buffer, 0, NLMSG_HDRLEN);
if (h) {
- nlmsghdr_to_string(h, &ds);
+ nlmsghdr_to_string(h, protocol, &ds);
if (h->nlmsg_type == NLMSG_ERROR) {
const struct nlmsgerr *e;
e = ofpbuf_at(buffer, NLMSG_HDRLEN,
ds_put_format(&ds, "(%s)", strerror(-e->error));
}
ds_put_cstr(&ds, ", in-reply-to(");
- nlmsghdr_to_string(&e->msg, &ds);
+ nlmsghdr_to_string(&e->msg, protocol, &ds);
ds_put_cstr(&ds, "))");
} else {
ds_put_cstr(&ds, " error(truncated)");
git://git.onelab.eu / sliver-openvswitch.git / blobdiff