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_class_is_planetlab(const struct dpif_class *class)
373 return class == &dpif_planetlab_class;
377 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
379 return strcmp(type, "internal") ? type
380 : dpif_netdev_class_is_planetlab(class) ? "pltap"
381 : dpif_netdev_class_is_dummy(class) ? "dummy"
386 create_dpif_netdev(struct dp_netdev *dp)
388 uint16_t netflow_id = hash_string(dp->name, 0);
389 struct dpif_netdev *dpif;
391 ovs_refcount_ref(&dp->ref_cnt);
393 dpif = xmalloc(sizeof *dpif);
394 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
396 dpif->last_port_seq = seq_read(dp->port_seq);
401 /* Choose an unused, non-zero port number and return it on success.
402 * Return ODPP_NONE on failure. */
404 choose_port(struct dp_netdev *dp, const char *name)
405 OVS_REQ_RDLOCK(dp->port_rwlock)
409 if (dp->class != &dpif_netdev_class &&
410 dp->class != &dpif_planetlab_class) {
414 /* If the port name begins with "br", start the number search at
415 * 100 to make writing tests easier. */
416 if (!strncmp(name, "br", 2)) {
420 /* If the port name contains a number, try to assign that port number.
421 * This can make writing unit tests easier because port numbers are
423 for (p = name; *p != '\0'; p++) {
424 if (isdigit((unsigned char) *p)) {
425 port_no = start_no + strtol(p, NULL, 10);
426 if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE)
427 && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
428 return u32_to_odp(port_no);
435 for (port_no = 1; port_no <= UINT16_MAX; port_no++) {
436 if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
437 return u32_to_odp(port_no);
445 create_dp_netdev(const char *name, const struct dpif_class *class,
446 struct dp_netdev **dpp)
447 OVS_REQUIRES(dp_netdev_mutex)
449 struct dp_netdev *dp;
453 dp = xzalloc(sizeof *dp);
454 shash_add(&dp_netdevs, name, dp);
456 *CONST_CAST(const struct dpif_class **, &dp->class) = class;
457 *CONST_CAST(const char **, &dp->name) = xstrdup(name);
458 ovs_refcount_init(&dp->ref_cnt);
459 atomic_flag_init(&dp->destroyed);
461 ovs_mutex_init(&dp->flow_mutex);
462 classifier_init(&dp->cls, NULL);
463 hmap_init(&dp->flow_table);
465 ovs_mutex_init(&dp->queue_mutex);
466 ovs_mutex_lock(&dp->queue_mutex);
467 for (i = 0; i < N_QUEUES; i++) {
468 dp->queues[i].head = dp->queues[i].tail = 0;
470 ovs_mutex_unlock(&dp->queue_mutex);
471 dp->queue_seq = seq_create();
473 dp->n_hit = ovsthread_counter_create();
474 dp->n_missed = ovsthread_counter_create();
475 dp->n_lost = ovsthread_counter_create();
477 ovs_rwlock_init(&dp->port_rwlock);
478 hmap_init(&dp->ports);
479 dp->port_seq = seq_create();
480 latch_init(&dp->exit_latch);
482 ovs_rwlock_wrlock(&dp->port_rwlock);
483 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
484 ovs_rwlock_unlock(&dp->port_rwlock);
489 dp_netdev_set_threads(dp, 2);
496 dpif_netdev_open(const struct dpif_class *class, const char *name,
497 bool create, struct dpif **dpifp)
499 struct dp_netdev *dp;
502 ovs_mutex_lock(&dp_netdev_mutex);
503 dp = shash_find_data(&dp_netdevs, name);
505 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
507 error = (dp->class != class ? EINVAL
512 *dpifp = create_dpif_netdev(dp);
514 ovs_mutex_unlock(&dp_netdev_mutex);
520 dp_netdev_purge_queues(struct dp_netdev *dp)
524 ovs_mutex_lock(&dp->queue_mutex);
525 for (i = 0; i < N_QUEUES; i++) {
526 struct dp_netdev_queue *q = &dp->queues[i];
528 while (q->tail != q->head) {
529 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
530 ofpbuf_uninit(&u->upcall.packet);
531 ofpbuf_uninit(&u->buf);
534 ovs_mutex_unlock(&dp->queue_mutex);
537 /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp'
538 * through the 'dp_netdevs' shash while freeing 'dp'. */
540 dp_netdev_free(struct dp_netdev *dp)
541 OVS_REQUIRES(dp_netdev_mutex)
543 struct dp_netdev_port *port, *next;
545 shash_find_and_delete(&dp_netdevs, dp->name);
547 dp_netdev_set_threads(dp, 0);
548 free(dp->forwarders);
550 dp_netdev_flow_flush(dp);
551 ovs_rwlock_wrlock(&dp->port_rwlock);
552 HMAP_FOR_EACH_SAFE (port, next, node, &dp->ports) {
553 do_del_port(dp, port->port_no);
555 ovs_rwlock_unlock(&dp->port_rwlock);
556 ovsthread_counter_destroy(dp->n_hit);
557 ovsthread_counter_destroy(dp->n_missed);
558 ovsthread_counter_destroy(dp->n_lost);
560 dp_netdev_purge_queues(dp);
561 seq_destroy(dp->queue_seq);
562 ovs_mutex_destroy(&dp->queue_mutex);
564 classifier_destroy(&dp->cls);
565 hmap_destroy(&dp->flow_table);
566 ovs_mutex_destroy(&dp->flow_mutex);
567 seq_destroy(dp->port_seq);
568 hmap_destroy(&dp->ports);
569 atomic_flag_destroy(&dp->destroyed);
570 ovs_refcount_destroy(&dp->ref_cnt);
571 latch_destroy(&dp->exit_latch);
572 free(CONST_CAST(char *, dp->name));
577 dp_netdev_unref(struct dp_netdev *dp)
580 /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't
581 * get a new reference to 'dp' through the 'dp_netdevs' shash. */
582 ovs_mutex_lock(&dp_netdev_mutex);
583 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
586 ovs_mutex_unlock(&dp_netdev_mutex);
591 dpif_netdev_close(struct dpif *dpif)
593 struct dp_netdev *dp = get_dp_netdev(dpif);
600 dpif_netdev_destroy(struct dpif *dpif)
602 struct dp_netdev *dp = get_dp_netdev(dpif);
604 if (!atomic_flag_test_and_set(&dp->destroyed)) {
605 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
606 /* Can't happen: 'dpif' still owns a reference to 'dp'. */
615 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
617 struct dp_netdev *dp = get_dp_netdev(dpif);
619 fat_rwlock_rdlock(&dp->cls.rwlock);
620 stats->n_flows = hmap_count(&dp->flow_table);
621 fat_rwlock_unlock(&dp->cls.rwlock);
623 stats->n_hit = ovsthread_counter_read(dp->n_hit);
624 stats->n_missed = ovsthread_counter_read(dp->n_missed);
625 stats->n_lost = ovsthread_counter_read(dp->n_lost);
626 stats->n_masks = UINT32_MAX;
627 stats->n_mask_hit = UINT64_MAX;
633 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
635 OVS_REQ_WRLOCK(dp->port_rwlock)
637 struct netdev_saved_flags *sf;
638 struct dp_netdev_port *port;
639 struct netdev *netdev;
640 struct netdev_rx *rx;
641 enum netdev_flags flags;
642 const char *open_type;
645 /* XXX reject devices already in some dp_netdev. */
647 /* Open and validate network device. */
648 open_type = dpif_netdev_port_open_type(dp->class, type);
649 error = netdev_open(devname, open_type, &netdev);
653 /* XXX reject non-Ethernet devices */
655 netdev_get_flags(netdev, &flags);
656 if (flags & NETDEV_LOOPBACK) {
657 VLOG_ERR("%s: cannot add a loopback device", devname);
658 netdev_close(netdev);
662 error = netdev_rx_open(netdev, &rx);
664 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
665 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
666 devname, ovs_strerror(errno));
667 netdev_close(netdev);
671 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
674 netdev_close(netdev);
678 port = xmalloc(sizeof *port);
679 port->port_no = port_no;
680 port->netdev = netdev;
683 port->type = xstrdup(type);
685 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
686 seq_change(dp->port_seq);
692 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
693 odp_port_t *port_nop)
695 struct dp_netdev *dp = get_dp_netdev(dpif);
696 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
697 const char *dpif_port;
701 ovs_rwlock_wrlock(&dp->port_rwlock);
702 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
703 if (*port_nop != ODPP_NONE) {
705 error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0;
707 port_no = choose_port(dp, dpif_port);
708 error = port_no == ODPP_NONE ? EFBIG : 0;
712 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
714 ovs_rwlock_unlock(&dp->port_rwlock);
720 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
722 struct dp_netdev *dp = get_dp_netdev(dpif);
725 ovs_rwlock_wrlock(&dp->port_rwlock);
726 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
727 ovs_rwlock_unlock(&dp->port_rwlock);
733 is_valid_port_number(odp_port_t port_no)
735 return port_no != ODPP_NONE;
738 static struct dp_netdev_port *
739 dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no)
740 OVS_REQ_RDLOCK(dp->port_rwlock)
742 struct dp_netdev_port *port;
744 HMAP_FOR_EACH_IN_BUCKET (port, node, hash_int(odp_to_u32(port_no), 0),
746 if (port->port_no == port_no) {
754 get_port_by_number(struct dp_netdev *dp,
755 odp_port_t port_no, struct dp_netdev_port **portp)
756 OVS_REQ_RDLOCK(dp->port_rwlock)
758 if (!is_valid_port_number(port_no)) {
762 *portp = dp_netdev_lookup_port(dp, port_no);
763 return *portp ? 0 : ENOENT;
768 get_port_by_name(struct dp_netdev *dp,
769 const char *devname, struct dp_netdev_port **portp)
770 OVS_REQ_RDLOCK(dp->port_rwlock)
772 struct dp_netdev_port *port;
774 HMAP_FOR_EACH (port, node, &dp->ports) {
775 if (!strcmp(netdev_get_name(port->netdev), devname)) {
784 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
785 OVS_REQ_WRLOCK(dp->port_rwlock)
787 struct dp_netdev_port *port;
790 error = get_port_by_number(dp, port_no, &port);
795 hmap_remove(&dp->ports, &port->node);
796 seq_change(dp->port_seq);
798 netdev_close(port->netdev);
799 netdev_restore_flags(port->sf);
800 netdev_rx_close(port->rx);
808 answer_port_query(const struct dp_netdev_port *port,
809 struct dpif_port *dpif_port)
811 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
812 dpif_port->type = xstrdup(port->type);
813 dpif_port->port_no = port->port_no;
817 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
818 struct dpif_port *dpif_port)
820 struct dp_netdev *dp = get_dp_netdev(dpif);
821 struct dp_netdev_port *port;
824 ovs_rwlock_rdlock(&dp->port_rwlock);
825 error = get_port_by_number(dp, port_no, &port);
826 if (!error && dpif_port) {
827 answer_port_query(port, dpif_port);
829 ovs_rwlock_unlock(&dp->port_rwlock);
835 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
836 struct dpif_port *dpif_port)
838 struct dp_netdev *dp = get_dp_netdev(dpif);
839 struct dp_netdev_port *port;
842 ovs_rwlock_rdlock(&dp->port_rwlock);
843 error = get_port_by_name(dp, devname, &port);
844 if (!error && dpif_port) {
845 answer_port_query(port, dpif_port);
847 ovs_rwlock_unlock(&dp->port_rwlock);
853 dp_netdev_remove_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
854 OVS_REQ_WRLOCK(dp->cls.rwlock)
855 OVS_REQUIRES(dp->flow_mutex)
857 struct cls_rule *cr = CONST_CAST(struct cls_rule *, &flow->cr);
858 struct hmap_node *node = CONST_CAST(struct hmap_node *, &flow->node);
860 classifier_remove(&dp->cls, cr);
861 hmap_remove(&dp->flow_table, node);
862 dp_netdev_flow_unref(flow);
865 static struct dp_netdev_flow *
866 dp_netdev_flow_ref(const struct dp_netdev_flow *flow_)
868 struct dp_netdev_flow *flow = CONST_CAST(struct dp_netdev_flow *, flow_);
870 ovs_refcount_ref(&flow->ref_cnt);
876 dp_netdev_flow_unref(struct dp_netdev_flow *flow)
878 if (flow && ovs_refcount_unref(&flow->ref_cnt) == 1) {
879 cls_rule_destroy(CONST_CAST(struct cls_rule *, &flow->cr));
880 ovs_mutex_lock(&flow->mutex);
881 dp_netdev_actions_unref(flow->actions);
882 ovs_refcount_destroy(&flow->ref_cnt);
883 ovs_mutex_unlock(&flow->mutex);
884 ovs_mutex_destroy(&flow->mutex);
890 dp_netdev_flow_flush(struct dp_netdev *dp)
892 struct dp_netdev_flow *netdev_flow, *next;
894 ovs_mutex_lock(&dp->flow_mutex);
895 fat_rwlock_wrlock(&dp->cls.rwlock);
896 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
897 dp_netdev_remove_flow(dp, netdev_flow);
899 fat_rwlock_unlock(&dp->cls.rwlock);
900 ovs_mutex_unlock(&dp->flow_mutex);
904 dpif_netdev_flow_flush(struct dpif *dpif)
906 struct dp_netdev *dp = get_dp_netdev(dpif);
908 dp_netdev_flow_flush(dp);
912 struct dp_netdev_port_state {
919 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
921 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
926 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
927 struct dpif_port *dpif_port)
929 struct dp_netdev_port_state *state = state_;
930 struct dp_netdev *dp = get_dp_netdev(dpif);
931 struct hmap_node *node;
934 ovs_rwlock_rdlock(&dp->port_rwlock);
935 node = hmap_at_position(&dp->ports, &state->bucket, &state->offset);
937 struct dp_netdev_port *port;
939 port = CONTAINER_OF(node, struct dp_netdev_port, node);
942 state->name = xstrdup(netdev_get_name(port->netdev));
943 dpif_port->name = state->name;
944 dpif_port->type = port->type;
945 dpif_port->port_no = port->port_no;
951 ovs_rwlock_unlock(&dp->port_rwlock);
957 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
959 struct dp_netdev_port_state *state = state_;
966 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
968 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
969 uint64_t new_port_seq;
972 new_port_seq = seq_read(dpif->dp->port_seq);
973 if (dpif->last_port_seq != new_port_seq) {
974 dpif->last_port_seq = new_port_seq;
984 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
986 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
988 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
991 static struct dp_netdev_flow *
992 dp_netdev_flow_cast(const struct cls_rule *cr)
994 return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL;
997 static struct dp_netdev_flow *
998 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
999 OVS_EXCLUDED(dp->cls.rwlock)
1001 struct dp_netdev_flow *netdev_flow;
1003 fat_rwlock_rdlock(&dp->cls.rwlock);
1004 netdev_flow = dp_netdev_flow_cast(classifier_lookup(&dp->cls, flow, NULL));
1005 dp_netdev_flow_ref(netdev_flow);
1006 fat_rwlock_unlock(&dp->cls.rwlock);
1011 static struct dp_netdev_flow *
1012 dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
1013 OVS_REQ_RDLOCK(dp->cls.rwlock)
1015 struct dp_netdev_flow *netdev_flow;
1017 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
1019 if (flow_equal(&netdev_flow->flow, flow)) {
1020 return dp_netdev_flow_ref(netdev_flow);
1028 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
1029 struct dpif_flow_stats *stats)
1030 OVS_REQ_RDLOCK(netdev_flow->mutex)
1032 stats->n_packets = netdev_flow->packet_count;
1033 stats->n_bytes = netdev_flow->byte_count;
1034 stats->used = netdev_flow->used;
1035 stats->tcp_flags = netdev_flow->tcp_flags;
1039 dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1040 const struct nlattr *mask_key,
1041 uint32_t mask_key_len, const struct flow *flow,
1045 enum odp_key_fitness fitness;
1047 fitness = odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow);
1049 /* This should not happen: it indicates that
1050 * odp_flow_key_from_mask() and odp_flow_key_to_mask()
1051 * disagree on the acceptable form of a mask. Log the problem
1052 * as an error, with enough details to enable debugging. */
1053 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1055 if (!VLOG_DROP_ERR(&rl)) {
1059 odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
1061 VLOG_ERR("internal error parsing flow mask %s (%s)",
1062 ds_cstr(&s), odp_key_fitness_to_string(fitness));
1068 /* Force unwildcard the in_port. */
1069 mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
1071 enum mf_field_id id;
1072 /* No mask key, unwildcard everything except fields whose
1073 * prerequisities are not met. */
1074 memset(mask, 0x0, sizeof *mask);
1076 for (id = 0; id < MFF_N_IDS; ++id) {
1077 /* Skip registers and metadata. */
1078 if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
1079 && id != MFF_METADATA) {
1080 const struct mf_field *mf = mf_from_id(id);
1081 if (mf_are_prereqs_ok(mf, flow)) {
1082 mf_mask_field(mf, mask);
1092 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1097 if (odp_flow_key_to_flow(key, key_len, flow)) {
1098 /* This should not happen: it indicates that odp_flow_key_from_flow()
1099 * and odp_flow_key_to_flow() disagree on the acceptable form of a
1100 * flow. Log the problem as an error, with enough details to enable
1102 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1104 if (!VLOG_DROP_ERR(&rl)) {
1108 odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
1109 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
1116 in_port = flow->in_port.odp_port;
1117 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
1125 dpif_netdev_flow_get(const struct dpif *dpif,
1126 const struct nlattr *nl_key, size_t nl_key_len,
1127 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
1129 struct dp_netdev *dp = get_dp_netdev(dpif);
1130 struct dp_netdev_flow *netdev_flow;
1134 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
1139 fat_rwlock_rdlock(&dp->cls.rwlock);
1140 netdev_flow = dp_netdev_find_flow(dp, &key);
1141 fat_rwlock_unlock(&dp->cls.rwlock);
1144 struct dp_netdev_actions *actions = NULL;
1146 ovs_mutex_lock(&netdev_flow->mutex);
1148 get_dpif_flow_stats(netdev_flow, stats);
1151 actions = dp_netdev_actions_ref(netdev_flow->actions);
1153 ovs_mutex_unlock(&netdev_flow->mutex);
1155 dp_netdev_flow_unref(netdev_flow);
1158 *actionsp = ofpbuf_clone_data(actions->actions, actions->size);
1159 dp_netdev_actions_unref(actions);
1169 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
1170 const struct flow_wildcards *wc,
1171 const struct nlattr *actions,
1173 OVS_REQUIRES(dp->flow_mutex)
1175 struct dp_netdev_flow *netdev_flow;
1178 netdev_flow = xzalloc(sizeof *netdev_flow);
1179 *CONST_CAST(struct flow *, &netdev_flow->flow) = *flow;
1180 ovs_refcount_init(&netdev_flow->ref_cnt);
1182 ovs_mutex_init(&netdev_flow->mutex);
1183 ovs_mutex_lock(&netdev_flow->mutex);
1185 netdev_flow->actions = dp_netdev_actions_create(actions, actions_len);
1187 match_init(&match, flow, wc);
1188 cls_rule_init(CONST_CAST(struct cls_rule *, &netdev_flow->cr),
1189 &match, NETDEV_RULE_PRIORITY);
1190 fat_rwlock_wrlock(&dp->cls.rwlock);
1191 classifier_insert(&dp->cls,
1192 CONST_CAST(struct cls_rule *, &netdev_flow->cr));
1193 hmap_insert(&dp->flow_table,
1194 CONST_CAST(struct hmap_node *, &netdev_flow->node),
1195 flow_hash(flow, 0));
1196 fat_rwlock_unlock(&dp->cls.rwlock);
1198 ovs_mutex_unlock(&netdev_flow->mutex);
1204 clear_stats(struct dp_netdev_flow *netdev_flow)
1205 OVS_REQUIRES(netdev_flow->mutex)
1207 netdev_flow->used = 0;
1208 netdev_flow->packet_count = 0;
1209 netdev_flow->byte_count = 0;
1210 netdev_flow->tcp_flags = 0;
1214 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
1216 struct dp_netdev *dp = get_dp_netdev(dpif);
1217 struct dp_netdev_flow *netdev_flow;
1219 struct flow_wildcards wc;
1222 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
1226 error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
1227 put->mask, put->mask_len,
1233 ovs_mutex_lock(&dp->flow_mutex);
1234 netdev_flow = dp_netdev_lookup_flow(dp, &flow);
1236 if (put->flags & DPIF_FP_CREATE) {
1237 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
1239 memset(put->stats, 0, sizeof *put->stats);
1241 error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
1250 if (put->flags & DPIF_FP_MODIFY
1251 && flow_equal(&flow, &netdev_flow->flow)) {
1252 struct dp_netdev_actions *new_actions;
1253 struct dp_netdev_actions *old_actions;
1255 new_actions = dp_netdev_actions_create(put->actions,
1258 ovs_mutex_lock(&netdev_flow->mutex);
1259 old_actions = netdev_flow->actions;
1260 netdev_flow->actions = new_actions;
1262 get_dpif_flow_stats(netdev_flow, put->stats);
1264 if (put->flags & DPIF_FP_ZERO_STATS) {
1265 clear_stats(netdev_flow);
1267 ovs_mutex_unlock(&netdev_flow->mutex);
1269 dp_netdev_actions_unref(old_actions);
1270 } else if (put->flags & DPIF_FP_CREATE) {
1273 /* Overlapping flow. */
1276 dp_netdev_flow_unref(netdev_flow);
1278 ovs_mutex_unlock(&dp->flow_mutex);
1284 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
1286 struct dp_netdev *dp = get_dp_netdev(dpif);
1287 struct dp_netdev_flow *netdev_flow;
1291 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
1296 ovs_mutex_lock(&dp->flow_mutex);
1297 fat_rwlock_wrlock(&dp->cls.rwlock);
1298 netdev_flow = dp_netdev_find_flow(dp, &key);
1301 ovs_mutex_lock(&netdev_flow->mutex);
1302 get_dpif_flow_stats(netdev_flow, del->stats);
1303 ovs_mutex_unlock(&netdev_flow->mutex);
1305 dp_netdev_remove_flow(dp, netdev_flow);
1306 dp_netdev_flow_unref(netdev_flow);
1310 fat_rwlock_unlock(&dp->cls.rwlock);
1311 ovs_mutex_unlock(&dp->flow_mutex);
1316 struct dp_netdev_flow_state {
1317 struct dp_netdev_actions *actions;
1318 struct odputil_keybuf keybuf;
1319 struct odputil_keybuf maskbuf;
1320 struct dpif_flow_stats stats;
1323 struct dp_netdev_flow_iter {
1327 struct ovs_mutex mutex;
1331 dpif_netdev_flow_dump_state_init(void **statep)
1333 struct dp_netdev_flow_state *state;
1335 *statep = state = xmalloc(sizeof *state);
1336 state->actions = NULL;
1340 dpif_netdev_flow_dump_state_uninit(void *state_)
1342 struct dp_netdev_flow_state *state = state_;
1344 dp_netdev_actions_unref(state->actions);
1349 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **iterp)
1351 struct dp_netdev_flow_iter *iter;
1353 *iterp = iter = xmalloc(sizeof *iter);
1357 ovs_mutex_init(&iter->mutex);
1362 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *iter_, void *state_,
1363 const struct nlattr **key, size_t *key_len,
1364 const struct nlattr **mask, size_t *mask_len,
1365 const struct nlattr **actions, size_t *actions_len,
1366 const struct dpif_flow_stats **stats)
1368 struct dp_netdev_flow_iter *iter = iter_;
1369 struct dp_netdev_flow_state *state = state_;
1370 struct dp_netdev *dp = get_dp_netdev(dpif);
1371 struct dp_netdev_flow *netdev_flow;
1374 ovs_mutex_lock(&iter->mutex);
1375 error = iter->status;
1377 struct hmap_node *node;
1379 fat_rwlock_rdlock(&dp->cls.rwlock);
1380 node = hmap_at_position(&dp->flow_table, &iter->bucket, &iter->offset);
1382 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
1383 dp_netdev_flow_ref(netdev_flow);
1385 fat_rwlock_unlock(&dp->cls.rwlock);
1387 iter->status = error = EOF;
1390 ovs_mutex_unlock(&iter->mutex);
1398 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
1399 odp_flow_key_from_flow(&buf, &netdev_flow->flow,
1400 netdev_flow->flow.in_port.odp_port);
1403 *key_len = buf.size;
1408 struct flow_wildcards wc;
1410 ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
1411 minimask_expand(&netdev_flow->cr.match.mask, &wc);
1412 odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
1413 odp_to_u32(wc.masks.in_port.odp_port),
1417 *mask_len = buf.size;
1420 if (actions || stats) {
1421 dp_netdev_actions_unref(state->actions);
1422 state->actions = NULL;
1424 ovs_mutex_lock(&netdev_flow->mutex);
1426 state->actions = dp_netdev_actions_ref(netdev_flow->actions);
1427 *actions = state->actions->actions;
1428 *actions_len = state->actions->size;
1431 get_dpif_flow_stats(netdev_flow, &state->stats);
1432 *stats = &state->stats;
1434 ovs_mutex_unlock(&netdev_flow->mutex);
1437 dp_netdev_flow_unref(netdev_flow);
1443 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *iter_)
1445 struct dp_netdev_flow_iter *iter = iter_;
1447 ovs_mutex_destroy(&iter->mutex);
1453 dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute)
1455 struct dp_netdev *dp = get_dp_netdev(dpif);
1456 struct pkt_metadata *md = &execute->md;
1459 if (execute->packet->size < ETH_HEADER_LEN ||
1460 execute->packet->size > UINT16_MAX) {
1464 /* Extract flow key. */
1465 flow_extract(execute->packet, md, &key);
1467 ovs_rwlock_rdlock(&dp->port_rwlock);
1468 dp_netdev_execute_actions(dp, &key, execute->packet, md, execute->actions,
1469 execute->actions_len);
1470 ovs_rwlock_unlock(&dp->port_rwlock);
1476 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1482 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1483 uint32_t queue_id, uint32_t *priority)
1485 *priority = queue_id;
1489 static struct dp_netdev_queue *
1490 find_nonempty_queue(struct dp_netdev *dp)
1491 OVS_REQUIRES(dp->queue_mutex)
1495 for (i = 0; i < N_QUEUES; i++) {
1496 struct dp_netdev_queue *q = &dp->queues[i];
1497 if (q->head != q->tail) {
1505 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1508 struct dp_netdev *dp = get_dp_netdev(dpif);
1509 struct dp_netdev_queue *q;
1512 ovs_mutex_lock(&dp->queue_mutex);
1513 q = find_nonempty_queue(dp);
1515 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1517 *upcall = u->upcall;
1526 ovs_mutex_unlock(&dp->queue_mutex);
1532 dpif_netdev_recv_wait(struct dpif *dpif)
1534 struct dp_netdev *dp = get_dp_netdev(dpif);
1537 ovs_mutex_lock(&dp->queue_mutex);
1538 seq = seq_read(dp->queue_seq);
1539 if (find_nonempty_queue(dp)) {
1540 poll_immediate_wake();
1542 seq_wait(dp->queue_seq, seq);
1544 ovs_mutex_unlock(&dp->queue_mutex);
1548 dpif_netdev_recv_purge(struct dpif *dpif)
1550 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1552 dp_netdev_purge_queues(dpif_netdev->dp);
1555 /* Creates and returns a new 'struct dp_netdev_actions', with a reference count
1556 * of 1, whose actions are a copy of from the 'ofpacts_len' bytes of
1558 struct dp_netdev_actions *
1559 dp_netdev_actions_create(const struct nlattr *actions, size_t size)
1561 struct dp_netdev_actions *netdev_actions;
1563 netdev_actions = xmalloc(sizeof *netdev_actions);
1564 ovs_refcount_init(&netdev_actions->ref_cnt);
1565 netdev_actions->actions = xmemdup(actions, size);
1566 netdev_actions->size = size;
1568 return netdev_actions;
1571 /* Increments 'actions''s refcount. */
1572 struct dp_netdev_actions *
1573 dp_netdev_actions_ref(const struct dp_netdev_actions *actions_)
1575 struct dp_netdev_actions *actions;
1577 actions = CONST_CAST(struct dp_netdev_actions *, actions_);
1579 ovs_refcount_ref(&actions->ref_cnt);
1584 /* Decrements 'actions''s refcount and frees 'actions' if the refcount reaches
1587 dp_netdev_actions_unref(struct dp_netdev_actions *actions)
1589 if (actions && ovs_refcount_unref(&actions->ref_cnt) == 1) {
1590 ovs_refcount_destroy(&actions->ref_cnt);
1591 free(actions->actions);
1597 dp_forwarder_main(void *f_)
1599 struct dp_forwarder *f = f_;
1600 struct dp_netdev *dp = f->dp;
1601 struct ofpbuf packet;
1603 f->name = xasprintf("forwarder_%u", ovsthread_id_self());
1604 set_subprogram_name("%s", f->name);
1606 ofpbuf_init(&packet, 0);
1607 while (!latch_is_set(&dp->exit_latch)) {
1608 bool received_anything;
1611 ovs_rwlock_rdlock(&dp->port_rwlock);
1612 for (i = 0; i < 50; i++) {
1613 struct dp_netdev_port *port;
1615 received_anything = false;
1616 HMAP_FOR_EACH (port, node, &f->dp->ports) {
1618 && port->node.hash >= f->min_hash
1619 && port->node.hash <= f->max_hash) {
1624 if (netdev_get_mtu(port->netdev, &mtu)) {
1625 mtu = ETH_PAYLOAD_MAX;
1627 buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + mtu;
1629 ofpbuf_clear(&packet);
1630 ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM,
1633 error = netdev_rx_recv(port->rx, &packet);
1635 struct pkt_metadata md
1636 = PKT_METADATA_INITIALIZER(port->port_no);
1638 dp_netdev_port_input(dp, &packet, &md);
1639 received_anything = true;
1640 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1641 static struct vlog_rate_limit rl
1642 = VLOG_RATE_LIMIT_INIT(1, 5);
1644 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1645 netdev_get_name(port->netdev),
1646 ovs_strerror(error));
1651 if (!received_anything) {
1656 if (received_anything) {
1657 poll_immediate_wake();
1659 struct dp_netdev_port *port;
1661 HMAP_FOR_EACH (port, node, &f->dp->ports)
1663 && port->node.hash >= f->min_hash
1664 && port->node.hash <= f->max_hash) {
1665 netdev_rx_wait(port->rx);
1667 seq_wait(dp->port_seq, seq_read(dp->port_seq));
1668 latch_wait(&dp->exit_latch);
1670 ovs_rwlock_unlock(&dp->port_rwlock);
1674 ofpbuf_uninit(&packet);
1682 dp_netdev_set_threads(struct dp_netdev *dp, int n)
1686 if (n == dp->n_forwarders) {
1690 /* Stop existing threads. */
1691 latch_set(&dp->exit_latch);
1692 for (i = 0; i < dp->n_forwarders; i++) {
1693 struct dp_forwarder *f = &dp->forwarders[i];
1695 xpthread_join(f->thread, NULL);
1697 latch_poll(&dp->exit_latch);
1698 free(dp->forwarders);
1700 /* Start new threads. */
1701 dp->forwarders = xmalloc(n * sizeof *dp->forwarders);
1702 dp->n_forwarders = n;
1703 for (i = 0; i < n; i++) {
1704 struct dp_forwarder *f = &dp->forwarders[i];
1707 f->min_hash = UINT32_MAX / n * i;
1708 f->max_hash = UINT32_MAX / n * (i + 1) - 1;
1710 f->max_hash = UINT32_MAX;
1712 xpthread_create(&f->thread, NULL, dp_forwarder_main, f);
1717 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1718 const struct ofpbuf *packet)
1719 OVS_REQUIRES(netdev_flow->mutex)
1721 netdev_flow->used = time_msec();
1722 netdev_flow->packet_count++;
1723 netdev_flow->byte_count += packet->size;
1724 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
1728 dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
1729 struct pkt_metadata *md)
1730 OVS_REQ_RDLOCK(dp->port_rwlock)
1732 struct dp_netdev_flow *netdev_flow;
1735 if (packet->size < ETH_HEADER_LEN) {
1738 flow_extract(packet, md, &key);
1739 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1741 struct dp_netdev_actions *actions;
1743 ovs_mutex_lock(&netdev_flow->mutex);
1744 dp_netdev_flow_used(netdev_flow, packet);
1745 actions = dp_netdev_actions_ref(netdev_flow->actions);
1746 ovs_mutex_unlock(&netdev_flow->mutex);
1748 dp_netdev_execute_actions(dp, &key, packet, md,
1749 actions->actions, actions->size);
1750 dp_netdev_actions_unref(actions);
1751 dp_netdev_flow_unref(netdev_flow);
1752 ovsthread_counter_inc(dp->n_hit, 1);
1754 ovsthread_counter_inc(dp->n_missed, 1);
1755 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1760 dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
1761 int queue_no, const struct flow *flow,
1762 const struct nlattr *userdata)
1763 OVS_EXCLUDED(dp->queue_mutex)
1765 struct dp_netdev_queue *q = &dp->queues[queue_no];
1768 ovs_mutex_lock(&dp->queue_mutex);
1769 if (q->head - q->tail < MAX_QUEUE_LEN) {
1770 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1771 struct dpif_upcall *upcall = &u->upcall;
1772 struct ofpbuf *buf = &u->buf;
1775 upcall->type = queue_no;
1777 /* Allocate buffer big enough for everything. */
1778 buf_size = ODPUTIL_FLOW_KEY_BYTES;
1780 buf_size += NLA_ALIGN(userdata->nla_len);
1782 ofpbuf_init(buf, buf_size);
1785 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1786 upcall->key = buf->data;
1787 upcall->key_len = buf->size;
1791 upcall->userdata = ofpbuf_put(buf, userdata,
1792 NLA_ALIGN(userdata->nla_len));
1795 /* Steal packet data. */
1796 ovs_assert(packet->source == OFPBUF_MALLOC);
1797 upcall->packet = *packet;
1798 ofpbuf_use(packet, NULL, 0);
1800 seq_change(dp->queue_seq);
1804 ovsthread_counter_inc(dp->n_lost, 1);
1807 ovs_mutex_unlock(&dp->queue_mutex);
1812 struct dp_netdev_execute_aux {
1813 struct dp_netdev *dp;
1814 const struct flow *key;
1818 dp_execute_cb(void *aux_, struct ofpbuf *packet,
1819 const struct pkt_metadata *md OVS_UNUSED,
1820 const struct nlattr *a, bool may_steal)
1821 OVS_NO_THREAD_SAFETY_ANALYSIS
1823 struct dp_netdev_execute_aux *aux = aux_;
1824 int type = nl_attr_type(a);
1825 struct dp_netdev_port *p;
1827 switch ((enum ovs_action_attr)type) {
1828 case OVS_ACTION_ATTR_OUTPUT:
1829 p = dp_netdev_lookup_port(aux->dp, u32_to_odp(nl_attr_get_u32(a)));
1831 netdev_send(p->netdev, packet);
1835 case OVS_ACTION_ATTR_USERSPACE: {
1836 const struct nlattr *userdata;
1838 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1840 /* Make a copy if we are not allowed to steal the packet's data. */
1842 packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
1844 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1847 ofpbuf_uninit(packet);
1851 case OVS_ACTION_ATTR_PUSH_VLAN:
1852 case OVS_ACTION_ATTR_POP_VLAN:
1853 case OVS_ACTION_ATTR_PUSH_MPLS:
1854 case OVS_ACTION_ATTR_POP_MPLS:
1855 case OVS_ACTION_ATTR_SET:
1856 case OVS_ACTION_ATTR_SAMPLE:
1857 case OVS_ACTION_ATTR_UNSPEC:
1858 case __OVS_ACTION_ATTR_MAX:
1864 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1865 struct ofpbuf *packet, struct pkt_metadata *md,
1866 const struct nlattr *actions, size_t actions_len)
1867 OVS_REQ_RDLOCK(dp->port_rwlock)
1869 struct dp_netdev_execute_aux aux = {dp, key};
1871 odp_execute_actions(&aux, packet, md, actions, actions_len, dp_execute_cb);
1874 #define DPIF_NETDEV_CLASS_FUNCTIONS \
1875 dpif_netdev_enumerate, \
1876 dpif_netdev_port_open_type, \
1878 dpif_netdev_close, \
1879 dpif_netdev_destroy, \
1882 dpif_netdev_get_stats, \
1883 dpif_netdev_port_add, \
1884 dpif_netdev_port_del, \
1885 dpif_netdev_port_query_by_number, \
1886 dpif_netdev_port_query_by_name, \
1887 NULL, /* port_get_pid */ \
1888 dpif_netdev_port_dump_start, \
1889 dpif_netdev_port_dump_next, \
1890 dpif_netdev_port_dump_done, \
1891 dpif_netdev_port_poll, \
1892 dpif_netdev_port_poll_wait, \
1893 dpif_netdev_flow_get, \
1894 dpif_netdev_flow_put, \
1895 dpif_netdev_flow_del, \
1896 dpif_netdev_flow_flush, \
1897 dpif_netdev_flow_dump_state_init, \
1898 dpif_netdev_flow_dump_start, \
1899 dpif_netdev_flow_dump_next, \
1901 dpif_netdev_flow_dump_done, \
1902 dpif_netdev_flow_dump_state_uninit, \
1903 dpif_netdev_execute, \
1904 NULL, /* operate */ \
1905 dpif_netdev_recv_set, \
1906 dpif_netdev_queue_to_priority, \
1908 dpif_netdev_recv_wait, \
1909 dpif_netdev_recv_purge, \
1911 const struct dpif_class dpif_netdev_class = {
1913 DPIF_NETDEV_CLASS_FUNCTIONS
1916 const struct dpif_class dpif_planetlab_class = {
1918 DPIF_NETDEV_CLASS_FUNCTIONS
1922 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1923 const char *argv[], void *aux OVS_UNUSED)
1925 struct dp_netdev_port *port;
1926 struct dp_netdev *dp;
1929 ovs_mutex_lock(&dp_netdev_mutex);
1930 dp = shash_find_data(&dp_netdevs, argv[1]);
1931 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1932 ovs_mutex_unlock(&dp_netdev_mutex);
1933 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1936 ovs_refcount_ref(&dp->ref_cnt);
1937 ovs_mutex_unlock(&dp_netdev_mutex);
1939 ovs_rwlock_wrlock(&dp->port_rwlock);
1940 if (get_port_by_name(dp, argv[2], &port)) {
1941 unixctl_command_reply_error(conn, "unknown port");
1945 port_no = u32_to_odp(atoi(argv[3]));
1946 if (!port_no || port_no == ODPP_NONE) {
1947 unixctl_command_reply_error(conn, "bad port number");
1950 if (dp_netdev_lookup_port(dp, port_no)) {
1951 unixctl_command_reply_error(conn, "port number already in use");
1954 hmap_remove(&dp->ports, &port->node);
1955 port->port_no = port_no;
1956 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
1957 seq_change(dp->port_seq);
1958 unixctl_command_reply(conn, NULL);
1961 ovs_rwlock_unlock(&dp->port_rwlock);
1962 dp_netdev_unref(dp);
1966 dpif_dummy_register__(const char *type)
1968 struct dpif_class *class;
1970 class = xmalloc(sizeof *class);
1971 *class = dpif_netdev_class;
1972 class->type = xstrdup(type);
1973 dp_register_provider(class);
1977 dpif_dummy_register(bool override)
1984 dp_enumerate_types(&types);
1985 SSET_FOR_EACH (type, &types) {
1986 if (!dp_unregister_provider(type)) {
1987 dpif_dummy_register__(type);
1990 sset_destroy(&types);
1993 dpif_dummy_register__("dummy");
1995 unixctl_command_register("dpif-dummy/change-port-number",
1996 "DP PORT NEW-NUMBER",
1997 3, 3, dpif_dummy_change_port_number, NULL);