2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
24 #include <netinet/in.h>
25 #include <sys/socket.h>
30 #include <sys/ioctl.h>
34 #include "classifier.h"
37 #include "dpif-provider.h"
39 #include "dynamic-string.h"
43 #include "meta-flow.h"
45 #include "netdev-vport.h"
47 #include "odp-execute.h"
49 #include "ofp-print.h"
52 #include "poll-loop.h"
62 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
64 /* By default, choose a priority in the middle. */
65 #define NETDEV_RULE_PRIORITY 0x8000
67 /* Configuration parameters. */
68 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
70 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
71 * headers to be aligned on a 4-byte boundary. */
72 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
75 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
76 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
77 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
78 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
80 /* Protects against changes to 'dp_netdevs'. */
81 static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
83 /* Contains all 'struct dp_netdev's. */
84 static struct shash dp_netdevs OVS_GUARDED_BY(dp_netdev_mutex)
85 = SHASH_INITIALIZER(&dp_netdevs);
87 struct dp_netdev_upcall {
88 struct dpif_upcall upcall; /* Queued upcall information. */
89 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
92 /* A queue passing packets from a struct dp_netdev to its clients.
98 * Any access at all requires the owning 'dp_netdev''s queue_mutex. */
99 struct dp_netdev_queue {
100 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN] OVS_GUARDED;
101 unsigned int head OVS_GUARDED;
102 unsigned int tail OVS_GUARDED;
105 /* Datapath based on the network device interface from netdev.h.
111 * Some members, marked 'const', are immutable. Accessing other members
112 * requires synchronization, as noted in more detail below.
114 * Acquisition order is, from outermost to innermost:
116 * dp_netdev_mutex (global)
123 const struct dpif_class *const class;
124 const char *const name;
125 struct ovs_refcount ref_cnt;
126 atomic_flag destroyed;
130 * Readers of 'cls' and 'flow_table' must take a 'cls->rwlock' read lock.
132 * Writers of 'cls' and 'flow_table' must take the 'flow_mutex' and then
133 * the 'cls->rwlock' write lock. (The outer 'flow_mutex' allows writers to
134 * atomically perform multiple operations on 'cls' and 'flow_table'.)
136 struct ovs_mutex flow_mutex;
137 struct classifier cls; /* Classifier. Protected by cls.rwlock. */
138 struct hmap flow_table OVS_GUARDED; /* Flow table. */
142 * Everything in 'queues' is protected by 'queue_mutex'. */
143 struct ovs_mutex queue_mutex;
144 struct dp_netdev_queue queues[N_QUEUES];
145 struct seq *queue_seq; /* Incremented whenever a packet is queued. */
149 * ovsthread_counter is internally synchronized. */
150 struct ovsthread_counter *n_hit; /* Number of flow table matches. */
151 struct ovsthread_counter *n_missed; /* Number of flow table misses. */
152 struct ovsthread_counter *n_lost; /* Number of misses not passed up. */
156 * Any lookup into 'ports' or any access to the dp_netdev_ports found
157 * through 'ports' requires taking 'port_rwlock'. */
158 struct ovs_rwlock port_rwlock;
159 struct hmap ports OVS_GUARDED;
160 struct seq *port_seq; /* Incremented whenever a port changes. */
163 static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *dp,
165 OVS_REQ_RDLOCK(dp->port_rwlock);
167 /* A port in a netdev-based datapath. */
168 struct dp_netdev_port {
169 struct hmap_node node; /* Node in dp_netdev's 'ports'. */
171 struct netdev *netdev;
172 struct netdev_saved_flags *sf;
173 struct netdev_rx *rx;
174 char *type; /* Port type as requested by user. */
177 /* A flow in dp_netdev's 'flow_table'.
183 * Except near the beginning or ending of its lifespan, rule 'rule' belongs to
184 * its dp_netdev's classifier. The text below calls this classifier 'cls'.
189 * The thread safety rules described here for "struct dp_netdev_flow" are
190 * motivated by two goals:
192 * - Prevent threads that read members of "struct dp_netdev_flow" from
193 * reading bad data due to changes by some thread concurrently modifying
196 * - Prevent two threads making changes to members of a given "struct
197 * dp_netdev_flow" from interfering with each other.
203 * A flow 'flow' may be accessed without a risk of being freed by code that
204 * holds a read-lock or write-lock on 'cls->rwlock' or that owns a reference to
205 * 'flow->ref_cnt' (or both). Code that needs to hold onto a flow for a while
206 * should take 'cls->rwlock', find the flow it needs, increment 'flow->ref_cnt'
207 * with dpif_netdev_flow_ref(), and drop 'cls->rwlock'.
209 * 'flow->ref_cnt' protects 'flow' from being freed. It doesn't protect the
210 * flow from being deleted from 'cls' (that's 'cls->rwlock') and it doesn't
211 * protect members of 'flow' from modification (that's 'flow->mutex').
213 * 'flow->mutex' protects the members of 'flow' from modification. It doesn't
214 * protect the flow from being deleted from 'cls' (that's 'cls->rwlock') and it
215 * doesn't prevent the flow from being freed (that's 'flow->ref_cnt').
217 * Some members, marked 'const', are immutable. Accessing other members
218 * requires synchronization, as noted in more detail below.
220 struct dp_netdev_flow {
221 /* Packet classification. */
222 const struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
224 /* Hash table index by unmasked flow. */
225 const struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
226 const struct flow flow; /* The flow that created this entry. */
228 /* Number of references.
229 * The classifier owns one reference.
230 * Any thread trying to keep a rule from being freed should hold its own
232 struct ovs_refcount ref_cnt;
234 /* Protects members marked OVS_GUARDED.
236 * Acquire after datapath's flow_mutex. */
237 struct ovs_mutex mutex OVS_ACQ_AFTER(dp_netdev_mutex);
241 * Reading or writing these members requires 'mutex'. */
242 long long int used OVS_GUARDED; /* Last used time, in monotonic msecs. */
243 long long int packet_count OVS_GUARDED; /* Number of packets matched. */
244 long long int byte_count OVS_GUARDED; /* Number of bytes matched. */
245 uint16_t tcp_flags OVS_GUARDED; /* Bitwise-OR of seen tcp_flags values. */
249 * Reading 'actions' requires 'mutex'.
250 * Writing 'actions' requires 'mutex' and (to allow for transactions) the
251 * datapath's flow_mutex. */
252 struct dp_netdev_actions *actions OVS_GUARDED;
255 static struct dp_netdev_flow *dp_netdev_flow_ref(
256 const struct dp_netdev_flow *);
257 static void dp_netdev_flow_unref(struct dp_netdev_flow *);
259 /* A set of datapath actions within a "struct dp_netdev_flow".
265 * A struct dp_netdev_actions 'actions' may be accessed without a risk of being
266 * freed by code that holds a read-lock or write-lock on 'flow->mutex' (where
267 * 'flow' is the dp_netdev_flow for which 'flow->actions == actions') or that
268 * owns a reference to 'actions->ref_cnt' (or both). */
269 struct dp_netdev_actions {
270 struct ovs_refcount ref_cnt;
272 /* These members are immutable: they do not change during the struct's
274 struct nlattr *actions; /* Sequence of OVS_ACTION_ATTR_* attributes. */
275 unsigned int size; /* Size of 'actions', in bytes. */
278 struct dp_netdev_actions *dp_netdev_actions_create(const struct nlattr *,
280 struct dp_netdev_actions *dp_netdev_actions_ref(
281 const struct dp_netdev_actions *);
282 void dp_netdev_actions_unref(struct dp_netdev_actions *);
284 /* Interface to netdev-based datapath. */
287 struct dp_netdev *dp;
288 uint64_t last_port_seq;
291 static int get_port_by_number(struct dp_netdev *dp, odp_port_t port_no,
292 struct dp_netdev_port **portp)
293 OVS_REQ_RDLOCK(dp->port_rwlock);
294 static int get_port_by_name(struct dp_netdev *dp, const char *devname,
295 struct dp_netdev_port **portp)
296 OVS_REQ_RDLOCK(dp->port_rwlock);
297 static void dp_netdev_free(struct dp_netdev *)
298 OVS_REQUIRES(dp_netdev_mutex);
299 static void dp_netdev_flow_flush(struct dp_netdev *);
300 static int do_add_port(struct dp_netdev *dp, const char *devname,
301 const char *type, odp_port_t port_no)
302 OVS_REQ_WRLOCK(dp->port_rwlock);
303 static int do_del_port(struct dp_netdev *dp, odp_port_t port_no)
304 OVS_REQ_WRLOCK(dp->port_rwlock);
305 static int dpif_netdev_open(const struct dpif_class *, const char *name,
306 bool create, struct dpif **);
307 static int dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *,
308 int queue_no, const struct flow *,
309 const struct nlattr *userdata)
310 OVS_EXCLUDED(dp->queue_mutex);
311 static void dp_netdev_execute_actions(struct dp_netdev *dp,
312 const struct flow *, struct ofpbuf *,
313 struct pkt_metadata *,
314 const struct nlattr *actions,
316 OVS_REQ_RDLOCK(dp->port_rwlock);
317 static void dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
318 struct pkt_metadata *)
319 OVS_REQ_RDLOCK(dp->port_rwlock);
321 static struct dpif_netdev *
322 dpif_netdev_cast(const struct dpif *dpif)
324 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
325 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
328 static struct dp_netdev *
329 get_dp_netdev(const struct dpif *dpif)
331 return dpif_netdev_cast(dpif)->dp;
335 dpif_netdev_enumerate(struct sset *all_dps)
337 struct shash_node *node;
339 ovs_mutex_lock(&dp_netdev_mutex);
340 SHASH_FOR_EACH(node, &dp_netdevs) {
341 sset_add(all_dps, node->name);
343 ovs_mutex_unlock(&dp_netdev_mutex);
349 dpif_netdev_class_is_dummy(const struct dpif_class *class)
351 return class != &dpif_netdev_class;
355 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
357 return strcmp(type, "internal") ? type
358 : dpif_netdev_class_is_dummy(class) ? "dummy"
363 create_dpif_netdev(struct dp_netdev *dp)
365 uint16_t netflow_id = hash_string(dp->name, 0);
366 struct dpif_netdev *dpif;
368 ovs_refcount_ref(&dp->ref_cnt);
370 dpif = xmalloc(sizeof *dpif);
371 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
373 dpif->last_port_seq = seq_read(dp->port_seq);
378 /* Choose an unused, non-zero port number and return it on success.
379 * Return ODPP_NONE on failure. */
381 choose_port(struct dp_netdev *dp, const char *name)
382 OVS_REQ_RDLOCK(dp->port_rwlock)
386 if (dp->class != &dpif_netdev_class) {
390 /* If the port name begins with "br", start the number search at
391 * 100 to make writing tests easier. */
392 if (!strncmp(name, "br", 2)) {
396 /* If the port name contains a number, try to assign that port number.
397 * This can make writing unit tests easier because port numbers are
399 for (p = name; *p != '\0'; p++) {
400 if (isdigit((unsigned char) *p)) {
401 port_no = start_no + strtol(p, NULL, 10);
402 if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE)
403 && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
404 return u32_to_odp(port_no);
411 for (port_no = 1; port_no <= UINT16_MAX; port_no++) {
412 if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
413 return u32_to_odp(port_no);
421 create_dp_netdev(const char *name, const struct dpif_class *class,
422 struct dp_netdev **dpp)
423 OVS_REQUIRES(dp_netdev_mutex)
425 struct dp_netdev *dp;
429 dp = xzalloc(sizeof *dp);
430 shash_add(&dp_netdevs, name, dp);
432 *CONST_CAST(const struct dpif_class **, &dp->class) = class;
433 *CONST_CAST(const char **, &dp->name) = xstrdup(name);
434 ovs_refcount_init(&dp->ref_cnt);
435 atomic_flag_init(&dp->destroyed);
437 ovs_mutex_init(&dp->flow_mutex);
438 classifier_init(&dp->cls, NULL);
439 hmap_init(&dp->flow_table);
441 ovs_mutex_init(&dp->queue_mutex);
442 ovs_mutex_lock(&dp->queue_mutex);
443 for (i = 0; i < N_QUEUES; i++) {
444 dp->queues[i].head = dp->queues[i].tail = 0;
446 ovs_mutex_unlock(&dp->queue_mutex);
447 dp->queue_seq = seq_create();
449 dp->n_hit = ovsthread_counter_create();
450 dp->n_missed = ovsthread_counter_create();
451 dp->n_lost = ovsthread_counter_create();
453 ovs_rwlock_init(&dp->port_rwlock);
454 hmap_init(&dp->ports);
455 dp->port_seq = seq_create();
457 ovs_rwlock_wrlock(&dp->port_rwlock);
458 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
459 ovs_rwlock_unlock(&dp->port_rwlock);
470 dpif_netdev_open(const struct dpif_class *class, const char *name,
471 bool create, struct dpif **dpifp)
473 struct dp_netdev *dp;
476 ovs_mutex_lock(&dp_netdev_mutex);
477 dp = shash_find_data(&dp_netdevs, name);
479 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
481 error = (dp->class != class ? EINVAL
486 *dpifp = create_dpif_netdev(dp);
488 ovs_mutex_unlock(&dp_netdev_mutex);
494 dp_netdev_purge_queues(struct dp_netdev *dp)
498 ovs_mutex_lock(&dp->queue_mutex);
499 for (i = 0; i < N_QUEUES; i++) {
500 struct dp_netdev_queue *q = &dp->queues[i];
502 while (q->tail != q->head) {
503 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
504 ofpbuf_uninit(&u->upcall.packet);
505 ofpbuf_uninit(&u->buf);
508 ovs_mutex_unlock(&dp->queue_mutex);
511 /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp'
512 * through the 'dp_netdevs' shash while freeing 'dp'. */
514 dp_netdev_free(struct dp_netdev *dp)
515 OVS_REQUIRES(dp_netdev_mutex)
517 struct dp_netdev_port *port, *next;
519 shash_find_and_delete(&dp_netdevs, dp->name);
521 dp_netdev_flow_flush(dp);
522 ovs_rwlock_wrlock(&dp->port_rwlock);
523 HMAP_FOR_EACH_SAFE (port, next, node, &dp->ports) {
524 do_del_port(dp, port->port_no);
526 ovs_rwlock_unlock(&dp->port_rwlock);
527 ovsthread_counter_destroy(dp->n_hit);
528 ovsthread_counter_destroy(dp->n_missed);
529 ovsthread_counter_destroy(dp->n_lost);
531 dp_netdev_purge_queues(dp);
532 seq_destroy(dp->queue_seq);
533 ovs_mutex_destroy(&dp->queue_mutex);
535 classifier_destroy(&dp->cls);
536 hmap_destroy(&dp->flow_table);
537 ovs_mutex_destroy(&dp->flow_mutex);
538 seq_destroy(dp->port_seq);
539 hmap_destroy(&dp->ports);
540 atomic_flag_destroy(&dp->destroyed);
541 ovs_refcount_destroy(&dp->ref_cnt);
542 free(CONST_CAST(char *, dp->name));
547 dp_netdev_unref(struct dp_netdev *dp)
550 /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't
551 * get a new reference to 'dp' through the 'dp_netdevs' shash. */
552 ovs_mutex_lock(&dp_netdev_mutex);
553 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
556 ovs_mutex_unlock(&dp_netdev_mutex);
561 dpif_netdev_close(struct dpif *dpif)
563 struct dp_netdev *dp = get_dp_netdev(dpif);
570 dpif_netdev_destroy(struct dpif *dpif)
572 struct dp_netdev *dp = get_dp_netdev(dpif);
574 if (!atomic_flag_test_and_set(&dp->destroyed)) {
575 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
576 /* Can't happen: 'dpif' still owns a reference to 'dp'. */
585 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
587 struct dp_netdev *dp = get_dp_netdev(dpif);
589 ovs_rwlock_rdlock(&dp->cls.rwlock);
590 stats->n_flows = hmap_count(&dp->flow_table);
591 ovs_rwlock_unlock(&dp->cls.rwlock);
593 stats->n_hit = ovsthread_counter_read(dp->n_hit);
594 stats->n_missed = ovsthread_counter_read(dp->n_missed);
595 stats->n_lost = ovsthread_counter_read(dp->n_lost);
596 stats->n_masks = UINT32_MAX;
597 stats->n_mask_hit = UINT64_MAX;
603 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
605 OVS_REQ_WRLOCK(dp->port_rwlock)
607 struct netdev_saved_flags *sf;
608 struct dp_netdev_port *port;
609 struct netdev *netdev;
610 struct netdev_rx *rx;
611 enum netdev_flags flags;
612 const char *open_type;
615 /* XXX reject devices already in some dp_netdev. */
617 /* Open and validate network device. */
618 open_type = dpif_netdev_port_open_type(dp->class, type);
619 error = netdev_open(devname, open_type, &netdev);
623 /* XXX reject non-Ethernet devices */
625 netdev_get_flags(netdev, &flags);
626 if (flags & NETDEV_LOOPBACK) {
627 VLOG_ERR("%s: cannot add a loopback device", devname);
628 netdev_close(netdev);
632 error = netdev_rx_open(netdev, &rx);
634 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
635 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
636 devname, ovs_strerror(errno));
637 netdev_close(netdev);
641 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
644 netdev_close(netdev);
648 port = xmalloc(sizeof *port);
649 port->port_no = port_no;
650 port->netdev = netdev;
653 port->type = xstrdup(type);
655 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
656 seq_change(dp->port_seq);
662 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
663 odp_port_t *port_nop)
665 struct dp_netdev *dp = get_dp_netdev(dpif);
666 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
667 const char *dpif_port;
671 ovs_rwlock_wrlock(&dp->port_rwlock);
672 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
673 if (*port_nop != ODPP_NONE) {
675 error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0;
677 port_no = choose_port(dp, dpif_port);
678 error = port_no == ODPP_NONE ? EFBIG : 0;
682 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
684 ovs_rwlock_unlock(&dp->port_rwlock);
690 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
692 struct dp_netdev *dp = get_dp_netdev(dpif);
695 ovs_rwlock_wrlock(&dp->port_rwlock);
696 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
697 ovs_rwlock_unlock(&dp->port_rwlock);
703 is_valid_port_number(odp_port_t port_no)
705 return port_no != ODPP_NONE;
708 static struct dp_netdev_port *
709 dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no)
710 OVS_REQ_RDLOCK(dp->port_rwlock)
712 struct dp_netdev_port *port;
714 HMAP_FOR_EACH_IN_BUCKET (port, node, hash_int(odp_to_u32(port_no), 0),
716 if (port->port_no == port_no) {
724 get_port_by_number(struct dp_netdev *dp,
725 odp_port_t port_no, struct dp_netdev_port **portp)
726 OVS_REQ_RDLOCK(dp->port_rwlock)
728 if (!is_valid_port_number(port_no)) {
732 *portp = dp_netdev_lookup_port(dp, port_no);
733 return *portp ? 0 : ENOENT;
738 get_port_by_name(struct dp_netdev *dp,
739 const char *devname, struct dp_netdev_port **portp)
740 OVS_REQ_RDLOCK(dp->port_rwlock)
742 struct dp_netdev_port *port;
744 HMAP_FOR_EACH (port, node, &dp->ports) {
745 if (!strcmp(netdev_get_name(port->netdev), devname)) {
754 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
755 OVS_REQ_WRLOCK(dp->port_rwlock)
757 struct dp_netdev_port *port;
760 error = get_port_by_number(dp, port_no, &port);
765 hmap_remove(&dp->ports, &port->node);
766 seq_change(dp->port_seq);
768 netdev_close(port->netdev);
769 netdev_restore_flags(port->sf);
770 netdev_rx_close(port->rx);
778 answer_port_query(const struct dp_netdev_port *port,
779 struct dpif_port *dpif_port)
781 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
782 dpif_port->type = xstrdup(port->type);
783 dpif_port->port_no = port->port_no;
787 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
788 struct dpif_port *dpif_port)
790 struct dp_netdev *dp = get_dp_netdev(dpif);
791 struct dp_netdev_port *port;
794 ovs_rwlock_rdlock(&dp->port_rwlock);
795 error = get_port_by_number(dp, port_no, &port);
796 if (!error && dpif_port) {
797 answer_port_query(port, dpif_port);
799 ovs_rwlock_unlock(&dp->port_rwlock);
805 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
806 struct dpif_port *dpif_port)
808 struct dp_netdev *dp = get_dp_netdev(dpif);
809 struct dp_netdev_port *port;
812 ovs_rwlock_rdlock(&dp->port_rwlock);
813 error = get_port_by_name(dp, devname, &port);
814 if (!error && dpif_port) {
815 answer_port_query(port, dpif_port);
817 ovs_rwlock_unlock(&dp->port_rwlock);
823 dp_netdev_remove_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
824 OVS_REQ_WRLOCK(dp->cls.rwlock)
825 OVS_REQUIRES(dp->flow_mutex)
827 struct cls_rule *cr = CONST_CAST(struct cls_rule *, &flow->cr);
828 struct hmap_node *node = CONST_CAST(struct hmap_node *, &flow->node);
830 classifier_remove(&dp->cls, cr);
831 hmap_remove(&dp->flow_table, node);
832 dp_netdev_flow_unref(flow);
835 static struct dp_netdev_flow *
836 dp_netdev_flow_ref(const struct dp_netdev_flow *flow_)
838 struct dp_netdev_flow *flow = CONST_CAST(struct dp_netdev_flow *, flow_);
840 ovs_refcount_ref(&flow->ref_cnt);
846 dp_netdev_flow_unref(struct dp_netdev_flow *flow)
848 if (flow && ovs_refcount_unref(&flow->ref_cnt) == 1) {
849 cls_rule_destroy(CONST_CAST(struct cls_rule *, &flow->cr));
850 ovs_mutex_lock(&flow->mutex);
851 dp_netdev_actions_unref(flow->actions);
852 ovs_mutex_unlock(&flow->mutex);
853 ovs_mutex_destroy(&flow->mutex);
859 dp_netdev_flow_flush(struct dp_netdev *dp)
861 struct dp_netdev_flow *netdev_flow, *next;
863 ovs_mutex_lock(&dp->flow_mutex);
864 ovs_rwlock_wrlock(&dp->cls.rwlock);
865 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
866 dp_netdev_remove_flow(dp, netdev_flow);
868 ovs_rwlock_unlock(&dp->cls.rwlock);
869 ovs_mutex_unlock(&dp->flow_mutex);
873 dpif_netdev_flow_flush(struct dpif *dpif)
875 struct dp_netdev *dp = get_dp_netdev(dpif);
877 dp_netdev_flow_flush(dp);
881 struct dp_netdev_port_state {
888 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
890 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
895 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
896 struct dpif_port *dpif_port)
898 struct dp_netdev_port_state *state = state_;
899 struct dp_netdev *dp = get_dp_netdev(dpif);
900 struct hmap_node *node;
903 ovs_rwlock_rdlock(&dp->port_rwlock);
904 node = hmap_at_position(&dp->ports, &state->bucket, &state->offset);
906 struct dp_netdev_port *port;
908 port = CONTAINER_OF(node, struct dp_netdev_port, node);
911 state->name = xstrdup(netdev_get_name(port->netdev));
912 dpif_port->name = state->name;
913 dpif_port->type = port->type;
914 dpif_port->port_no = port->port_no;
920 ovs_rwlock_unlock(&dp->port_rwlock);
926 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
928 struct dp_netdev_port_state *state = state_;
935 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
937 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
938 uint64_t new_port_seq;
941 new_port_seq = seq_read(dpif->dp->port_seq);
942 if (dpif->last_port_seq != new_port_seq) {
943 dpif->last_port_seq = new_port_seq;
953 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
955 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
957 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
960 static struct dp_netdev_flow *
961 dp_netdev_flow_cast(const struct cls_rule *cr)
963 return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL;
966 static struct dp_netdev_flow *
967 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
968 OVS_EXCLUDED(dp->cls.rwlock)
970 struct dp_netdev_flow *netdev_flow;
972 ovs_rwlock_rdlock(&dp->cls.rwlock);
973 netdev_flow = dp_netdev_flow_cast(classifier_lookup(&dp->cls, flow, NULL));
974 dp_netdev_flow_ref(netdev_flow);
975 ovs_rwlock_unlock(&dp->cls.rwlock);
980 static struct dp_netdev_flow *
981 dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
982 OVS_REQ_RDLOCK(dp->cls.rwlock)
984 struct dp_netdev_flow *netdev_flow;
986 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
988 if (flow_equal(&netdev_flow->flow, flow)) {
989 return dp_netdev_flow_ref(netdev_flow);
997 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
998 struct dpif_flow_stats *stats)
999 OVS_REQ_RDLOCK(netdev_flow->mutex)
1001 stats->n_packets = netdev_flow->packet_count;
1002 stats->n_bytes = netdev_flow->byte_count;
1003 stats->used = netdev_flow->used;
1004 stats->tcp_flags = netdev_flow->tcp_flags;
1008 dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1009 const struct nlattr *mask_key,
1010 uint32_t mask_key_len, const struct flow *flow,
1014 if (odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow)) {
1015 /* This should not happen: it indicates that
1016 * odp_flow_key_from_mask() and odp_flow_key_to_mask()
1017 * disagree on the acceptable form of a mask. Log the problem
1018 * as an error, with enough details to enable debugging. */
1019 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1021 if (!VLOG_DROP_ERR(&rl)) {
1025 odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
1027 VLOG_ERR("internal error parsing flow mask %s", ds_cstr(&s));
1033 /* Force unwildcard the in_port. */
1034 mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
1036 enum mf_field_id id;
1037 /* No mask key, unwildcard everything except fields whose
1038 * prerequisities are not met. */
1039 memset(mask, 0x0, sizeof *mask);
1041 for (id = 0; id < MFF_N_IDS; ++id) {
1042 /* Skip registers and metadata. */
1043 if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
1044 && id != MFF_METADATA) {
1045 const struct mf_field *mf = mf_from_id(id);
1046 if (mf_are_prereqs_ok(mf, flow)) {
1047 mf_mask_field(mf, mask);
1057 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1062 if (odp_flow_key_to_flow(key, key_len, flow)) {
1063 /* This should not happen: it indicates that odp_flow_key_from_flow()
1064 * and odp_flow_key_to_flow() disagree on the acceptable form of a
1065 * flow. Log the problem as an error, with enough details to enable
1067 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1069 if (!VLOG_DROP_ERR(&rl)) {
1073 odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
1074 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
1081 in_port = flow->in_port.odp_port;
1082 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
1090 dpif_netdev_flow_get(const struct dpif *dpif,
1091 const struct nlattr *nl_key, size_t nl_key_len,
1092 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
1094 struct dp_netdev *dp = get_dp_netdev(dpif);
1095 struct dp_netdev_flow *netdev_flow;
1099 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
1104 ovs_rwlock_rdlock(&dp->cls.rwlock);
1105 netdev_flow = dp_netdev_find_flow(dp, &key);
1106 ovs_rwlock_unlock(&dp->cls.rwlock);
1109 struct dp_netdev_actions *actions = NULL;
1111 ovs_mutex_lock(&netdev_flow->mutex);
1113 get_dpif_flow_stats(netdev_flow, stats);
1116 actions = dp_netdev_actions_ref(netdev_flow->actions);
1118 ovs_mutex_unlock(&netdev_flow->mutex);
1120 dp_netdev_flow_unref(netdev_flow);
1123 *actionsp = ofpbuf_clone_data(actions->actions, actions->size);
1124 dp_netdev_actions_unref(actions);
1134 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
1135 const struct flow_wildcards *wc,
1136 const struct nlattr *actions,
1138 OVS_REQUIRES(dp->flow_mutex)
1140 struct dp_netdev_flow *netdev_flow;
1143 netdev_flow = xzalloc(sizeof *netdev_flow);
1144 *CONST_CAST(struct flow *, &netdev_flow->flow) = *flow;
1145 ovs_refcount_init(&netdev_flow->ref_cnt);
1147 ovs_mutex_init(&netdev_flow->mutex);
1148 ovs_mutex_lock(&netdev_flow->mutex);
1150 netdev_flow->actions = dp_netdev_actions_create(actions, actions_len);
1152 match_init(&match, flow, wc);
1153 cls_rule_init(CONST_CAST(struct cls_rule *, &netdev_flow->cr),
1154 &match, NETDEV_RULE_PRIORITY);
1155 ovs_rwlock_wrlock(&dp->cls.rwlock);
1156 classifier_insert(&dp->cls,
1157 CONST_CAST(struct cls_rule *, &netdev_flow->cr));
1158 hmap_insert(&dp->flow_table,
1159 CONST_CAST(struct hmap_node *, &netdev_flow->node),
1160 flow_hash(flow, 0));
1161 ovs_rwlock_unlock(&dp->cls.rwlock);
1163 ovs_mutex_unlock(&netdev_flow->mutex);
1169 clear_stats(struct dp_netdev_flow *netdev_flow)
1170 OVS_REQUIRES(netdev_flow->mutex)
1172 netdev_flow->used = 0;
1173 netdev_flow->packet_count = 0;
1174 netdev_flow->byte_count = 0;
1175 netdev_flow->tcp_flags = 0;
1179 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
1181 struct dp_netdev *dp = get_dp_netdev(dpif);
1182 struct dp_netdev_flow *netdev_flow;
1184 struct flow_wildcards wc;
1187 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
1191 error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
1192 put->mask, put->mask_len,
1198 ovs_mutex_lock(&dp->flow_mutex);
1199 netdev_flow = dp_netdev_lookup_flow(dp, &flow);
1201 if (put->flags & DPIF_FP_CREATE) {
1202 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
1204 memset(put->stats, 0, sizeof *put->stats);
1206 error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
1215 if (put->flags & DPIF_FP_MODIFY
1216 && flow_equal(&flow, &netdev_flow->flow)) {
1217 struct dp_netdev_actions *new_actions;
1218 struct dp_netdev_actions *old_actions;
1220 new_actions = dp_netdev_actions_create(put->actions,
1223 ovs_mutex_lock(&netdev_flow->mutex);
1224 old_actions = netdev_flow->actions;
1225 netdev_flow->actions = new_actions;
1227 get_dpif_flow_stats(netdev_flow, put->stats);
1229 if (put->flags & DPIF_FP_ZERO_STATS) {
1230 clear_stats(netdev_flow);
1232 ovs_mutex_unlock(&netdev_flow->mutex);
1234 dp_netdev_actions_unref(old_actions);
1235 } else if (put->flags & DPIF_FP_CREATE) {
1238 /* Overlapping flow. */
1241 dp_netdev_flow_unref(netdev_flow);
1243 ovs_mutex_unlock(&dp->flow_mutex);
1249 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
1251 struct dp_netdev *dp = get_dp_netdev(dpif);
1252 struct dp_netdev_flow *netdev_flow;
1256 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
1261 ovs_mutex_lock(&dp->flow_mutex);
1262 ovs_rwlock_wrlock(&dp->cls.rwlock);
1263 netdev_flow = dp_netdev_find_flow(dp, &key);
1266 ovs_mutex_lock(&netdev_flow->mutex);
1267 get_dpif_flow_stats(netdev_flow, del->stats);
1268 ovs_mutex_unlock(&netdev_flow->mutex);
1270 dp_netdev_remove_flow(dp, netdev_flow);
1274 ovs_rwlock_unlock(&dp->cls.rwlock);
1275 ovs_mutex_unlock(&dp->flow_mutex);
1280 struct dp_netdev_flow_state {
1283 struct dp_netdev_actions *actions;
1284 struct odputil_keybuf keybuf;
1285 struct odputil_keybuf maskbuf;
1286 struct dpif_flow_stats stats;
1290 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
1292 struct dp_netdev_flow_state *state;
1294 *statep = state = xmalloc(sizeof *state);
1297 state->actions = NULL;
1302 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
1303 const struct nlattr **key, size_t *key_len,
1304 const struct nlattr **mask, size_t *mask_len,
1305 const struct nlattr **actions, size_t *actions_len,
1306 const struct dpif_flow_stats **stats)
1308 struct dp_netdev_flow_state *state = state_;
1309 struct dp_netdev *dp = get_dp_netdev(dpif);
1310 struct dp_netdev_flow *netdev_flow;
1311 struct hmap_node *node;
1313 ovs_rwlock_rdlock(&dp->cls.rwlock);
1314 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
1316 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
1317 dp_netdev_flow_ref(netdev_flow);
1319 ovs_rwlock_unlock(&dp->cls.rwlock);
1327 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
1328 odp_flow_key_from_flow(&buf, &netdev_flow->flow,
1329 netdev_flow->flow.in_port.odp_port);
1332 *key_len = buf.size;
1337 struct flow_wildcards wc;
1339 ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
1340 minimask_expand(&netdev_flow->cr.match.mask, &wc);
1341 odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
1342 odp_to_u32(wc.masks.in_port.odp_port));
1345 *mask_len = buf.size;
1348 if (actions || stats) {
1349 dp_netdev_actions_unref(state->actions);
1350 state->actions = NULL;
1352 ovs_mutex_lock(&netdev_flow->mutex);
1354 state->actions = dp_netdev_actions_ref(netdev_flow->actions);
1355 *actions = state->actions->actions;
1356 *actions_len = state->actions->size;
1359 get_dpif_flow_stats(netdev_flow, &state->stats);
1360 *stats = &state->stats;
1362 ovs_mutex_unlock(&netdev_flow->mutex);
1365 dp_netdev_flow_unref(netdev_flow);
1371 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1373 struct dp_netdev_flow_state *state = state_;
1375 dp_netdev_actions_unref(state->actions);
1381 dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute)
1383 struct dp_netdev *dp = get_dp_netdev(dpif);
1384 struct pkt_metadata *md = &execute->md;
1387 if (execute->packet->size < ETH_HEADER_LEN ||
1388 execute->packet->size > UINT16_MAX) {
1392 /* Extract flow key. */
1393 flow_extract(execute->packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1394 (union flow_in_port *)&md->in_port, &key);
1396 ovs_rwlock_rdlock(&dp->port_rwlock);
1397 dp_netdev_execute_actions(dp, &key, execute->packet, md, execute->actions,
1398 execute->actions_len);
1399 ovs_rwlock_unlock(&dp->port_rwlock);
1405 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1411 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1412 uint32_t queue_id, uint32_t *priority)
1414 *priority = queue_id;
1418 static struct dp_netdev_queue *
1419 find_nonempty_queue(struct dp_netdev *dp)
1420 OVS_REQUIRES(dp->queue_mutex)
1424 for (i = 0; i < N_QUEUES; i++) {
1425 struct dp_netdev_queue *q = &dp->queues[i];
1426 if (q->head != q->tail) {
1434 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1437 struct dp_netdev *dp = get_dp_netdev(dpif);
1438 struct dp_netdev_queue *q;
1441 ovs_mutex_lock(&dp->queue_mutex);
1442 q = find_nonempty_queue(dp);
1444 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1446 *upcall = u->upcall;
1455 ovs_mutex_unlock(&dp->queue_mutex);
1461 dpif_netdev_recv_wait(struct dpif *dpif)
1463 struct dp_netdev *dp = get_dp_netdev(dpif);
1466 ovs_mutex_lock(&dp->queue_mutex);
1467 seq = seq_read(dp->queue_seq);
1468 if (find_nonempty_queue(dp)) {
1469 poll_immediate_wake();
1471 seq_wait(dp->queue_seq, seq);
1473 ovs_mutex_unlock(&dp->queue_mutex);
1477 dpif_netdev_recv_purge(struct dpif *dpif)
1479 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1481 dp_netdev_purge_queues(dpif_netdev->dp);
1484 /* Creates and returns a new 'struct dp_netdev_actions', with a reference count
1485 * of 1, whose actions are a copy of from the 'ofpacts_len' bytes of
1487 struct dp_netdev_actions *
1488 dp_netdev_actions_create(const struct nlattr *actions, size_t size)
1490 struct dp_netdev_actions *netdev_actions;
1492 netdev_actions = xmalloc(sizeof *netdev_actions);
1493 ovs_refcount_init(&netdev_actions->ref_cnt);
1494 netdev_actions->actions = xmemdup(actions, size);
1495 netdev_actions->size = size;
1497 return netdev_actions;
1500 /* Increments 'actions''s refcount. */
1501 struct dp_netdev_actions *
1502 dp_netdev_actions_ref(const struct dp_netdev_actions *actions_)
1504 struct dp_netdev_actions *actions;
1506 actions = CONST_CAST(struct dp_netdev_actions *, actions_);
1508 ovs_refcount_ref(&actions->ref_cnt);
1513 /* Decrements 'actions''s refcount and frees 'actions' if the refcount reaches
1516 dp_netdev_actions_unref(struct dp_netdev_actions *actions)
1518 if (actions && ovs_refcount_unref(&actions->ref_cnt) == 1) {
1519 free(actions->actions);
1525 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1526 const struct ofpbuf *packet)
1527 OVS_REQUIRES(netdev_flow->mutex)
1529 netdev_flow->used = time_msec();
1530 netdev_flow->packet_count++;
1531 netdev_flow->byte_count += packet->size;
1532 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
1536 dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
1537 struct pkt_metadata *md)
1538 OVS_REQ_RDLOCK(dp->port_rwlock)
1540 struct dp_netdev_flow *netdev_flow;
1543 if (packet->size < ETH_HEADER_LEN) {
1546 flow_extract(packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1547 (union flow_in_port *)&md->in_port, &key);
1548 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1550 struct dp_netdev_actions *actions;
1552 ovs_mutex_lock(&netdev_flow->mutex);
1553 dp_netdev_flow_used(netdev_flow, packet);
1554 actions = dp_netdev_actions_ref(netdev_flow->actions);
1555 ovs_mutex_unlock(&netdev_flow->mutex);
1557 dp_netdev_execute_actions(dp, &key, packet, md,
1558 actions->actions, actions->size);
1559 dp_netdev_actions_unref(actions);
1560 ovsthread_counter_inc(dp->n_hit, 1);
1562 ovsthread_counter_inc(dp->n_missed, 1);
1563 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1568 dpif_netdev_run(struct dpif *dpif)
1570 struct dp_netdev_port *port;
1571 struct dp_netdev *dp;
1572 struct ofpbuf packet;
1574 dp = get_dp_netdev(dpif);
1575 ofpbuf_init(&packet, 0);
1577 ovs_rwlock_rdlock(&dp->port_rwlock);
1578 HMAP_FOR_EACH (port, node, &dp->ports) {
1583 error = netdev_get_mtu(port->netdev, &mtu);
1585 mtu = ETH_PAYLOAD_MAX;
1587 buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + mtu;
1589 ofpbuf_clear(&packet);
1590 ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM, buf_size);
1592 error = port->rx ? netdev_rx_recv(port->rx, &packet) : EOPNOTSUPP;
1594 struct pkt_metadata md = PKT_METADATA_INITIALIZER(port->port_no);
1595 dp_netdev_port_input(dp, &packet, &md);
1596 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1597 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1599 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1600 netdev_get_name(port->netdev), ovs_strerror(error));
1603 ovs_rwlock_unlock(&dp->port_rwlock);
1605 ofpbuf_uninit(&packet);
1609 dpif_netdev_wait(struct dpif *dpif)
1611 struct dp_netdev_port *port;
1613 /* There is a race here, if thread A calls dpif_netdev_wait(dpif) and
1614 * thread B calls dpif_port_add(dpif) or dpif_port_remove(dpif) before
1615 * A makes it to poll_block().
1617 * But I think it doesn't matter:
1619 * - In the dpif_port_add() case, A will not wake up when a packet
1620 * arrives on the new port, but this would also happen if the
1621 * ordering were reversed.
1623 * - In the dpif_port_remove() case, A might wake up spuriously, but
1624 * that is harmless. */
1626 ovs_mutex_lock(&dp_netdev_mutex);
1627 HMAP_FOR_EACH (port, node, &get_dp_netdev(dpif)->ports) {
1629 netdev_rx_wait(port->rx);
1632 ovs_mutex_unlock(&dp_netdev_mutex);
1636 dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
1637 int queue_no, const struct flow *flow,
1638 const struct nlattr *userdata)
1639 OVS_EXCLUDED(dp->queue_mutex)
1641 struct dp_netdev_queue *q = &dp->queues[queue_no];
1644 ovs_mutex_lock(&dp->queue_mutex);
1645 if (q->head - q->tail < MAX_QUEUE_LEN) {
1646 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1647 struct dpif_upcall *upcall = &u->upcall;
1648 struct ofpbuf *buf = &u->buf;
1651 upcall->type = queue_no;
1653 /* Allocate buffer big enough for everything. */
1654 buf_size = ODPUTIL_FLOW_KEY_BYTES;
1656 buf_size += NLA_ALIGN(userdata->nla_len);
1658 ofpbuf_init(buf, buf_size);
1661 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1662 upcall->key = buf->data;
1663 upcall->key_len = buf->size;
1667 upcall->userdata = ofpbuf_put(buf, userdata,
1668 NLA_ALIGN(userdata->nla_len));
1671 /* Steal packet data. */
1672 ovs_assert(packet->source == OFPBUF_MALLOC);
1673 upcall->packet = *packet;
1674 ofpbuf_use(packet, NULL, 0);
1676 seq_change(dp->queue_seq);
1680 ovsthread_counter_inc(dp->n_lost, 1);
1683 ovs_mutex_unlock(&dp->queue_mutex);
1688 struct dp_netdev_execute_aux {
1689 struct dp_netdev *dp;
1690 const struct flow *key;
1694 dp_execute_cb(void *aux_, struct ofpbuf *packet,
1695 const struct pkt_metadata *md OVS_UNUSED,
1696 const struct nlattr *a, bool may_steal)
1697 OVS_NO_THREAD_SAFETY_ANALYSIS
1699 struct dp_netdev_execute_aux *aux = aux_;
1700 int type = nl_attr_type(a);
1701 struct dp_netdev_port *p;
1703 switch ((enum ovs_action_attr)type) {
1704 case OVS_ACTION_ATTR_OUTPUT:
1705 p = dp_netdev_lookup_port(aux->dp, u32_to_odp(nl_attr_get_u32(a)));
1707 netdev_send(p->netdev, packet);
1711 case OVS_ACTION_ATTR_USERSPACE: {
1712 const struct nlattr *userdata;
1714 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1716 /* Make a copy if we are not allowed to steal the packet's data. */
1718 packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
1720 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1723 ofpbuf_uninit(packet);
1727 case OVS_ACTION_ATTR_PUSH_VLAN:
1728 case OVS_ACTION_ATTR_POP_VLAN:
1729 case OVS_ACTION_ATTR_PUSH_MPLS:
1730 case OVS_ACTION_ATTR_POP_MPLS:
1731 case OVS_ACTION_ATTR_SET:
1732 case OVS_ACTION_ATTR_SAMPLE:
1733 case OVS_ACTION_ATTR_UNSPEC:
1734 case __OVS_ACTION_ATTR_MAX:
1740 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1741 struct ofpbuf *packet, struct pkt_metadata *md,
1742 const struct nlattr *actions, size_t actions_len)
1743 OVS_REQ_RDLOCK(dp->port_rwlock)
1745 struct dp_netdev_execute_aux aux = {dp, key};
1747 odp_execute_actions(&aux, packet, md, actions, actions_len, dp_execute_cb);
1750 const struct dpif_class dpif_netdev_class = {
1752 dpif_netdev_enumerate,
1753 dpif_netdev_port_open_type,
1756 dpif_netdev_destroy,
1759 dpif_netdev_get_stats,
1760 dpif_netdev_port_add,
1761 dpif_netdev_port_del,
1762 dpif_netdev_port_query_by_number,
1763 dpif_netdev_port_query_by_name,
1764 NULL, /* port_get_pid */
1765 dpif_netdev_port_dump_start,
1766 dpif_netdev_port_dump_next,
1767 dpif_netdev_port_dump_done,
1768 dpif_netdev_port_poll,
1769 dpif_netdev_port_poll_wait,
1770 dpif_netdev_flow_get,
1771 dpif_netdev_flow_put,
1772 dpif_netdev_flow_del,
1773 dpif_netdev_flow_flush,
1774 dpif_netdev_flow_dump_start,
1775 dpif_netdev_flow_dump_next,
1776 dpif_netdev_flow_dump_done,
1777 dpif_netdev_execute,
1779 dpif_netdev_recv_set,
1780 dpif_netdev_queue_to_priority,
1782 dpif_netdev_recv_wait,
1783 dpif_netdev_recv_purge,
1787 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1788 const char *argv[], void *aux OVS_UNUSED)
1790 struct dp_netdev_port *port;
1791 struct dp_netdev *dp;
1794 ovs_mutex_lock(&dp_netdev_mutex);
1795 dp = shash_find_data(&dp_netdevs, argv[1]);
1796 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1797 ovs_mutex_unlock(&dp_netdev_mutex);
1798 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1801 ovs_refcount_ref(&dp->ref_cnt);
1802 ovs_mutex_unlock(&dp_netdev_mutex);
1804 ovs_rwlock_wrlock(&dp->port_rwlock);
1805 if (get_port_by_name(dp, argv[2], &port)) {
1806 unixctl_command_reply_error(conn, "unknown port");
1810 port_no = u32_to_odp(atoi(argv[3]));
1811 if (!port_no || port_no == ODPP_NONE) {
1812 unixctl_command_reply_error(conn, "bad port number");
1815 if (dp_netdev_lookup_port(dp, port_no)) {
1816 unixctl_command_reply_error(conn, "port number already in use");
1819 hmap_remove(&dp->ports, &port->node);
1820 port->port_no = port_no;
1821 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
1822 seq_change(dp->port_seq);
1823 unixctl_command_reply(conn, NULL);
1826 ovs_rwlock_unlock(&dp->port_rwlock);
1827 dp_netdev_unref(dp);
1831 dpif_dummy_register__(const char *type)
1833 struct dpif_class *class;
1835 class = xmalloc(sizeof *class);
1836 *class = dpif_netdev_class;
1837 class->type = xstrdup(type);
1838 dp_register_provider(class);
1842 dpif_dummy_register(bool override)
1849 dp_enumerate_types(&types);
1850 SSET_FOR_EACH (type, &types) {
1851 if (!dp_unregister_provider(type)) {
1852 dpif_dummy_register__(type);
1855 sset_destroy(&types);
1858 dpif_dummy_register__("dummy");
1860 unixctl_command_register("dpif-dummy/change-port-number",
1861 "DP PORT NEW-NUMBER",
1862 3, 3, dpif_dummy_change_port_number, NULL);