2 * Copyright (c) 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
24 #include <netinet/in.h>
25 #include <sys/socket.h>
30 #include <sys/ioctl.h>
34 #include "classifier.h"
37 #include "dpif-provider.h"
39 #include "dynamic-string.h"
44 #include "meta-flow.h"
46 #include "netdev-vport.h"
48 #include "odp-execute.h"
50 #include "ofp-print.h"
53 #include "poll-loop.h"
63 VLOG_DEFINE_THIS_MODULE(dpif_netdev);
65 /* By default, choose a priority in the middle. */
66 #define NETDEV_RULE_PRIORITY 0x8000
68 /* Configuration parameters. */
69 enum { MAX_FLOWS = 65536 }; /* Maximum number of flows in flow table. */
71 /* Enough headroom to add a vlan tag, plus an extra 2 bytes to allow IP
72 * headers to be aligned on a 4-byte boundary. */
73 enum { DP_NETDEV_HEADROOM = 2 + VLAN_HEADER_LEN };
76 enum { N_QUEUES = 2 }; /* Number of queues for dpif_recv(). */
77 enum { MAX_QUEUE_LEN = 128 }; /* Maximum number of packets per queue. */
78 enum { QUEUE_MASK = MAX_QUEUE_LEN - 1 };
79 BUILD_ASSERT_DECL(IS_POW2(MAX_QUEUE_LEN));
81 /* Protects against changes to 'dp_netdevs'. */
82 static struct ovs_mutex dp_netdev_mutex = OVS_MUTEX_INITIALIZER;
84 /* Contains all 'struct dp_netdev's. */
85 static struct shash dp_netdevs OVS_GUARDED_BY(dp_netdev_mutex)
86 = SHASH_INITIALIZER(&dp_netdevs);
88 struct dp_netdev_upcall {
89 struct dpif_upcall upcall; /* Queued upcall information. */
90 struct ofpbuf buf; /* ofpbuf instance for upcall.packet. */
93 /* A queue passing packets from a struct dp_netdev to its clients.
99 * Any access at all requires the owning 'dp_netdev''s queue_mutex. */
100 struct dp_netdev_queue {
101 struct dp_netdev_upcall upcalls[MAX_QUEUE_LEN] OVS_GUARDED;
102 unsigned int head OVS_GUARDED;
103 unsigned int tail OVS_GUARDED;
106 /* Datapath based on the network device interface from netdev.h.
112 * Some members, marked 'const', are immutable. Accessing other members
113 * requires synchronization, as noted in more detail below.
115 * Acquisition order is, from outermost to innermost:
117 * dp_netdev_mutex (global)
124 const struct dpif_class *const class;
125 const char *const name;
126 struct ovs_refcount ref_cnt;
127 atomic_flag destroyed;
131 * Readers of 'cls' and 'flow_table' must take a 'cls->rwlock' read lock.
133 * Writers of 'cls' and 'flow_table' must take the 'flow_mutex' and then
134 * the 'cls->rwlock' write lock. (The outer 'flow_mutex' allows writers to
135 * atomically perform multiple operations on 'cls' and 'flow_table'.)
137 struct ovs_mutex flow_mutex;
138 struct classifier cls; /* Classifier. Protected by cls.rwlock. */
139 struct hmap flow_table OVS_GUARDED; /* Flow table. */
143 * Everything in 'queues' is protected by 'queue_mutex'. */
144 struct ovs_mutex queue_mutex;
145 struct dp_netdev_queue queues[N_QUEUES];
146 struct seq *queue_seq; /* Incremented whenever a packet is queued. */
150 * ovsthread_counter is internally synchronized. */
151 struct ovsthread_counter *n_hit; /* Number of flow table matches. */
152 struct ovsthread_counter *n_missed; /* Number of flow table misses. */
153 struct ovsthread_counter *n_lost; /* Number of misses not passed up. */
157 * Any lookup into 'ports' or any access to the dp_netdev_ports found
158 * through 'ports' requires taking 'port_rwlock'. */
159 struct ovs_rwlock port_rwlock;
160 struct hmap ports OVS_GUARDED;
161 struct seq *port_seq; /* Incremented whenever a port changes. */
163 /* Forwarding threads. */
164 struct latch exit_latch;
165 struct dp_forwarder *forwarders;
169 static struct dp_netdev_port *dp_netdev_lookup_port(const struct dp_netdev *dp,
171 OVS_REQ_RDLOCK(dp->port_rwlock);
173 /* A port in a netdev-based datapath. */
174 struct dp_netdev_port {
175 struct hmap_node node; /* Node in dp_netdev's 'ports'. */
177 struct netdev *netdev;
178 struct netdev_saved_flags *sf;
179 struct netdev_rx *rx;
180 char *type; /* Port type as requested by user. */
183 /* A flow in dp_netdev's 'flow_table'.
189 * Except near the beginning or ending of its lifespan, rule 'rule' belongs to
190 * its dp_netdev's classifier. The text below calls this classifier 'cls'.
195 * The thread safety rules described here for "struct dp_netdev_flow" are
196 * motivated by two goals:
198 * - Prevent threads that read members of "struct dp_netdev_flow" from
199 * reading bad data due to changes by some thread concurrently modifying
202 * - Prevent two threads making changes to members of a given "struct
203 * dp_netdev_flow" from interfering with each other.
209 * A flow 'flow' may be accessed without a risk of being freed by code that
210 * holds a read-lock or write-lock on 'cls->rwlock' or that owns a reference to
211 * 'flow->ref_cnt' (or both). Code that needs to hold onto a flow for a while
212 * should take 'cls->rwlock', find the flow it needs, increment 'flow->ref_cnt'
213 * with dpif_netdev_flow_ref(), and drop 'cls->rwlock'.
215 * 'flow->ref_cnt' protects 'flow' from being freed. It doesn't protect the
216 * flow from being deleted from 'cls' (that's 'cls->rwlock') and it doesn't
217 * protect members of 'flow' from modification (that's 'flow->mutex').
219 * 'flow->mutex' protects the members of 'flow' from modification. It doesn't
220 * protect the flow from being deleted from 'cls' (that's 'cls->rwlock') and it
221 * doesn't prevent the flow from being freed (that's 'flow->ref_cnt').
223 * Some members, marked 'const', are immutable. Accessing other members
224 * requires synchronization, as noted in more detail below.
226 struct dp_netdev_flow {
227 /* Packet classification. */
228 const struct cls_rule cr; /* In owning dp_netdev's 'cls'. */
230 /* Hash table index by unmasked flow. */
231 const struct hmap_node node; /* In owning dp_netdev's 'flow_table'. */
232 const struct flow flow; /* The flow that created this entry. */
234 /* Number of references.
235 * The classifier owns one reference.
236 * Any thread trying to keep a rule from being freed should hold its own
238 struct ovs_refcount ref_cnt;
240 /* Protects members marked OVS_GUARDED.
242 * Acquire after datapath's flow_mutex. */
243 struct ovs_mutex mutex OVS_ACQ_AFTER(dp_netdev_mutex);
247 * Reading or writing these members requires 'mutex'. */
248 long long int used OVS_GUARDED; /* Last used time, in monotonic msecs. */
249 long long int packet_count OVS_GUARDED; /* Number of packets matched. */
250 long long int byte_count OVS_GUARDED; /* Number of bytes matched. */
251 uint16_t tcp_flags OVS_GUARDED; /* Bitwise-OR of seen tcp_flags values. */
255 * Reading 'actions' requires 'mutex'.
256 * Writing 'actions' requires 'mutex' and (to allow for transactions) the
257 * datapath's flow_mutex. */
258 struct dp_netdev_actions *actions OVS_GUARDED;
261 static struct dp_netdev_flow *dp_netdev_flow_ref(
262 const struct dp_netdev_flow *);
263 static void dp_netdev_flow_unref(struct dp_netdev_flow *);
265 /* A set of datapath actions within a "struct dp_netdev_flow".
271 * A struct dp_netdev_actions 'actions' may be accessed without a risk of being
272 * freed by code that holds a read-lock or write-lock on 'flow->mutex' (where
273 * 'flow' is the dp_netdev_flow for which 'flow->actions == actions') or that
274 * owns a reference to 'actions->ref_cnt' (or both). */
275 struct dp_netdev_actions {
276 struct ovs_refcount ref_cnt;
278 /* These members are immutable: they do not change during the struct's
280 struct nlattr *actions; /* Sequence of OVS_ACTION_ATTR_* attributes. */
281 unsigned int size; /* Size of 'actions', in bytes. */
284 struct dp_netdev_actions *dp_netdev_actions_create(const struct nlattr *,
286 struct dp_netdev_actions *dp_netdev_actions_ref(
287 const struct dp_netdev_actions *);
288 void dp_netdev_actions_unref(struct dp_netdev_actions *);
290 /* A thread that receives packets from some ports, looks them up in the flow
291 * table, and executes the actions it finds. */
292 struct dp_forwarder {
293 struct dp_netdev *dp;
296 uint32_t min_hash, max_hash;
299 /* Interface to netdev-based datapath. */
302 struct dp_netdev *dp;
303 uint64_t last_port_seq;
306 static int get_port_by_number(struct dp_netdev *dp, odp_port_t port_no,
307 struct dp_netdev_port **portp)
308 OVS_REQ_RDLOCK(dp->port_rwlock);
309 static int get_port_by_name(struct dp_netdev *dp, const char *devname,
310 struct dp_netdev_port **portp)
311 OVS_REQ_RDLOCK(dp->port_rwlock);
312 static void dp_netdev_free(struct dp_netdev *)
313 OVS_REQUIRES(dp_netdev_mutex);
314 static void dp_netdev_flow_flush(struct dp_netdev *);
315 static int do_add_port(struct dp_netdev *dp, const char *devname,
316 const char *type, odp_port_t port_no)
317 OVS_REQ_WRLOCK(dp->port_rwlock);
318 static int do_del_port(struct dp_netdev *dp, odp_port_t port_no)
319 OVS_REQ_WRLOCK(dp->port_rwlock);
320 static int dpif_netdev_open(const struct dpif_class *, const char *name,
321 bool create, struct dpif **);
322 static int dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *,
323 int queue_no, const struct flow *,
324 const struct nlattr *userdata)
325 OVS_EXCLUDED(dp->queue_mutex);
326 static void dp_netdev_execute_actions(struct dp_netdev *dp,
327 const struct flow *, struct ofpbuf *,
328 struct pkt_metadata *,
329 const struct nlattr *actions,
331 OVS_REQ_RDLOCK(dp->port_rwlock);
332 static void dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
333 struct pkt_metadata *)
334 OVS_REQ_RDLOCK(dp->port_rwlock);
335 static void dp_netdev_set_threads(struct dp_netdev *, int n);
337 static struct dpif_netdev *
338 dpif_netdev_cast(const struct dpif *dpif)
340 ovs_assert(dpif->dpif_class->open == dpif_netdev_open);
341 return CONTAINER_OF(dpif, struct dpif_netdev, dpif);
344 static struct dp_netdev *
345 get_dp_netdev(const struct dpif *dpif)
347 return dpif_netdev_cast(dpif)->dp;
351 dpif_netdev_enumerate(struct sset *all_dps)
353 struct shash_node *node;
355 ovs_mutex_lock(&dp_netdev_mutex);
356 SHASH_FOR_EACH(node, &dp_netdevs) {
357 sset_add(all_dps, node->name);
359 ovs_mutex_unlock(&dp_netdev_mutex);
365 dpif_netdev_class_is_dummy(const struct dpif_class *class)
367 return class != &dpif_netdev_class;
371 dpif_netdev_port_open_type(const struct dpif_class *class, const char *type)
373 return strcmp(type, "internal") ? type
374 : dpif_netdev_class_is_dummy(class) ? "dummy"
379 create_dpif_netdev(struct dp_netdev *dp)
381 uint16_t netflow_id = hash_string(dp->name, 0);
382 struct dpif_netdev *dpif;
384 ovs_refcount_ref(&dp->ref_cnt);
386 dpif = xmalloc(sizeof *dpif);
387 dpif_init(&dpif->dpif, dp->class, dp->name, netflow_id >> 8, netflow_id);
389 dpif->last_port_seq = seq_read(dp->port_seq);
394 /* Choose an unused, non-zero port number and return it on success.
395 * Return ODPP_NONE on failure. */
397 choose_port(struct dp_netdev *dp, const char *name)
398 OVS_REQ_RDLOCK(dp->port_rwlock)
402 if (dp->class != &dpif_netdev_class) {
406 /* If the port name begins with "br", start the number search at
407 * 100 to make writing tests easier. */
408 if (!strncmp(name, "br", 2)) {
412 /* If the port name contains a number, try to assign that port number.
413 * This can make writing unit tests easier because port numbers are
415 for (p = name; *p != '\0'; p++) {
416 if (isdigit((unsigned char) *p)) {
417 port_no = start_no + strtol(p, NULL, 10);
418 if (port_no > 0 && port_no != odp_to_u32(ODPP_NONE)
419 && !dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
420 return u32_to_odp(port_no);
427 for (port_no = 1; port_no <= UINT16_MAX; port_no++) {
428 if (!dp_netdev_lookup_port(dp, u32_to_odp(port_no))) {
429 return u32_to_odp(port_no);
437 create_dp_netdev(const char *name, const struct dpif_class *class,
438 struct dp_netdev **dpp)
439 OVS_REQUIRES(dp_netdev_mutex)
441 struct dp_netdev *dp;
445 dp = xzalloc(sizeof *dp);
446 shash_add(&dp_netdevs, name, dp);
448 *CONST_CAST(const struct dpif_class **, &dp->class) = class;
449 *CONST_CAST(const char **, &dp->name) = xstrdup(name);
450 ovs_refcount_init(&dp->ref_cnt);
451 atomic_flag_init(&dp->destroyed);
453 ovs_mutex_init(&dp->flow_mutex);
454 classifier_init(&dp->cls, NULL);
455 hmap_init(&dp->flow_table);
457 ovs_mutex_init(&dp->queue_mutex);
458 ovs_mutex_lock(&dp->queue_mutex);
459 for (i = 0; i < N_QUEUES; i++) {
460 dp->queues[i].head = dp->queues[i].tail = 0;
462 ovs_mutex_unlock(&dp->queue_mutex);
463 dp->queue_seq = seq_create();
465 dp->n_hit = ovsthread_counter_create();
466 dp->n_missed = ovsthread_counter_create();
467 dp->n_lost = ovsthread_counter_create();
469 ovs_rwlock_init(&dp->port_rwlock);
470 hmap_init(&dp->ports);
471 dp->port_seq = seq_create();
472 latch_init(&dp->exit_latch);
474 ovs_rwlock_wrlock(&dp->port_rwlock);
475 error = do_add_port(dp, name, "internal", ODPP_LOCAL);
476 ovs_rwlock_unlock(&dp->port_rwlock);
481 dp_netdev_set_threads(dp, 2);
488 dpif_netdev_open(const struct dpif_class *class, const char *name,
489 bool create, struct dpif **dpifp)
491 struct dp_netdev *dp;
494 ovs_mutex_lock(&dp_netdev_mutex);
495 dp = shash_find_data(&dp_netdevs, name);
497 error = create ? create_dp_netdev(name, class, &dp) : ENODEV;
499 error = (dp->class != class ? EINVAL
504 *dpifp = create_dpif_netdev(dp);
506 ovs_mutex_unlock(&dp_netdev_mutex);
512 dp_netdev_purge_queues(struct dp_netdev *dp)
516 ovs_mutex_lock(&dp->queue_mutex);
517 for (i = 0; i < N_QUEUES; i++) {
518 struct dp_netdev_queue *q = &dp->queues[i];
520 while (q->tail != q->head) {
521 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
522 ofpbuf_uninit(&u->upcall.packet);
523 ofpbuf_uninit(&u->buf);
526 ovs_mutex_unlock(&dp->queue_mutex);
529 /* Requires dp_netdev_mutex so that we can't get a new reference to 'dp'
530 * through the 'dp_netdevs' shash while freeing 'dp'. */
532 dp_netdev_free(struct dp_netdev *dp)
533 OVS_REQUIRES(dp_netdev_mutex)
535 struct dp_netdev_port *port, *next;
537 shash_find_and_delete(&dp_netdevs, dp->name);
539 dp_netdev_set_threads(dp, 0);
540 free(dp->forwarders);
542 dp_netdev_flow_flush(dp);
543 ovs_rwlock_wrlock(&dp->port_rwlock);
544 HMAP_FOR_EACH_SAFE (port, next, node, &dp->ports) {
545 do_del_port(dp, port->port_no);
547 ovs_rwlock_unlock(&dp->port_rwlock);
548 ovsthread_counter_destroy(dp->n_hit);
549 ovsthread_counter_destroy(dp->n_missed);
550 ovsthread_counter_destroy(dp->n_lost);
552 dp_netdev_purge_queues(dp);
553 seq_destroy(dp->queue_seq);
554 ovs_mutex_destroy(&dp->queue_mutex);
556 classifier_destroy(&dp->cls);
557 hmap_destroy(&dp->flow_table);
558 ovs_mutex_destroy(&dp->flow_mutex);
559 seq_destroy(dp->port_seq);
560 hmap_destroy(&dp->ports);
561 atomic_flag_destroy(&dp->destroyed);
562 ovs_refcount_destroy(&dp->ref_cnt);
563 latch_destroy(&dp->exit_latch);
564 free(CONST_CAST(char *, dp->name));
569 dp_netdev_unref(struct dp_netdev *dp)
572 /* Take dp_netdev_mutex so that, if dp->ref_cnt falls to zero, we can't
573 * get a new reference to 'dp' through the 'dp_netdevs' shash. */
574 ovs_mutex_lock(&dp_netdev_mutex);
575 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
578 ovs_mutex_unlock(&dp_netdev_mutex);
583 dpif_netdev_close(struct dpif *dpif)
585 struct dp_netdev *dp = get_dp_netdev(dpif);
592 dpif_netdev_destroy(struct dpif *dpif)
594 struct dp_netdev *dp = get_dp_netdev(dpif);
596 if (!atomic_flag_test_and_set(&dp->destroyed)) {
597 if (ovs_refcount_unref(&dp->ref_cnt) == 1) {
598 /* Can't happen: 'dpif' still owns a reference to 'dp'. */
607 dpif_netdev_get_stats(const struct dpif *dpif, struct dpif_dp_stats *stats)
609 struct dp_netdev *dp = get_dp_netdev(dpif);
611 fat_rwlock_rdlock(&dp->cls.rwlock);
612 stats->n_flows = hmap_count(&dp->flow_table);
613 fat_rwlock_unlock(&dp->cls.rwlock);
615 stats->n_hit = ovsthread_counter_read(dp->n_hit);
616 stats->n_missed = ovsthread_counter_read(dp->n_missed);
617 stats->n_lost = ovsthread_counter_read(dp->n_lost);
618 stats->n_masks = UINT32_MAX;
619 stats->n_mask_hit = UINT64_MAX;
625 do_add_port(struct dp_netdev *dp, const char *devname, const char *type,
627 OVS_REQ_WRLOCK(dp->port_rwlock)
629 struct netdev_saved_flags *sf;
630 struct dp_netdev_port *port;
631 struct netdev *netdev;
632 struct netdev_rx *rx;
633 enum netdev_flags flags;
634 const char *open_type;
637 /* XXX reject devices already in some dp_netdev. */
639 /* Open and validate network device. */
640 open_type = dpif_netdev_port_open_type(dp->class, type);
641 error = netdev_open(devname, open_type, &netdev);
645 /* XXX reject non-Ethernet devices */
647 netdev_get_flags(netdev, &flags);
648 if (flags & NETDEV_LOOPBACK) {
649 VLOG_ERR("%s: cannot add a loopback device", devname);
650 netdev_close(netdev);
654 error = netdev_rx_open(netdev, &rx);
656 && !(error == EOPNOTSUPP && dpif_netdev_class_is_dummy(dp->class))) {
657 VLOG_ERR("%s: cannot receive packets on this network device (%s)",
658 devname, ovs_strerror(errno));
659 netdev_close(netdev);
663 error = netdev_turn_flags_on(netdev, NETDEV_PROMISC, &sf);
666 netdev_close(netdev);
670 port = xmalloc(sizeof *port);
671 port->port_no = port_no;
672 port->netdev = netdev;
675 port->type = xstrdup(type);
677 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
678 seq_change(dp->port_seq);
684 dpif_netdev_port_add(struct dpif *dpif, struct netdev *netdev,
685 odp_port_t *port_nop)
687 struct dp_netdev *dp = get_dp_netdev(dpif);
688 char namebuf[NETDEV_VPORT_NAME_BUFSIZE];
689 const char *dpif_port;
693 ovs_rwlock_wrlock(&dp->port_rwlock);
694 dpif_port = netdev_vport_get_dpif_port(netdev, namebuf, sizeof namebuf);
695 if (*port_nop != ODPP_NONE) {
697 error = dp_netdev_lookup_port(dp, *port_nop) ? EBUSY : 0;
699 port_no = choose_port(dp, dpif_port);
700 error = port_no == ODPP_NONE ? EFBIG : 0;
704 error = do_add_port(dp, dpif_port, netdev_get_type(netdev), port_no);
706 ovs_rwlock_unlock(&dp->port_rwlock);
712 dpif_netdev_port_del(struct dpif *dpif, odp_port_t port_no)
714 struct dp_netdev *dp = get_dp_netdev(dpif);
717 ovs_rwlock_wrlock(&dp->port_rwlock);
718 error = port_no == ODPP_LOCAL ? EINVAL : do_del_port(dp, port_no);
719 ovs_rwlock_unlock(&dp->port_rwlock);
725 is_valid_port_number(odp_port_t port_no)
727 return port_no != ODPP_NONE;
730 static struct dp_netdev_port *
731 dp_netdev_lookup_port(const struct dp_netdev *dp, odp_port_t port_no)
732 OVS_REQ_RDLOCK(dp->port_rwlock)
734 struct dp_netdev_port *port;
736 HMAP_FOR_EACH_IN_BUCKET (port, node, hash_int(odp_to_u32(port_no), 0),
738 if (port->port_no == port_no) {
746 get_port_by_number(struct dp_netdev *dp,
747 odp_port_t port_no, struct dp_netdev_port **portp)
748 OVS_REQ_RDLOCK(dp->port_rwlock)
750 if (!is_valid_port_number(port_no)) {
754 *portp = dp_netdev_lookup_port(dp, port_no);
755 return *portp ? 0 : ENOENT;
760 get_port_by_name(struct dp_netdev *dp,
761 const char *devname, struct dp_netdev_port **portp)
762 OVS_REQ_RDLOCK(dp->port_rwlock)
764 struct dp_netdev_port *port;
766 HMAP_FOR_EACH (port, node, &dp->ports) {
767 if (!strcmp(netdev_get_name(port->netdev), devname)) {
776 do_del_port(struct dp_netdev *dp, odp_port_t port_no)
777 OVS_REQ_WRLOCK(dp->port_rwlock)
779 struct dp_netdev_port *port;
782 error = get_port_by_number(dp, port_no, &port);
787 hmap_remove(&dp->ports, &port->node);
788 seq_change(dp->port_seq);
790 netdev_close(port->netdev);
791 netdev_restore_flags(port->sf);
792 netdev_rx_close(port->rx);
800 answer_port_query(const struct dp_netdev_port *port,
801 struct dpif_port *dpif_port)
803 dpif_port->name = xstrdup(netdev_get_name(port->netdev));
804 dpif_port->type = xstrdup(port->type);
805 dpif_port->port_no = port->port_no;
809 dpif_netdev_port_query_by_number(const struct dpif *dpif, odp_port_t port_no,
810 struct dpif_port *dpif_port)
812 struct dp_netdev *dp = get_dp_netdev(dpif);
813 struct dp_netdev_port *port;
816 ovs_rwlock_rdlock(&dp->port_rwlock);
817 error = get_port_by_number(dp, port_no, &port);
818 if (!error && dpif_port) {
819 answer_port_query(port, dpif_port);
821 ovs_rwlock_unlock(&dp->port_rwlock);
827 dpif_netdev_port_query_by_name(const struct dpif *dpif, const char *devname,
828 struct dpif_port *dpif_port)
830 struct dp_netdev *dp = get_dp_netdev(dpif);
831 struct dp_netdev_port *port;
834 ovs_rwlock_rdlock(&dp->port_rwlock);
835 error = get_port_by_name(dp, devname, &port);
836 if (!error && dpif_port) {
837 answer_port_query(port, dpif_port);
839 ovs_rwlock_unlock(&dp->port_rwlock);
845 dp_netdev_remove_flow(struct dp_netdev *dp, struct dp_netdev_flow *flow)
846 OVS_REQ_WRLOCK(dp->cls.rwlock)
847 OVS_REQUIRES(dp->flow_mutex)
849 struct cls_rule *cr = CONST_CAST(struct cls_rule *, &flow->cr);
850 struct hmap_node *node = CONST_CAST(struct hmap_node *, &flow->node);
852 classifier_remove(&dp->cls, cr);
853 hmap_remove(&dp->flow_table, node);
854 dp_netdev_flow_unref(flow);
857 static struct dp_netdev_flow *
858 dp_netdev_flow_ref(const struct dp_netdev_flow *flow_)
860 struct dp_netdev_flow *flow = CONST_CAST(struct dp_netdev_flow *, flow_);
862 ovs_refcount_ref(&flow->ref_cnt);
868 dp_netdev_flow_unref(struct dp_netdev_flow *flow)
870 if (flow && ovs_refcount_unref(&flow->ref_cnt) == 1) {
871 cls_rule_destroy(CONST_CAST(struct cls_rule *, &flow->cr));
872 ovs_mutex_lock(&flow->mutex);
873 dp_netdev_actions_unref(flow->actions);
874 ovs_refcount_destroy(&flow->ref_cnt);
875 ovs_mutex_unlock(&flow->mutex);
876 ovs_mutex_destroy(&flow->mutex);
882 dp_netdev_flow_flush(struct dp_netdev *dp)
884 struct dp_netdev_flow *netdev_flow, *next;
886 ovs_mutex_lock(&dp->flow_mutex);
887 fat_rwlock_wrlock(&dp->cls.rwlock);
888 HMAP_FOR_EACH_SAFE (netdev_flow, next, node, &dp->flow_table) {
889 dp_netdev_remove_flow(dp, netdev_flow);
891 fat_rwlock_unlock(&dp->cls.rwlock);
892 ovs_mutex_unlock(&dp->flow_mutex);
896 dpif_netdev_flow_flush(struct dpif *dpif)
898 struct dp_netdev *dp = get_dp_netdev(dpif);
900 dp_netdev_flow_flush(dp);
904 struct dp_netdev_port_state {
911 dpif_netdev_port_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
913 *statep = xzalloc(sizeof(struct dp_netdev_port_state));
918 dpif_netdev_port_dump_next(const struct dpif *dpif, void *state_,
919 struct dpif_port *dpif_port)
921 struct dp_netdev_port_state *state = state_;
922 struct dp_netdev *dp = get_dp_netdev(dpif);
923 struct hmap_node *node;
926 ovs_rwlock_rdlock(&dp->port_rwlock);
927 node = hmap_at_position(&dp->ports, &state->bucket, &state->offset);
929 struct dp_netdev_port *port;
931 port = CONTAINER_OF(node, struct dp_netdev_port, node);
934 state->name = xstrdup(netdev_get_name(port->netdev));
935 dpif_port->name = state->name;
936 dpif_port->type = port->type;
937 dpif_port->port_no = port->port_no;
943 ovs_rwlock_unlock(&dp->port_rwlock);
949 dpif_netdev_port_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
951 struct dp_netdev_port_state *state = state_;
958 dpif_netdev_port_poll(const struct dpif *dpif_, char **devnamep OVS_UNUSED)
960 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
961 uint64_t new_port_seq;
964 new_port_seq = seq_read(dpif->dp->port_seq);
965 if (dpif->last_port_seq != new_port_seq) {
966 dpif->last_port_seq = new_port_seq;
976 dpif_netdev_port_poll_wait(const struct dpif *dpif_)
978 struct dpif_netdev *dpif = dpif_netdev_cast(dpif_);
980 seq_wait(dpif->dp->port_seq, dpif->last_port_seq);
983 static struct dp_netdev_flow *
984 dp_netdev_flow_cast(const struct cls_rule *cr)
986 return cr ? CONTAINER_OF(cr, struct dp_netdev_flow, cr) : NULL;
989 static struct dp_netdev_flow *
990 dp_netdev_lookup_flow(const struct dp_netdev *dp, const struct flow *flow)
991 OVS_EXCLUDED(dp->cls.rwlock)
993 struct dp_netdev_flow *netdev_flow;
995 fat_rwlock_rdlock(&dp->cls.rwlock);
996 netdev_flow = dp_netdev_flow_cast(classifier_lookup(&dp->cls, flow, NULL));
997 dp_netdev_flow_ref(netdev_flow);
998 fat_rwlock_unlock(&dp->cls.rwlock);
1003 static struct dp_netdev_flow *
1004 dp_netdev_find_flow(const struct dp_netdev *dp, const struct flow *flow)
1005 OVS_REQ_RDLOCK(dp->cls.rwlock)
1007 struct dp_netdev_flow *netdev_flow;
1009 HMAP_FOR_EACH_WITH_HASH (netdev_flow, node, flow_hash(flow, 0),
1011 if (flow_equal(&netdev_flow->flow, flow)) {
1012 return dp_netdev_flow_ref(netdev_flow);
1020 get_dpif_flow_stats(struct dp_netdev_flow *netdev_flow,
1021 struct dpif_flow_stats *stats)
1022 OVS_REQ_RDLOCK(netdev_flow->mutex)
1024 stats->n_packets = netdev_flow->packet_count;
1025 stats->n_bytes = netdev_flow->byte_count;
1026 stats->used = netdev_flow->used;
1027 stats->tcp_flags = netdev_flow->tcp_flags;
1031 dpif_netdev_mask_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1032 const struct nlattr *mask_key,
1033 uint32_t mask_key_len, const struct flow *flow,
1037 enum odp_key_fitness fitness;
1039 fitness = odp_flow_key_to_mask(mask_key, mask_key_len, mask, flow);
1041 /* This should not happen: it indicates that
1042 * odp_flow_key_from_mask() and odp_flow_key_to_mask()
1043 * disagree on the acceptable form of a mask. Log the problem
1044 * as an error, with enough details to enable debugging. */
1045 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1047 if (!VLOG_DROP_ERR(&rl)) {
1051 odp_flow_format(key, key_len, mask_key, mask_key_len, NULL, &s,
1053 VLOG_ERR("internal error parsing flow mask %s (%s)",
1054 ds_cstr(&s), odp_key_fitness_to_string(fitness));
1060 /* Force unwildcard the in_port. */
1061 mask->in_port.odp_port = u32_to_odp(UINT32_MAX);
1063 enum mf_field_id id;
1064 /* No mask key, unwildcard everything except fields whose
1065 * prerequisities are not met. */
1066 memset(mask, 0x0, sizeof *mask);
1068 for (id = 0; id < MFF_N_IDS; ++id) {
1069 /* Skip registers and metadata. */
1070 if (!(id >= MFF_REG0 && id < MFF_REG0 + FLOW_N_REGS)
1071 && id != MFF_METADATA) {
1072 const struct mf_field *mf = mf_from_id(id);
1073 if (mf_are_prereqs_ok(mf, flow)) {
1074 mf_mask_field(mf, mask);
1084 dpif_netdev_flow_from_nlattrs(const struct nlattr *key, uint32_t key_len,
1089 if (odp_flow_key_to_flow(key, key_len, flow)) {
1090 /* This should not happen: it indicates that odp_flow_key_from_flow()
1091 * and odp_flow_key_to_flow() disagree on the acceptable form of a
1092 * flow. Log the problem as an error, with enough details to enable
1094 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
1096 if (!VLOG_DROP_ERR(&rl)) {
1100 odp_flow_format(key, key_len, NULL, 0, NULL, &s, true);
1101 VLOG_ERR("internal error parsing flow key %s", ds_cstr(&s));
1108 in_port = flow->in_port.odp_port;
1109 if (!is_valid_port_number(in_port) && in_port != ODPP_NONE) {
1117 dpif_netdev_flow_get(const struct dpif *dpif,
1118 const struct nlattr *nl_key, size_t nl_key_len,
1119 struct ofpbuf **actionsp, struct dpif_flow_stats *stats)
1121 struct dp_netdev *dp = get_dp_netdev(dpif);
1122 struct dp_netdev_flow *netdev_flow;
1126 error = dpif_netdev_flow_from_nlattrs(nl_key, nl_key_len, &key);
1131 fat_rwlock_rdlock(&dp->cls.rwlock);
1132 netdev_flow = dp_netdev_find_flow(dp, &key);
1133 fat_rwlock_unlock(&dp->cls.rwlock);
1136 struct dp_netdev_actions *actions = NULL;
1138 ovs_mutex_lock(&netdev_flow->mutex);
1140 get_dpif_flow_stats(netdev_flow, stats);
1143 actions = dp_netdev_actions_ref(netdev_flow->actions);
1145 ovs_mutex_unlock(&netdev_flow->mutex);
1147 dp_netdev_flow_unref(netdev_flow);
1150 *actionsp = ofpbuf_clone_data(actions->actions, actions->size);
1151 dp_netdev_actions_unref(actions);
1161 dp_netdev_flow_add(struct dp_netdev *dp, const struct flow *flow,
1162 const struct flow_wildcards *wc,
1163 const struct nlattr *actions,
1165 OVS_REQUIRES(dp->flow_mutex)
1167 struct dp_netdev_flow *netdev_flow;
1170 netdev_flow = xzalloc(sizeof *netdev_flow);
1171 *CONST_CAST(struct flow *, &netdev_flow->flow) = *flow;
1172 ovs_refcount_init(&netdev_flow->ref_cnt);
1174 ovs_mutex_init(&netdev_flow->mutex);
1175 ovs_mutex_lock(&netdev_flow->mutex);
1177 netdev_flow->actions = dp_netdev_actions_create(actions, actions_len);
1179 match_init(&match, flow, wc);
1180 cls_rule_init(CONST_CAST(struct cls_rule *, &netdev_flow->cr),
1181 &match, NETDEV_RULE_PRIORITY);
1182 fat_rwlock_wrlock(&dp->cls.rwlock);
1183 classifier_insert(&dp->cls,
1184 CONST_CAST(struct cls_rule *, &netdev_flow->cr));
1185 hmap_insert(&dp->flow_table,
1186 CONST_CAST(struct hmap_node *, &netdev_flow->node),
1187 flow_hash(flow, 0));
1188 fat_rwlock_unlock(&dp->cls.rwlock);
1190 ovs_mutex_unlock(&netdev_flow->mutex);
1196 clear_stats(struct dp_netdev_flow *netdev_flow)
1197 OVS_REQUIRES(netdev_flow->mutex)
1199 netdev_flow->used = 0;
1200 netdev_flow->packet_count = 0;
1201 netdev_flow->byte_count = 0;
1202 netdev_flow->tcp_flags = 0;
1206 dpif_netdev_flow_put(struct dpif *dpif, const struct dpif_flow_put *put)
1208 struct dp_netdev *dp = get_dp_netdev(dpif);
1209 struct dp_netdev_flow *netdev_flow;
1211 struct flow_wildcards wc;
1214 error = dpif_netdev_flow_from_nlattrs(put->key, put->key_len, &flow);
1218 error = dpif_netdev_mask_from_nlattrs(put->key, put->key_len,
1219 put->mask, put->mask_len,
1225 ovs_mutex_lock(&dp->flow_mutex);
1226 netdev_flow = dp_netdev_lookup_flow(dp, &flow);
1228 if (put->flags & DPIF_FP_CREATE) {
1229 if (hmap_count(&dp->flow_table) < MAX_FLOWS) {
1231 memset(put->stats, 0, sizeof *put->stats);
1233 error = dp_netdev_flow_add(dp, &flow, &wc, put->actions,
1242 if (put->flags & DPIF_FP_MODIFY
1243 && flow_equal(&flow, &netdev_flow->flow)) {
1244 struct dp_netdev_actions *new_actions;
1245 struct dp_netdev_actions *old_actions;
1247 new_actions = dp_netdev_actions_create(put->actions,
1250 ovs_mutex_lock(&netdev_flow->mutex);
1251 old_actions = netdev_flow->actions;
1252 netdev_flow->actions = new_actions;
1254 get_dpif_flow_stats(netdev_flow, put->stats);
1256 if (put->flags & DPIF_FP_ZERO_STATS) {
1257 clear_stats(netdev_flow);
1259 ovs_mutex_unlock(&netdev_flow->mutex);
1261 dp_netdev_actions_unref(old_actions);
1262 } else if (put->flags & DPIF_FP_CREATE) {
1265 /* Overlapping flow. */
1268 dp_netdev_flow_unref(netdev_flow);
1270 ovs_mutex_unlock(&dp->flow_mutex);
1276 dpif_netdev_flow_del(struct dpif *dpif, const struct dpif_flow_del *del)
1278 struct dp_netdev *dp = get_dp_netdev(dpif);
1279 struct dp_netdev_flow *netdev_flow;
1283 error = dpif_netdev_flow_from_nlattrs(del->key, del->key_len, &key);
1288 ovs_mutex_lock(&dp->flow_mutex);
1289 fat_rwlock_wrlock(&dp->cls.rwlock);
1290 netdev_flow = dp_netdev_find_flow(dp, &key);
1293 ovs_mutex_lock(&netdev_flow->mutex);
1294 get_dpif_flow_stats(netdev_flow, del->stats);
1295 ovs_mutex_unlock(&netdev_flow->mutex);
1297 dp_netdev_remove_flow(dp, netdev_flow);
1298 dp_netdev_flow_unref(netdev_flow);
1302 fat_rwlock_unlock(&dp->cls.rwlock);
1303 ovs_mutex_unlock(&dp->flow_mutex);
1308 struct dp_netdev_flow_state {
1311 struct dp_netdev_actions *actions;
1312 struct odputil_keybuf keybuf;
1313 struct odputil_keybuf maskbuf;
1314 struct dpif_flow_stats stats;
1318 dpif_netdev_flow_dump_start(const struct dpif *dpif OVS_UNUSED, void **statep)
1320 struct dp_netdev_flow_state *state;
1322 *statep = state = xmalloc(sizeof *state);
1325 state->actions = NULL;
1330 dpif_netdev_flow_dump_next(const struct dpif *dpif, void *state_,
1331 const struct nlattr **key, size_t *key_len,
1332 const struct nlattr **mask, size_t *mask_len,
1333 const struct nlattr **actions, size_t *actions_len,
1334 const struct dpif_flow_stats **stats)
1336 struct dp_netdev_flow_state *state = state_;
1337 struct dp_netdev *dp = get_dp_netdev(dpif);
1338 struct dp_netdev_flow *netdev_flow;
1339 struct hmap_node *node;
1341 fat_rwlock_rdlock(&dp->cls.rwlock);
1342 node = hmap_at_position(&dp->flow_table, &state->bucket, &state->offset);
1344 netdev_flow = CONTAINER_OF(node, struct dp_netdev_flow, node);
1345 dp_netdev_flow_ref(netdev_flow);
1347 fat_rwlock_unlock(&dp->cls.rwlock);
1355 ofpbuf_use_stack(&buf, &state->keybuf, sizeof state->keybuf);
1356 odp_flow_key_from_flow(&buf, &netdev_flow->flow,
1357 netdev_flow->flow.in_port.odp_port);
1360 *key_len = buf.size;
1365 struct flow_wildcards wc;
1367 ofpbuf_use_stack(&buf, &state->maskbuf, sizeof state->maskbuf);
1368 minimask_expand(&netdev_flow->cr.match.mask, &wc);
1369 odp_flow_key_from_mask(&buf, &wc.masks, &netdev_flow->flow,
1370 odp_to_u32(wc.masks.in_port.odp_port),
1374 *mask_len = buf.size;
1377 if (actions || stats) {
1378 dp_netdev_actions_unref(state->actions);
1379 state->actions = NULL;
1381 ovs_mutex_lock(&netdev_flow->mutex);
1383 state->actions = dp_netdev_actions_ref(netdev_flow->actions);
1384 *actions = state->actions->actions;
1385 *actions_len = state->actions->size;
1388 get_dpif_flow_stats(netdev_flow, &state->stats);
1389 *stats = &state->stats;
1391 ovs_mutex_unlock(&netdev_flow->mutex);
1394 dp_netdev_flow_unref(netdev_flow);
1400 dpif_netdev_flow_dump_done(const struct dpif *dpif OVS_UNUSED, void *state_)
1402 struct dp_netdev_flow_state *state = state_;
1404 dp_netdev_actions_unref(state->actions);
1410 dpif_netdev_execute(struct dpif *dpif, struct dpif_execute *execute)
1412 struct dp_netdev *dp = get_dp_netdev(dpif);
1413 struct pkt_metadata *md = &execute->md;
1416 if (execute->packet->size < ETH_HEADER_LEN ||
1417 execute->packet->size > UINT16_MAX) {
1421 /* Extract flow key. */
1422 flow_extract(execute->packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1423 (union flow_in_port *)&md->in_port, &key);
1425 ovs_rwlock_rdlock(&dp->port_rwlock);
1426 dp_netdev_execute_actions(dp, &key, execute->packet, md, execute->actions,
1427 execute->actions_len);
1428 ovs_rwlock_unlock(&dp->port_rwlock);
1434 dpif_netdev_recv_set(struct dpif *dpif OVS_UNUSED, bool enable OVS_UNUSED)
1440 dpif_netdev_queue_to_priority(const struct dpif *dpif OVS_UNUSED,
1441 uint32_t queue_id, uint32_t *priority)
1443 *priority = queue_id;
1447 static struct dp_netdev_queue *
1448 find_nonempty_queue(struct dp_netdev *dp)
1449 OVS_REQUIRES(dp->queue_mutex)
1453 for (i = 0; i < N_QUEUES; i++) {
1454 struct dp_netdev_queue *q = &dp->queues[i];
1455 if (q->head != q->tail) {
1463 dpif_netdev_recv(struct dpif *dpif, struct dpif_upcall *upcall,
1466 struct dp_netdev *dp = get_dp_netdev(dpif);
1467 struct dp_netdev_queue *q;
1470 ovs_mutex_lock(&dp->queue_mutex);
1471 q = find_nonempty_queue(dp);
1473 struct dp_netdev_upcall *u = &q->upcalls[q->tail++ & QUEUE_MASK];
1475 *upcall = u->upcall;
1484 ovs_mutex_unlock(&dp->queue_mutex);
1490 dpif_netdev_recv_wait(struct dpif *dpif)
1492 struct dp_netdev *dp = get_dp_netdev(dpif);
1495 ovs_mutex_lock(&dp->queue_mutex);
1496 seq = seq_read(dp->queue_seq);
1497 if (find_nonempty_queue(dp)) {
1498 poll_immediate_wake();
1500 seq_wait(dp->queue_seq, seq);
1502 ovs_mutex_unlock(&dp->queue_mutex);
1506 dpif_netdev_recv_purge(struct dpif *dpif)
1508 struct dpif_netdev *dpif_netdev = dpif_netdev_cast(dpif);
1510 dp_netdev_purge_queues(dpif_netdev->dp);
1513 /* Creates and returns a new 'struct dp_netdev_actions', with a reference count
1514 * of 1, whose actions are a copy of from the 'ofpacts_len' bytes of
1516 struct dp_netdev_actions *
1517 dp_netdev_actions_create(const struct nlattr *actions, size_t size)
1519 struct dp_netdev_actions *netdev_actions;
1521 netdev_actions = xmalloc(sizeof *netdev_actions);
1522 ovs_refcount_init(&netdev_actions->ref_cnt);
1523 netdev_actions->actions = xmemdup(actions, size);
1524 netdev_actions->size = size;
1526 return netdev_actions;
1529 /* Increments 'actions''s refcount. */
1530 struct dp_netdev_actions *
1531 dp_netdev_actions_ref(const struct dp_netdev_actions *actions_)
1533 struct dp_netdev_actions *actions;
1535 actions = CONST_CAST(struct dp_netdev_actions *, actions_);
1537 ovs_refcount_ref(&actions->ref_cnt);
1542 /* Decrements 'actions''s refcount and frees 'actions' if the refcount reaches
1545 dp_netdev_actions_unref(struct dp_netdev_actions *actions)
1547 if (actions && ovs_refcount_unref(&actions->ref_cnt) == 1) {
1548 ovs_refcount_destroy(&actions->ref_cnt);
1549 free(actions->actions);
1555 dp_forwarder_main(void *f_)
1557 struct dp_forwarder *f = f_;
1558 struct dp_netdev *dp = f->dp;
1559 struct ofpbuf packet;
1561 f->name = xasprintf("forwarder_%u", ovsthread_id_self());
1562 set_subprogram_name("%s", f->name);
1564 ofpbuf_init(&packet, 0);
1565 while (!latch_is_set(&dp->exit_latch)) {
1566 bool received_anything;
1569 ovs_rwlock_rdlock(&dp->port_rwlock);
1570 for (i = 0; i < 50; i++) {
1571 struct dp_netdev_port *port;
1573 received_anything = false;
1574 HMAP_FOR_EACH (port, node, &f->dp->ports) {
1576 && port->node.hash >= f->min_hash
1577 && port->node.hash <= f->max_hash) {
1582 if (netdev_get_mtu(port->netdev, &mtu)) {
1583 mtu = ETH_PAYLOAD_MAX;
1585 buf_size = DP_NETDEV_HEADROOM + VLAN_ETH_HEADER_LEN + mtu;
1587 ofpbuf_clear(&packet);
1588 ofpbuf_reserve_with_tailroom(&packet, DP_NETDEV_HEADROOM,
1591 error = netdev_rx_recv(port->rx, &packet);
1593 struct pkt_metadata md
1594 = PKT_METADATA_INITIALIZER(port->port_no);
1595 dp_netdev_port_input(dp, &packet, &md);
1597 received_anything = true;
1598 } else if (error != EAGAIN && error != EOPNOTSUPP) {
1599 static struct vlog_rate_limit rl
1600 = VLOG_RATE_LIMIT_INIT(1, 5);
1602 VLOG_ERR_RL(&rl, "error receiving data from %s: %s",
1603 netdev_get_name(port->netdev),
1604 ovs_strerror(error));
1609 if (!received_anything) {
1614 if (received_anything) {
1615 poll_immediate_wake();
1617 struct dp_netdev_port *port;
1619 HMAP_FOR_EACH (port, node, &f->dp->ports)
1621 && port->node.hash >= f->min_hash
1622 && port->node.hash <= f->max_hash) {
1623 netdev_rx_wait(port->rx);
1625 seq_wait(dp->port_seq, seq_read(dp->port_seq));
1626 latch_wait(&dp->exit_latch);
1628 ovs_rwlock_unlock(&dp->port_rwlock);
1632 ofpbuf_uninit(&packet);
1640 dp_netdev_set_threads(struct dp_netdev *dp, int n)
1644 if (n == dp->n_forwarders) {
1648 /* Stop existing threads. */
1649 latch_set(&dp->exit_latch);
1650 for (i = 0; i < dp->n_forwarders; i++) {
1651 struct dp_forwarder *f = &dp->forwarders[i];
1653 xpthread_join(f->thread, NULL);
1655 latch_poll(&dp->exit_latch);
1656 free(dp->forwarders);
1658 /* Start new threads. */
1659 dp->forwarders = xmalloc(n * sizeof *dp->forwarders);
1660 dp->n_forwarders = n;
1661 for (i = 0; i < n; i++) {
1662 struct dp_forwarder *f = &dp->forwarders[i];
1665 f->min_hash = UINT32_MAX / n * i;
1666 f->max_hash = UINT32_MAX / n * (i + 1) - 1;
1668 f->max_hash = UINT32_MAX;
1670 xpthread_create(&f->thread, NULL, dp_forwarder_main, f);
1675 dp_netdev_flow_used(struct dp_netdev_flow *netdev_flow,
1676 const struct ofpbuf *packet)
1677 OVS_REQUIRES(netdev_flow->mutex)
1679 netdev_flow->used = time_msec();
1680 netdev_flow->packet_count++;
1681 netdev_flow->byte_count += packet->size;
1682 netdev_flow->tcp_flags |= packet_get_tcp_flags(packet, &netdev_flow->flow);
1686 dp_netdev_port_input(struct dp_netdev *dp, struct ofpbuf *packet,
1687 struct pkt_metadata *md)
1688 OVS_REQ_RDLOCK(dp->port_rwlock)
1690 struct dp_netdev_flow *netdev_flow;
1693 if (packet->size < ETH_HEADER_LEN) {
1696 flow_extract(packet, md->skb_priority, md->pkt_mark, &md->tunnel,
1697 (union flow_in_port *)&md->in_port, &key);
1698 netdev_flow = dp_netdev_lookup_flow(dp, &key);
1700 struct dp_netdev_actions *actions;
1702 ovs_mutex_lock(&netdev_flow->mutex);
1703 dp_netdev_flow_used(netdev_flow, packet);
1704 actions = dp_netdev_actions_ref(netdev_flow->actions);
1705 ovs_mutex_unlock(&netdev_flow->mutex);
1707 dp_netdev_execute_actions(dp, &key, packet, md,
1708 actions->actions, actions->size);
1709 dp_netdev_actions_unref(actions);
1710 dp_netdev_flow_unref(netdev_flow);
1711 ovsthread_counter_inc(dp->n_hit, 1);
1713 ovsthread_counter_inc(dp->n_missed, 1);
1714 dp_netdev_output_userspace(dp, packet, DPIF_UC_MISS, &key, NULL);
1719 dp_netdev_output_userspace(struct dp_netdev *dp, struct ofpbuf *packet,
1720 int queue_no, const struct flow *flow,
1721 const struct nlattr *userdata)
1722 OVS_EXCLUDED(dp->queue_mutex)
1724 struct dp_netdev_queue *q = &dp->queues[queue_no];
1727 ovs_mutex_lock(&dp->queue_mutex);
1728 if (q->head - q->tail < MAX_QUEUE_LEN) {
1729 struct dp_netdev_upcall *u = &q->upcalls[q->head++ & QUEUE_MASK];
1730 struct dpif_upcall *upcall = &u->upcall;
1731 struct ofpbuf *buf = &u->buf;
1734 upcall->type = queue_no;
1736 /* Allocate buffer big enough for everything. */
1737 buf_size = ODPUTIL_FLOW_KEY_BYTES;
1739 buf_size += NLA_ALIGN(userdata->nla_len);
1741 ofpbuf_init(buf, buf_size);
1744 odp_flow_key_from_flow(buf, flow, flow->in_port.odp_port);
1745 upcall->key = buf->data;
1746 upcall->key_len = buf->size;
1750 upcall->userdata = ofpbuf_put(buf, userdata,
1751 NLA_ALIGN(userdata->nla_len));
1754 /* Steal packet data. */
1755 ovs_assert(packet->source == OFPBUF_MALLOC);
1756 upcall->packet = *packet;
1757 ofpbuf_use(packet, NULL, 0);
1759 seq_change(dp->queue_seq);
1763 ovsthread_counter_inc(dp->n_lost, 1);
1766 ovs_mutex_unlock(&dp->queue_mutex);
1771 struct dp_netdev_execute_aux {
1772 struct dp_netdev *dp;
1773 const struct flow *key;
1777 dp_execute_cb(void *aux_, struct ofpbuf *packet,
1778 const struct pkt_metadata *md OVS_UNUSED,
1779 const struct nlattr *a, bool may_steal)
1780 OVS_NO_THREAD_SAFETY_ANALYSIS
1782 struct dp_netdev_execute_aux *aux = aux_;
1783 int type = nl_attr_type(a);
1784 struct dp_netdev_port *p;
1786 switch ((enum ovs_action_attr)type) {
1787 case OVS_ACTION_ATTR_OUTPUT:
1788 p = dp_netdev_lookup_port(aux->dp, u32_to_odp(nl_attr_get_u32(a)));
1790 netdev_send(p->netdev, packet);
1794 case OVS_ACTION_ATTR_USERSPACE: {
1795 const struct nlattr *userdata;
1797 userdata = nl_attr_find_nested(a, OVS_USERSPACE_ATTR_USERDATA);
1799 /* Make a copy if we are not allowed to steal the packet's data. */
1801 packet = ofpbuf_clone_with_headroom(packet, DP_NETDEV_HEADROOM);
1803 dp_netdev_output_userspace(aux->dp, packet, DPIF_UC_ACTION, aux->key,
1806 ofpbuf_uninit(packet);
1810 case OVS_ACTION_ATTR_PUSH_VLAN:
1811 case OVS_ACTION_ATTR_POP_VLAN:
1812 case OVS_ACTION_ATTR_PUSH_MPLS:
1813 case OVS_ACTION_ATTR_POP_MPLS:
1814 case OVS_ACTION_ATTR_SET:
1815 case OVS_ACTION_ATTR_SAMPLE:
1816 case OVS_ACTION_ATTR_UNSPEC:
1817 case __OVS_ACTION_ATTR_MAX:
1823 dp_netdev_execute_actions(struct dp_netdev *dp, const struct flow *key,
1824 struct ofpbuf *packet, struct pkt_metadata *md,
1825 const struct nlattr *actions, size_t actions_len)
1826 OVS_REQ_RDLOCK(dp->port_rwlock)
1828 struct dp_netdev_execute_aux aux = {dp, key};
1830 odp_execute_actions(&aux, packet, md, actions, actions_len, dp_execute_cb);
1833 const struct dpif_class dpif_netdev_class = {
1835 dpif_netdev_enumerate,
1836 dpif_netdev_port_open_type,
1839 dpif_netdev_destroy,
1842 dpif_netdev_get_stats,
1843 dpif_netdev_port_add,
1844 dpif_netdev_port_del,
1845 dpif_netdev_port_query_by_number,
1846 dpif_netdev_port_query_by_name,
1847 NULL, /* port_get_pid */
1848 dpif_netdev_port_dump_start,
1849 dpif_netdev_port_dump_next,
1850 dpif_netdev_port_dump_done,
1851 dpif_netdev_port_poll,
1852 dpif_netdev_port_poll_wait,
1853 dpif_netdev_flow_get,
1854 dpif_netdev_flow_put,
1855 dpif_netdev_flow_del,
1856 dpif_netdev_flow_flush,
1857 dpif_netdev_flow_dump_start,
1858 dpif_netdev_flow_dump_next,
1859 dpif_netdev_flow_dump_done,
1860 dpif_netdev_execute,
1862 dpif_netdev_recv_set,
1863 dpif_netdev_queue_to_priority,
1865 dpif_netdev_recv_wait,
1866 dpif_netdev_recv_purge,
1870 dpif_dummy_change_port_number(struct unixctl_conn *conn, int argc OVS_UNUSED,
1871 const char *argv[], void *aux OVS_UNUSED)
1873 struct dp_netdev_port *port;
1874 struct dp_netdev *dp;
1877 ovs_mutex_lock(&dp_netdev_mutex);
1878 dp = shash_find_data(&dp_netdevs, argv[1]);
1879 if (!dp || !dpif_netdev_class_is_dummy(dp->class)) {
1880 ovs_mutex_unlock(&dp_netdev_mutex);
1881 unixctl_command_reply_error(conn, "unknown datapath or not a dummy");
1884 ovs_refcount_ref(&dp->ref_cnt);
1885 ovs_mutex_unlock(&dp_netdev_mutex);
1887 ovs_rwlock_wrlock(&dp->port_rwlock);
1888 if (get_port_by_name(dp, argv[2], &port)) {
1889 unixctl_command_reply_error(conn, "unknown port");
1893 port_no = u32_to_odp(atoi(argv[3]));
1894 if (!port_no || port_no == ODPP_NONE) {
1895 unixctl_command_reply_error(conn, "bad port number");
1898 if (dp_netdev_lookup_port(dp, port_no)) {
1899 unixctl_command_reply_error(conn, "port number already in use");
1902 hmap_remove(&dp->ports, &port->node);
1903 port->port_no = port_no;
1904 hmap_insert(&dp->ports, &port->node, hash_int(odp_to_u32(port_no), 0));
1905 seq_change(dp->port_seq);
1906 unixctl_command_reply(conn, NULL);
1909 ovs_rwlock_unlock(&dp->port_rwlock);
1910 dp_netdev_unref(dp);
1914 dpif_dummy_register__(const char *type)
1916 struct dpif_class *class;
1918 class = xmalloc(sizeof *class);
1919 *class = dpif_netdev_class;
1920 class->type = xstrdup(type);
1921 dp_register_provider(class);
1925 dpif_dummy_register(bool override)
1932 dp_enumerate_types(&types);
1933 SSET_FOR_EACH (type, &types) {
1934 if (!dp_unregister_provider(type)) {
1935 dpif_dummy_register__(type);
1938 sset_destroy(&types);
1941 dpif_dummy_register__("dummy");
1943 unixctl_command_register("dpif-dummy/change-port-number",
1944 "DP PORT NEW-NUMBER",
1945 3, 3, dpif_dummy_change_port_number, NULL);