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"
44 #include "meta-flow.h"
46 #include "netdev-vport.h"
48 #include "odp-execute.h"
50 #include "ofp-print.h"
53 #include "poll-loop.h"
63 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
65 /* By default, choose a priority in the middle. */
66 #define NETDEV_RULE_PRIORITY 0x8000
68 /* Configuration parameters. */
69 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
71 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
72 * headers to be aligned on a 4-byte boundary. */
73 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
76 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
77 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
78 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
79 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
81 /* Protects against changes to 'dp_netdevs'. */
82 static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
84 /* Contains all 'struct dp_netdev's. */
85 static struct shash dp_netdevs OVS_GUARDED_BY(dp_netdev_mutex)
86 = SHASH_INITIALIZER(&dp_netdevs);
88 struct dp_netdev_upcall {
89 struct dpif_upcall upcall; /* Queued upcall information. */
90 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
93 /* A queue passing packets from a struct dp_netdev to its clients.
99 * Any access at all requires the owning 'dp_netdev''s queue_mutex. */
100 struct dp_netdev_queue {
101 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN] OVS_GUARDED;
102 unsigned int head OVS_GUARDED;
103 unsigned int tail OVS_GUARDED;
106 /* Datapath based on the network device interface from netdev.h.
112 * Some members, marked 'const', are immutable. Accessing other members
113 * requires synchronization, as noted in more detail below.
115 * Acquisition order is, from outermost to innermost:
117 * dp_netdev_mutex (global)
124 const struct dpif_class *const class;
125 const char *const name;
126 struct ovs_refcount ref_cnt;
127 atomic_flag destroyed;
131 * Readers of 'cls' and 'flow_table' must take a 'cls->rwlock' read lock.
133 * Writers of 'cls' and 'flow_table' must take the 'flow_mutex' and then
134 * the 'cls->rwlock' write lock. (The outer 'flow_mutex' allows writers to
135 * atomically perform multiple operations on 'cls' and 'flow_table'.)
137 struct ovs_mutex flow_mutex;
138 struct classifier cls; /* Classifier. Protected by cls.rwlock. */
139 struct hmap flow_table OVS_GUARDED; /* Flow table. */
143 * Everything in 'queues' is protected by 'queue_mutex'. */
144 struct ovs_mutex queue_mutex;
145 struct dp_netdev_queue queues[N_QUEUES];
146 struct seq *queue_seq; /* Incremented whenever a packet is queued. */
150 * ovsthread_counter is internally synchronized. */
151 struct ovsthread_counter *n_hit; /* Number of flow table matches. */
152 struct ovsthread_counter *n_missed; /* Number of flow table misses. */
153 struct ovsthread_counter *n_lost; /* Number of misses not passed up. */
157 * Any lookup into 'ports' or any access to the dp_netdev_ports found
158 * through 'ports' requires taking 'port_rwlock'. */
159 struct ovs_rwlock port_rwlock;
160 struct hmap ports OVS_GUARDED;
161 struct seq *port_seq; /* Incremented whenever a port changes. */
163 /* Forwarding threads. */
164 struct latch exit_latch;
165 struct dp_forwarder *forwarders;
169 static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *dp,
171 OVS_REQ_RDLOCK(dp->port_rwlock);
173 /* A port in a netdev-based datapath. */
174 struct dp_netdev_port {
175 struct hmap_node node; /* Node in dp_netdev's 'ports'. */
177 struct netdev *netdev;
178 struct netdev_saved_flags *sf;
179 struct netdev_rx *rx;
180 char *type; /* Port type as requested by user. */
183 /* A flow in dp_netdev's 'flow_table'.
189 * Except near the beginning or ending of its lifespan, rule 'rule' belongs to
190 * its dp_netdev's classifier. The text below calls this classifier 'cls'.
195 * The thread safety rules described here for "struct dp_netdev_flow" are
196 * motivated by two goals:
198 * - Prevent threads that read members of "struct dp_netdev_flow" from
199 * reading bad data due to changes by some thread concurrently modifying
202 * - Prevent two threads making changes to members of a given "struct
203 * dp_netdev_flow" from interfering with each other.
209 * A flow 'flow' may be accessed without a risk of being freed by code that
210 * holds a read-lock or write-lock on 'cls->rwlock' or that owns a reference to
211 * 'flow->ref_cnt' (or both). Code that needs to hold onto a flow for a while
212 * should take 'cls->rwlock', find the flow it needs, increment 'flow->ref_cnt'
213 * with dpif_netdev_flow_ref(), and drop 'cls->rwlock'.
215 * 'flow->ref_cnt' protects 'flow' from being freed. It doesn't protect the
216 * flow from being deleted from 'cls' (that's 'cls->rwlock') and it doesn't
217 * protect members of 'flow' from modification (that's 'flow->mutex').
219 * 'flow->mutex' protects the members of 'flow' from modification. It doesn't
220 * protect the flow from being deleted from 'cls' (that's 'cls->rwlock') and it
221 * doesn't prevent the flow from being freed (that's 'flow->ref_cnt').
223 * Some members, marked 'const', are immutable. Accessing other members
224 * requires synchronization, as noted in more detail below.
226 struct dp_netdev_flow {
227 /* Packet classification. */
228 const struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
230 /* Hash table index by unmasked flow. */
231 const struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
232 const struct flow flow; /* The flow that created this entry. */
234 /* Number of references.
235 * The classifier owns one reference.
236 * Any thread trying to keep a rule from being freed should hold its own
238 struct ovs_refcount ref_cnt;
240 /* Protects members marked OVS_GUARDED.
242 * Acquire after datapath's flow_mutex. */
243 struct ovs_mutex mutex OVS_ACQ_AFTER(dp_netdev_mutex);
247 * Reading or writing these members requires 'mutex'. */
248 long long int used OVS_GUARDED; /* Last used time, in monotonic msecs. */
249 long long int packet_count OVS_GUARDED; /* Number of packets matched. */
250 long long int byte_count OVS_GUARDED; /* Number of bytes matched. */
251 uint16_t tcp_flags OVS_GUARDED; /* Bitwise-OR of seen tcp_flags values. */
255 * Reading 'actions' requires 'mutex'.
256 * Writing 'actions' requires 'mutex' and (to allow for transactions) the
257 * datapath's flow_mutex. */
258 struct dp_netdev_actions *actions OVS_GUARDED;
261 static struct dp_netdev_flow *dp_netdev_flow_ref(
262 const struct dp_netdev_flow *);
263 static void dp_netdev_flow_unref(struct dp_netdev_flow *);
265 /* A set of datapath actions within a "struct dp_netdev_flow".
271 * A struct dp_netdev_actions 'actions' may be accessed without a risk of being
272 * freed by code that holds a read-lock or write-lock on 'flow->mutex' (where
273 * 'flow' is the dp_netdev_flow for which 'flow->actions == actions') or that
274 * owns a reference to 'actions->ref_cnt' (or both). */
275 struct dp_netdev_actions {
276 struct ovs_refcount ref_cnt;
278 /* These members are immutable: they do not change during the struct's
280 struct nlattr *actions; /* Sequence of OVS_ACTION_ATTR_* attributes. */
281 unsigned int size; /* Size of 'actions', in bytes. */
284 struct dp_netdev_actions *dp_netdev_actions_create(const struct nlattr *,
286 struct dp_netdev_actions *dp_netdev_actions_ref(
287 const struct dp_netdev_actions *);
288 void dp_netdev_actions_unref(struct dp_netdev_actions *);
290 /* A thread that receives packets from some ports, looks them up in the flow
291 * table, and executes the actions it finds. */
292 struct dp_forwarder {
293 struct dp_netdev *dp;
296 uint32_t min_hash, max_hash;
299 /* Interface to netdev-based datapath. */
302 struct dp_netdev *dp;
303 uint64_t last_port_seq;
306 static int get_port_by_number(struct dp_netdev *dp, odp_port_t port_no,
307 struct dp_netdev_port **portp)
308 OVS_REQ_RDLOCK(dp->port_rwlock);
309 static int get_port_by_name(struct dp_netdev *dp, const char *devname,
310 struct dp_netdev_port **portp)
311 OVS_REQ_RDLOCK(dp->port_rwlock);
312 static void dp_netdev_free(struct dp_netdev *)
313 OVS_REQUIRES(dp_netdev_mutex);
314 static void dp_netdev_flow_flush(struct dp_netdev *);
315 static int do_add_port(struct dp_netdev *dp, const char *devname,
316 const char *type, odp_port_t port_no)
317 OVS_REQ_WRLOCK(dp->port_rwlock);
318 static int do_del_port(struct dp_netdev *dp, odp_port_t port_no)
319 OVS_REQ_WRLOCK(dp->port_rwlock);
320 static int dpif_netdev_open(const struct dpif_class *, const char *name,
321 bool create, struct dpif **);
322 static int dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *,
323 int queue_no, const struct flow *,
324 const struct nlattr *userdata)
325 OVS_EXCLUDED(dp->queue_mutex);
326 static void dp_netdev_execute_actions(struct dp_netdev *dp,
327 const struct flow *, struct ofpbuf *,
328 struct pkt_metadata *,
329 const struct nlattr *actions,
331 OVS_REQ_RDLOCK(dp->port_rwlock);
332 static void dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
333 struct pkt_metadata *)
334 OVS_REQ_RDLOCK(dp->port_rwlock);
335 static void dp_netdev_set_threads(struct dp_netdev *, int n);
337 static struct dpif_netdev *
338 dpif_netdev_cast(const struct dpif *dpif)
340 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
341 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
344 static struct dp_netdev *
345 get_dp_netdev(const struct dpif *dpif)
347 return dpif_netdev_cast(dpif)->dp;
351 dpif_netdev_enumerate(struct sset *all_dps)
353 struct shash_node *node;
355 ovs_mutex_lock(&dp_netdev_mutex);
356 SHASH_FOR_EACH(node, &dp_netdevs) {
357 sset_add(all_dps, node->name);
359 ovs_mutex_unlock(&dp_netdev_mutex);
365 dpif_netdev_class_is_dummy(const struct dpif_class *class)
367 return class != &dpif_netdev_class;
371 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
373 return strcmp(type, "internal") ? type
374 : dpif_netdev_class_is_dummy(class) ? "dummy"
379 create_dpif_netdev(struct dp_netdev *dp)
381 uint16_t netflow_id = hash_string(dp->name, 0);
382 struct dpif_netdev *dpif;
384 ovs_refcount_ref(&dp->ref_cnt);
386 dpif = xmalloc(sizeof *dpif);
387 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
389 dpif->last_port_seq = seq_read(dp->port_seq);
394 /* Choose an unused, non-zero port number and return it on success.
395 * Return ODPP_NONE on failure. */
397 choose_port(struct dp_netdev *dp, const char *name)
398 OVS_REQ_RDLOCK(dp->port_rwlock)
402 if (dp->class != &dpif_netdev_class) {
406 /* If the port name begins with "br", start the number search at
407 * 100 to make writing tests easier. */
408 if (!strncmp(name, "br", 2)) {
412 /* If the port name contains a number, try to assign that port number.
413 * This can make writing unit tests easier because port numbers are
415 for (p = name; *p != '\0'; p++) {
416 if (isdigit((unsigned char) *p)) {
417 port_no = start_no + strtol(p, NULL, 10);
418 if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE)
419 && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
420 return u32_to_odp(port_no);
427 for (port_no = 1; port_no <= UINT16_MAX; port_no++) {
428 if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
429 return u32_to_odp(port_no);
437 create_dp_netdev(const char *name, const struct dpif_class *class,
438 struct dp_netdev **dpp)
439 OVS_REQUIRES(dp_netdev_mutex)
441 struct dp_netdev *dp;
445 dp = xzalloc(sizeof *dp);
446 shash_add(&dp_netdevs, name, dp);
448 *CONST_CAST(const struct dpif_class **, &dp->class) = class;
449 *CONST_CAST(const char **, &dp->name) = xstrdup(name);
450 ovs_refcount_init(&dp->ref_cnt);
451 atomic_flag_init(&dp->destroyed);
453 ovs_mutex_init(&dp->flow_mutex);
454 classifier_init(&dp->cls, NULL);
455 hmap_init(&dp->flow_table);
457 ovs_mutex_init(&dp->queue_mutex);
458 ovs_mutex_lock(&dp->queue_mutex);
459 for (i = 0; i < N_QUEUES; i++) {
460 dp->queues[i].head = dp->queues[i].tail = 0;
462 ovs_mutex_unlock(&dp->queue_mutex);
463 dp->queue_seq = seq_create();
465 dp->n_hit = ovsthread_counter_create();
466 dp->n_missed = ovsthread_counter_create();
467 dp->n_lost = ovsthread_counter_create();
469 ovs_rwlock_init(&dp->port_rwlock);
470 hmap_init(&dp->ports);
471 dp->port_seq = seq_create();
472 latch_init(&dp->exit_latch);
474 ovs_rwlock_wrlock(&dp->port_rwlock);
475 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
476 ovs_rwlock_unlock(&dp->port_rwlock);
481 dp_netdev_set_threads(dp, 2);
488 dpif_netdev_open(const struct dpif_class *class, const char *name,
489 bool create, struct dpif **dpifp)
491 struct dp_netdev *dp;
494 ovs_mutex_lock(&dp_netdev_mutex);
495 dp = shash_find_data(&dp_netdevs, name);
497 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
499 error = (dp->class != class ? EINVAL
504 *dpifp = create_dpif_netdev(dp);
506 ovs_mutex_unlock(&dp_netdev_mutex);
512 dp_netdev_purge_queues(struct dp_netdev *dp)
516 ovs_mutex_lock(&dp->queue_mutex);
517 for (i = 0; i < N_QUEUES; i++) {
518 struct dp_netdev_queue *q = &dp->queues[i];
520 while (q->tail != q->head) {
521 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
522 ofpbuf_uninit(&u->upcall.packet);
523 ofpbuf_uninit(&u->buf);
526 ovs_mutex_unlock(&dp->queue_mutex);
529 /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp'
530 * through the 'dp_netdevs' shash while freeing 'dp'. */
532 dp_netdev_free(struct dp_netdev *dp)
533 OVS_REQUIRES(dp_netdev_mutex)
535 struct dp_netdev_port *port, *next;
537 shash_find_and_delete(&dp_netdevs, dp->name);
539 dp_netdev_set_threads(dp, 0);
540 free(dp->forwarders);
542 dp_netdev_flow_flush(dp);
543 ovs_rwlock_wrlock(&dp->port_rwlock);
544 HMAP_FOR_EACH_SAFE (port, next, node, &dp->ports) {
545 do_del_port(dp, port->port_no);
547 ovs_rwlock_unlock(&dp->port_rwlock);
548 ovsthread_counter_destroy(dp->n_hit);
549 ovsthread_counter_destroy(dp->n_missed);
550 ovsthread_counter_destroy(dp->n_lost);
552 dp_netdev_purge_queues(dp);
553 seq_destroy(dp->queue_seq);
554 ovs_mutex_destroy(&dp->queue_mutex);
556 classifier_destroy(&dp->cls);
557 hmap_destroy(&dp->flow_table);
558 ovs_mutex_destroy(&dp->flow_mutex);
559 seq_destroy(dp->port_seq);
560 hmap_destroy(&dp->ports);
561 atomic_flag_destroy(&dp->destroyed);
562 ovs_refcount_destroy(&dp->ref_cnt);
563 latch_destroy(&dp->exit_latch);
564 free(CONST_CAST(char *, dp->name));
569 dp_netdev_unref(struct dp_netdev *dp)
572 /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't
573 * get a new reference to 'dp' through the 'dp_netdevs' shash. */
574 ovs_mutex_lock(&dp_netdev_mutex);
575 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
578 ovs_mutex_unlock(&dp_netdev_mutex);
583 dpif_netdev_close(struct dpif *dpif)
585 struct dp_netdev *dp = get_dp_netdev(dpif);
592 dpif_netdev_destroy(struct dpif *dpif)
594 struct dp_netdev *dp = get_dp_netdev(dpif);
596 if (!atomic_flag_test_and_set(&dp->destroyed)) {
597 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
598 /* Can't happen: 'dpif' still owns a reference to 'dp'. */
607 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
609 struct dp_netdev *dp = get_dp_netdev(dpif);
611 fat_rwlock_rdlock(&dp->cls.rwlock);
612 stats->n_flows = hmap_count(&dp->flow_table);
613 fat_rwlock_unlock(&dp->cls.rwlock);
615 stats->n_hit = ovsthread_counter_read(dp->n_hit);
616 stats->n_missed = ovsthread_counter_read(dp->n_missed);
617 stats->n_lost = ovsthread_counter_read(dp->n_lost);
618 stats->n_masks = UINT32_MAX;
619 stats->n_mask_hit = UINT64_MAX;
625 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
627 OVS_REQ_WRLOCK(dp->port_rwlock)
629 struct netdev_saved_flags *sf;
630 struct dp_netdev_port *port;
631 struct netdev *netdev;
632 struct netdev_rx *rx;
633 enum netdev_flags flags;
634 const char *open_type;
637 /* XXX reject devices already in some dp_netdev. */
639 /* Open and validate network device. */
640 open_type = dpif_netdev_port_open_type(dp->class, type);
641 error = netdev_open(devname, open_type, &netdev);
645 /* XXX reject non-Ethernet devices */
647 netdev_get_flags(netdev, &flags);
648 if (flags & NETDEV_LOOPBACK) {
649 VLOG_ERR("%s: cannot add a loopback device", devname);
650 netdev_close(netdev);
654 error = netdev_rx_open(netdev, &rx);
656 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
657 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
658 devname, ovs_strerror(errno));
659 netdev_close(netdev);
663 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
666 netdev_close(netdev);
670 port = xmalloc(sizeof *port);
671 port->port_no = port_no;
672 port->netdev = netdev;
675 port->type = xstrdup(type);
677 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
678 seq_change(dp->port_seq);
684 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
685 odp_port_t *port_nop)
687 struct dp_netdev *dp = get_dp_netdev(dpif);
688 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
689 const char *dpif_port;
693 ovs_rwlock_wrlock(&dp->port_rwlock);
694 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
695 if (*port_nop != ODPP_NONE) {
697 error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0;
699 port_no = choose_port(dp, dpif_port);
700 error = port_no == ODPP_NONE ? EFBIG : 0;
704 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
706 ovs_rwlock_unlock(&dp->port_rwlock);
712 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
714 struct dp_netdev *dp = get_dp_netdev(dpif);
717 ovs_rwlock_wrlock(&dp->port_rwlock);
718 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
719 ovs_rwlock_unlock(&dp->port_rwlock);
725 is_valid_port_number(odp_port_t port_no)
727 return port_no != ODPP_NONE;
730 static struct dp_netdev_port *
731 dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no)
732 OVS_REQ_RDLOCK(dp->port_rwlock)
734 struct dp_netdev_port *port;
736 HMAP_FOR_EACH_IN_BUCKET (port, node, hash_int(odp_to_u32(port_no), 0),
738 if (port->port_no == port_no) {
746 get_port_by_number(struct dp_netdev *dp,
747 odp_port_t port_no, struct dp_netdev_port **portp)
748 OVS_REQ_RDLOCK(dp->port_rwlock)
750 if (!is_valid_port_number(port_no)) {
754 *portp = dp_netdev_lookup_port(dp, port_no);
755 return *portp ? 0 : ENOENT;
760 get_port_by_name(struct dp_netdev *dp,
761 const char *devname, struct dp_netdev_port **portp)
762 OVS_REQ_RDLOCK(dp->port_rwlock)
764 struct dp_netdev_port *port;
766 HMAP_FOR_EACH (port, node, &dp->ports) {
767 if (!strcmp(netdev_get_name(port->netdev), devname)) {
776 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
777 OVS_REQ_WRLOCK(dp->port_rwlock)
779 struct dp_netdev_port *port;
782 error = get_port_by_number(dp, port_no, &port);
787 hmap_remove(&dp->ports, &port->node);
788 seq_change(dp->port_seq);
790 netdev_close(port->netdev);
791 netdev_restore_flags(port->sf);
792 netdev_rx_close(port->rx);
800 answer_port_query(const struct dp_netdev_port *port,
801 struct dpif_port *dpif_port)
803 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
804 dpif_port->type = xstrdup(port->type);
805 dpif_port->port_no = port->port_no;
809 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
810 struct dpif_port *dpif_port)
812 struct dp_netdev *dp = get_dp_netdev(dpif);
813 struct dp_netdev_port *port;
816 ovs_rwlock_rdlock(&dp->port_rwlock);
817 error = get_port_by_number(dp, port_no, &port);
818 if (!error && dpif_port) {
819 answer_port_query(port, dpif_port);
821 ovs_rwlock_unlock(&dp->port_rwlock);
827 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
828 struct dpif_port *dpif_port)
830 struct dp_netdev *dp = get_dp_netdev(dpif);
831 struct dp_netdev_port *port;
834 ovs_rwlock_rdlock(&dp->port_rwlock);
835 error = get_port_by_name(dp, devname, &port);
836 if (!error && dpif_port) {
837 answer_port_query(port, dpif_port);
839 ovs_rwlock_unlock(&dp->port_rwlock);
845 dp_netdev_remove_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
846 OVS_REQ_WRLOCK(dp->cls.rwlock)
847 OVS_REQUIRES(dp->flow_mutex)
849 struct cls_rule *cr = CONST_CAST(struct cls_rule *, &flow->cr);
850 struct hmap_node *node = CONST_CAST(struct hmap_node *, &flow->node);
852 classifier_remove(&dp->cls, cr);
853 hmap_remove(&dp->flow_table, node);
854 dp_netdev_flow_unref(flow);
857 static struct dp_netdev_flow *
858 dp_netdev_flow_ref(const struct dp_netdev_flow *flow_)
860 struct dp_netdev_flow *flow = CONST_CAST(struct dp_netdev_flow *, flow_);
862 ovs_refcount_ref(&flow->ref_cnt);
868 dp_netdev_flow_unref(struct dp_netdev_flow *flow)
870 if (flow && ovs_refcount_unref(&flow->ref_cnt) == 1) {
871 cls_rule_destroy(CONST_CAST(struct cls_rule *, &flow->cr));
872 ovs_mutex_lock(&flow->mutex);
873 dp_netdev_actions_unref(flow->actions);
874 ovs_mutex_unlock(&flow->mutex);
875 ovs_mutex_destroy(&flow->mutex);
881 dp_netdev_flow_flush(struct dp_netdev *dp)
883 struct dp_netdev_flow *netdev_flow, *next;
885 ovs_mutex_lock(&dp->flow_mutex);
886 fat_rwlock_wrlock(&dp->cls.rwlock);
887 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
888 dp_netdev_remove_flow(dp, netdev_flow);
890 fat_rwlock_unlock(&dp->cls.rwlock);
891 ovs_mutex_unlock(&dp->flow_mutex);
895 dpif_netdev_flow_flush(struct dpif *dpif)
897 struct dp_netdev *dp = get_dp_netdev(dpif);
899 dp_netdev_flow_flush(dp);
903 struct dp_netdev_port_state {
910 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
912 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
917 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
918 struct dpif_port *dpif_port)
920 struct dp_netdev_port_state *state = state_;
921 struct dp_netdev *dp = get_dp_netdev(dpif);
922 struct hmap_node *node;
925 ovs_rwlock_rdlock(&dp->port_rwlock);
926 node = hmap_at_position(&dp->ports, &state->bucket, &state->offset);
928 struct dp_netdev_port *port;
930 port = CONTAINER_OF(node, struct dp_netdev_port, node);
933 state->name = xstrdup(netdev_get_name(port->netdev));
934 dpif_port->name = state->name;
935 dpif_port->type = port->type;
936 dpif_port->port_no = port->port_no;
942 ovs_rwlock_unlock(&dp->port_rwlock);
948 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
950 struct dp_netdev_port_state *state = state_;
957 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
959 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
960 uint64_t new_port_seq;
963 new_port_seq = seq_read(dpif->dp->port_seq);
964 if (dpif->last_port_seq != new_port_seq) {
965 dpif->last_port_seq = new_port_seq;
975 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
977 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
979 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
982 static struct dp_netdev_flow *
983 dp_netdev_flow_cast(const struct cls_rule *cr)
985 return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL;
988 static struct dp_netdev_flow *
989 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
990 OVS_EXCLUDED(dp->cls.rwlock)
992 struct dp_netdev_flow *netdev_flow;
994 fat_rwlock_rdlock(&dp->cls.rwlock);
995 netdev_flow = dp_netdev_flow_cast(classifier_lookup(&dp->cls, flow, NULL));
996 dp_netdev_flow_ref(netdev_flow);
997 fat_rwlock_unlock(&dp->cls.rwlock);
1002 static struct dp_netdev_flow *
1003 dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
1004 OVS_REQ_RDLOCK(dp->cls.rwlock)
1006 struct dp_netdev_flow *netdev_flow;
1008 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
1010 if (flow_equal(&netdev_flow->flow, flow)) {
1011 return dp_netdev_flow_ref(netdev_flow);
1019 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
1020 struct dpif_flow_stats *stats)
1021 OVS_REQ_RDLOCK(netdev_flow->mutex)
1023 stats->n_packets = netdev_flow->packet_count;
1024 stats->n_bytes = netdev_flow->byte_count;
1025 stats->used = netdev_flow->used;
1026 stats->tcp_flags = netdev_flow->tcp_flags;
1030 dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1031 const struct nlattr *mask_key,
1032 uint32_t mask_key_len, const struct flow *flow,
1036 if (odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow)) {
1037 /* This should not happen: it indicates that
1038 * odp_flow_key_from_mask() and odp_flow_key_to_mask()
1039 * disagree on the acceptable form of a mask. Log the problem
1040 * as an error, with enough details to enable debugging. */
1041 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1043 if (!VLOG_DROP_ERR(&rl)) {
1047 odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
1049 VLOG_ERR("internal error parsing flow mask %s", ds_cstr(&s));
1055 /* Force unwildcard the in_port. */
1056 mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
1058 enum mf_field_id id;
1059 /* No mask key, unwildcard everything except fields whose
1060 * prerequisities are not met. */
1061 memset(mask, 0x0, sizeof *mask);
1063 for (id = 0; id < MFF_N_IDS; ++id) {
1064 /* Skip registers and metadata. */
1065 if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
1066 && id != MFF_METADATA) {
1067 const struct mf_field *mf = mf_from_id(id);
1068 if (mf_are_prereqs_ok(mf, flow)) {
1069 mf_mask_field(mf, mask);
1079 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1084 if (odp_flow_key_to_flow(key, key_len, flow)) {
1085 /* This should not happen: it indicates that odp_flow_key_from_flow()
1086 * and odp_flow_key_to_flow() disagree on the acceptable form of a
1087 * flow. Log the problem as an error, with enough details to enable
1089 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1091 if (!VLOG_DROP_ERR(&rl)) {
1095 odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
1096 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
1103 in_port = flow->in_port.odp_port;
1104 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
1112 dpif_netdev_flow_get(const struct dpif *dpif,
1113 const struct nlattr *nl_key, size_t nl_key_len,
1114 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
1116 struct dp_netdev *dp = get_dp_netdev(dpif);
1117 struct dp_netdev_flow *netdev_flow;
1121 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
1126 fat_rwlock_rdlock(&dp->cls.rwlock);
1127 netdev_flow = dp_netdev_find_flow(dp, &key);
1128 fat_rwlock_unlock(&dp->cls.rwlock);
1131 struct dp_netdev_actions *actions = NULL;
1133 ovs_mutex_lock(&netdev_flow->mutex);
1135 get_dpif_flow_stats(netdev_flow, stats);
1138 actions = dp_netdev_actions_ref(netdev_flow->actions);
1140 ovs_mutex_unlock(&netdev_flow->mutex);
1142 dp_netdev_flow_unref(netdev_flow);
1145 *actionsp = ofpbuf_clone_data(actions->actions, actions->size);
1146 dp_netdev_actions_unref(actions);
1156 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
1157 const struct flow_wildcards *wc,
1158 const struct nlattr *actions,
1160 OVS_REQUIRES(dp->flow_mutex)
1162 struct dp_netdev_flow *netdev_flow;
1165 netdev_flow = xzalloc(sizeof *netdev_flow);
1166 *CONST_CAST(struct flow *, &netdev_flow->flow) = *flow;
1167 ovs_refcount_init(&netdev_flow->ref_cnt);
1169 ovs_mutex_init(&netdev_flow->mutex);
1170 ovs_mutex_lock(&netdev_flow->mutex);
1172 netdev_flow->actions = dp_netdev_actions_create(actions, actions_len);
1174 match_init(&match, flow, wc);
1175 cls_rule_init(CONST_CAST(struct cls_rule *, &netdev_flow->cr),
1176 &match, NETDEV_RULE_PRIORITY);
1177 fat_rwlock_wrlock(&dp->cls.rwlock);
1178 classifier_insert(&dp->cls,
1179 CONST_CAST(struct cls_rule *, &netdev_flow->cr));
1180 hmap_insert(&dp->flow_table,
1181 CONST_CAST(struct hmap_node *, &netdev_flow->node),
1182 flow_hash(flow, 0));
1183 fat_rwlock_unlock(&dp->cls.rwlock);
1185 ovs_mutex_unlock(&netdev_flow->mutex);
1191 clear_stats(struct dp_netdev_flow *netdev_flow)
1192 OVS_REQUIRES(netdev_flow->mutex)
1194 netdev_flow->used = 0;
1195 netdev_flow->packet_count = 0;
1196 netdev_flow->byte_count = 0;
1197 netdev_flow->tcp_flags = 0;
1201 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
1203 struct dp_netdev *dp = get_dp_netdev(dpif);
1204 struct dp_netdev_flow *netdev_flow;
1206 struct flow_wildcards wc;
1209 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
1213 error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
1214 put->mask, put->mask_len,
1220 ovs_mutex_lock(&dp->flow_mutex);
1221 netdev_flow = dp_netdev_lookup_flow(dp, &flow);
1223 if (put->flags & DPIF_FP_CREATE) {
1224 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
1226 memset(put->stats, 0, sizeof *put->stats);
1228 error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
1237 if (put->flags & DPIF_FP_MODIFY
1238 && flow_equal(&flow, &netdev_flow->flow)) {
1239 struct dp_netdev_actions *new_actions;
1240 struct dp_netdev_actions *old_actions;
1242 new_actions = dp_netdev_actions_create(put->actions,
1245 ovs_mutex_lock(&netdev_flow->mutex);
1246 old_actions = netdev_flow->actions;
1247 netdev_flow->actions = new_actions;
1249 get_dpif_flow_stats(netdev_flow, put->stats);
1251 if (put->flags & DPIF_FP_ZERO_STATS) {
1252 clear_stats(netdev_flow);
1254 ovs_mutex_unlock(&netdev_flow->mutex);
1256 dp_netdev_actions_unref(old_actions);
1257 } else if (put->flags & DPIF_FP_CREATE) {
1260 /* Overlapping flow. */
1263 dp_netdev_flow_unref(netdev_flow);
1265 ovs_mutex_unlock(&dp->flow_mutex);
1271 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
1273 struct dp_netdev *dp = get_dp_netdev(dpif);
1274 struct dp_netdev_flow *netdev_flow;
1278 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
1283 ovs_mutex_lock(&dp->flow_mutex);
1284 fat_rwlock_wrlock(&dp->cls.rwlock);
1285 netdev_flow = dp_netdev_find_flow(dp, &key);
1288 ovs_mutex_lock(&netdev_flow->mutex);
1289 get_dpif_flow_stats(netdev_flow, del->stats);
1290 ovs_mutex_unlock(&netdev_flow->mutex);
1292 dp_netdev_remove_flow(dp, netdev_flow);
1296 fat_rwlock_unlock(&dp->cls.rwlock);
1297 ovs_mutex_unlock(&dp->flow_mutex);
1302 struct dp_netdev_flow_state {
1305 struct dp_netdev_actions *actions;
1306 struct odputil_keybuf keybuf;
1307 struct odputil_keybuf maskbuf;
1308 struct dpif_flow_stats stats;
1312 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
1314 struct dp_netdev_flow_state *state;
1316 *statep = state = xmalloc(sizeof *state);
1319 state->actions = NULL;
1324 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
1325 const struct nlattr **key, size_t *key_len,
1326 const struct nlattr **mask, size_t *mask_len,
1327 const struct nlattr **actions, size_t *actions_len,
1328 const struct dpif_flow_stats **stats)
1330 struct dp_netdev_flow_state *state = state_;
1331 struct dp_netdev *dp = get_dp_netdev(dpif);
1332 struct dp_netdev_flow *netdev_flow;
1333 struct hmap_node *node;
1335 fat_rwlock_rdlock(&dp->cls.rwlock);
1336 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
1338 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
1339 dp_netdev_flow_ref(netdev_flow);
1341 fat_rwlock_unlock(&dp->cls.rwlock);
1349 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
1350 odp_flow_key_from_flow(&buf, &netdev_flow->flow,
1351 netdev_flow->flow.in_port.odp_port);
1354 *key_len = buf.size;
1359 struct flow_wildcards wc;
1361 ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
1362 minimask_expand(&netdev_flow->cr.match.mask, &wc);
1363 odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
1364 odp_to_u32(wc.masks.in_port.odp_port));
1367 *mask_len = buf.size;
1370 if (actions || stats) {
1371 dp_netdev_actions_unref(state->actions);
1372 state->actions = NULL;
1374 ovs_mutex_lock(&netdev_flow->mutex);
1376 state->actions = dp_netdev_actions_ref(netdev_flow->actions);
1377 *actions = state->actions->actions;
1378 *actions_len = state->actions->size;
1381 get_dpif_flow_stats(netdev_flow, &state->stats);
1382 *stats = &state->stats;
1384 ovs_mutex_unlock(&netdev_flow->mutex);
1387 dp_netdev_flow_unref(netdev_flow);
1393 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1395 struct dp_netdev_flow_state *state = state_;
1397 dp_netdev_actions_unref(state->actions);
1403 dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute)
1405 struct dp_netdev *dp = get_dp_netdev(dpif);
1406 struct pkt_metadata *md = &execute->md;
1409 if (execute->packet->size < ETH_HEADER_LEN ||
1410 execute->packet->size > UINT16_MAX) {
1414 /* Extract flow key. */
1415 flow_extract(execute->packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1416 (union flow_in_port *)&md->in_port, &key);
1418 ovs_rwlock_rdlock(&dp->port_rwlock);
1419 dp_netdev_execute_actions(dp, &key, execute->packet, md, execute->actions,
1420 execute->actions_len);
1421 ovs_rwlock_unlock(&dp->port_rwlock);
1427 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1433 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1434 uint32_t queue_id, uint32_t *priority)
1436 *priority = queue_id;
1440 static struct dp_netdev_queue *
1441 find_nonempty_queue(struct dp_netdev *dp)
1442 OVS_REQUIRES(dp->queue_mutex)
1446 for (i = 0; i < N_QUEUES; i++) {
1447 struct dp_netdev_queue *q = &dp->queues[i];
1448 if (q->head != q->tail) {
1456 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1459 struct dp_netdev *dp = get_dp_netdev(dpif);
1460 struct dp_netdev_queue *q;
1463 ovs_mutex_lock(&dp->queue_mutex);
1464 q = find_nonempty_queue(dp);
1466 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1468 *upcall = u->upcall;
1477 ovs_mutex_unlock(&dp->queue_mutex);
1483 dpif_netdev_recv_wait(struct dpif *dpif)
1485 struct dp_netdev *dp = get_dp_netdev(dpif);
1488 ovs_mutex_lock(&dp->queue_mutex);
1489 seq = seq_read(dp->queue_seq);
1490 if (find_nonempty_queue(dp)) {
1491 poll_immediate_wake();
1493 seq_wait(dp->queue_seq, seq);
1495 ovs_mutex_unlock(&dp->queue_mutex);
1499 dpif_netdev_recv_purge(struct dpif *dpif)
1501 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1503 dp_netdev_purge_queues(dpif_netdev->dp);
1506 /* Creates and returns a new 'struct dp_netdev_actions', with a reference count
1507 * of 1, whose actions are a copy of from the 'ofpacts_len' bytes of
1509 struct dp_netdev_actions *
1510 dp_netdev_actions_create(const struct nlattr *actions, size_t size)
1512 struct dp_netdev_actions *netdev_actions;
1514 netdev_actions = xmalloc(sizeof *netdev_actions);
1515 ovs_refcount_init(&netdev_actions->ref_cnt);
1516 netdev_actions->actions = xmemdup(actions, size);
1517 netdev_actions->size = size;
1519 return netdev_actions;
1522 /* Increments 'actions''s refcount. */
1523 struct dp_netdev_actions *
1524 dp_netdev_actions_ref(const struct dp_netdev_actions *actions_)
1526 struct dp_netdev_actions *actions;
1528 actions = CONST_CAST(struct dp_netdev_actions *, actions_);
1530 ovs_refcount_ref(&actions->ref_cnt);
1535 /* Decrements 'actions''s refcount and frees 'actions' if the refcount reaches
1538 dp_netdev_actions_unref(struct dp_netdev_actions *actions)
1540 if (actions && ovs_refcount_unref(&actions->ref_cnt) == 1) {
1541 free(actions->actions);
1547 dp_forwarder_main(void *f_)
1549 struct dp_forwarder *f = f_;
1550 struct dp_netdev *dp = f->dp;
1551 struct ofpbuf packet;
1553 f->name = xasprintf("forwarder_%u", ovsthread_id_self());
1554 set_subprogram_name("%s", f->name);
1556 ofpbuf_init(&packet, 0);
1557 while (!latch_is_set(&dp->exit_latch)) {
1558 bool received_anything;
1561 ovs_rwlock_rdlock(&dp->port_rwlock);
1562 for (i = 0; i < 50; i++) {
1563 struct dp_netdev_port *port;
1565 received_anything = false;
1566 HMAP_FOR_EACH (port, node, &f->dp->ports) {
1568 && port->node.hash >= f->min_hash
1569 && port->node.hash <= f->max_hash) {
1574 if (netdev_get_mtu(port->netdev, &mtu)) {
1575 mtu = ETH_PAYLOAD_MAX;
1577 buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + mtu;
1579 ofpbuf_clear(&packet);
1580 ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM,
1583 error = netdev_rx_recv(port->rx, &packet);
1585 struct pkt_metadata md
1586 = PKT_METADATA_INITIALIZER(port->port_no);
1587 dp_netdev_port_input(dp, &packet, &md);
1589 received_anything = true;
1590 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1591 static struct vlog_rate_limit rl
1592 = VLOG_RATE_LIMIT_INIT(1, 5);
1594 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1595 netdev_get_name(port->netdev),
1596 ovs_strerror(error));
1601 if (!received_anything) {
1606 if (received_anything) {
1607 poll_immediate_wake();
1609 struct dp_netdev_port *port;
1611 HMAP_FOR_EACH (port, node, &f->dp->ports)
1613 && port->node.hash >= f->min_hash
1614 && port->node.hash <= f->max_hash) {
1615 netdev_rx_wait(port->rx);
1617 seq_wait(dp->port_seq, seq_read(dp->port_seq));
1618 latch_wait(&dp->exit_latch);
1620 ovs_rwlock_unlock(&dp->port_rwlock);
1624 ofpbuf_uninit(&packet);
1632 dp_netdev_set_threads(struct dp_netdev *dp, int n)
1636 if (n == dp->n_forwarders) {
1640 /* Stop existing threads. */
1641 latch_set(&dp->exit_latch);
1642 for (i = 0; i < dp->n_forwarders; i++) {
1643 struct dp_forwarder *f = &dp->forwarders[i];
1645 xpthread_join(f->thread, NULL);
1647 latch_poll(&dp->exit_latch);
1648 free(dp->forwarders);
1650 /* Start new threads. */
1651 dp->forwarders = xmalloc(n * sizeof *dp->forwarders);
1652 dp->n_forwarders = n;
1653 for (i = 0; i < n; i++) {
1654 struct dp_forwarder *f = &dp->forwarders[i];
1657 f->min_hash = UINT32_MAX / n * i;
1658 f->max_hash = UINT32_MAX / n * (i + 1) - 1;
1660 f->max_hash = UINT32_MAX;
1662 xpthread_create(&f->thread, NULL, dp_forwarder_main, f);
1667 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1668 const struct ofpbuf *packet)
1669 OVS_REQUIRES(netdev_flow->mutex)
1671 netdev_flow->used = time_msec();
1672 netdev_flow->packet_count++;
1673 netdev_flow->byte_count += packet->size;
1674 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
1678 dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
1679 struct pkt_metadata *md)
1680 OVS_REQ_RDLOCK(dp->port_rwlock)
1682 struct dp_netdev_flow *netdev_flow;
1685 if (packet->size < ETH_HEADER_LEN) {
1688 flow_extract(packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1689 (union flow_in_port *)&md->in_port, &key);
1690 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1692 struct dp_netdev_actions *actions;
1694 ovs_mutex_lock(&netdev_flow->mutex);
1695 dp_netdev_flow_used(netdev_flow, packet);
1696 actions = dp_netdev_actions_ref(netdev_flow->actions);
1697 ovs_mutex_unlock(&netdev_flow->mutex);
1699 dp_netdev_execute_actions(dp, &key, packet, md,
1700 actions->actions, actions->size);
1701 dp_netdev_actions_unref(actions);
1702 ovsthread_counter_inc(dp->n_hit, 1);
1704 ovsthread_counter_inc(dp->n_missed, 1);
1705 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1710 dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
1711 int queue_no, const struct flow *flow,
1712 const struct nlattr *userdata)
1713 OVS_EXCLUDED(dp->queue_mutex)
1715 struct dp_netdev_queue *q = &dp->queues[queue_no];
1718 ovs_mutex_lock(&dp->queue_mutex);
1719 if (q->head - q->tail < MAX_QUEUE_LEN) {
1720 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1721 struct dpif_upcall *upcall = &u->upcall;
1722 struct ofpbuf *buf = &u->buf;
1725 upcall->type = queue_no;
1727 /* Allocate buffer big enough for everything. */
1728 buf_size = ODPUTIL_FLOW_KEY_BYTES;
1730 buf_size += NLA_ALIGN(userdata->nla_len);
1732 ofpbuf_init(buf, buf_size);
1735 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1736 upcall->key = buf->data;
1737 upcall->key_len = buf->size;
1741 upcall->userdata = ofpbuf_put(buf, userdata,
1742 NLA_ALIGN(userdata->nla_len));
1745 /* Steal packet data. */
1746 ovs_assert(packet->source == OFPBUF_MALLOC);
1747 upcall->packet = *packet;
1748 ofpbuf_use(packet, NULL, 0);
1750 seq_change(dp->queue_seq);
1754 ovsthread_counter_inc(dp->n_lost, 1);
1757 ovs_mutex_unlock(&dp->queue_mutex);
1762 struct dp_netdev_execute_aux {
1763 struct dp_netdev *dp;
1764 const struct flow *key;
1768 dp_execute_cb(void *aux_, struct ofpbuf *packet,
1769 const struct pkt_metadata *md OVS_UNUSED,
1770 const struct nlattr *a, bool may_steal)
1771 OVS_NO_THREAD_SAFETY_ANALYSIS
1773 struct dp_netdev_execute_aux *aux = aux_;
1774 int type = nl_attr_type(a);
1775 struct dp_netdev_port *p;
1777 switch ((enum ovs_action_attr)type) {
1778 case OVS_ACTION_ATTR_OUTPUT:
1779 p = dp_netdev_lookup_port(aux->dp, u32_to_odp(nl_attr_get_u32(a)));
1781 netdev_send(p->netdev, packet);
1785 case OVS_ACTION_ATTR_USERSPACE: {
1786 const struct nlattr *userdata;
1788 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1790 /* Make a copy if we are not allowed to steal the packet's data. */
1792 packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
1794 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1797 ofpbuf_uninit(packet);
1801 case OVS_ACTION_ATTR_PUSH_VLAN:
1802 case OVS_ACTION_ATTR_POP_VLAN:
1803 case OVS_ACTION_ATTR_PUSH_MPLS:
1804 case OVS_ACTION_ATTR_POP_MPLS:
1805 case OVS_ACTION_ATTR_SET:
1806 case OVS_ACTION_ATTR_SAMPLE:
1807 case OVS_ACTION_ATTR_UNSPEC:
1808 case __OVS_ACTION_ATTR_MAX:
1814 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1815 struct ofpbuf *packet, struct pkt_metadata *md,
1816 const struct nlattr *actions, size_t actions_len)
1817 OVS_REQ_RDLOCK(dp->port_rwlock)
1819 struct dp_netdev_execute_aux aux = {dp, key};
1821 odp_execute_actions(&aux, packet, md, actions, actions_len, dp_execute_cb);
1824 const struct dpif_class dpif_netdev_class = {
1826 dpif_netdev_enumerate,
1827 dpif_netdev_port_open_type,
1830 dpif_netdev_destroy,
1833 dpif_netdev_get_stats,
1834 dpif_netdev_port_add,
1835 dpif_netdev_port_del,
1836 dpif_netdev_port_query_by_number,
1837 dpif_netdev_port_query_by_name,
1838 NULL, /* port_get_pid */
1839 dpif_netdev_port_dump_start,
1840 dpif_netdev_port_dump_next,
1841 dpif_netdev_port_dump_done,
1842 dpif_netdev_port_poll,
1843 dpif_netdev_port_poll_wait,
1844 dpif_netdev_flow_get,
1845 dpif_netdev_flow_put,
1846 dpif_netdev_flow_del,
1847 dpif_netdev_flow_flush,
1848 dpif_netdev_flow_dump_start,
1849 dpif_netdev_flow_dump_next,
1850 dpif_netdev_flow_dump_done,
1851 dpif_netdev_execute,
1853 dpif_netdev_recv_set,
1854 dpif_netdev_queue_to_priority,
1856 dpif_netdev_recv_wait,
1857 dpif_netdev_recv_purge,
1861 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1862 const char *argv[], void *aux OVS_UNUSED)
1864 struct dp_netdev_port *port;
1865 struct dp_netdev *dp;
1868 ovs_mutex_lock(&dp_netdev_mutex);
1869 dp = shash_find_data(&dp_netdevs, argv[1]);
1870 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1871 ovs_mutex_unlock(&dp_netdev_mutex);
1872 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1875 ovs_refcount_ref(&dp->ref_cnt);
1876 ovs_mutex_unlock(&dp_netdev_mutex);
1878 ovs_rwlock_wrlock(&dp->port_rwlock);
1879 if (get_port_by_name(dp, argv[2], &port)) {
1880 unixctl_command_reply_error(conn, "unknown port");
1884 port_no = u32_to_odp(atoi(argv[3]));
1885 if (!port_no || port_no == ODPP_NONE) {
1886 unixctl_command_reply_error(conn, "bad port number");
1889 if (dp_netdev_lookup_port(dp, port_no)) {
1890 unixctl_command_reply_error(conn, "port number already in use");
1893 hmap_remove(&dp->ports, &port->node);
1894 port->port_no = port_no;
1895 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
1896 seq_change(dp->port_seq);
1897 unixctl_command_reply(conn, NULL);
1900 ovs_rwlock_unlock(&dp->port_rwlock);
1901 dp_netdev_unref(dp);
1905 dpif_dummy_register__(const char *type)
1907 struct dpif_class *class;
1909 class = xmalloc(sizeof *class);
1910 *class = dpif_netdev_class;
1911 class->type = xstrdup(type);
1912 dp_register_provider(class);
1916 dpif_dummy_register(bool override)
1923 dp_enumerate_types(&types);
1924 SSET_FOR_EACH (type, &types) {
1925 if (!dp_unregister_provider(type)) {
1926 dpif_dummy_register__(type);
1929 sset_destroy(&types);
1932 dpif_dummy_register__("dummy");
1934 unixctl_command_register("dpif-dummy/change-port-number",
1935 "DP PORT NEW-NUMBER",
1936 3, 3, dpif_dummy_change_port_number, NULL);