X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev.c;h=9aaf10757c1deaf8652d740416a8a48fa405ab1b;hb=41ca1e0afb4b261a217c9fdaf672ef606e8434f9;hp=185d3dfcec492ce23bf9e3f30f5aea976812b522;hpb=6c0386119d614d5e26ca08cc3d8e527806b87ef9;p=sliver-openvswitch.git diff --git a/lib/netdev.c b/lib/netdev.c index 185d3dfce..9aaf10757 100644 --- a/lib/netdev.c +++ b/lib/netdev.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010, 2011, 2012 Nicira Networks. + * Copyright (c) 2008, 2009, 2010, 2011, 2012, 2013, 2014 Nicira, Inc. * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. @@ -17,7 +17,6 @@ #include #include "netdev.h" -#include #include #include #include @@ -26,10 +25,12 @@ #include #include "coverage.h" +#include "dpif.h" #include "dynamic-string.h" #include "fatal-signal.h" #include "hash.h" #include "list.h" +#include "netdev-dpdk.h" #include "netdev-provider.h" #include "netdev-vport.h" #include "ofpbuf.h" @@ -37,6 +38,7 @@ #include "packets.h" #include "poll-loop.h" #include "shash.h" +#include "smap.h" #include "sset.h" #include "svec.h" #include "vlog.h" @@ -48,38 +50,94 @@ COVERAGE_DEFINE(netdev_sent); COVERAGE_DEFINE(netdev_add_router); COVERAGE_DEFINE(netdev_get_stats); -static struct shash netdev_classes = SHASH_INITIALIZER(&netdev_classes); +struct netdev_saved_flags { + struct netdev *netdev; + struct list node; /* In struct netdev's saved_flags_list. */ + enum netdev_flags saved_flags; + enum netdev_flags saved_values; +}; + +/* Protects 'netdev_shash' and the mutable members of struct netdev. */ +static struct ovs_mutex netdev_mutex = OVS_MUTEX_INITIALIZER; /* All created network devices. */ -static struct shash netdev_dev_shash = SHASH_INITIALIZER(&netdev_dev_shash); +static struct shash netdev_shash OVS_GUARDED_BY(netdev_mutex) + = SHASH_INITIALIZER(&netdev_shash); + +/* Protects 'netdev_classes' against insertions or deletions. + * + * This is a recursive mutex to allow recursive acquisition when calling into + * providers. For example, netdev_run() calls into provider 'run' functions, + * which might reasonably want to call one of the netdev functions that takes + * netdev_class_mutex. */ +static struct ovs_mutex netdev_class_mutex OVS_ACQ_BEFORE(netdev_mutex); + +/* Contains 'struct netdev_registered_class'es. */ +static struct hmap netdev_classes OVS_GUARDED_BY(netdev_class_mutex) + = HMAP_INITIALIZER(&netdev_classes); -/* All open network devices. */ -static struct list netdev_list = LIST_INITIALIZER(&netdev_list); +struct netdev_registered_class { + struct hmap_node hmap_node; /* In 'netdev_classes', by class->type. */ + const struct netdev_class *class; + atomic_int ref_cnt; /* Number of 'struct netdev's of this class. */ +}; /* This is set pretty low because we probably won't learn anything from the * additional log messages. */ static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); -static void close_all_netdevs(void *aux OVS_UNUSED); -static int restore_flags(struct netdev *netdev); -void update_device_args(struct netdev_dev *, const struct shash *args); +static void restore_all_flags(void *aux OVS_UNUSED); +void update_device_args(struct netdev *, const struct shash *args); + +int +netdev_n_rxq(const struct netdev *netdev) +{ + return netdev->n_rxq; +} + +bool +netdev_is_pmd(const struct netdev *netdev) +{ + return !strcmp(netdev->netdev_class->type, "dpdk"); +} + +static void +netdev_class_mutex_initialize(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) +{ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + + if (ovsthread_once_start(&once)) { + ovs_mutex_init_recursive(&netdev_class_mutex); + ovsthread_once_done(&once); + } +} static void netdev_initialize(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - static bool inited; + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; - if (!inited) { - inited = true; + if (ovsthread_once_start(&once)) { + netdev_class_mutex_initialize(); - fatal_signal_add_hook(close_all_netdevs, NULL, NULL, true); + fatal_signal_add_hook(restore_all_flags, NULL, NULL, true); + netdev_vport_patch_register(); -#ifdef HAVE_NETLINK +#ifdef __linux__ netdev_register_provider(&netdev_linux_class); netdev_register_provider(&netdev_internal_class); netdev_register_provider(&netdev_tap_class); - netdev_vport_register(); + netdev_vport_tunnel_register(); +#endif +#if defined(__FreeBSD__) || defined(__NetBSD__) + netdev_register_provider(&netdev_tap_class); + netdev_register_provider(&netdev_bsd_class); #endif + netdev_dpdk_register(); + + ovsthread_once_done(&once); } } @@ -89,14 +147,18 @@ netdev_initialize(void) * main poll loop. */ void netdev_run(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - struct shash_node *node; - SHASH_FOR_EACH(node, &netdev_classes) { - const struct netdev_class *netdev_class = node->data; - if (netdev_class->run) { - netdev_class->run(); + struct netdev_registered_class *rc; + + netdev_initialize(); + ovs_mutex_lock(&netdev_class_mutex); + HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { + if (rc->class->run) { + rc->class->run(); } } + ovs_mutex_unlock(&netdev_class_mutex); } /* Arranges for poll_block() to wake up when netdev_run() needs to be called. @@ -105,39 +167,66 @@ netdev_run(void) * main poll loop. */ void netdev_wait(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - struct shash_node *node; - SHASH_FOR_EACH(node, &netdev_classes) { - const struct netdev_class *netdev_class = node->data; - if (netdev_class->wait) { - netdev_class->wait(); + struct netdev_registered_class *rc; + + ovs_mutex_lock(&netdev_class_mutex); + HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { + if (rc->class->wait) { + rc->class->wait(); + } + } + ovs_mutex_unlock(&netdev_class_mutex); +} + +static struct netdev_registered_class * +netdev_lookup_class(const char *type) + OVS_REQ_RDLOCK(netdev_class_mutex) +{ + struct netdev_registered_class *rc; + + HMAP_FOR_EACH_WITH_HASH (rc, hmap_node, hash_string(type, 0), + &netdev_classes) { + if (!strcmp(type, rc->class->type)) { + return rc; } } + return NULL; } /* Initializes and registers a new netdev provider. After successful * registration, new netdevs of that type can be opened using netdev_open(). */ int netdev_register_provider(const struct netdev_class *new_class) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - if (shash_find(&netdev_classes, new_class->type)) { + int error; + + netdev_class_mutex_initialize(); + ovs_mutex_lock(&netdev_class_mutex); + if (netdev_lookup_class(new_class->type)) { VLOG_WARN("attempted to register duplicate netdev provider: %s", new_class->type); - return EEXIST; - } - - if (new_class->init) { - int error = new_class->init(); - if (error) { + error = EEXIST; + } else { + error = new_class->init ? new_class->init() : 0; + if (!error) { + struct netdev_registered_class *rc; + + rc = xmalloc(sizeof *rc); + hmap_insert(&netdev_classes, &rc->hmap_node, + hash_string(new_class->type, 0)); + rc->class = new_class; + atomic_init(&rc->ref_cnt, 0); + } else { VLOG_ERR("failed to initialize %s network device class: %s", - new_class->type, strerror(error)); - return error; + new_class->type, ovs_strerror(error)); } } + ovs_mutex_unlock(&netdev_class_mutex); - shash_add(&netdev_classes, new_class->type, new_class); - - return 0; + return error; } /* Unregisters a netdev provider. 'type' must have been previously @@ -145,51 +234,93 @@ netdev_register_provider(const struct netdev_class *new_class) * new netdevs of that type cannot be opened using netdev_open(). */ int netdev_unregister_provider(const char *type) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - struct shash_node *del_node, *netdev_dev_node; + struct netdev_registered_class *rc; + int error; - del_node = shash_find(&netdev_classes, type); - if (!del_node) { + ovs_mutex_lock(&netdev_class_mutex); + rc = netdev_lookup_class(type); + if (!rc) { VLOG_WARN("attempted to unregister a netdev provider that is not " "registered: %s", type); - return EAFNOSUPPORT; - } - - SHASH_FOR_EACH(netdev_dev_node, &netdev_dev_shash) { - struct netdev_dev *netdev_dev = netdev_dev_node->data; - if (!strcmp(netdev_dev->netdev_class->type, type)) { + error = EAFNOSUPPORT; + } else { + int ref_cnt; + + atomic_read(&rc->ref_cnt, &ref_cnt); + if (!ref_cnt) { + hmap_remove(&netdev_classes, &rc->hmap_node); + free(rc); + error = 0; + } else { VLOG_WARN("attempted to unregister in use netdev provider: %s", type); - return EBUSY; + error = EBUSY; } } + ovs_mutex_unlock(&netdev_class_mutex); - shash_delete(&netdev_classes, del_node); - - return 0; -} - -const struct netdev_class * -netdev_lookup_provider(const char *type) -{ - netdev_initialize(); - return shash_find_data(&netdev_classes, type && type[0] ? type : "system"); + return error; } /* Clears 'types' and enumerates the types of all currently registered netdev * providers into it. The caller must first initialize the sset. */ void netdev_enumerate_types(struct sset *types) + OVS_EXCLUDED(netdev_mutex) { - struct shash_node *node; + struct netdev_registered_class *rc; netdev_initialize(); sset_clear(types); - SHASH_FOR_EACH(node, &netdev_classes) { - const struct netdev_class *netdev_class = node->data; - sset_add(types, netdev_class->type); + ovs_mutex_lock(&netdev_class_mutex); + HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { + sset_add(types, rc->class->type); + } + ovs_mutex_unlock(&netdev_class_mutex); +} + +/* Check that the network device name is not the same as any of the registered + * vport providers' dpif_port name (dpif_port is NULL if the vport provider + * does not define it) or the datapath internal port name (e.g. ovs-system). + * + * Returns true if there is a name conflict, false otherwise. */ +bool +netdev_is_reserved_name(const char *name) + OVS_EXCLUDED(netdev_mutex) +{ + struct netdev_registered_class *rc; + + netdev_initialize(); + + ovs_mutex_lock(&netdev_class_mutex); + HMAP_FOR_EACH (rc, hmap_node, &netdev_classes) { + const char *dpif_port = netdev_vport_class_get_dpif_port(rc->class); + if (dpif_port && !strcmp(dpif_port, name)) { + ovs_mutex_unlock(&netdev_class_mutex); + return true; + } } + ovs_mutex_unlock(&netdev_class_mutex); + + if (!strncmp(name, "ovs-", 4)) { + struct sset types; + const char *type; + + sset_init(&types); + dp_enumerate_types(&types); + SSET_FOR_EACH (type, &types) { + if (!strcmp(name+4, type)) { + sset_destroy(&types); + return true; + } + } + sset_destroy(&types); + } + + return false; } /* Opens the network device named 'name' (e.g. "eth0") of the specified 'type' @@ -201,82 +332,132 @@ netdev_enumerate_types(struct sset *types) * before they can be used. */ int netdev_open(const char *name, const char *type, struct netdev **netdevp) + OVS_EXCLUDED(netdev_mutex) { - struct netdev_dev *netdev_dev; + struct netdev *netdev; int error; - *netdevp = NULL; netdev_initialize(); - netdev_dev = shash_find_data(&netdev_dev_shash, name); - - if (!netdev_dev) { - const struct netdev_class *class; - - class = netdev_lookup_provider(type); - if (!class) { + ovs_mutex_lock(&netdev_class_mutex); + ovs_mutex_lock(&netdev_mutex); + netdev = shash_find_data(&netdev_shash, name); + if (!netdev) { + struct netdev_registered_class *rc; + + rc = netdev_lookup_class(type && type[0] ? type : "system"); + if (rc) { + netdev = rc->class->alloc(); + if (netdev) { + memset(netdev, 0, sizeof *netdev); + netdev->netdev_class = rc->class; + netdev->name = xstrdup(name); + netdev->change_seq = 1; + netdev->node = shash_add(&netdev_shash, name, netdev); + + /* By default enable one rx queue per netdev. */ + if (netdev->netdev_class->rxq_alloc) { + netdev->n_rxq = 1; + } else { + netdev->n_rxq = 0; + } + list_init(&netdev->saved_flags_list); + + error = rc->class->construct(netdev); + if (!error) { + int old_ref_cnt; + + atomic_add(&rc->ref_cnt, 1, &old_ref_cnt); + netdev_change_seq_changed(netdev); + } else { + free(netdev->name); + ovs_assert(list_is_empty(&netdev->saved_flags_list)); + shash_delete(&netdev_shash, netdev->node); + rc->class->dealloc(netdev); + } + } else { + error = ENOMEM; + } + } else { VLOG_WARN("could not create netdev %s of unknown type %s", name, type); - return EAFNOSUPPORT; + error = EAFNOSUPPORT; } - error = class->create(class, name, &netdev_dev); - if (error) { - return error; - } - assert(netdev_dev->netdev_class == class); - + } else { + error = 0; } - error = netdev_dev->netdev_class->open(netdev_dev, netdevp); + ovs_mutex_unlock(&netdev_mutex); + ovs_mutex_unlock(&netdev_class_mutex); if (!error) { - netdev_dev->ref_cnt++; + netdev->ref_cnt++; + *netdevp = netdev; } else { - if (!netdev_dev->ref_cnt) { - netdev_dev_uninit(netdev_dev, true); - } + *netdevp = NULL; } - return error; } +/* Returns a reference to 'netdev_' for the caller to own. Returns null if + * 'netdev_' is null. */ +struct netdev * +netdev_ref(const struct netdev *netdev_) + OVS_EXCLUDED(netdev_mutex) +{ + struct netdev *netdev = CONST_CAST(struct netdev *, netdev_); + + if (netdev) { + ovs_mutex_lock(&netdev_mutex); + ovs_assert(netdev->ref_cnt > 0); + netdev->ref_cnt++; + ovs_mutex_unlock(&netdev_mutex); + } + return netdev; +} + /* Reconfigures the device 'netdev' with 'args'. 'args' may be empty * or NULL if none are needed. */ int -netdev_set_config(struct netdev *netdev, const struct shash *args) +netdev_set_config(struct netdev *netdev, const struct smap *args) + OVS_EXCLUDED(netdev_mutex) { - struct netdev_dev *netdev_dev = netdev_get_dev(netdev); + if (netdev->netdev_class->set_config) { + const struct smap no_args = SMAP_INITIALIZER(&no_args); + int error; - if (netdev_dev->netdev_class->set_config) { - struct shash no_args = SHASH_INITIALIZER(&no_args); - return netdev_dev->netdev_class->set_config(netdev_dev, - args ? args : &no_args); - } else if (args && !shash_is_empty(args)) { + error = netdev->netdev_class->set_config(netdev, + args ? args : &no_args); + if (error) { + VLOG_WARN("%s: could not set configuration (%s)", + netdev_get_name(netdev), ovs_strerror(error)); + } + return error; + } else if (args && !smap_is_empty(args)) { VLOG_WARN("%s: arguments provided to device that is not configurable", netdev_get_name(netdev)); } - return 0; } /* Returns the current configuration for 'netdev' in 'args'. The caller must - * have already initialized 'args' with shash_init(). Returns 0 on success, in + * have already initialized 'args' with smap_init(). Returns 0 on success, in * which case 'args' will be filled with 'netdev''s configuration. On failure * returns a positive errno value, in which case 'args' will be empty. * * The caller owns 'args' and its contents and must eventually free them with - * shash_destroy_free_data(). */ + * smap_destroy(). */ int -netdev_get_config(const struct netdev *netdev, struct shash *args) +netdev_get_config(const struct netdev *netdev, struct smap *args) + OVS_EXCLUDED(netdev_mutex) { - struct netdev_dev *netdev_dev = netdev_get_dev(netdev); int error; - shash_clear_free_data(args); - if (netdev_dev->netdev_class->get_config) { - error = netdev_dev->netdev_class->get_config(netdev_dev, args); + smap_clear(args); + if (netdev->netdev_class->get_config) { + error = netdev->netdev_class->get_config(netdev, args); if (error) { - shash_clear_free_data(args); + smap_clear(args); } } else { error = 0; @@ -285,51 +466,53 @@ netdev_get_config(const struct netdev *netdev, struct shash *args) return error; } -/* Closes and destroys 'netdev'. */ -void -netdev_close(struct netdev *netdev) +const struct netdev_tunnel_config * +netdev_get_tunnel_config(const struct netdev *netdev) + OVS_EXCLUDED(netdev_mutex) { - if (netdev) { - struct netdev_dev *netdev_dev = netdev_get_dev(netdev); - - assert(netdev_dev->ref_cnt); - netdev_dev->ref_cnt--; - netdev_uninit(netdev, true); - - /* If the reference count for the netdev device is zero, destroy it. */ - if (!netdev_dev->ref_cnt) { - netdev_dev_uninit(netdev_dev, true); - } + if (netdev->netdev_class->get_tunnel_config) { + return netdev->netdev_class->get_tunnel_config(netdev); + } else { + return NULL; } } -/* Returns true if a network device named 'name' exists and may be opened, - * otherwise false. */ -bool -netdev_exists(const char *name) -{ - struct netdev *netdev; - int error; - - error = netdev_open(name, "system", &netdev); - if (!error) { - netdev_close(netdev); - return true; +static void +netdev_unref(struct netdev *dev) + OVS_RELEASES(netdev_mutex) +{ + ovs_assert(dev->ref_cnt); + if (!--dev->ref_cnt) { + const struct netdev_class *class = dev->netdev_class; + struct netdev_registered_class *rc; + int old_ref_cnt; + + dev->netdev_class->destruct(dev); + + shash_delete(&netdev_shash, dev->node); + free(dev->name); + dev->netdev_class->dealloc(dev); + ovs_mutex_unlock(&netdev_mutex); + + ovs_mutex_lock(&netdev_class_mutex); + rc = netdev_lookup_class(class->type); + atomic_sub(&rc->ref_cnt, 1, &old_ref_cnt); + ovs_assert(old_ref_cnt > 0); + ovs_mutex_unlock(&netdev_class_mutex); } else { - if (error != ENODEV) { - VLOG_WARN("failed to open network device %s: %s", - name, strerror(error)); - } - return false; + ovs_mutex_unlock(&netdev_mutex); } } -/* Returns true if a network device named 'name' is currently opened, - * otherwise false. */ -bool -netdev_is_open(const char *name) +/* Closes and destroys 'netdev'. */ +void +netdev_close(struct netdev *netdev) + OVS_EXCLUDED(netdev_mutex) { - return !!shash_find_data(&netdev_dev_shash, name); + if (netdev) { + ovs_mutex_lock(&netdev_mutex); + netdev_unref(netdev); + } } /* Parses 'netdev_name_', which is of the form [type@]name into its component @@ -351,82 +534,99 @@ netdev_parse_name(const char *netdev_name_, char **name, char **type) } } -/* Attempts to set up 'netdev' for receiving packets with netdev_recv(). - * Returns 0 if successful, otherwise a positive errno value. EOPNOTSUPP - * indicates that the network device does not implement packet reception - * through this interface. */ +/* Attempts to open a netdev_rxq handle for obtaining packets received on + * 'netdev'. On success, returns 0 and stores a nonnull 'netdev_rxq *' into + * '*rxp'. On failure, returns a positive errno value and stores NULL into + * '*rxp'. + * + * Some kinds of network devices might not support receiving packets. This + * function returns EOPNOTSUPP in that case.*/ int -netdev_listen(struct netdev *netdev) +netdev_rxq_open(struct netdev *netdev, struct netdev_rxq **rxp, int id) + OVS_EXCLUDED(netdev_mutex) { - int (*listen)(struct netdev *); + int error; + + if (netdev->netdev_class->rxq_alloc && id < netdev->n_rxq) { + struct netdev_rxq *rx = netdev->netdev_class->rxq_alloc(); + if (rx) { + rx->netdev = netdev; + rx->queue_id = id; + error = netdev->netdev_class->rxq_construct(rx); + if (!error) { + ovs_mutex_lock(&netdev_mutex); + netdev->ref_cnt++; + ovs_mutex_unlock(&netdev_mutex); + + *rxp = rx; + return 0; + } + netdev->netdev_class->rxq_dealloc(rx); + } else { + error = ENOMEM; + } + } else { + error = EOPNOTSUPP; + } - listen = netdev_get_dev(netdev)->netdev_class->listen; - return listen ? (listen)(netdev) : EOPNOTSUPP; + *rxp = NULL; + return error; +} + +/* Closes 'rx'. */ +void +netdev_rxq_close(struct netdev_rxq *rx) + OVS_EXCLUDED(netdev_mutex) +{ + if (rx) { + struct netdev *netdev = rx->netdev; + netdev->netdev_class->rxq_destruct(rx); + netdev->netdev_class->rxq_dealloc(rx); + netdev_close(netdev); + } } -/* Attempts to receive a packet from 'netdev' into 'buffer', which the caller - * must have initialized with sufficient room for the packet. The space - * required to receive any packet is ETH_HEADER_LEN bytes, plus VLAN_HEADER_LEN - * bytes, plus the device's MTU (which may be retrieved via netdev_get_mtu()). - * (Some devices do not allow for a VLAN header, in which case VLAN_HEADER_LEN - * need not be included.) +/* Attempts to receive batch of packets from 'rx'. * - * This function can only be expected to return a packet if ->listen() has - * been called successfully. + * Returns EAGAIN immediately if no packet is ready to be received. * - * If a packet is successfully retrieved, returns 0. In this case 'buffer' is - * guaranteed to contain at least ETH_TOTAL_MIN bytes. Otherwise, returns a - * positive errno value. Returns EAGAIN immediately if no packet is ready to - * be returned. + * Returns EMSGSIZE, and discards the packet, if the received packet is longer + * than 'ofpbuf_tailroom(buffer)'. * - * Some network devices may not implement support for this function. In such - * cases this function will always return EOPNOTSUPP. */ + * It is advised that the tailroom of 'buffer' should be + * VLAN_HEADER_LEN bytes longer than the MTU to allow space for an + * out-of-band VLAN header to be added to the packet. At the very least, + * 'buffer' must have at least ETH_TOTAL_MIN bytes of tailroom. + * + * This function may be set to null if it would always return EOPNOTSUPP + * anyhow. */ int -netdev_recv(struct netdev *netdev, struct ofpbuf *buffer) +netdev_rxq_recv(struct netdev_rxq *rx, struct ofpbuf **buffers, int *cnt) { - int (*recv)(struct netdev *, void *, size_t); int retval; - assert(buffer->size == 0); - assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN); - - recv = netdev_get_dev(netdev)->netdev_class->recv; - retval = (recv - ? (recv)(netdev, buffer->data, ofpbuf_tailroom(buffer)) - : -EOPNOTSUPP); - if (retval >= 0) { + retval = rx->netdev->netdev_class->rxq_recv(rx, buffers, cnt); + if (!retval) { COVERAGE_INC(netdev_received); - buffer->size += retval; - if (buffer->size < ETH_TOTAL_MIN) { - ofpbuf_put_zeros(buffer, ETH_TOTAL_MIN - buffer->size); - } - return 0; - } else { - return -retval; } + return retval; } -/* Registers with the poll loop to wake up from the next call to poll_block() - * when a packet is ready to be received with netdev_recv() on 'netdev'. */ +/* Arranges for poll_block() to wake up when a packet is ready to be received + * on 'rx'. */ void -netdev_recv_wait(struct netdev *netdev) +netdev_rxq_wait(struct netdev_rxq *rx) { - void (*recv_wait)(struct netdev *); - - recv_wait = netdev_get_dev(netdev)->netdev_class->recv_wait; - if (recv_wait) { - recv_wait(netdev); - } + rx->netdev->netdev_class->rxq_wait(rx); } -/* Discards all packets waiting to be received from 'netdev'. */ +/* Discards any packets ready to be received on 'rx'. */ int -netdev_drain(struct netdev *netdev) +netdev_rxq_drain(struct netdev_rxq *rx) { - int (*drain)(struct netdev *); - - drain = netdev_get_dev(netdev)->netdev_class->drain; - return drain ? drain(netdev) : 0; + return (rx->netdev->netdev_class->rxq_drain + ? rx->netdev->netdev_class->rxq_drain(rx) + : 0); } /* Sends 'buffer' on 'netdev'. Returns 0 if successful, otherwise a positive @@ -434,7 +634,7 @@ netdev_drain(struct netdev *netdev) * immediately. Returns EMSGSIZE if a partial packet was transmitted or if * the packet is too big or too small to transmit on the device. * - * The caller retains ownership of 'buffer' in all cases. + * To retain ownership of 'buffer' caller can set may_steal to false. * * The kernel maintains a packet transmission queue, so the caller is not * expected to do additional queuing of packets. @@ -442,13 +642,13 @@ netdev_drain(struct netdev *netdev) * Some network devices may not implement support for this function. In such * cases this function will always return EOPNOTSUPP. */ int -netdev_send(struct netdev *netdev, const struct ofpbuf *buffer) +netdev_send(struct netdev *netdev, struct ofpbuf *buffer, bool may_steal) { - int (*send)(struct netdev *, const void *, size_t); int error; - send = netdev_get_dev(netdev)->netdev_class->send; - error = send ? (send)(netdev, buffer->data, buffer->size) : EOPNOTSUPP; + error = (netdev->netdev_class->send + ? netdev->netdev_class->send(netdev, buffer, may_steal) + : EOPNOTSUPP); if (!error) { COVERAGE_INC(netdev_sent); } @@ -465,11 +665,8 @@ netdev_send(struct netdev *netdev, const struct ofpbuf *buffer) void netdev_send_wait(struct netdev *netdev) { - void (*send_wait)(struct netdev *); - - send_wait = netdev_get_dev(netdev)->netdev_class->send_wait; - if (send_wait) { - send_wait(netdev); + if (netdev->netdev_class->send_wait) { + netdev->netdev_class->send_wait(netdev); } } @@ -478,7 +675,7 @@ netdev_send_wait(struct netdev *netdev) int netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN]) { - return netdev_get_dev(netdev)->netdev_class->set_etheraddr(netdev, mac); + return netdev->netdev_class->set_etheraddr(netdev, mac); } /* Retrieves 'netdev''s MAC address. If successful, returns 0 and copies the @@ -487,7 +684,7 @@ netdev_set_etheraddr(struct netdev *netdev, const uint8_t mac[ETH_ADDR_LEN]) int netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN]) { - return netdev_get_dev(netdev)->netdev_class->get_etheraddr(netdev, mac); + return netdev->netdev_class->get_etheraddr(netdev, mac); } /* Returns the name of the network device that 'netdev' represents, @@ -495,7 +692,7 @@ netdev_get_etheraddr(const struct netdev *netdev, uint8_t mac[ETH_ADDR_LEN]) const char * netdev_get_name(const struct netdev *netdev) { - return netdev_get_dev(netdev)->name; + return netdev->name; } /* Retrieves the MTU of 'netdev'. The MTU is the maximum size of transmitted @@ -509,15 +706,15 @@ netdev_get_name(const struct netdev *netdev) int netdev_get_mtu(const struct netdev *netdev, int *mtup) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; int error; error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP; if (error) { *mtup = 0; if (error != EOPNOTSUPP) { - VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: " - "%s", netdev_get_name(netdev), strerror(error)); + VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: " + "%s", netdev_get_name(netdev), ovs_strerror(error)); } } return error; @@ -532,13 +729,13 @@ netdev_get_mtu(const struct netdev *netdev, int *mtup) int netdev_set_mtu(const struct netdev *netdev, int mtu) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; int error; error = class->set_mtu ? class->set_mtu(netdev, mtu) : EOPNOTSUPP; if (error && error != EOPNOTSUPP) { - VLOG_WARN_RL(&rl, "failed to set MTU for network device %s: %s", - netdev_get_name(netdev), strerror(error)); + VLOG_DBG_RL(&rl, "failed to set MTU for network device %s: %s", + netdev_get_name(netdev), ovs_strerror(error)); } return error; @@ -561,7 +758,7 @@ netdev_get_ifindex(const struct netdev *netdev) { int (*get_ifindex)(const struct netdev *); - get_ifindex = netdev_get_dev(netdev)->netdev_class->get_ifindex; + get_ifindex = netdev->netdev_class->get_ifindex; return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP; } @@ -602,9 +799,10 @@ netdev_get_features(const struct netdev *netdev, peer = &dummy[3]; } - get_features = netdev_get_dev(netdev)->netdev_class->get_features; + get_features = netdev->netdev_class->get_features; error = get_features - ? get_features(netdev, current, advertised, supported, peer) + ? get_features(netdev, current, advertised, supported, + peer) : EOPNOTSUPP; if (error) { *current = *advertised = *supported = *peer = 0; @@ -614,9 +812,10 @@ netdev_get_features(const struct netdev *netdev, /* Returns the maximum speed of a network connection that has the NETDEV_F_* * bits in 'features', in bits per second. If no bits that indicate a speed - * are set in 'features', assumes 100Mbps. */ + * are set in 'features', returns 'default_bps'. */ uint64_t -netdev_features_to_bps(enum netdev_features features) +netdev_features_to_bps(enum netdev_features features, + uint64_t default_bps) { enum { F_1000000MB = NETDEV_F_1TB_FD, @@ -635,7 +834,7 @@ netdev_features_to_bps(enum netdev_features features) : features & F_1000MB ? UINT64_C(1000000000) : features & F_100MB ? UINT64_C(100000000) : features & F_10MB ? UINT64_C(10000000) - : UINT64_C(100000000)); + : default_bps); } /* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link @@ -654,8 +853,8 @@ int netdev_set_advertisements(struct netdev *netdev, enum netdev_features advertise) { - return (netdev_get_dev(netdev)->netdev_class->set_advertisements - ? netdev_get_dev(netdev)->netdev_class->set_advertisements( + return (netdev->netdev_class->set_advertisements + ? netdev->netdev_class->set_advertisements( netdev, advertise) : EOPNOTSUPP); } @@ -680,8 +879,8 @@ netdev_get_in4(const struct netdev *netdev, struct in_addr netmask; int error; - error = (netdev_get_dev(netdev)->netdev_class->get_in4 - ? netdev_get_dev(netdev)->netdev_class->get_in4(netdev, + error = (netdev->netdev_class->get_in4 + ? netdev->netdev_class->get_in4(netdev, &address, &netmask) : EOPNOTSUPP); if (address_) { @@ -699,8 +898,8 @@ netdev_get_in4(const struct netdev *netdev, int netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask) { - return (netdev_get_dev(netdev)->netdev_class->set_in4 - ? netdev_get_dev(netdev)->netdev_class->set_in4(netdev, addr, mask) + return (netdev->netdev_class->set_in4 + ? netdev->netdev_class->set_in4(netdev, addr, mask) : EOPNOTSUPP); } @@ -730,8 +929,8 @@ int netdev_add_router(struct netdev *netdev, struct in_addr router) { COVERAGE_INC(netdev_add_router); - return (netdev_get_dev(netdev)->netdev_class->add_router - ? netdev_get_dev(netdev)->netdev_class->add_router(netdev, router) + return (netdev->netdev_class->add_router + ? netdev->netdev_class->add_router(netdev, router) : EOPNOTSUPP); } @@ -747,8 +946,8 @@ netdev_get_next_hop(const struct netdev *netdev, const struct in_addr *host, struct in_addr *next_hop, char **netdev_name) { - int error = (netdev_get_dev(netdev)->netdev_class->get_next_hop - ? netdev_get_dev(netdev)->netdev_class->get_next_hop( + int error = (netdev->netdev_class->get_next_hop + ? netdev->netdev_class->get_next_hop( host, next_hop, netdev_name) : EOPNOTSUPP); if (error) { @@ -758,18 +957,16 @@ netdev_get_next_hop(const struct netdev *netdev, return error; } -/* Populates 'sh' with status information. +/* Populates 'smap' with status information. * - * Populates 'sh' with 'netdev' specific status information. This information - * may be used to populate the status column of the Interface table as defined - * in ovs-vswitchd.conf.db(5). */ + * Populates 'smap' with 'netdev' specific status information. This + * information may be used to populate the status column of the Interface table + * as defined in ovs-vswitchd.conf.db(5). */ int -netdev_get_status(const struct netdev *netdev, struct shash *sh) +netdev_get_status(const struct netdev *netdev, struct smap *smap) { - struct netdev_dev *dev = netdev_get_dev(netdev); - - return (dev->netdev_class->get_status - ? dev->netdev_class->get_status(netdev, sh) + return (netdev->netdev_class->get_status + ? netdev->netdev_class->get_status(netdev, smap) : EOPNOTSUPP); } @@ -790,8 +987,8 @@ netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6) struct in6_addr dummy; int error; - error = (netdev_get_dev(netdev)->netdev_class->get_in6 - ? netdev_get_dev(netdev)->netdev_class->get_in6(netdev, + error = (netdev->netdev_class->get_in6 + ? netdev->netdev_class->get_in6(netdev, in6 ? in6 : &dummy) : EOPNOTSUPP); if (error && in6) { @@ -801,37 +998,47 @@ netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6) } /* On 'netdev', turns off the flags in 'off' and then turns on the flags in - * 'on'. If 'permanent' is true, the changes will persist; otherwise, they - * will be reverted when 'netdev' is closed or the program exits. Returns 0 if - * successful, otherwise a positive errno value. */ + * 'on'. Returns 0 if successful, otherwise a positive errno value. */ static int do_update_flags(struct netdev *netdev, enum netdev_flags off, enum netdev_flags on, enum netdev_flags *old_flagsp, - bool permanent) + struct netdev_saved_flags **sfp) + OVS_EXCLUDED(netdev_mutex) { + struct netdev_saved_flags *sf = NULL; enum netdev_flags old_flags; int error; - error = netdev_get_dev(netdev)->netdev_class->update_flags(netdev, - off & ~on, on, &old_flags); + error = netdev->netdev_class->update_flags(netdev, off & ~on, on, + &old_flags); if (error) { VLOG_WARN_RL(&rl, "failed to %s flags for network device %s: %s", off || on ? "set" : "get", netdev_get_name(netdev), - strerror(error)); + ovs_strerror(error)); old_flags = 0; - } else if ((off || on) && !permanent) { + } else if ((off || on) && sfp) { enum netdev_flags new_flags = (old_flags & ~off) | on; enum netdev_flags changed_flags = old_flags ^ new_flags; if (changed_flags) { - if (!netdev->changed_flags) { - netdev->save_flags = old_flags; - } - netdev->changed_flags |= changed_flags; + ovs_mutex_lock(&netdev_mutex); + *sfp = sf = xmalloc(sizeof *sf); + sf->netdev = netdev; + list_push_front(&netdev->saved_flags_list, &sf->node); + sf->saved_flags = changed_flags; + sf->saved_values = changed_flags & new_flags; + + netdev->ref_cnt++; + ovs_mutex_unlock(&netdev_mutex); } } + if (old_flagsp) { *old_flagsp = old_flags; } + if (sfp) { + *sfp = sf; + } + return error; } @@ -841,41 +1048,66 @@ do_update_flags(struct netdev *netdev, enum netdev_flags off, int netdev_get_flags(const struct netdev *netdev_, enum netdev_flags *flagsp) { - struct netdev *netdev = (struct netdev *) netdev_; - return do_update_flags(netdev, 0, 0, flagsp, false); + struct netdev *netdev = CONST_CAST(struct netdev *, netdev_); + return do_update_flags(netdev, 0, 0, flagsp, NULL); } /* Sets the flags for 'netdev' to 'flags'. - * If 'permanent' is true, the changes will persist; otherwise, they - * will be reverted when 'netdev' is closed or the program exits. * Returns 0 if successful, otherwise a positive errno value. */ int netdev_set_flags(struct netdev *netdev, enum netdev_flags flags, - bool permanent) + struct netdev_saved_flags **sfp) { - return do_update_flags(netdev, -1, flags, NULL, permanent); + return do_update_flags(netdev, -1, flags, NULL, sfp); } -/* Turns on the specified 'flags' on 'netdev'. - * If 'permanent' is true, the changes will persist; otherwise, they - * will be reverted when 'netdev' is closed or the program exits. - * Returns 0 if successful, otherwise a positive errno value. */ +/* Turns on the specified 'flags' on 'netdev': + * + * - On success, returns 0. If 'sfp' is nonnull, sets '*sfp' to a newly + * allocated 'struct netdev_saved_flags *' that may be passed to + * netdev_restore_flags() to restore the original values of 'flags' on + * 'netdev' (this will happen automatically at program termination if + * netdev_restore_flags() is never called) , or to NULL if no flags were + * actually changed. + * + * - On failure, returns a positive errno value. If 'sfp' is nonnull, sets + * '*sfp' to NULL. */ int netdev_turn_flags_on(struct netdev *netdev, enum netdev_flags flags, - bool permanent) + struct netdev_saved_flags **sfp) { - return do_update_flags(netdev, 0, flags, NULL, permanent); + return do_update_flags(netdev, 0, flags, NULL, sfp); } -/* Turns off the specified 'flags' on 'netdev'. - * If 'permanent' is true, the changes will persist; otherwise, they - * will be reverted when 'netdev' is closed or the program exits. - * Returns 0 if successful, otherwise a positive errno value. */ +/* Turns off the specified 'flags' on 'netdev'. See netdev_turn_flags_on() for + * details of the interface. */ int netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags, - bool permanent) + struct netdev_saved_flags **sfp) +{ + return do_update_flags(netdev, flags, 0, NULL, sfp); +} + +/* Restores the flags that were saved in 'sf', and destroys 'sf'. + * Does nothing if 'sf' is NULL. */ +void +netdev_restore_flags(struct netdev_saved_flags *sf) + OVS_EXCLUDED(netdev_mutex) { - return do_update_flags(netdev, flags, 0, NULL, permanent); + if (sf) { + struct netdev *netdev = sf->netdev; + enum netdev_flags old_flags; + + netdev->netdev_class->update_flags(netdev, + sf->saved_flags & sf->saved_values, + sf->saved_flags & ~sf->saved_values, + &old_flags); + + ovs_mutex_lock(&netdev_mutex); + list_remove(&sf->node); + free(sf); + netdev_unref(netdev); + } } /* Looks up the ARP table entry for 'ip' on 'netdev'. If one exists and can be @@ -886,9 +1118,8 @@ int netdev_arp_lookup(const struct netdev *netdev, ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN]) { - int error = (netdev_get_dev(netdev)->netdev_class->arp_lookup - ? netdev_get_dev(netdev)->netdev_class->arp_lookup(netdev, - ip, mac) + int error = (netdev->netdev_class->arp_lookup + ? netdev->netdev_class->arp_lookup(netdev, ip, mac) : EOPNOTSUPP); if (error) { memset(mac, 0, ETH_ADDR_LEN); @@ -909,15 +1140,14 @@ netdev_get_carrier(const struct netdev *netdev) return false; } - if (!netdev_get_dev(netdev)->netdev_class->get_carrier) { + if (!netdev->netdev_class->get_carrier) { return true; } - error = netdev_get_dev(netdev)->netdev_class->get_carrier(netdev, - &carrier); + error = netdev->netdev_class->get_carrier(netdev, &carrier); if (error) { VLOG_DBG("%s: failed to get network device carrier status, assuming " - "down: %s", netdev_get_name(netdev), strerror(error)); + "down: %s", netdev_get_name(netdev), ovs_strerror(error)); carrier = false; } @@ -928,8 +1158,8 @@ netdev_get_carrier(const struct netdev *netdev) long long int netdev_get_carrier_resets(const struct netdev *netdev) { - return (netdev_get_dev(netdev)->netdev_class->get_carrier_resets - ? netdev_get_dev(netdev)->netdev_class->get_carrier_resets(netdev) + return (netdev->netdev_class->get_carrier_resets + ? netdev->netdev_class->get_carrier_resets(netdev) : 0); } @@ -944,9 +1174,8 @@ netdev_get_carrier_resets(const struct netdev *netdev) int netdev_set_miimon_interval(struct netdev *netdev, long long int interval) { - struct netdev_dev *netdev_dev = netdev_get_dev(netdev); - return (netdev_dev->netdev_class->set_miimon_interval - ? netdev_dev->netdev_class->set_miimon_interval(netdev, interval) + return (netdev->netdev_class->set_miimon_interval + ? netdev->netdev_class->set_miimon_interval(netdev, interval) : EOPNOTSUPP); } @@ -957,8 +1186,8 @@ netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats) int error; COVERAGE_INC(netdev_get_stats); - error = (netdev_get_dev(netdev)->netdev_class->get_stats - ? netdev_get_dev(netdev)->netdev_class->get_stats(netdev, stats) + error = (netdev->netdev_class->get_stats + ? netdev->netdev_class->get_stats(netdev, stats) : EOPNOTSUPP); if (error) { memset(stats, 0xff, sizeof *stats); @@ -974,8 +1203,8 @@ netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats) int netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats) { - return (netdev_get_dev(netdev)->netdev_class->set_stats - ? netdev_get_dev(netdev)->netdev_class->set_stats(netdev, stats) + return (netdev->netdev_class->set_stats + ? netdev->netdev_class->set_stats(netdev, stats) : EOPNOTSUPP); } @@ -986,8 +1215,8 @@ int netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate, uint32_t kbits_burst) { - return (netdev_get_dev(netdev)->netdev_class->set_policing - ? netdev_get_dev(netdev)->netdev_class->set_policing(netdev, + return (netdev->netdev_class->set_policing + ? netdev->netdev_class->set_policing(netdev, kbits_rate, kbits_burst) : EOPNOTSUPP); } @@ -1008,7 +1237,7 @@ netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate, int netdev_get_qos_types(const struct netdev *netdev, struct sset *types) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; return (class->get_qos_types ? class->get_qos_types(netdev, types) : 0); @@ -1023,7 +1252,7 @@ int netdev_get_qos_capabilities(const struct netdev *netdev, const char *type, struct netdev_qos_capabilities *caps) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; if (*type) { int retval = (class->get_qos_capabilities @@ -1064,10 +1293,9 @@ netdev_get_n_queues(const struct netdev *netdev, * * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'. * - * The caller must initialize 'details' as an empty shash (e.g. with - * shash_init()) before calling this function. The caller must free 'details', - * including 'data' members, when it is no longer needed (e.g. with - * shash_destroy_free_data()). + * The caller must initialize 'details' as an empty smap (e.g. with + * smap_init()) before calling this function. The caller must free 'details' + * when it is no longer needed (e.g. with smap_destroy()). * * The caller must not modify or free '*typep'. * @@ -1077,16 +1305,16 @@ netdev_get_n_queues(const struct netdev *netdev, * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */ int netdev_get_qos(const struct netdev *netdev, - const char **typep, struct shash *details) + const char **typep, struct smap *details) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; int retval; if (class->get_qos) { retval = class->get_qos(netdev, typep, details); if (retval) { *typep = NULL; - shash_clear_free_data(details); + smap_clear(details); } return retval; } else { @@ -1116,9 +1344,9 @@ netdev_get_qos(const struct netdev *netdev, * details. */ int netdev_set_qos(struct netdev *netdev, - const char *type, const struct shash *details) + const char *type, const struct smap *details) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; if (!type) { type = ""; @@ -1126,7 +1354,7 @@ netdev_set_qos(struct netdev *netdev, if (class->set_qos) { if (!details) { - static struct shash empty = SHASH_INITIALIZER(&empty); + static const struct smap empty = SMAP_INITIALIZER(&empty); details = ∅ } return class->set_qos(netdev, type, details); @@ -1146,21 +1374,21 @@ netdev_set_qos(struct netdev *netdev, * given 'type' in the "other_config" column in the "Queue" table in * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). * - * The caller must initialize 'details' (e.g. with shash_init()) before calling - * this function. The caller must free 'details', including 'data' members, - * when it is no longer needed (e.g. with shash_destroy_free_data()). */ + * The caller must initialize 'details' (e.g. with smap_init()) before calling + * this function. The caller must free 'details' when it is no longer needed + * (e.g. with smap_destroy()). */ int netdev_get_queue(const struct netdev *netdev, - unsigned int queue_id, struct shash *details) + unsigned int queue_id, struct smap *details) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; int retval; retval = (class->get_queue ? class->get_queue(netdev, queue_id, details) : EOPNOTSUPP); if (retval) { - shash_clear_free_data(details); + smap_clear(details); } return retval; } @@ -1179,9 +1407,9 @@ netdev_get_queue(const struct netdev *netdev, * it. */ int netdev_set_queue(struct netdev *netdev, - unsigned int queue_id, const struct shash *details) + unsigned int queue_id, const struct smap *details) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; return (class->set_queue ? class->set_queue(netdev, queue_id, details) : EOPNOTSUPP); @@ -1200,7 +1428,7 @@ netdev_set_queue(struct netdev *netdev, int netdev_delete_queue(struct netdev *netdev, unsigned int queue_id) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; return (class->delete_queue ? class->delete_queue(netdev, queue_id) : EOPNOTSUPP); @@ -1209,43 +1437,99 @@ netdev_delete_queue(struct netdev *netdev, unsigned int queue_id) /* Obtains statistics about 'queue_id' on 'netdev'. On success, returns 0 and * fills 'stats' with the queue's statistics; individual members of 'stats' may * be set to all-1-bits if the statistic is unavailable. On failure, returns a - * positive errno value and fills 'stats' with all-1-bits. */ + * positive errno value and fills 'stats' with values indicating unsupported + * statistics. */ int netdev_get_queue_stats(const struct netdev *netdev, unsigned int queue_id, struct netdev_queue_stats *stats) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; int retval; retval = (class->get_queue_stats ? class->get_queue_stats(netdev, queue_id, stats) : EOPNOTSUPP); if (retval) { - memset(stats, 0xff, sizeof *stats); + stats->tx_bytes = UINT64_MAX; + stats->tx_packets = UINT64_MAX; + stats->tx_errors = UINT64_MAX; + stats->created = LLONG_MIN; } return retval; } -/* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID, - * its configuration, and the 'aux' specified by the caller. The order of - * iteration is unspecified, but (when successful) each queue is visited - * exactly once. +/* Initializes 'dump' to begin dumping the queues in a netdev. + * + * This function provides no status indication. An error status for the entire + * dump operation is provided when it is completed by calling + * netdev_queue_dump_done(). + */ +void +netdev_queue_dump_start(struct netdev_queue_dump *dump, + const struct netdev *netdev) +{ + dump->netdev = netdev_ref(netdev); + if (netdev->netdev_class->queue_dump_start) { + dump->error = netdev->netdev_class->queue_dump_start(netdev, + &dump->state); + } else { + dump->error = EOPNOTSUPP; + } +} + +/* Attempts to retrieve another queue from 'dump', which must have been + * initialized with netdev_queue_dump_start(). On success, stores a new queue + * ID into '*queue_id', fills 'details' with configuration details for the + * queue, and returns true. On failure, returns false. * - * Calling this function may be more efficient than calling netdev_get_queue() - * for every queue. + * Queues are not necessarily dumped in increasing order of queue ID (or any + * other predictable order). * - * 'cb' must not modify or free the 'details' argument passed in. + * Failure might indicate an actual error or merely that the last queue has + * been dumped. An error status for the entire dump operation is provided when + * it is completed by calling netdev_queue_dump_done(). * - * Returns 0 if successful, otherwise a positive errno value. On error, some - * configured queues may not have been included in the iteration. */ + * The returned contents of 'details' should be documented as valid for the + * given 'type' in the "other_config" column in the "Queue" table in + * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). + * + * The caller must initialize 'details' (e.g. with smap_init()) before calling + * this function. This function will clear and replace its contents. The + * caller must free 'details' when it is no longer needed (e.g. with + * smap_destroy()). */ +bool +netdev_queue_dump_next(struct netdev_queue_dump *dump, + unsigned int *queue_id, struct smap *details) +{ + const struct netdev *netdev = dump->netdev; + + if (dump->error) { + return false; + } + + dump->error = netdev->netdev_class->queue_dump_next(netdev, dump->state, + queue_id, details); + + if (dump->error) { + netdev->netdev_class->queue_dump_done(netdev, dump->state); + return false; + } + return true; +} + +/* Completes queue table dump operation 'dump', which must have been + * initialized with netdev_queue_dump_start(). Returns 0 if the dump operation + * was error-free, otherwise a positive errno value describing the problem. */ int -netdev_dump_queues(const struct netdev *netdev, - netdev_dump_queues_cb *cb, void *aux) +netdev_queue_dump_done(struct netdev_queue_dump *dump) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; - return (class->dump_queues - ? class->dump_queues(netdev, cb, aux) - : EOPNOTSUPP); + const struct netdev *netdev = dump->netdev; + if (!dump->error && netdev->netdev_class->queue_dump_done) { + dump->error = netdev->netdev_class->queue_dump_done(netdev, + dump->state); + } + netdev_close(dump->netdev); + return dump->error == EOF ? 0 : dump->error; } /* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID, @@ -1264,193 +1548,158 @@ int netdev_dump_queue_stats(const struct netdev *netdev, netdev_dump_queue_stats_cb *cb, void *aux) { - const struct netdev_class *class = netdev_get_dev(netdev)->netdev_class; + const struct netdev_class *class = netdev->netdev_class; return (class->dump_queue_stats ? class->dump_queue_stats(netdev, cb, aux) : EOPNOTSUPP); } -/* Returns a sequence number which indicates changes in one of 'netdev''s - * properties. The returned sequence will be nonzero so that callers have a - * value which they may use as a reset when tracking 'netdev'. - * - * The returned sequence number will change whenever 'netdev''s flags, - * features, ethernet address, or carrier changes. It may change for other - * reasons as well, or no reason at all. */ -unsigned int -netdev_change_seq(const struct netdev *netdev) -{ - return netdev_get_dev(netdev)->netdev_class->change_seq(netdev); -} -/* Initializes 'netdev_dev' as a netdev device named 'name' of the specified - * 'netdev_class'. This function is ordinarily called from a netdev provider's - * 'create' function. - * - * This function adds 'netdev_dev' to a netdev-owned shash, so it is - * very important that 'netdev_dev' only be freed after calling - * the refcount drops to zero. */ -void -netdev_dev_init(struct netdev_dev *netdev_dev, const char *name, - const struct netdev_class *netdev_class) -{ - assert(!shash_find(&netdev_dev_shash, name)); - - memset(netdev_dev, 0, sizeof *netdev_dev); - netdev_dev->netdev_class = netdev_class; - netdev_dev->name = xstrdup(name); - netdev_dev->node = shash_add(&netdev_dev_shash, name, netdev_dev); -} - -/* Undoes the results of initialization. - * - * Normally this function does not need to be called as netdev_close has - * the same effect when the refcount drops to zero. - * However, it may be called by providers due to an error on creation - * that occurs after initialization. It this case netdev_close() would - * never be called. */ -void -netdev_dev_uninit(struct netdev_dev *netdev_dev, bool destroy) -{ - char *name = netdev_dev->name; - - assert(!netdev_dev->ref_cnt); - - shash_delete(&netdev_dev_shash, netdev_dev->node); - - if (destroy) { - netdev_dev->netdev_class->destroy(netdev_dev); - } - free(name); -} - -/* Returns the class type of 'netdev_dev'. +/* Returns the class type of 'netdev'. * * The caller must not free the returned value. */ const char * -netdev_dev_get_type(const struct netdev_dev *netdev_dev) +netdev_get_type(const struct netdev *netdev) { - return netdev_dev->netdev_class->type; + return netdev->netdev_class->type; } -/* Returns the class associated with 'netdev_dev'. */ +/* Returns the class associated with 'netdev'. */ const struct netdev_class * -netdev_dev_get_class(const struct netdev_dev *netdev_dev) +netdev_get_class(const struct netdev *netdev) { - return netdev_dev->netdev_class; + return netdev->netdev_class; } -/* Returns the name of 'netdev_dev'. +/* Returns the netdev with 'name' or NULL if there is none. * - * The caller must not free the returned value. */ -const char * -netdev_dev_get_name(const struct netdev_dev *netdev_dev) + * The caller must free the returned netdev with netdev_close(). */ +struct netdev * +netdev_from_name(const char *name) + OVS_EXCLUDED(netdev_mutex) { - return netdev_dev->name; -} + struct netdev *netdev; -/* Returns the netdev_dev with 'name' or NULL if there is none. - * - * The caller must not free the returned value. */ -struct netdev_dev * -netdev_dev_from_name(const char *name) -{ - return shash_find_data(&netdev_dev_shash, name); + ovs_mutex_lock(&netdev_mutex); + netdev = shash_find_data(&netdev_shash, name); + if (netdev) { + netdev->ref_cnt++; + } + ovs_mutex_unlock(&netdev_mutex); + + return netdev; } /* Fills 'device_list' with devices that match 'netdev_class'. * - * The caller is responsible for initializing and destroying 'device_list' - * but the contained netdev_devs must not be freed. */ + * The caller is responsible for initializing and destroying 'device_list' and + * must close each device on the list. */ void -netdev_dev_get_devices(const struct netdev_class *netdev_class, - struct shash *device_list) +netdev_get_devices(const struct netdev_class *netdev_class, + struct shash *device_list) + OVS_EXCLUDED(netdev_mutex) { struct shash_node *node; - SHASH_FOR_EACH (node, &netdev_dev_shash) { - struct netdev_dev *dev = node->data; + + ovs_mutex_lock(&netdev_mutex); + SHASH_FOR_EACH (node, &netdev_shash) { + struct netdev *dev = node->data; if (dev->netdev_class == netdev_class) { + dev->ref_cnt++; shash_add(device_list, node->name, node->data); } } + ovs_mutex_unlock(&netdev_mutex); } -/* Initializes 'netdev' as a instance of the netdev_dev. +/* Extracts pointers to all 'netdev-vports' into an array 'vports' + * and returns it. Stores the size of the array into '*size'. * - * This function adds 'netdev' to a netdev-owned linked list, so it is very - * important that 'netdev' only be freed after calling netdev_close(). */ -void -netdev_init(struct netdev *netdev, struct netdev_dev *netdev_dev) + * The caller is responsible for freeing 'vports' and must close + * each 'netdev-vport' in the list. */ +struct netdev ** +netdev_get_vports(size_t *size) + OVS_EXCLUDED(netdev_mutex) { - memset(netdev, 0, sizeof *netdev); - netdev->netdev_dev = netdev_dev; - list_push_back(&netdev_list, &netdev->node); -} + struct netdev **vports; + struct shash_node *node; + size_t n = 0; -/* Undoes the results of initialization. - * - * Normally this function only needs to be called from netdev_close(). - * However, it may be called by providers due to an error on opening - * that occurs after initialization. It this case netdev_close() would - * never be called. */ -void -netdev_uninit(struct netdev *netdev, bool close) -{ - /* Restore flags that we changed, if any. */ - int error = restore_flags(netdev); - list_remove(&netdev->node); - if (error) { - VLOG_WARN("failed to restore network device flags on %s: %s", - netdev_get_name(netdev), strerror(error)); + if (!size) { + return NULL; } - if (close) { - netdev_get_dev(netdev)->netdev_class->close(netdev); + /* Explicitly allocates big enough chunk of memory. */ + vports = xmalloc(shash_count(&netdev_shash) * sizeof *vports); + ovs_mutex_lock(&netdev_mutex); + SHASH_FOR_EACH (node, &netdev_shash) { + struct netdev *dev = node->data; + + if (netdev_vport_is_vport_class(dev->netdev_class)) { + dev->ref_cnt++; + vports[n] = dev; + n++; + } } -} + ovs_mutex_unlock(&netdev_mutex); + *size = n; + return vports; +} -/* Returns the class type of 'netdev'. - * - * The caller must not free the returned value. */ const char * -netdev_get_type(const struct netdev *netdev) +netdev_get_type_from_name(const char *name) { - return netdev_get_dev(netdev)->netdev_class->type; + struct netdev *dev = netdev_from_name(name); + const char *type = dev ? netdev_get_type(dev) : NULL; + netdev_close(dev); + return type; } - -struct netdev_dev * -netdev_get_dev(const struct netdev *netdev) + +struct netdev * +netdev_rxq_get_netdev(const struct netdev_rxq *rx) { - return netdev->netdev_dev; + ovs_assert(rx->netdev->ref_cnt > 0); + return rx->netdev; } - -/* Restore the network device flags on 'netdev' to those that were active - * before we changed them. Returns 0 if successful, otherwise a positive - * errno value. - * - * To avoid reentry, the caller must ensure that fatal signals are blocked. */ -static int -restore_flags(struct netdev *netdev) + +const char * +netdev_rxq_get_name(const struct netdev_rxq *rx) { - if (netdev->changed_flags) { - enum netdev_flags restore = netdev->save_flags & netdev->changed_flags; - enum netdev_flags old_flags; - return netdev_get_dev(netdev)->netdev_class->update_flags(netdev, - netdev->changed_flags & ~restore, - restore, &old_flags); - } - return 0; + return netdev_get_name(netdev_rxq_get_netdev(rx)); } -/* Close all netdevs on shutdown so they can do any needed cleanup such as - * destroying devices, restoring flags, etc. */ static void -close_all_netdevs(void *aux OVS_UNUSED) +restore_all_flags(void *aux OVS_UNUSED) { - struct netdev *netdev, *next; - LIST_FOR_EACH_SAFE(netdev, next, node, &netdev_list) { - netdev_close(netdev); + struct shash_node *node; + + SHASH_FOR_EACH (node, &netdev_shash) { + struct netdev *netdev = node->data; + const struct netdev_saved_flags *sf; + enum netdev_flags saved_values; + enum netdev_flags saved_flags; + + saved_values = saved_flags = 0; + LIST_FOR_EACH (sf, node, &netdev->saved_flags_list) { + saved_flags |= sf->saved_flags; + saved_values &= ~sf->saved_flags; + saved_values |= sf->saved_flags & sf->saved_values; + } + if (saved_flags) { + enum netdev_flags old_flags; + + netdev->netdev_class->update_flags(netdev, + saved_flags & saved_values, + saved_flags & ~saved_values, + &old_flags); + } } } + +uint64_t +netdev_get_change_seq(const struct netdev *netdev) +{ + return netdev->change_seq; +}