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_stats is internally synchronized. */
151 struct ovsthread_stats stats; /* Contains 'struct dp_netdev_stats *'. */
155 * Any lookup into 'ports' or any access to the dp_netdev_ports found
156 * through 'ports' requires taking 'port_rwlock'. */
157 struct ovs_rwlock port_rwlock;
158 struct hmap ports OVS_GUARDED;
159 struct seq *port_seq; /* Incremented whenever a port changes. */
161 /* Forwarding threads. */
162 struct latch exit_latch;
163 struct dp_forwarder *forwarders;
167 static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *dp,
169 OVS_REQ_RDLOCK(dp->port_rwlock);
172 DP_STAT_HIT, /* Packets that matched in the flow table. */
173 DP_STAT_MISS, /* Packets that did not match. */
174 DP_STAT_LOST, /* Packets not passed up to the client. */
178 /* Contained by struct dp_netdev's 'stats' member. */
179 struct dp_netdev_stats {
180 struct ovs_mutex mutex; /* Protects 'n'. */
182 /* Indexed by DP_STAT_*, protected by 'mutex'. */
183 unsigned long long int n[DP_N_STATS] OVS_GUARDED;
187 /* A port in a netdev-based datapath. */
188 struct dp_netdev_port {
189 struct hmap_node node; /* Node in dp_netdev's 'ports'. */
191 struct netdev *netdev;
192 struct netdev_saved_flags *sf;
193 struct netdev_rx *rx;
194 char *type; /* Port type as requested by user. */
197 /* A flow in dp_netdev's 'flow_table'.
203 * Except near the beginning or ending of its lifespan, rule 'rule' belongs to
204 * its dp_netdev's classifier. The text below calls this classifier 'cls'.
209 * The thread safety rules described here for "struct dp_netdev_flow" are
210 * motivated by two goals:
212 * - Prevent threads that read members of "struct dp_netdev_flow" from
213 * reading bad data due to changes by some thread concurrently modifying
216 * - Prevent two threads making changes to members of a given "struct
217 * dp_netdev_flow" from interfering with each other.
223 * A flow 'flow' may be accessed without a risk of being freed by code that
224 * holds a read-lock or write-lock on 'cls->rwlock' or that owns a reference to
225 * 'flow->ref_cnt' (or both). Code that needs to hold onto a flow for a while
226 * should take 'cls->rwlock', find the flow it needs, increment 'flow->ref_cnt'
227 * with dpif_netdev_flow_ref(), and drop 'cls->rwlock'.
229 * 'flow->ref_cnt' protects 'flow' from being freed. It doesn't protect the
230 * flow from being deleted from 'cls' (that's 'cls->rwlock') and it doesn't
231 * protect members of 'flow' from modification (that's 'flow->mutex').
233 * 'flow->mutex' protects the members of 'flow' from modification. It doesn't
234 * protect the flow from being deleted from 'cls' (that's 'cls->rwlock') and it
235 * doesn't prevent the flow from being freed (that's 'flow->ref_cnt').
237 * Some members, marked 'const', are immutable. Accessing other members
238 * requires synchronization, as noted in more detail below.
240 struct dp_netdev_flow {
241 /* Packet classification. */
242 const struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
244 /* Hash table index by unmasked flow. */
245 const struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
246 const struct flow flow; /* The flow that created this entry. */
248 /* Number of references.
249 * The classifier owns one reference.
250 * Any thread trying to keep a rule from being freed should hold its own
252 struct ovs_refcount ref_cnt;
254 /* Protects members marked OVS_GUARDED.
256 * Acquire after datapath's flow_mutex. */
257 struct ovs_mutex mutex OVS_ACQ_AFTER(dp_netdev_mutex);
261 * Reading or writing these members requires 'mutex'. */
262 long long int used OVS_GUARDED; /* Last used time, in monotonic msecs. */
263 long long int packet_count OVS_GUARDED; /* Number of packets matched. */
264 long long int byte_count OVS_GUARDED; /* Number of bytes matched. */
265 uint16_t tcp_flags OVS_GUARDED; /* Bitwise-OR of seen tcp_flags values. */
269 * Reading 'actions' requires 'mutex'.
270 * Writing 'actions' requires 'mutex' and (to allow for transactions) the
271 * datapath's flow_mutex. */
272 struct dp_netdev_actions *actions OVS_GUARDED;
275 static struct dp_netdev_flow *dp_netdev_flow_ref(
276 const struct dp_netdev_flow *);
277 static void dp_netdev_flow_unref(struct dp_netdev_flow *);
279 /* A set of datapath actions within a "struct dp_netdev_flow".
285 * A struct dp_netdev_actions 'actions' may be accessed without a risk of being
286 * freed by code that holds a read-lock or write-lock on 'flow->mutex' (where
287 * 'flow' is the dp_netdev_flow for which 'flow->actions == actions') or that
288 * owns a reference to 'actions->ref_cnt' (or both). */
289 struct dp_netdev_actions {
290 struct ovs_refcount ref_cnt;
292 /* These members are immutable: they do not change during the struct's
294 struct nlattr *actions; /* Sequence of OVS_ACTION_ATTR_* attributes. */
295 unsigned int size; /* Size of 'actions', in bytes. */
298 struct dp_netdev_actions *dp_netdev_actions_create(const struct nlattr *,
300 struct dp_netdev_actions *dp_netdev_actions_ref(
301 const struct dp_netdev_actions *);
302 void dp_netdev_actions_unref(struct dp_netdev_actions *);
304 /* A thread that receives packets from some ports, looks them up in the flow
305 * table, and executes the actions it finds. */
306 struct dp_forwarder {
307 struct dp_netdev *dp;
310 uint32_t min_hash, max_hash;
313 /* Interface to netdev-based datapath. */
316 struct dp_netdev *dp;
317 uint64_t last_port_seq;
320 static int get_port_by_number(struct dp_netdev *dp, odp_port_t port_no,
321 struct dp_netdev_port **portp)
322 OVS_REQ_RDLOCK(dp->port_rwlock);
323 static int get_port_by_name(struct dp_netdev *dp, const char *devname,
324 struct dp_netdev_port **portp)
325 OVS_REQ_RDLOCK(dp->port_rwlock);
326 static void dp_netdev_free(struct dp_netdev *)
327 OVS_REQUIRES(dp_netdev_mutex);
328 static void dp_netdev_flow_flush(struct dp_netdev *);
329 static int do_add_port(struct dp_netdev *dp, const char *devname,
330 const char *type, odp_port_t port_no)
331 OVS_REQ_WRLOCK(dp->port_rwlock);
332 static int do_del_port(struct dp_netdev *dp, odp_port_t port_no)
333 OVS_REQ_WRLOCK(dp->port_rwlock);
334 static int dpif_netdev_open(const struct dpif_class *, const char *name,
335 bool create, struct dpif **);
336 static int dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *,
337 int queue_no, const struct flow *,
338 const struct nlattr *userdata)
339 OVS_EXCLUDED(dp->queue_mutex);
340 static void dp_netdev_execute_actions(struct dp_netdev *dp,
341 const struct flow *, struct ofpbuf *,
342 struct pkt_metadata *,
343 const struct nlattr *actions,
345 OVS_REQ_RDLOCK(dp->port_rwlock);
346 static void dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
347 struct pkt_metadata *)
348 OVS_REQ_RDLOCK(dp->port_rwlock);
349 static void dp_netdev_set_threads(struct dp_netdev *, int n);
351 static struct dpif_netdev *
352 dpif_netdev_cast(const struct dpif *dpif)
354 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
355 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
358 static struct dp_netdev *
359 get_dp_netdev(const struct dpif *dpif)
361 return dpif_netdev_cast(dpif)->dp;
365 dpif_netdev_enumerate(struct sset *all_dps)
367 struct shash_node *node;
369 ovs_mutex_lock(&dp_netdev_mutex);
370 SHASH_FOR_EACH(node, &dp_netdevs) {
371 sset_add(all_dps, node->name);
373 ovs_mutex_unlock(&dp_netdev_mutex);
379 dpif_netdev_class_is_dummy(const struct dpif_class *class)
381 return class != &dpif_netdev_class;
385 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
387 return strcmp(type, "internal") ? type
388 : dpif_netdev_class_is_dummy(class) ? "dummy"
393 create_dpif_netdev(struct dp_netdev *dp)
395 uint16_t netflow_id = hash_string(dp->name, 0);
396 struct dpif_netdev *dpif;
398 ovs_refcount_ref(&dp->ref_cnt);
400 dpif = xmalloc(sizeof *dpif);
401 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
403 dpif->last_port_seq = seq_read(dp->port_seq);
408 /* Choose an unused, non-zero port number and return it on success.
409 * Return ODPP_NONE on failure. */
411 choose_port(struct dp_netdev *dp, const char *name)
412 OVS_REQ_RDLOCK(dp->port_rwlock)
416 if (dp->class != &dpif_netdev_class) {
420 /* If the port name begins with "br", start the number search at
421 * 100 to make writing tests easier. */
422 if (!strncmp(name, "br", 2)) {
426 /* If the port name contains a number, try to assign that port number.
427 * This can make writing unit tests easier because port numbers are
429 for (p = name; *p != '\0'; p++) {
430 if (isdigit((unsigned char) *p)) {
431 port_no = start_no + strtol(p, NULL, 10);
432 if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE)
433 && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
434 return u32_to_odp(port_no);
441 for (port_no = 1; port_no <= UINT16_MAX; port_no++) {
442 if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
443 return u32_to_odp(port_no);
451 create_dp_netdev(const char *name, const struct dpif_class *class,
452 struct dp_netdev **dpp)
453 OVS_REQUIRES(dp_netdev_mutex)
455 struct dp_netdev *dp;
459 dp = xzalloc(sizeof *dp);
460 shash_add(&dp_netdevs, name, dp);
462 *CONST_CAST(const struct dpif_class **, &dp->class) = class;
463 *CONST_CAST(const char **, &dp->name) = xstrdup(name);
464 ovs_refcount_init(&dp->ref_cnt);
465 atomic_flag_clear(&dp->destroyed);
467 ovs_mutex_init(&dp->flow_mutex);
468 classifier_init(&dp->cls, NULL);
469 hmap_init(&dp->flow_table);
471 ovs_mutex_init(&dp->queue_mutex);
472 ovs_mutex_lock(&dp->queue_mutex);
473 for (i = 0; i < N_QUEUES; i++) {
474 dp->queues[i].head = dp->queues[i].tail = 0;
476 ovs_mutex_unlock(&dp->queue_mutex);
477 dp->queue_seq = seq_create();
479 ovsthread_stats_init(&dp->stats);
481 ovs_rwlock_init(&dp->port_rwlock);
482 hmap_init(&dp->ports);
483 dp->port_seq = seq_create();
484 latch_init(&dp->exit_latch);
486 ovs_rwlock_wrlock(&dp->port_rwlock);
487 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
488 ovs_rwlock_unlock(&dp->port_rwlock);
493 dp_netdev_set_threads(dp, 2);
500 dpif_netdev_open(const struct dpif_class *class, const char *name,
501 bool create, struct dpif **dpifp)
503 struct dp_netdev *dp;
506 ovs_mutex_lock(&dp_netdev_mutex);
507 dp = shash_find_data(&dp_netdevs, name);
509 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
511 error = (dp->class != class ? EINVAL
516 *dpifp = create_dpif_netdev(dp);
518 ovs_mutex_unlock(&dp_netdev_mutex);
524 dp_netdev_purge_queues(struct dp_netdev *dp)
528 ovs_mutex_lock(&dp->queue_mutex);
529 for (i = 0; i < N_QUEUES; i++) {
530 struct dp_netdev_queue *q = &dp->queues[i];
532 while (q->tail != q->head) {
533 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
534 ofpbuf_uninit(&u->upcall.packet);
535 ofpbuf_uninit(&u->buf);
538 ovs_mutex_unlock(&dp->queue_mutex);
541 /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp'
542 * through the 'dp_netdevs' shash while freeing 'dp'. */
544 dp_netdev_free(struct dp_netdev *dp)
545 OVS_REQUIRES(dp_netdev_mutex)
547 struct dp_netdev_port *port, *next;
548 struct dp_netdev_stats *bucket;
551 shash_find_and_delete(&dp_netdevs, dp->name);
553 dp_netdev_set_threads(dp, 0);
554 free(dp->forwarders);
556 dp_netdev_flow_flush(dp);
557 ovs_rwlock_wrlock(&dp->port_rwlock);
558 HMAP_FOR_EACH_SAFE (port, next, node, &dp->ports) {
559 do_del_port(dp, port->port_no);
561 ovs_rwlock_unlock(&dp->port_rwlock);
563 OVSTHREAD_STATS_FOR_EACH_BUCKET (bucket, i, &dp->stats) {
564 ovs_mutex_destroy(&bucket->mutex);
565 free_cacheline(bucket);
567 ovsthread_stats_destroy(&dp->stats);
569 dp_netdev_purge_queues(dp);
570 seq_destroy(dp->queue_seq);
571 ovs_mutex_destroy(&dp->queue_mutex);
573 classifier_destroy(&dp->cls);
574 hmap_destroy(&dp->flow_table);
575 ovs_mutex_destroy(&dp->flow_mutex);
576 seq_destroy(dp->port_seq);
577 hmap_destroy(&dp->ports);
578 latch_destroy(&dp->exit_latch);
579 free(CONST_CAST(char *, dp->name));
584 dp_netdev_unref(struct dp_netdev *dp)
587 /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't
588 * get a new reference to 'dp' through the 'dp_netdevs' shash. */
589 ovs_mutex_lock(&dp_netdev_mutex);
590 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
593 ovs_mutex_unlock(&dp_netdev_mutex);
598 dpif_netdev_close(struct dpif *dpif)
600 struct dp_netdev *dp = get_dp_netdev(dpif);
607 dpif_netdev_destroy(struct dpif *dpif)
609 struct dp_netdev *dp = get_dp_netdev(dpif);
611 if (!atomic_flag_test_and_set(&dp->destroyed)) {
612 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
613 /* Can't happen: 'dpif' still owns a reference to 'dp'. */
622 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
624 struct dp_netdev *dp = get_dp_netdev(dpif);
625 struct dp_netdev_stats *bucket;
628 fat_rwlock_rdlock(&dp->cls.rwlock);
629 stats->n_flows = hmap_count(&dp->flow_table);
630 fat_rwlock_unlock(&dp->cls.rwlock);
632 stats->n_hit = stats->n_missed = stats->n_lost = 0;
633 OVSTHREAD_STATS_FOR_EACH_BUCKET (bucket, i, &dp->stats) {
634 ovs_mutex_lock(&bucket->mutex);
635 stats->n_hit += bucket->n[DP_STAT_HIT];
636 stats->n_missed += bucket->n[DP_STAT_MISS];
637 stats->n_lost += bucket->n[DP_STAT_LOST];
638 ovs_mutex_unlock(&bucket->mutex);
640 stats->n_masks = UINT32_MAX;
641 stats->n_mask_hit = UINT64_MAX;
647 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
649 OVS_REQ_WRLOCK(dp->port_rwlock)
651 struct netdev_saved_flags *sf;
652 struct dp_netdev_port *port;
653 struct netdev *netdev;
654 struct netdev_rx *rx;
655 enum netdev_flags flags;
656 const char *open_type;
659 /* XXX reject devices already in some dp_netdev. */
661 /* Open and validate network device. */
662 open_type = dpif_netdev_port_open_type(dp->class, type);
663 error = netdev_open(devname, open_type, &netdev);
667 /* XXX reject non-Ethernet devices */
669 netdev_get_flags(netdev, &flags);
670 if (flags & NETDEV_LOOPBACK) {
671 VLOG_ERR("%s: cannot add a loopback device", devname);
672 netdev_close(netdev);
676 error = netdev_rx_open(netdev, &rx);
678 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
679 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
680 devname, ovs_strerror(errno));
681 netdev_close(netdev);
685 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
688 netdev_close(netdev);
692 port = xmalloc(sizeof *port);
693 port->port_no = port_no;
694 port->netdev = netdev;
697 port->type = xstrdup(type);
699 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
700 seq_change(dp->port_seq);
706 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
707 odp_port_t *port_nop)
709 struct dp_netdev *dp = get_dp_netdev(dpif);
710 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
711 const char *dpif_port;
715 ovs_rwlock_wrlock(&dp->port_rwlock);
716 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
717 if (*port_nop != ODPP_NONE) {
719 error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0;
721 port_no = choose_port(dp, dpif_port);
722 error = port_no == ODPP_NONE ? EFBIG : 0;
726 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
728 ovs_rwlock_unlock(&dp->port_rwlock);
734 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
736 struct dp_netdev *dp = get_dp_netdev(dpif);
739 ovs_rwlock_wrlock(&dp->port_rwlock);
740 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
741 ovs_rwlock_unlock(&dp->port_rwlock);
747 is_valid_port_number(odp_port_t port_no)
749 return port_no != ODPP_NONE;
752 static struct dp_netdev_port *
753 dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no)
754 OVS_REQ_RDLOCK(dp->port_rwlock)
756 struct dp_netdev_port *port;
758 HMAP_FOR_EACH_IN_BUCKET (port, node, hash_int(odp_to_u32(port_no), 0),
760 if (port->port_no == port_no) {
768 get_port_by_number(struct dp_netdev *dp,
769 odp_port_t port_no, struct dp_netdev_port **portp)
770 OVS_REQ_RDLOCK(dp->port_rwlock)
772 if (!is_valid_port_number(port_no)) {
776 *portp = dp_netdev_lookup_port(dp, port_no);
777 return *portp ? 0 : ENOENT;
782 get_port_by_name(struct dp_netdev *dp,
783 const char *devname, struct dp_netdev_port **portp)
784 OVS_REQ_RDLOCK(dp->port_rwlock)
786 struct dp_netdev_port *port;
788 HMAP_FOR_EACH (port, node, &dp->ports) {
789 if (!strcmp(netdev_get_name(port->netdev), devname)) {
798 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
799 OVS_REQ_WRLOCK(dp->port_rwlock)
801 struct dp_netdev_port *port;
804 error = get_port_by_number(dp, port_no, &port);
809 hmap_remove(&dp->ports, &port->node);
810 seq_change(dp->port_seq);
812 netdev_close(port->netdev);
813 netdev_restore_flags(port->sf);
814 netdev_rx_close(port->rx);
822 answer_port_query(const struct dp_netdev_port *port,
823 struct dpif_port *dpif_port)
825 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
826 dpif_port->type = xstrdup(port->type);
827 dpif_port->port_no = port->port_no;
831 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
832 struct dpif_port *dpif_port)
834 struct dp_netdev *dp = get_dp_netdev(dpif);
835 struct dp_netdev_port *port;
838 ovs_rwlock_rdlock(&dp->port_rwlock);
839 error = get_port_by_number(dp, port_no, &port);
840 if (!error && dpif_port) {
841 answer_port_query(port, dpif_port);
843 ovs_rwlock_unlock(&dp->port_rwlock);
849 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
850 struct dpif_port *dpif_port)
852 struct dp_netdev *dp = get_dp_netdev(dpif);
853 struct dp_netdev_port *port;
856 ovs_rwlock_rdlock(&dp->port_rwlock);
857 error = get_port_by_name(dp, devname, &port);
858 if (!error && dpif_port) {
859 answer_port_query(port, dpif_port);
861 ovs_rwlock_unlock(&dp->port_rwlock);
867 dp_netdev_remove_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
868 OVS_REQ_WRLOCK(dp->cls.rwlock)
869 OVS_REQUIRES(dp->flow_mutex)
871 struct cls_rule *cr = CONST_CAST(struct cls_rule *, &flow->cr);
872 struct hmap_node *node = CONST_CAST(struct hmap_node *, &flow->node);
874 classifier_remove(&dp->cls, cr);
875 hmap_remove(&dp->flow_table, node);
876 dp_netdev_flow_unref(flow);
879 static struct dp_netdev_flow *
880 dp_netdev_flow_ref(const struct dp_netdev_flow *flow_)
882 struct dp_netdev_flow *flow = CONST_CAST(struct dp_netdev_flow *, flow_);
884 ovs_refcount_ref(&flow->ref_cnt);
890 dp_netdev_flow_unref(struct dp_netdev_flow *flow)
892 if (flow && ovs_refcount_unref(&flow->ref_cnt) == 1) {
893 cls_rule_destroy(CONST_CAST(struct cls_rule *, &flow->cr));
894 ovs_mutex_lock(&flow->mutex);
895 dp_netdev_actions_unref(flow->actions);
896 ovs_mutex_unlock(&flow->mutex);
897 ovs_mutex_destroy(&flow->mutex);
903 dp_netdev_flow_flush(struct dp_netdev *dp)
905 struct dp_netdev_flow *netdev_flow, *next;
907 ovs_mutex_lock(&dp->flow_mutex);
908 fat_rwlock_wrlock(&dp->cls.rwlock);
909 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
910 dp_netdev_remove_flow(dp, netdev_flow);
912 fat_rwlock_unlock(&dp->cls.rwlock);
913 ovs_mutex_unlock(&dp->flow_mutex);
917 dpif_netdev_flow_flush(struct dpif *dpif)
919 struct dp_netdev *dp = get_dp_netdev(dpif);
921 dp_netdev_flow_flush(dp);
925 struct dp_netdev_port_state {
932 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
934 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
939 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
940 struct dpif_port *dpif_port)
942 struct dp_netdev_port_state *state = state_;
943 struct dp_netdev *dp = get_dp_netdev(dpif);
944 struct hmap_node *node;
947 ovs_rwlock_rdlock(&dp->port_rwlock);
948 node = hmap_at_position(&dp->ports, &state->bucket, &state->offset);
950 struct dp_netdev_port *port;
952 port = CONTAINER_OF(node, struct dp_netdev_port, node);
955 state->name = xstrdup(netdev_get_name(port->netdev));
956 dpif_port->name = state->name;
957 dpif_port->type = port->type;
958 dpif_port->port_no = port->port_no;
964 ovs_rwlock_unlock(&dp->port_rwlock);
970 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
972 struct dp_netdev_port_state *state = state_;
979 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
981 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
982 uint64_t new_port_seq;
985 new_port_seq = seq_read(dpif->dp->port_seq);
986 if (dpif->last_port_seq != new_port_seq) {
987 dpif->last_port_seq = new_port_seq;
997 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
999 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
1001 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
1004 static struct dp_netdev_flow *
1005 dp_netdev_flow_cast(const struct cls_rule *cr)
1007 return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL;
1010 static struct dp_netdev_flow *
1011 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
1012 OVS_EXCLUDED(dp->cls.rwlock)
1014 struct dp_netdev_flow *netdev_flow;
1016 fat_rwlock_rdlock(&dp->cls.rwlock);
1017 netdev_flow = dp_netdev_flow_cast(classifier_lookup(&dp->cls, flow, NULL));
1018 dp_netdev_flow_ref(netdev_flow);
1019 fat_rwlock_unlock(&dp->cls.rwlock);
1024 static struct dp_netdev_flow *
1025 dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
1026 OVS_REQ_RDLOCK(dp->cls.rwlock)
1028 struct dp_netdev_flow *netdev_flow;
1030 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
1032 if (flow_equal(&netdev_flow->flow, flow)) {
1033 return dp_netdev_flow_ref(netdev_flow);
1041 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
1042 struct dpif_flow_stats *stats)
1043 OVS_REQ_RDLOCK(netdev_flow->mutex)
1045 stats->n_packets = netdev_flow->packet_count;
1046 stats->n_bytes = netdev_flow->byte_count;
1047 stats->used = netdev_flow->used;
1048 stats->tcp_flags = netdev_flow->tcp_flags;
1052 dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1053 const struct nlattr *mask_key,
1054 uint32_t mask_key_len, const struct flow *flow,
1058 enum odp_key_fitness fitness;
1060 fitness = odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow);
1062 /* This should not happen: it indicates that
1063 * odp_flow_key_from_mask() and odp_flow_key_to_mask()
1064 * disagree on the acceptable form of a mask. Log the problem
1065 * as an error, with enough details to enable debugging. */
1066 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1068 if (!VLOG_DROP_ERR(&rl)) {
1072 odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
1074 VLOG_ERR("internal error parsing flow mask %s (%s)",
1075 ds_cstr(&s), odp_key_fitness_to_string(fitness));
1081 /* Force unwildcard the in_port. */
1082 mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
1084 enum mf_field_id id;
1085 /* No mask key, unwildcard everything except fields whose
1086 * prerequisities are not met. */
1087 memset(mask, 0x0, sizeof *mask);
1089 for (id = 0; id < MFF_N_IDS; ++id) {
1090 /* Skip registers and metadata. */
1091 if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
1092 && id != MFF_METADATA) {
1093 const struct mf_field *mf = mf_from_id(id);
1094 if (mf_are_prereqs_ok(mf, flow)) {
1095 mf_mask_field(mf, mask);
1105 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1110 if (odp_flow_key_to_flow(key, key_len, flow)) {
1111 /* This should not happen: it indicates that odp_flow_key_from_flow()
1112 * and odp_flow_key_to_flow() disagree on the acceptable form of a
1113 * flow. Log the problem as an error, with enough details to enable
1115 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1117 if (!VLOG_DROP_ERR(&rl)) {
1121 odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
1122 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
1129 in_port = flow->in_port.odp_port;
1130 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
1138 dpif_netdev_flow_get(const struct dpif *dpif,
1139 const struct nlattr *nl_key, size_t nl_key_len,
1140 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
1142 struct dp_netdev *dp = get_dp_netdev(dpif);
1143 struct dp_netdev_flow *netdev_flow;
1147 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
1152 fat_rwlock_rdlock(&dp->cls.rwlock);
1153 netdev_flow = dp_netdev_find_flow(dp, &key);
1154 fat_rwlock_unlock(&dp->cls.rwlock);
1157 struct dp_netdev_actions *actions = NULL;
1159 ovs_mutex_lock(&netdev_flow->mutex);
1161 get_dpif_flow_stats(netdev_flow, stats);
1164 actions = dp_netdev_actions_ref(netdev_flow->actions);
1166 ovs_mutex_unlock(&netdev_flow->mutex);
1168 dp_netdev_flow_unref(netdev_flow);
1171 *actionsp = ofpbuf_clone_data(actions->actions, actions->size);
1172 dp_netdev_actions_unref(actions);
1182 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
1183 const struct flow_wildcards *wc,
1184 const struct nlattr *actions,
1186 OVS_REQUIRES(dp->flow_mutex)
1188 struct dp_netdev_flow *netdev_flow;
1191 netdev_flow = xzalloc(sizeof *netdev_flow);
1192 *CONST_CAST(struct flow *, &netdev_flow->flow) = *flow;
1193 ovs_refcount_init(&netdev_flow->ref_cnt);
1195 ovs_mutex_init(&netdev_flow->mutex);
1196 ovs_mutex_lock(&netdev_flow->mutex);
1198 netdev_flow->actions = dp_netdev_actions_create(actions, actions_len);
1200 match_init(&match, flow, wc);
1201 cls_rule_init(CONST_CAST(struct cls_rule *, &netdev_flow->cr),
1202 &match, NETDEV_RULE_PRIORITY);
1203 fat_rwlock_wrlock(&dp->cls.rwlock);
1204 classifier_insert(&dp->cls,
1205 CONST_CAST(struct cls_rule *, &netdev_flow->cr));
1206 hmap_insert(&dp->flow_table,
1207 CONST_CAST(struct hmap_node *, &netdev_flow->node),
1208 flow_hash(flow, 0));
1209 fat_rwlock_unlock(&dp->cls.rwlock);
1211 ovs_mutex_unlock(&netdev_flow->mutex);
1217 clear_stats(struct dp_netdev_flow *netdev_flow)
1218 OVS_REQUIRES(netdev_flow->mutex)
1220 netdev_flow->used = 0;
1221 netdev_flow->packet_count = 0;
1222 netdev_flow->byte_count = 0;
1223 netdev_flow->tcp_flags = 0;
1227 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
1229 struct dp_netdev *dp = get_dp_netdev(dpif);
1230 struct dp_netdev_flow *netdev_flow;
1232 struct flow_wildcards wc;
1235 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
1239 error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
1240 put->mask, put->mask_len,
1246 ovs_mutex_lock(&dp->flow_mutex);
1247 netdev_flow = dp_netdev_lookup_flow(dp, &flow);
1249 if (put->flags & DPIF_FP_CREATE) {
1250 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
1252 memset(put->stats, 0, sizeof *put->stats);
1254 error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
1263 if (put->flags & DPIF_FP_MODIFY
1264 && flow_equal(&flow, &netdev_flow->flow)) {
1265 struct dp_netdev_actions *new_actions;
1266 struct dp_netdev_actions *old_actions;
1268 new_actions = dp_netdev_actions_create(put->actions,
1271 ovs_mutex_lock(&netdev_flow->mutex);
1272 old_actions = netdev_flow->actions;
1273 netdev_flow->actions = new_actions;
1275 get_dpif_flow_stats(netdev_flow, put->stats);
1277 if (put->flags & DPIF_FP_ZERO_STATS) {
1278 clear_stats(netdev_flow);
1280 ovs_mutex_unlock(&netdev_flow->mutex);
1282 dp_netdev_actions_unref(old_actions);
1283 } else if (put->flags & DPIF_FP_CREATE) {
1286 /* Overlapping flow. */
1289 dp_netdev_flow_unref(netdev_flow);
1291 ovs_mutex_unlock(&dp->flow_mutex);
1297 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
1299 struct dp_netdev *dp = get_dp_netdev(dpif);
1300 struct dp_netdev_flow *netdev_flow;
1304 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
1309 ovs_mutex_lock(&dp->flow_mutex);
1310 fat_rwlock_wrlock(&dp->cls.rwlock);
1311 netdev_flow = dp_netdev_find_flow(dp, &key);
1314 ovs_mutex_lock(&netdev_flow->mutex);
1315 get_dpif_flow_stats(netdev_flow, del->stats);
1316 ovs_mutex_unlock(&netdev_flow->mutex);
1318 dp_netdev_remove_flow(dp, netdev_flow);
1319 dp_netdev_flow_unref(netdev_flow);
1323 fat_rwlock_unlock(&dp->cls.rwlock);
1324 ovs_mutex_unlock(&dp->flow_mutex);
1329 struct dp_netdev_flow_state {
1330 struct dp_netdev_actions *actions;
1331 struct odputil_keybuf keybuf;
1332 struct odputil_keybuf maskbuf;
1333 struct dpif_flow_stats stats;
1336 struct dp_netdev_flow_iter {
1340 struct ovs_mutex mutex;
1344 dpif_netdev_flow_dump_state_init(void **statep)
1346 struct dp_netdev_flow_state *state;
1348 *statep = state = xmalloc(sizeof *state);
1349 state->actions = NULL;
1353 dpif_netdev_flow_dump_state_uninit(void *state_)
1355 struct dp_netdev_flow_state *state = state_;
1357 dp_netdev_actions_unref(state->actions);
1362 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **iterp)
1364 struct dp_netdev_flow_iter *iter;
1366 *iterp = iter = xmalloc(sizeof *iter);
1370 ovs_mutex_init(&iter->mutex);
1375 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *iter_, void *state_,
1376 const struct nlattr **key, size_t *key_len,
1377 const struct nlattr **mask, size_t *mask_len,
1378 const struct nlattr **actions, size_t *actions_len,
1379 const struct dpif_flow_stats **stats)
1381 struct dp_netdev_flow_iter *iter = iter_;
1382 struct dp_netdev_flow_state *state = state_;
1383 struct dp_netdev *dp = get_dp_netdev(dpif);
1384 struct dp_netdev_flow *netdev_flow;
1387 ovs_mutex_lock(&iter->mutex);
1388 error = iter->status;
1390 struct hmap_node *node;
1392 fat_rwlock_rdlock(&dp->cls.rwlock);
1393 node = hmap_at_position(&dp->flow_table, &iter->bucket, &iter->offset);
1395 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
1396 dp_netdev_flow_ref(netdev_flow);
1398 fat_rwlock_unlock(&dp->cls.rwlock);
1400 iter->status = error = EOF;
1403 ovs_mutex_unlock(&iter->mutex);
1411 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
1412 odp_flow_key_from_flow(&buf, &netdev_flow->flow,
1413 netdev_flow->flow.in_port.odp_port);
1416 *key_len = buf.size;
1421 struct flow_wildcards wc;
1423 ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
1424 minimask_expand(&netdev_flow->cr.match.mask, &wc);
1425 odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
1426 odp_to_u32(wc.masks.in_port.odp_port),
1430 *mask_len = buf.size;
1433 if (actions || stats) {
1434 dp_netdev_actions_unref(state->actions);
1435 state->actions = NULL;
1437 ovs_mutex_lock(&netdev_flow->mutex);
1439 state->actions = dp_netdev_actions_ref(netdev_flow->actions);
1440 *actions = state->actions->actions;
1441 *actions_len = state->actions->size;
1444 get_dpif_flow_stats(netdev_flow, &state->stats);
1445 *stats = &state->stats;
1447 ovs_mutex_unlock(&netdev_flow->mutex);
1450 dp_netdev_flow_unref(netdev_flow);
1456 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *iter_)
1458 struct dp_netdev_flow_iter *iter = iter_;
1460 ovs_mutex_destroy(&iter->mutex);
1466 dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute)
1468 struct dp_netdev *dp = get_dp_netdev(dpif);
1469 struct pkt_metadata *md = &execute->md;
1472 if (execute->packet->size < ETH_HEADER_LEN ||
1473 execute->packet->size > UINT16_MAX) {
1477 /* Extract flow key. */
1478 flow_extract(execute->packet, md, &key);
1480 ovs_rwlock_rdlock(&dp->port_rwlock);
1481 dp_netdev_execute_actions(dp, &key, execute->packet, md, execute->actions,
1482 execute->actions_len);
1483 ovs_rwlock_unlock(&dp->port_rwlock);
1489 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1495 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1496 uint32_t queue_id, uint32_t *priority)
1498 *priority = queue_id;
1502 static struct dp_netdev_queue *
1503 find_nonempty_queue(struct dp_netdev *dp)
1504 OVS_REQUIRES(dp->queue_mutex)
1508 for (i = 0; i < N_QUEUES; i++) {
1509 struct dp_netdev_queue *q = &dp->queues[i];
1510 if (q->head != q->tail) {
1518 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1521 struct dp_netdev *dp = get_dp_netdev(dpif);
1522 struct dp_netdev_queue *q;
1525 ovs_mutex_lock(&dp->queue_mutex);
1526 q = find_nonempty_queue(dp);
1528 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1530 *upcall = u->upcall;
1539 ovs_mutex_unlock(&dp->queue_mutex);
1545 dpif_netdev_recv_wait(struct dpif *dpif)
1547 struct dp_netdev *dp = get_dp_netdev(dpif);
1550 ovs_mutex_lock(&dp->queue_mutex);
1551 seq = seq_read(dp->queue_seq);
1552 if (find_nonempty_queue(dp)) {
1553 poll_immediate_wake();
1555 seq_wait(dp->queue_seq, seq);
1557 ovs_mutex_unlock(&dp->queue_mutex);
1561 dpif_netdev_recv_purge(struct dpif *dpif)
1563 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1565 dp_netdev_purge_queues(dpif_netdev->dp);
1568 /* Creates and returns a new 'struct dp_netdev_actions', with a reference count
1569 * of 1, whose actions are a copy of from the 'ofpacts_len' bytes of
1571 struct dp_netdev_actions *
1572 dp_netdev_actions_create(const struct nlattr *actions, size_t size)
1574 struct dp_netdev_actions *netdev_actions;
1576 netdev_actions = xmalloc(sizeof *netdev_actions);
1577 ovs_refcount_init(&netdev_actions->ref_cnt);
1578 netdev_actions->actions = xmemdup(actions, size);
1579 netdev_actions->size = size;
1581 return netdev_actions;
1584 /* Increments 'actions''s refcount. */
1585 struct dp_netdev_actions *
1586 dp_netdev_actions_ref(const struct dp_netdev_actions *actions_)
1588 struct dp_netdev_actions *actions;
1590 actions = CONST_CAST(struct dp_netdev_actions *, actions_);
1592 ovs_refcount_ref(&actions->ref_cnt);
1597 /* Decrements 'actions''s refcount and frees 'actions' if the refcount reaches
1600 dp_netdev_actions_unref(struct dp_netdev_actions *actions)
1602 if (actions && ovs_refcount_unref(&actions->ref_cnt) == 1) {
1603 free(actions->actions);
1609 dp_forwarder_main(void *f_)
1611 struct dp_forwarder *f = f_;
1612 struct dp_netdev *dp = f->dp;
1613 struct ofpbuf packet;
1615 f->name = xasprintf("forwarder_%u", ovsthread_id_self());
1616 set_subprogram_name("%s", f->name);
1618 ofpbuf_init(&packet, 0);
1619 while (!latch_is_set(&dp->exit_latch)) {
1620 bool received_anything;
1623 ovs_rwlock_rdlock(&dp->port_rwlock);
1624 for (i = 0; i < 50; i++) {
1625 struct dp_netdev_port *port;
1627 received_anything = false;
1628 HMAP_FOR_EACH (port, node, &f->dp->ports) {
1630 && port->node.hash >= f->min_hash
1631 && port->node.hash <= f->max_hash) {
1636 if (netdev_get_mtu(port->netdev, &mtu)) {
1637 mtu = ETH_PAYLOAD_MAX;
1639 buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + mtu;
1641 ofpbuf_clear(&packet);
1642 ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM,
1645 error = netdev_rx_recv(port->rx, &packet);
1647 struct pkt_metadata md
1648 = PKT_METADATA_INITIALIZER(port->port_no);
1650 dp_netdev_port_input(dp, &packet, &md);
1651 received_anything = true;
1652 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1653 static struct vlog_rate_limit rl
1654 = VLOG_RATE_LIMIT_INIT(1, 5);
1656 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1657 netdev_get_name(port->netdev),
1658 ovs_strerror(error));
1663 if (!received_anything) {
1668 if (received_anything) {
1669 poll_immediate_wake();
1671 struct dp_netdev_port *port;
1673 HMAP_FOR_EACH (port, node, &f->dp->ports)
1675 && port->node.hash >= f->min_hash
1676 && port->node.hash <= f->max_hash) {
1677 netdev_rx_wait(port->rx);
1679 seq_wait(dp->port_seq, seq_read(dp->port_seq));
1680 latch_wait(&dp->exit_latch);
1682 ovs_rwlock_unlock(&dp->port_rwlock);
1686 ofpbuf_uninit(&packet);
1694 dp_netdev_set_threads(struct dp_netdev *dp, int n)
1698 if (n == dp->n_forwarders) {
1702 /* Stop existing threads. */
1703 latch_set(&dp->exit_latch);
1704 for (i = 0; i < dp->n_forwarders; i++) {
1705 struct dp_forwarder *f = &dp->forwarders[i];
1707 xpthread_join(f->thread, NULL);
1709 latch_poll(&dp->exit_latch);
1710 free(dp->forwarders);
1712 /* Start new threads. */
1713 dp->forwarders = xmalloc(n * sizeof *dp->forwarders);
1714 dp->n_forwarders = n;
1715 for (i = 0; i < n; i++) {
1716 struct dp_forwarder *f = &dp->forwarders[i];
1719 f->min_hash = UINT32_MAX / n * i;
1720 f->max_hash = UINT32_MAX / n * (i + 1) - 1;
1722 f->max_hash = UINT32_MAX;
1724 xpthread_create(&f->thread, NULL, dp_forwarder_main, f);
1729 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1730 const struct ofpbuf *packet)
1731 OVS_REQUIRES(netdev_flow->mutex)
1733 netdev_flow->used = time_msec();
1734 netdev_flow->packet_count++;
1735 netdev_flow->byte_count += packet->size;
1736 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
1740 dp_netdev_stats_new_cb(void)
1742 struct dp_netdev_stats *bucket = xzalloc_cacheline(sizeof *bucket);
1743 ovs_mutex_init(&bucket->mutex);
1748 dp_netdev_count_packet(struct dp_netdev *dp, enum dp_stat_type type)
1750 struct dp_netdev_stats *bucket;
1752 bucket = ovsthread_stats_bucket_get(&dp->stats, dp_netdev_stats_new_cb);
1753 ovs_mutex_lock(&bucket->mutex);
1755 ovs_mutex_unlock(&bucket->mutex);
1759 dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
1760 struct pkt_metadata *md)
1761 OVS_REQ_RDLOCK(dp->port_rwlock)
1763 struct dp_netdev_flow *netdev_flow;
1766 if (packet->size < ETH_HEADER_LEN) {
1769 flow_extract(packet, md, &key);
1770 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1772 struct dp_netdev_actions *actions;
1774 ovs_mutex_lock(&netdev_flow->mutex);
1775 dp_netdev_flow_used(netdev_flow, packet);
1776 actions = dp_netdev_actions_ref(netdev_flow->actions);
1777 ovs_mutex_unlock(&netdev_flow->mutex);
1779 dp_netdev_execute_actions(dp, &key, packet, md,
1780 actions->actions, actions->size);
1781 dp_netdev_actions_unref(actions);
1782 dp_netdev_flow_unref(netdev_flow);
1783 dp_netdev_count_packet(dp, DP_STAT_HIT);
1785 dp_netdev_count_packet(dp, DP_STAT_MISS);
1786 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1791 dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
1792 int queue_no, const struct flow *flow,
1793 const struct nlattr *userdata)
1794 OVS_EXCLUDED(dp->queue_mutex)
1796 struct dp_netdev_queue *q = &dp->queues[queue_no];
1799 ovs_mutex_lock(&dp->queue_mutex);
1800 if (q->head - q->tail < MAX_QUEUE_LEN) {
1801 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1802 struct dpif_upcall *upcall = &u->upcall;
1803 struct ofpbuf *buf = &u->buf;
1806 upcall->type = queue_no;
1808 /* Allocate buffer big enough for everything. */
1809 buf_size = ODPUTIL_FLOW_KEY_BYTES;
1811 buf_size += NLA_ALIGN(userdata->nla_len);
1813 ofpbuf_init(buf, buf_size);
1816 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1817 upcall->key = buf->data;
1818 upcall->key_len = buf->size;
1822 upcall->userdata = ofpbuf_put(buf, userdata,
1823 NLA_ALIGN(userdata->nla_len));
1826 /* Steal packet data. */
1827 ovs_assert(packet->source == OFPBUF_MALLOC);
1828 upcall->packet = *packet;
1829 ofpbuf_use(packet, NULL, 0);
1831 seq_change(dp->queue_seq);
1835 dp_netdev_count_packet(dp, DP_STAT_LOST);
1838 ovs_mutex_unlock(&dp->queue_mutex);
1843 struct dp_netdev_execute_aux {
1844 struct dp_netdev *dp;
1845 const struct flow *key;
1849 dp_execute_cb(void *aux_, struct ofpbuf *packet,
1850 const struct pkt_metadata *md OVS_UNUSED,
1851 const struct nlattr *a, bool may_steal)
1852 OVS_NO_THREAD_SAFETY_ANALYSIS
1854 struct dp_netdev_execute_aux *aux = aux_;
1855 int type = nl_attr_type(a);
1856 struct dp_netdev_port *p;
1858 switch ((enum ovs_action_attr)type) {
1859 case OVS_ACTION_ATTR_OUTPUT:
1860 p = dp_netdev_lookup_port(aux->dp, u32_to_odp(nl_attr_get_u32(a)));
1862 netdev_send(p->netdev, packet);
1866 case OVS_ACTION_ATTR_USERSPACE: {
1867 const struct nlattr *userdata;
1869 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1871 /* Make a copy if we are not allowed to steal the packet's data. */
1873 packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
1875 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1878 ofpbuf_uninit(packet);
1882 case OVS_ACTION_ATTR_PUSH_VLAN:
1883 case OVS_ACTION_ATTR_POP_VLAN:
1884 case OVS_ACTION_ATTR_PUSH_MPLS:
1885 case OVS_ACTION_ATTR_POP_MPLS:
1886 case OVS_ACTION_ATTR_SET:
1887 case OVS_ACTION_ATTR_SAMPLE:
1888 case OVS_ACTION_ATTR_UNSPEC:
1889 case __OVS_ACTION_ATTR_MAX:
1895 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1896 struct ofpbuf *packet, struct pkt_metadata *md,
1897 const struct nlattr *actions, size_t actions_len)
1898 OVS_REQ_RDLOCK(dp->port_rwlock)
1900 struct dp_netdev_execute_aux aux = {dp, key};
1902 odp_execute_actions(&aux, packet, md, actions, actions_len, dp_execute_cb);
1905 const struct dpif_class dpif_netdev_class = {
1907 dpif_netdev_enumerate,
1908 dpif_netdev_port_open_type,
1911 dpif_netdev_destroy,
1914 dpif_netdev_get_stats,
1915 dpif_netdev_port_add,
1916 dpif_netdev_port_del,
1917 dpif_netdev_port_query_by_number,
1918 dpif_netdev_port_query_by_name,
1919 NULL, /* port_get_pid */
1920 dpif_netdev_port_dump_start,
1921 dpif_netdev_port_dump_next,
1922 dpif_netdev_port_dump_done,
1923 dpif_netdev_port_poll,
1924 dpif_netdev_port_poll_wait,
1925 dpif_netdev_flow_get,
1926 dpif_netdev_flow_put,
1927 dpif_netdev_flow_del,
1928 dpif_netdev_flow_flush,
1929 dpif_netdev_flow_dump_state_init,
1930 dpif_netdev_flow_dump_start,
1931 dpif_netdev_flow_dump_next,
1933 dpif_netdev_flow_dump_done,
1934 dpif_netdev_flow_dump_state_uninit,
1935 dpif_netdev_execute,
1937 dpif_netdev_recv_set,
1938 dpif_netdev_queue_to_priority,
1940 dpif_netdev_recv_wait,
1941 dpif_netdev_recv_purge,
1945 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1946 const char *argv[], void *aux OVS_UNUSED)
1948 struct dp_netdev_port *port;
1949 struct dp_netdev *dp;
1952 ovs_mutex_lock(&dp_netdev_mutex);
1953 dp = shash_find_data(&dp_netdevs, argv[1]);
1954 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1955 ovs_mutex_unlock(&dp_netdev_mutex);
1956 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1959 ovs_refcount_ref(&dp->ref_cnt);
1960 ovs_mutex_unlock(&dp_netdev_mutex);
1962 ovs_rwlock_wrlock(&dp->port_rwlock);
1963 if (get_port_by_name(dp, argv[2], &port)) {
1964 unixctl_command_reply_error(conn, "unknown port");
1968 port_no = u32_to_odp(atoi(argv[3]));
1969 if (!port_no || port_no == ODPP_NONE) {
1970 unixctl_command_reply_error(conn, "bad port number");
1973 if (dp_netdev_lookup_port(dp, port_no)) {
1974 unixctl_command_reply_error(conn, "port number already in use");
1977 hmap_remove(&dp->ports, &port->node);
1978 port->port_no = port_no;
1979 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
1980 seq_change(dp->port_seq);
1981 unixctl_command_reply(conn, NULL);
1984 ovs_rwlock_unlock(&dp->port_rwlock);
1985 dp_netdev_unref(dp);
1989 dpif_dummy_register__(const char *type)
1991 struct dpif_class *class;
1993 class = xmalloc(sizeof *class);
1994 *class = dpif_netdev_class;
1995 class->type = xstrdup(type);
1996 dp_register_provider(class);
2000 dpif_dummy_register(bool override)
2007 dp_enumerate_types(&types);
2008 SSET_FOR_EACH (type, &types) {
2009 if (!dp_unregister_provider(type)) {
2010 dpif_dummy_register__(type);
2013 sset_destroy(&types);
2016 dpif_dummy_register__("dummy");
2018 unixctl_command_register("dpif-dummy/change-port-number",
2019 "DP PORT NEW-NUMBER",
2020 3, 3, dpif_dummy_change_port_number, NULL);