Page MenuHomeFreeBSD
Differential D36002 Diff 110128 sys/netlink/netlink_message_writer.c

Changeset View

Standalone View

View Options

sys/netlink/netlink_message_writer.c

  • This file was added.
/*-
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
* Copyright (c) 2022 Alexander V. Chernikov
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include <sys/param.h>
#include <sys/malloc.h>
#include <sys/lock.h>
#include <sys/rmlock.h>
#include <sys/mbuf.h>
#include <sys/ck.h>
#include <sys/socket.h>
#include <sys/socketvar.h>
#include <sys/syslog.h>
#include <netlink/netlink.h>
#include <netlink/netlink_ctl.h>
#include <netlink/netlink_linux.h>
#include <netlink/netlink_var.h>
#define DEBUG_MOD_NAME nl_message
#define DEBUG_MAX_LEVEL LOG_DEBUG3
#include <netlink/netlink_debug.h>
_DECLARE_DEBUG(LOG_DEBUG);
/*
* The goal of this file is to provide convenient message writing KPI on top of
* different storage methods (mbufs, uio, temporary memory chunks).
*
* The main KPI guarantee is the the (last) message always resides in the contiguous
* memory buffer, so one is able to update the header after writing the entire message.
*
* This guarantee comes with a side effect of potentially reallocating underlying
* buffer, so one needs to update the desired pointers after something is added
* to the header.
*
* Messaging layer contains hooks performing transparent Linux translation for the messages.
*
* There are 3 types of supported targets:
* * socket (adds mbufs to the socket buffer, used for message replies)
* * group (sends mbuf/chain to the specified groups, used for the notifications)
* * chain (returns mbuf chain, used in Linux message translation code)
*
* There are 3 types of storage:
* * NS_WRITER_TYPE_MBUF (mbuf-based, most efficient, used when a single message
* fits in MCLBYTES)
* * NS_WRITER_TYPE_BUF (fallback, malloc-based, used when a single message needs
* to be larger than one supported by NS_WRITER_TYPE_MBUF)
* * NS_WRITER_TYPE_LBUF (malloc-based, similar to NS_WRITER_TYPE_BUF, used for
* Linux sockets, calls translation hook prior to sending messages to the socket).
*
* Internally, KPI switches between different types of storage when memory requirements
* change. It happens transparently to the caller.
*/
typedef bool nlwriter_op_init(struct nlmsg_state *ns, int size, bool waitok);
typedef bool nlwriter_op_write(struct nlmsg_state *ns, void *buf, int buflen, int cnt);
struct nlwriter_ops {
nlwriter_op_init *init;
nlwriter_op_write *write_socket;
nlwriter_op_write *write_group;
nlwriter_op_write *write_chain;
};
/*
* NS_WRITER_TYPE_BUF
* Writes message to a temporary memory buffer,
* flushing to the socket/group when buffer size limit is reached
*/
static bool
nlmsg_get_ns_buf(struct nlmsg_state *ns, int size, bool waitok)
{
int mflag = waitok ? M_WAITOK : M_NOWAIT;
ns->_storage = malloc(size, M_NETLINK, mflag | M_ZERO);
if (__predict_false(ns->_storage == NULL))
return (false);
ns->alloc_len = size;
ns->offset = 0;
ns->hdr = NULL;
ns->data = ns->_storage;
ns->writer_type = NS_WRITER_TYPE_BUF;
ns->malloc_flag = mflag;
ns->num_messages = 0;
ns->enomem = false;
return (true);
}
static bool
nlmsg_write_socket_buf(struct nlmsg_state *ns, void *buf, int datalen, int cnt)
{
NL_LOG(LOG_DEBUG2, "IN: ptr: %p len: %d arg: %p", buf, datalen, ns);
if (__predict_false(datalen == 0)) {
free(buf, M_NETLINK);
return (true);
}
struct mbuf *m = m_getm2(NULL, datalen, ns->malloc_flag, MT_DATA, M_PKTHDR);
if (__predict_false(m == NULL)) {
/* XXX: should we set sorcverr? */
free(buf, M_NETLINK);
return (false);
}
m_append(m, datalen, buf);
free(buf, M_NETLINK);
int io_flags = (ns->ignore_limit) ? NL_IOF_IGNORE_LIMIT : 0;
return (nl_send_one(m, (struct nlpcb *)(ns->arg_ptr), cnt, io_flags));
}
static bool
nlmsg_write_group_buf(struct nlmsg_state *ns, void *buf, int datalen, int cnt)
{
NL_LOG(LOG_DEBUG2, "IN: ptr: %p len: %d arg: %p", buf, datalen, ns->arg_ptr);
if (__predict_false(datalen == 0)) {
free(buf, M_NETLINK);
return (true);
}
struct mbuf *m = m_getm2(NULL, datalen, ns->malloc_flag, MT_DATA, M_PKTHDR);
if (__predict_false(m == NULL)) {
free(buf, M_NETLINK);
return (false);
}
bool success = m_append(m, datalen, buf) != 0;
free(buf, M_NETLINK);
if (!success)
return (false);
nl_send_group(m, cnt, ns->arg_uint >> 16, ns->arg_uint & 0xFFFF);
return (true);
}
static bool
nlmsg_write_chain_buf(struct nlmsg_state *ns, void *buf, int datalen, int cnt)
{
struct mbuf **m0 = (struct mbuf **)(ns->arg_ptr);
NL_LOG(LOG_DEBUG2, "IN: ptr: %p len: %d arg: %p", buf, datalen, ns->arg_ptr);
if (__predict_false(datalen == 0)) {
free(buf, M_NETLINK);
return (true);
}
if (*m0 == NULL) {
struct mbuf *m;
m = m_getm2(NULL, datalen, ns->malloc_flag, MT_DATA, M_PKTHDR);
if (__predict_false(m == NULL)) {
free(buf, M_NETLINK);
return (false);
}
*m0 = m;
}
if (__predict_false(m_append(*m0, datalen, buf) == 0)) {
free(buf, M_NETLINK);
return (false);
}
return (true);
}
/*
* NS_WRITER_TYPE_MBUF
* Writes message to the allocated mbuf,
* flushing to socket/group when mbuf size limit is reached.
* This is the most efficient mechanism as it avoids double-copying.
*
* Allocates a single mbuf suitable to store up to @size bytes of data.
* If size < MHLEN (around 160 bytes), allocates mbuf with pkghdr
* If size <= MCLBYTES (2k), allocate a single mbuf cluster
* Otherwise, return NULL.
*/
static bool
nlmsg_get_ns_mbuf(struct nlmsg_state *ns, int size, bool waitok)
{
struct mbuf *m;
int mflag = waitok ? M_WAITOK : M_NOWAIT;
m = m_get2(size, mflag, MT_DATA, M_PKTHDR);
if (__predict_false(m == NULL))
return (false);
ns->alloc_len = M_TRAILINGSPACE(m);
ns->offset = 0;
ns->hdr = NULL;
ns->_storage = (void *)m;
ns->data = mtod(m, void *);
ns->writer_type = NS_WRITER_TYPE_MBUF;
ns->malloc_flag = mflag;
ns->num_messages = 0;
ns->enomem = false;
NL_LOG(LOG_DEBUG2, "alloc mbuf %p req_len %d alloc_len %d data_ptr %p",
m, size, ns->alloc_len, ns->data);
return (true);
}
static bool
nlmsg_write_socket_mbuf(struct nlmsg_state *ns, void *buf, int datalen, int cnt)
{
struct mbuf *m = (struct mbuf *)buf;
NL_LOG(LOG_DEBUG2, "IN: ptr: %p len: %d arg: %p", buf, datalen, ns->arg_ptr);
if (__predict_false(datalen == 0)) {
m_freem(m);
return (true);
}
m->m_pkthdr.len = datalen;
m->m_len = datalen;
int io_flags = (ns->ignore_limit) ? NL_IOF_IGNORE_LIMIT : 0;
return (nl_send_one(m, (struct nlpcb *)(ns->arg_ptr), cnt, io_flags));
}
static bool
nlmsg_write_group_mbuf(struct nlmsg_state *ns, void *buf, int datalen, int cnt)
{
struct mbuf *m = (struct mbuf *)buf;
NL_LOG(LOG_DEBUG2, "IN: ptr: %p len: %d arg: %p", buf, datalen, ns->arg_ptr);
if (__predict_false(datalen == 0)) {
m_freem(m);
return (true);
}
m->m_pkthdr.len = datalen;
m->m_len = datalen;
nl_send_group(m, cnt, ns->arg_uint >> 16, ns->arg_uint & 0xFFFF);
return (true);
}
static bool
nlmsg_write_chain_mbuf(struct nlmsg_state *ns, void *buf, int datalen, int cnt)
{
struct mbuf *m_new = (struct mbuf *)buf;
struct mbuf **m0 = (struct mbuf **)(ns->arg_ptr);
NL_LOG(LOG_DEBUG2, "IN: ptr: %p len: %d arg: %p", buf, datalen, ns->arg_ptr);
if (__predict_false(datalen == 0)) {
m_freem(m_new);
return (true);
}
m_new->m_pkthdr.len = datalen;
m_new->m_len = datalen;
if (*m0 == NULL) {
*m0 = m_new;
} else {
struct mbuf *m_last;
for (m_last = *m0; m_last->m_next != NULL; m_last = m_last->m_next)
;
m_last->m_next = m_new;
(*m0)->m_pkthdr.len += datalen;
}
return (true);
}
/*
* NS_WRITER_TYPE_LBUF
* Writes message to the allocated memory buffer,
* flushing to socket/group when mbuf size limit is reached.
* Calls linux handler to rewrite messages before sending to the socket.
*/
static bool
nlmsg_get_ns_lbuf(struct nlmsg_state *ns, int size, bool waitok)
{
int mflag = waitok ? M_WAITOK : M_NOWAIT;
size = roundup2(size, sizeof(void *));
int add_size = sizeof(struct linear_buffer) + SCRATCH_BUFFER_SIZE;
char *buf = malloc(add_size + size * 2, M_NETLINK, mflag | M_ZERO);
if (__predict_false(buf == NULL))
return (false);
/* Fill buffer header first */
struct linear_buffer *lb = (struct linear_buffer *)buf;
lb->base = &buf[sizeof(struct linear_buffer) + size];
lb->size = size + SCRATCH_BUFFER_SIZE;
ns->alloc_len = size;
ns->offset = 0;
ns->hdr = NULL;
ns->_storage = buf;
ns->data = (char *)(lb + 1);
ns->malloc_flag = mflag;
ns->writer_type = NS_WRITER_TYPE_LBUF;
ns->num_messages = 0;
ns->enomem = false;
return (true);
}
static bool
nlmsg_write_socket_lbuf(struct nlmsg_state *ns, void *buf, int datalen, int cnt)
{
struct linear_buffer *lb = (struct linear_buffer *)buf;
char *data = (char *)(lb + 1);
struct nlpcb *nlp = (struct nlpcb *)(ns->arg_ptr);
if (__predict_false(datalen == 0)) {
free(buf, M_NETLINK);
return (true);
}
struct mbuf *m = NULL;
if (linux_netlink_p != NULL)
m = linux_netlink_p->msgs_to_linux(nlp->nl_proto, data, datalen, nlp);
free(buf, M_NETLINK);
if (__predict_false(m == NULL)) {
/* XXX: should we set sorcverr? */
return (false);
}
int io_flags = (ns->ignore_limit) ? NL_IOF_IGNORE_LIMIT : 0;
return (nl_send_one(m, nlp, cnt, io_flags));
}
/* Shouldn't be called (maybe except Linux code originating message) */
static bool
nlmsg_write_group_lbuf(struct nlmsg_state *ns, void *buf, int datalen, int cnt)
{
struct linear_buffer *lb = (struct linear_buffer *)buf;
char *data = (char *)(lb + 1);
if (__predict_false(datalen == 0)) {
free(buf, M_NETLINK);
return (true);
}
struct mbuf *m = m_getm2(NULL, datalen, ns->malloc_flag, MT_DATA, M_PKTHDR);
if (__predict_false(m == NULL)) {
free(buf, M_NETLINK);
return (false);
}
m_append(m, datalen, data);
free(buf, M_NETLINK);
nl_send_group(m, cnt, ns->arg_uint >> 16, ns->arg_uint & 0xFFFF);
return (true);
}
struct nlwriter_ops nlmsg_writers[] = {
/* NS_WRITER_TYPE_MBUF */
{
.init = nlmsg_get_ns_mbuf,
.write_socket = nlmsg_write_socket_mbuf,
.write_group = nlmsg_write_group_mbuf,
.write_chain = nlmsg_write_chain_mbuf,
},
/* NS_WRITER_TYPE_BUF */
{
.init = nlmsg_get_ns_buf,
.write_socket = nlmsg_write_socket_buf,
.write_group = nlmsg_write_group_buf,
.write_chain = nlmsg_write_chain_buf,
},
/* NS_WRITER_TYPE_LBUF */
{
.init = nlmsg_get_ns_lbuf,
.write_socket = nlmsg_write_socket_lbuf,
.write_group = nlmsg_write_group_lbuf,
},
};
static void
nlmsg_set_callback(struct nlmsg_state *ns)
{
struct nlwriter_ops *pops = &nlmsg_writers[ns->writer_type];
switch (ns->writer_target) {
case NS_WRITER_TARGET_SOCKET:
ns->cb = pops->write_socket;
break;
case NS_WRITER_TARGET_GROUP:
ns->cb = pops->write_group;
break;
case NS_WRITER_TARGET_CHAIN:
ns->cb = pops->write_chain;
break;
default:
panic("not implemented");
}
}
static bool
nlmsg_get_buf_type(struct nlmsg_state *ns, int size, int type, bool waitok)
{
MPASS(type + 1 <= sizeof(nlmsg_writers) / sizeof(nlmsg_writers[0]));
NL_LOG(LOG_DEBUG3, "Setting up ns %p size %d type %d", ns, size, type);
return (nlmsg_writers[type].init(ns, size, waitok));
}
static bool
nlmsg_get_buf(struct nlmsg_state *ns, int size, bool waitok, bool is_linux)
{
int type;
if (!is_linux) {
if (__predict_true(size <= MCLBYTES))
type = NS_WRITER_TYPE_MBUF;
else
type = NS_WRITER_TYPE_BUF;
} else
type = NS_WRITER_TYPE_LBUF;
return (nlmsg_get_buf_type(ns, size, type, waitok));
}
bool
nlmsg_get_socket_writer(int size, struct nlpcb *nlp, struct nlmsg_state *ns)
{
if (!nlmsg_get_buf(ns, size, false, nlp->nl_linux))
return (false);
ns->arg_ptr = (void *)nlp;
ns->writer_target = NS_WRITER_TARGET_SOCKET;
nlmsg_set_callback(ns);
return (true);
}
bool
nlmsg_get_group_writer(int size, int protocol, int group_id, struct nlmsg_state *ns)
{
if (!nlmsg_get_buf(ns, size, false, false))
return (false);
ns->arg_uint = (uint64_t)protocol << 16 | (uint64_t)group_id;
ns->writer_target = NS_WRITER_TARGET_GROUP;
nlmsg_set_callback(ns);
return (true);
}
bool
nlmsg_get_chain_writer(int size, struct mbuf **pm, struct nlmsg_state *ns)
{
if (!nlmsg_get_buf(ns, size, false, false))
return (false);
*pm = NULL;
ns->arg_ptr = (void *)pm;
ns->writer_target = NS_WRITER_TARGET_CHAIN;
nlmsg_set_callback(ns);
NL_LOG(LOG_DEBUG3, "setup cb %p (need %p)", ns->cb, &nlmsg_write_chain_mbuf);
return (true);
}
void
nlmsg_ignore_limit(struct nlmsg_state *ns)
{
ns->ignore_limit = true;
}
bool
nlmsg_flush(struct nlmsg_state *ns)
{
if (__predict_false(ns->hdr != NULL)) {
/* Last message has not been completed, skip it. */
int completed_len = (char *)ns->hdr - ns->data;
/* Send completed messages */
ns->offset -= ns->offset - completed_len;
ns->hdr = NULL;
}
NL_LOG(LOG_DEBUG2, "OUT");
bool result = ns->cb(ns, ns->_storage, ns->offset, ns->num_messages);
ns->_storage = NULL;
if (!result) {
NL_LOG(LOG_DEBUG, "ns %p offset %d: flush with %p() failed", ns, ns->offset, ns->cb);
}
return (result);
}
/*
* Flushes previous data and allocates new underlying storage
* sufficient for holding at least @required_len bytes.
* Return true on success.
*/
bool
nlmsg_refill_buffer(struct nlmsg_state *ns, int required_len)
{
struct nlmsg_state ns_new = {};
int completed_len, new_len;
if (ns->enomem)
return (false);
NL_LOG(LOG_DEBUG3, "no space at offset %d/%d (want %d), trying to reclaim",
ns->offset, ns->alloc_len, required_len);
/* Calculated new buffer size and allocate it s*/
completed_len = (ns->hdr != NULL) ? (char *)ns->hdr - ns->data : ns->offset;
if (completed_len > 0 && required_len < MCLBYTES) {
/* We already ran out of space, use the largest effective size */
new_len = max(ns->alloc_len, MCLBYTES);
} else {
if (ns->alloc_len < MCLBYTES)
new_len = MCLBYTES;
else
new_len = ns->alloc_len * 2;
while (new_len < required_len)
new_len *= 2;
}
bool waitok = (ns->malloc_flag == M_WAITOK);
bool is_linux = (ns->writer_type == NS_WRITER_TYPE_LBUF);
if (!nlmsg_get_buf(&ns_new, new_len, waitok, is_linux)) {
ns->enomem = true;
NL_LOG(LOG_DEBUG, "getting new buf failed, setting ENOMEM");
return (false);
}
if (ns->ignore_limit)
nlmsg_ignore_limit(&ns_new);
/* Update callback data */
ns_new.writer_target = ns->writer_target;
nlmsg_set_callback(&ns_new);
ns_new.arg_uint = ns->arg_uint;
/* Copy last (unfinished) header to the new storage */
int last_len = ns->offset - completed_len;
if (last_len > 0) {
memcpy(ns_new.data, ns->hdr, last_len);
ns_new.hdr = (struct nlmsghdr *)ns_new.data;
ns_new.offset = last_len;
}
NL_LOG(LOG_DEBUG2, "completed: %d bytes, copied: %d bytes", completed_len, last_len);
/* Flush completed headers & switch to the new ns */
nlmsg_flush(ns);
memcpy(ns, &ns_new, sizeof(struct nlmsg_state));
NL_LOG(LOG_DEBUG2, "switched buffer: used %d/%d bytes", ns->offset, ns->alloc_len);
return (true);
}
bool
nlmsg_add(struct nlmsg_state *ns, uint32_t portid, uint32_t seq, uint16_t type,
uint16_t flags, uint32_t len)
{
struct nlmsghdr *hdr;
MPASS(ns->hdr == NULL);
int required_len = NETLINK_ALIGN(len + sizeof(struct nlmsghdr));
if (__predict_false(ns->offset + required_len > ns->alloc_len)) {
if (!nlmsg_refill_buffer(ns, required_len))
return (false);
}
hdr = (struct nlmsghdr *)(&ns->data[ns->offset]);
hdr->nlmsg_len = len;
hdr->nlmsg_type = type;
hdr->nlmsg_flags = flags;
hdr->nlmsg_seq = seq;
hdr->nlmsg_pid = portid;
ns->hdr = hdr;
ns->offset += sizeof(struct nlmsghdr);
return (true);
}
bool
nlmsg_end(struct nlmsg_state *ns)
{
MPASS(ns->hdr != NULL);
if (ns->enomem) {
NL_LOG(LOG_DEBUG, "ENOMEM when dumping message");
nlmsg_abort(ns);
return (false);
}
ns->hdr->nlmsg_len = (uint32_t)(ns->data + ns->offset - (char *)ns->hdr);
ns->hdr = NULL;
ns->num_messages++;
return (true);
}
void
nlmsg_abort(struct nlmsg_state *ns)
{
if (ns->hdr != NULL) {
ns->offset = (uint32_t)((char *)ns->hdr - ns->data);
ns->hdr = NULL;
}
}
void
nlmsg_ack(struct nlpcb *nlp, int error, struct nlmsghdr *hdr,
struct netlink_parse_tracker *npt)
{
struct nlmsgerr *errmsg;
int payload_len;
uint32_t flags = nlp->nl_flags;
struct nlmsg_state *ns = npt->ns;
bool cap_ack;
payload_len = sizeof(struct nlmsgerr);
/*
* The only case when we send the full message in the
* reply is when there is an error and NETLINK_CAP_ACK
* is not set.
*/
cap_ack = (error == 0) || (flags & NLF_CAP_ACK);
if (!cap_ack)
payload_len += hdr->nlmsg_len - sizeof(struct nlmsghdr);
payload_len = NETLINK_ALIGN(payload_len);
uint16_t nl_flags = cap_ack ? NLM_F_CAPPED : 0;
if ((npt->err_msg || npt->err_off) && nlp->nl_flags & NLF_EXT_ACK)
nl_flags |= NLM_F_ACK_TLVS;
/*
* TODO: handle cookies
*/
NL_LOG(LOG_DEBUG3, "acknowledging message type %d seq %d",
hdr->nlmsg_type, hdr->nlmsg_seq);
if (!nlmsg_add(ns, nlp->nl_port, hdr->nlmsg_seq, NLMSG_ERROR, nl_flags, payload_len))
goto enomem;
errmsg = nlmsg_reserve_data(ns, payload_len, struct nlmsgerr);
errmsg->error = error;
/* In case of error copy the whole message, else just the header */
memcpy(&errmsg->msg, hdr, cap_ack ? sizeof(*hdr) : hdr->nlmsg_len);
if (npt->err_msg != NULL && nlp->nl_flags & NLF_EXT_ACK)
nlattr_add_string(ns, NLMSGERR_ATTR_MSG, npt->err_msg);
if (npt->err_off != 0 && nlp->nl_flags & NLF_EXT_ACK)
nlattr_add_u32(ns, NLMSGERR_ATTR_OFFS, npt->err_off);
if (nlmsg_end(ns))
return;
enomem:
NLP_LOG(LOG_DEBUG, nlp, "error allocating ack data for message %d seq %u",
hdr->nlmsg_type, hdr->nlmsg_seq);
nlmsg_abort(ns);
}
bool
nlmsg_end_dump(struct nlmsg_state *ns, int error, struct nlmsghdr *hdr)
{
if (!nlmsg_add(ns, hdr->nlmsg_pid, hdr->nlmsg_seq, NLMSG_DONE, 0, sizeof(int))) {
NL_LOG(LOG_DEBUG, "Error finalizing table dump");
return (false);
}
/* Save operation result */
int *perror = nlmsg_reserve_object(ns, int);
NL_LOG(LOG_DEBUG2, "record error=%d at off %d (%p)", error,
ns->offset, perror);
*perror = error;
nlmsg_end(ns);
return (true);
}