#include "poll-loop.h"
#include "timeval.h"
#include "util.h"
-
#include "vlog.h"
-#define THIS_MODULE VLM_netlink
+
+VLOG_DEFINE_THIS_MODULE(netlink);
+
+COVERAGE_DEFINE(netlink_overflow);
+COVERAGE_DEFINE(netlink_received);
+COVERAGE_DEFINE(netlink_recv_retry);
+COVERAGE_DEFINE(netlink_send);
+COVERAGE_DEFINE(netlink_sent);
/* Linux header file confusion causes this to be undefined. */
#ifndef SOL_NETLINK
};
/* Next nlmsghdr sequence number.
- *
+ *
* This implementation uses sequence numbers that are unique process-wide, to
* avoid a hypothetical race: send request, close socket, open new socket that
* reuses the old socket's PID value, send request on new socket, receive reply
/* Destroys netlink socket 'sock'. */
void
-nl_sock_destroy(struct nl_sock *sock)
+nl_sock_destroy(struct nl_sock *sock)
{
if (sock) {
close(sock->fd);
* '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)
+nl_sock_send(struct nl_sock *sock, const struct ofpbuf *msg, bool wait)
{
struct nlmsghdr *nlmsg = nl_msg_nlmsghdr(msg);
int error;
* 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)
+ bool wait)
{
struct msghdr msg;
int error;
* If 'wait' is true, nl_sock_recv waits for a message to be ready; otherwise,
* returns EAGAIN if the 'sock' receive buffer is empty. */
int
-nl_sock_recv(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
+nl_sock_recv(struct nl_sock *sock, struct ofpbuf **bufp, bool wait)
{
uint8_t tmp;
ssize_t bufsize = 2048;
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.
+ * and doubling the buffer size each time.
*/
nlmsghdr = ofpbuf_put_uninit(buf, bufsize);
iov.iov_base = nlmsghdr;
iov.iov_len = bufsize;
do {
- nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | MSG_PEEK);
+ nbytes = recvmsg(sock->fd, &msg, (wait ? 0 : MSG_DONTWAIT) | MSG_PEEK);
} while (nbytes < 0 && errno == EINTR);
if (nbytes < 0) {
ofpbuf_delete(buf);
*
* Bare Netlink is an unreliable transport protocol. This function layers
* reliable delivery and reply semantics on top of bare Netlink.
- *
+ *
* In Netlink, sending a request to the kernel is reliable enough, because the
* kernel will tell us if the message cannot be queued (and we will in that
* case put it on the transmit queue and wait until it can be delivered).
- *
+ *
* Receiving the reply is the real problem: if the socket buffer is full when
* the kernel tries to send the reply, the reply will be dropped. However, the
* kernel sets a flag that a reply has been dropped. The next call to recv
*/
int
nl_sock_transact(struct nl_sock *sock,
- const struct ofpbuf *request, struct ofpbuf **replyp)
+ const struct ofpbuf *request, struct ofpbuf **replyp)
{
uint32_t seq = nl_msg_nlmsghdr(request)->nlmsg_seq;
struct nlmsghdr *nlmsghdr;
/* 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;
-
+
send:
retval = nl_sock_send(sock, request, true);
if (retval) {
return 0;
}
+/* Starts a Netlink "dump" operation, by sending 'request' to the kernel via
+ * 'sock', 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 'sock''s pid, before the message is sent. NLM_F_DUMP and
+ * NLM_F_ACK will be set in nlmsg_flags.
+ *
+ * The properties of Netlink make dump operations reliable as long as all of
+ * the following are true:
+ *
+ * - At most a single dump is in progress at a time on a given nl_sock.
+ *
+ * - The nl_sock is not subscribed to any multicast groups.
+ *
+ * - The nl_sock is not used to send any other messages before the dump
+ * operation is complete.
+ *
+ * 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().
+ *
+ * The caller is responsible for destroying 'request'. The caller must not
+ * close 'sock' before it completes the dump operation (by calling
+ * nl_dump_done()).
+ */
+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->sock = sock;
+ dump->status = nl_sock_send(sock, request, true);
+ dump->buffer = NULL;
+}
+
+/* Helper function for nl_dump_next(). */
+static int
+nl_dump_recv(struct nl_dump *dump, struct ofpbuf **bufferp)
+{
+ struct nlmsghdr *nlmsghdr;
+ struct ofpbuf *buffer;
+ int retval;
+
+ retval = nl_sock_recv(dump->sock, bufferp, true);
+ if (retval) {
+ return retval == EINTR ? EAGAIN : retval;
+ }
+ buffer = *bufferp;
+
+ nlmsghdr = nl_msg_nlmsghdr(buffer);
+ if (dump->seq != nlmsghdr->nlmsg_seq) {
+ VLOG_DBG_RL(&rl, "ignoring seq %"PRIu32" != expected %"PRIu32,
+ nlmsghdr->nlmsg_seq, dump->seq);
+ return EAGAIN;
+ }
+
+ if (nl_msg_nlmsgerr(buffer, &retval)) {
+ VLOG_INFO_RL(&rl, "netlink dump request error (%s)",
+ strerror(retval));
+ return retval && retval != EAGAIN ? retval : EPROTO;
+ }
+
+ return 0;
+}
+
+/* Attempts to retrieve another reply from 'dump', which must have been
+ * initialized with nl_dump_start().
+ *
+ * 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).
+ *
+ * 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().
+ */
+bool
+nl_dump_next(struct nl_dump *dump, struct ofpbuf *reply)
+{
+ struct nlmsghdr *nlmsghdr;
+
+ reply->data = NULL;
+ reply->size = 0;
+ if (dump->status) {
+ 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);
+ if (retval) {
+ ofpbuf_delete(dump->buffer);
+ dump->buffer = NULL;
+ if (retval != EAGAIN) {
+ dump->status = retval;
+ return false;
+ }
+ }
+ }
+
+ nlmsghdr = nl_msg_next(dump->buffer, reply);
+ if (!nlmsghdr) {
+ VLOG_WARN_RL(&rl, "netlink dump reply contains message fragment");
+ dump->status = EPROTO;
+ return false;
+ } else if (nlmsghdr->nlmsg_type == NLMSG_DONE) {
+ dump->status = EOF;
+ return false;
+ }
+
+ return true;
+}
+
+/* Completes Netlink dump operation 'dump', which must have been initialized
+ * with nl_dump_start(). Returns 0 if the dump operation was error-free,
+ * otherwise a positive errno value describing the problem. */
+int
+nl_dump_done(struct nl_dump *dump)
+{
+ /* Drain any remaining messages that the client didn't read. Otherwise the
+ * kernel will continue to queue them up and waste buffer space. */
+ while (!dump->status) {
+ struct ofpbuf reply;
+ if (!nl_dump_next(dump, &reply)) {
+ assert(dump->status);
+ }
+ }
+
+ ofpbuf_delete(dump->buffer);
+ return dump->status == EOF ? 0 : dump->status;
+}
+
/* Causes poll_block() to wake up when any of the specified 'events' (which is
* a OR'd combination of POLLIN, POLLOUT, etc.) occur on 'sock'. */
void
*
* 'msg' must be at least as large as a nlmsghdr. */
struct nlmsghdr *
-nl_msg_nlmsghdr(const struct ofpbuf *msg)
+nl_msg_nlmsghdr(const struct ofpbuf *msg)
{
return ofpbuf_at_assert(msg, 0, NLMSG_HDRLEN);
}
* Returns a null pointer if 'msg' is not large enough to contain an nlmsghdr
* and a genlmsghdr. */
struct genlmsghdr *
-nl_msg_genlmsghdr(const struct ofpbuf *msg)
+nl_msg_genlmsghdr(const struct ofpbuf *msg)
{
return ofpbuf_at(msg, NLMSG_HDRLEN, GENL_HDRLEN);
}
*
* 'msg' must be at least as large as a nlmsghdr. */
bool
-nl_msg_nlmsgerr(const struct ofpbuf *msg, int *errorp)
+nl_msg_nlmsgerr(const struct ofpbuf *msg, int *errorp)
{
if (nl_msg_nlmsghdr(msg)->nlmsg_type == NLMSG_ERROR) {
struct nlmsgerr *err = ofpbuf_at(msg, NLMSG_HDRLEN, sizeof *err);
/* Ensures that 'b' has room for at least 'size' bytes plus netlink padding at
* its tail end, reallocating and copying its data if necessary. */
void
-nl_msg_reserve(struct ofpbuf *msg, size_t size)
+nl_msg_reserve(struct ofpbuf *msg, size_t size)
{
ofpbuf_prealloc_tailroom(msg, NLMSG_ALIGN(size));
}
* message. */
void
nl_msg_put_nlmsghdr(struct ofpbuf *msg,
- size_t expected_payload, uint32_t type, uint32_t flags)
+ size_t expected_payload, uint32_t type, uint32_t flags)
{
struct nlmsghdr *nlmsghdr;
* the tail end of 'msg'. Data in 'msg' is reallocated and copied if
* necessary. */
void
-nl_msg_put(struct ofpbuf *msg, const void *data, size_t size)
+nl_msg_put(struct ofpbuf *msg, const void *data, size_t size)
{
memcpy(nl_msg_put_uninit(msg, size), data, size);
}
* end of 'msg', reallocating and copying its data if necessary. Returns a
* pointer to the first byte of the new data, which is left uninitialized. */
void *
-nl_msg_put_uninit(struct ofpbuf *msg, size_t size)
+nl_msg_put_uninit(struct ofpbuf *msg, size_t size)
{
size_t pad = NLMSG_ALIGN(size) - size;
char *p = ofpbuf_put_uninit(msg, size + pad);
if (pad) {
- memset(p + size, 0, pad);
+ memset(p + size, 0, pad);
}
return p;
}
* 'msg', reallocating and copying its data if necessary. Returns a pointer to
* the first byte of data in the attribute, which is left uninitialized. */
void *
-nl_msg_put_unspec_uninit(struct ofpbuf *msg, uint16_t type, size_t size)
+nl_msg_put_unspec_uninit(struct ofpbuf *msg, uint16_t type, size_t size)
{
size_t total_size = NLA_HDRLEN + size;
struct nlattr* nla = nl_msg_put_uninit(msg, total_size);
* attribute, which is left uninitialized. */
void
nl_msg_put_unspec(struct ofpbuf *msg, uint16_t type,
- const void *data, size_t size)
+ const void *data, size_t size)
{
memcpy(nl_msg_put_unspec_uninit(msg, type, size), data, size);
}
* (Some Netlink protocols use the presence or absence of an attribute as a
* Boolean flag.) */
void
-nl_msg_put_flag(struct ofpbuf *msg, uint16_t type)
+nl_msg_put_flag(struct ofpbuf *msg, uint16_t type)
{
nl_msg_put_unspec(msg, type, NULL, 0);
}
/* Appends a Netlink attribute of the given 'type' and the given 8-bit 'value'
* to 'msg'. */
void
-nl_msg_put_u8(struct ofpbuf *msg, uint16_t type, uint8_t value)
+nl_msg_put_u8(struct ofpbuf *msg, uint16_t type, uint8_t value)
{
nl_msg_put_unspec(msg, type, &value, sizeof value);
}
nl_msg_put_unspec(msg, type, value, strlen(value) + 1);
}
-/* Appends a Netlink attribute of the given 'type' and the given buffered
- * netlink message in 'nested_msg' to 'msg'. The nlmsg_len field in
- * 'nested_msg' is finalized to match 'nested_msg->size'. */
+/* Adds the header for nested Netlink attributes to 'msg', with the specified
+ * 'type', and returns the header's offset within 'msg'. The caller should add
+ * the content for the nested Netlink attribute to 'msg' (e.g. using the other
+ * nl_msg_*() functions), and then pass the returned offset to
+ * nl_msg_end_nested() to finish up the nested attributes. */
+size_t
+nl_msg_start_nested(struct ofpbuf *msg, uint16_t type)
+{
+ size_t offset = msg->size;
+ nl_msg_put_unspec(msg, type, NULL, 0);
+ return offset;
+}
+
+/* Finalizes a nested Netlink attribute in 'msg'. 'offset' should be the value
+ * returned by nl_msg_start_nested(). */
+void
+nl_msg_end_nested(struct ofpbuf *msg, size_t offset)
+{
+ struct nlattr *attr = ofpbuf_at_assert(msg, offset, sizeof *attr);
+ attr->nla_len = msg->size - offset;
+}
+
+/* Appends a nested Netlink attribute of the given 'type', with the 'size'
+ * bytes of content starting at 'data', to 'msg'. */
void
nl_msg_put_nested(struct ofpbuf *msg,
- uint16_t type, struct ofpbuf *nested_msg)
+ uint16_t type, const void *data, size_t size)
{
- nl_msg_nlmsghdr(nested_msg)->nlmsg_len = nested_msg->size;
- nl_msg_put_unspec(msg, type, nested_msg->data, nested_msg->size);
+ size_t offset = nl_msg_start_nested(msg, type);
+ nl_msg_put(msg, data, size);
+ nl_msg_end_nested(msg, offset);
}
+/* If 'buffer' begins with a valid "struct nlmsghdr", pulls the header and its
+ * payload off 'buffer', stores header and payload in 'msg->data' and
+ * 'msg->size', and returns a pointer to the header.
+ *
+ * If 'buffer' does not begin with a "struct nlmsghdr" or begins with one that
+ * is invalid, returns NULL without modifying 'buffer'. */
+struct nlmsghdr *
+nl_msg_next(struct ofpbuf *buffer, struct ofpbuf *msg)
+{
+ if (buffer->size >= sizeof(struct nlmsghdr)) {
+ struct nlmsghdr *nlmsghdr = nl_msg_nlmsghdr(buffer);
+ size_t len = nlmsghdr->nlmsg_len;
+ if (len >= sizeof *nlmsghdr && len <= buffer->size) {
+ msg->data = nlmsghdr;
+ msg->size = len;
+ ofpbuf_pull(buffer, len);
+ return nlmsghdr;
+ }
+ }
+
+ msg->data = NULL;
+ msg->size = 0;
+ return NULL;
+}
+\f
+/* Attributes. */
+
/* Returns the first byte in the payload of attribute 'nla'. */
const void *
-nl_attr_get(const struct nlattr *nla)
+nl_attr_get(const struct nlattr *nla)
{
assert(nla->nla_len >= NLA_HDRLEN);
return nla + 1;
/* Returns the number of bytes in the payload of attribute 'nla'. */
size_t
-nl_attr_get_size(const struct nlattr *nla)
+nl_attr_get_size(const struct nlattr *nla)
{
assert(nla->nla_len >= NLA_HDRLEN);
return nla->nla_len - NLA_HDRLEN;
/* Asserts that 'nla''s payload is at least 'size' bytes long, and returns the
* first byte of the payload. */
const void *
-nl_attr_get_unspec(const struct nlattr *nla, size_t size)
+nl_attr_get_unspec(const struct nlattr *nla, size_t size)
{
assert(nla->nla_len >= NLA_HDRLEN + size);
return nla + 1;
/* Returns true if 'nla' is nonnull. (Some Netlink protocols use the presence
* or absence of an attribute as a Boolean flag.) */
bool
-nl_attr_get_flag(const struct nlattr *nla)
+nl_attr_get_flag(const struct nlattr *nla)
{
return nla != NULL;
}
*
* Asserts that 'nla''s payload is at least 1 byte long. */
uint8_t
-nl_attr_get_u8(const struct nlattr *nla)
+nl_attr_get_u8(const struct nlattr *nla)
{
return NL_ATTR_GET_AS(nla, uint8_t);
}
*
* Asserts that 'nla''s payload is at least 2 bytes long. */
uint16_t
-nl_attr_get_u16(const struct nlattr *nla)
+nl_attr_get_u16(const struct nlattr *nla)
{
return NL_ATTR_GET_AS(nla, uint16_t);
}
*
* Asserts that 'nla''s payload is at least 4 bytes long. */
uint32_t
-nl_attr_get_u32(const struct nlattr *nla)
+nl_attr_get_u32(const struct nlattr *nla)
{
return NL_ATTR_GET_AS(nla, uint32_t);
}
*
* Asserts that 'nla''s payload is at least 8 bytes long. */
uint64_t
-nl_attr_get_u64(const struct nlattr *nla)
+nl_attr_get_u64(const struct nlattr *nla)
{
return NL_ATTR_GET_AS(nla, uint64_t);
}
*
* Asserts that 'nla''s payload contains a null-terminated string. */
const char *
-nl_attr_get_string(const struct nlattr *nla)
+nl_attr_get_string(const struct nlattr *nla)
{
assert(nla->nla_len > NLA_HDRLEN);
assert(memchr(nl_attr_get(nla), '\0', nla->nla_len - NLA_HDRLEN) != NULL);
return nl_attr_get(nla);
}
+/* Initializes 'nested' to the payload of 'nla'. Doesn't initialize every
+ * field in 'nested', but enough to poke around with it in a read-only way. */
+void
+nl_attr_get_nested(const struct nlattr *nla, struct ofpbuf *nested)
+{
+ nested->data = (void *) nl_attr_get(nla);
+ nested->size = nl_attr_get_size(nla);
+}
+
/* Default minimum and maximum payload sizes for each type of attribute. */
static const size_t attr_len_range[][2] = {
[0 ... N_NL_ATTR_TYPES - 1] = { 0, SIZE_MAX },
type = nla->nla_type;
if (type < n_attrs && policy[type].type != NL_A_NO_ATTR) {
- const struct nl_policy *p = &policy[type];
+ const struct nl_policy *e = &policy[type];
size_t min_len, max_len;
/* Validate length and content. */
- min_len = p->min_len ? p->min_len : attr_len_range[p->type][0];
- max_len = p->max_len ? p->max_len : attr_len_range[p->type][1];
+ min_len = e->min_len ? e->min_len : attr_len_range[e->type][0];
+ max_len = e->max_len ? e->max_len : attr_len_range[e->type][1];
if (len < min_len || len > max_len) {
VLOG_DBG_RL(&rl, "%zu: attr %"PRIu16" length %zu not in "
"allowed range %zu...%zu",
offset, type, len, min_len, max_len);
return false;
}
- if (p->type == NL_A_STRING) {
+ if (e->type == NL_A_STRING) {
if (((char *) nla)[nla->nla_len - 1]) {
VLOG_DBG_RL(&rl, "%zu: attr %"PRIu16" lacks null at end",
offset, type);
return false;
}
}
- if (!p->optional && attrs[type] == NULL) {
+ if (!e->optional && attrs[type] == NULL) {
assert(n_required > 0);
--n_required;
}
}
return true;
}
+
+/* Parses the Netlink attributes within 'nla'. 'policy[i]', for 0 <= i <
+ * n_attrs, specifies how the attribute with nla_type == i is parsed; a pointer
+ * to attribute i is stored in attrs[i]. Returns true if successful, false on
+ * failure. */
+bool
+nl_parse_nested(const struct nlattr *nla, const struct nl_policy policy[],
+ struct nlattr *attrs[], size_t n_attrs)
+{
+ struct ofpbuf buf;
+
+ nl_attr_get_nested(nla, &buf);
+ return nl_policy_parse(&buf, 0, policy, attrs, n_attrs);
+}
\f
/* Miscellaneous. */
-static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
+static const struct nl_policy family_policy[CTRL_ATTR_MAX + 1] = {
[CTRL_ATTR_FAMILY_ID] = {.type = NL_A_U16},
};
-static int do_lookup_genl_family(const char *name)
+static int do_lookup_genl_family(const char *name)
{
struct nl_sock *sock;
struct ofpbuf request, *reply;
* may use '*number' as the family number. On failure, returns a positive
* errno value and '*number' caches the errno value. */
int
-nl_lookup_genl_family(const char *name, int *number)
+nl_lookup_genl_family(const char *name, int *number)
{
if (*number == 0) {
*number = do_lookup_genl_family(name);
unsigned int bits;
const char *name;
};
- static const struct nlmsg_flag flags[] = {
+ static const struct nlmsg_flag flags[] = {
{ NLM_F_REQUEST, "REQUEST" },
{ NLM_F_MULTI, "MULTI" },
{ NLM_F_ACK, "ACK" },