X-Git-Url: http://git.onelab.eu/?a=blobdiff_plain;f=lib%2Fnetdev.c;h=a15315a79019ca5062a07b7796d0ff3476fad97f;hb=41ca1e0afb4b261a217c9fdaf672ef606e8434f9;hp=82a490419778edcd8601f60569d55f46a6d38059;hpb=67a4917b07031b387beafaedce413b4207214059;p=sliver-openvswitch.git diff --git a/lib/netdev.c b/lib/netdev.c index 82a490419..9aaf10757 100644 --- a/lib/netdev.c +++ b/lib/netdev.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2008, 2009, 2010 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 @@ -26,75 +25,120 @@ #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" #include "openflow/openflow.h" #include "packets.h" #include "poll-loop.h" #include "shash.h" +#include "smap.h" +#include "sset.h" #include "svec.h" - -#define THIS_MODULE VLM_netdev #include "vlog.h" -static const struct netdev_class *netdev_classes[] = { - &netdev_linux_class, - &netdev_tap_class, +VLOG_DEFINE_THIS_MODULE(netdev); + +COVERAGE_DEFINE(netdev_received); +COVERAGE_DEFINE(netdev_sent); +COVERAGE_DEFINE(netdev_add_router); +COVERAGE_DEFINE(netdev_get_stats); + +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; }; -static int n_netdev_classes = ARRAY_SIZE(netdev_classes); + +/* 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_obj_shash = SHASH_INITIALIZER(&netdev_obj_shash); +static struct shash netdev_shash OVS_GUARDED_BY(netdev_mutex) + = SHASH_INITIALIZER(&netdev_shash); -/* All open network devices. */ -static struct list netdev_list = LIST_INITIALIZER(&netdev_list); +/* 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); + +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 restore_all_flags(void *aux); -static int restore_flags(struct netdev *netdev); +static void restore_all_flags(void *aux OVS_UNUSED); +void update_device_args(struct netdev *, const struct shash *args); -/* Attempts to initialize the netdev module. Returns 0 if successful, - * otherwise a positive errno value. - * - * Calling this function is optional. If not called explicitly, it will - * automatically be called upon the first attempt to open or create a - * network device. */ int -netdev_initialize(void) +netdev_n_rxq(const struct netdev *netdev) +{ + return netdev->n_rxq; +} + +bool +netdev_is_pmd(const struct netdev *netdev) { - static int status = -1; - if (status < 0) { - int i, j; + return !strcmp(netdev->netdev_class->type, "dpdk"); +} - fatal_signal_add_hook(restore_all_flags, NULL, true); +static void +netdev_class_mutex_initialize(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) +{ + static struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; - status = 0; - for (i = j = 0; i < n_netdev_classes; i++) { - const struct netdev_class *class = netdev_classes[i]; - if (class->init) { - int retval = class->init(); - if (!retval) { - netdev_classes[j++] = class; - } else { - VLOG_ERR("failed to initialize %s network device " - "class: %s", class->type, strerror(retval)); - if (!status) { - status = retval; - } - } - } else { - netdev_classes[j++] = class; - } - } - n_netdev_classes = j; + 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 struct ovsthread_once once = OVSTHREAD_ONCE_INITIALIZER; + + if (ovsthread_once_start(&once)) { + netdev_class_mutex_initialize(); + + fatal_signal_add_hook(restore_all_flags, NULL, NULL, true); + netdev_vport_patch_register(); + +#ifdef __linux__ + netdev_register_provider(&netdev_linux_class); + netdev_register_provider(&netdev_internal_class); + netdev_register_provider(&netdev_tap_class); + 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); } - return status; } /* Performs periodic work needed by all the various kinds of netdevs. @@ -103,14 +147,18 @@ netdev_initialize(void) * main poll loop. */ void netdev_run(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - int i; - for (i = 0; i < n_netdev_classes; i++) { - const struct netdev_class *class = netdev_classes[i]; - if (class->run) { - 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. @@ -119,285 +167,466 @@ netdev_run(void) * main poll loop. */ void netdev_wait(void) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - int i; - for (i = 0; i < n_netdev_classes; i++) { - const struct netdev_class *class = netdev_classes[i]; - if (class->wait) { - 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); } -/* Attempts to create a network device object of 'type' with 'name'. 'type' - * corresponds to the 'type' field used in the netdev_class * structure. - * Arguments for creation are provided in 'args', which may be empty or NULL - * if none are needed. */ -int -netdev_create(const char *name, const char *type, const struct shash *args) +static struct netdev_registered_class * +netdev_lookup_class(const char *type) + OVS_REQ_RDLOCK(netdev_class_mutex) { - struct shash empty_args = SHASH_INITIALIZER(&empty_args); - int i; - - netdev_initialize(); + struct netdev_registered_class *rc; - if (!args) { - args = &empty_args; + HMAP_FOR_EACH_WITH_HASH (rc, hmap_node, hash_string(type, 0), + &netdev_classes) { + if (!strcmp(type, rc->class->type)) { + return rc; + } } + return NULL; +} - if (shash_find(&netdev_obj_shash, name)) { - VLOG_WARN("attempted to create a netdev object with bound name: %s", - name); - return EEXIST; - } +/* 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) +{ + int error; - for (i = 0; i < n_netdev_classes; i++) { - const struct netdev_class *class = netdev_classes[i]; - if (!strcmp(type, class->type)) { - return class->create(name, type, args, true); + 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); + 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, ovs_strerror(error)); } } + ovs_mutex_unlock(&netdev_class_mutex); - VLOG_WARN("could not create netdev object of unknown type: %s", type); - - return EINVAL; + return error; } -/* Destroys netdev object 'name'. Netdev objects maintain a reference count - * which is incremented on netdev_open() and decremented on netdev_close(). - * If 'name' has a non-zero reference count, it will not destroy the object - * and return EBUSY. */ +/* Unregisters a netdev provider. 'type' must have been previously + * registered and not currently be in use by any netdevs. After unregistration + * new netdevs of that type cannot be opened using netdev_open(). */ int -netdev_destroy(const char *name) +netdev_unregister_provider(const char *type) + OVS_EXCLUDED(netdev_class_mutex, netdev_mutex) { - struct shash_node *node; - struct netdev_obj *netdev_obj; + struct netdev_registered_class *rc; + int error; - node = shash_find(&netdev_obj_shash, name); - if (!node) { - return ENODEV; - } + 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); + error = EAFNOSUPPORT; + } else { + int ref_cnt; - netdev_obj = node->data; - if (netdev_obj->ref_cnt != 0) { - VLOG_WARN("attempt to destroy open netdev object (%d): %s", - netdev_obj->ref_cnt, name); - return EBUSY; + 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); + error = EBUSY; + } } + ovs_mutex_unlock(&netdev_class_mutex); - shash_delete(&netdev_obj_shash, node); - netdev_obj->netdev_class->destroy(netdev_obj); - - return 0; + return error; } -/* Reconfigures the device object 'name' with 'args'. 'args' may be empty - * or NULL if none are needed. */ -int -netdev_reconfigure(const char *name, const struct shash *args) +/* 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 empty_args = SHASH_INITIALIZER(&empty_args); - struct netdev_obj *netdev_obj; + struct netdev_registered_class *rc; - if (!args) { - args = &empty_args; - } + netdev_initialize(); + sset_clear(types); - netdev_obj = shash_find_data(&netdev_obj_shash, name); - if (!netdev_obj) { - return ENODEV; + 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); +} - if (netdev_obj->netdev_class->reconfigure) { - return netdev_obj->netdev_class->reconfigure(netdev_obj, args); +/* 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 0; + return false; } -/* Opens the network device named 'name' (e.g. "eth0") and returns zero if - * successful, otherwise a positive errno value. On success, sets '*netdevp' - * to the new network device, otherwise to null. +/* Opens the network device named 'name' (e.g. "eth0") of the specified 'type' + * (e.g. "system") and returns zero if successful, otherwise a positive errno + * value. On success, sets '*netdevp' to the new network device, otherwise to + * null. * - * 'ethertype' may be a 16-bit Ethernet protocol value in host byte order to - * capture frames of that type received on the device. It may also be one of - * the 'enum netdev_pseudo_ethertype' values to receive frames in one of those - * categories. */ + * Some network devices may need to be configured (with netdev_set_config()) + * before they can be used. */ int -netdev_open(const char *name, int ethertype, struct netdev **netdevp) +netdev_open(const char *name, const char *type, struct netdev **netdevp) + OVS_EXCLUDED(netdev_mutex) { - struct netdev_obj *netdev_obj; - struct netdev *netdev = NULL; + struct netdev *netdev; int error; - int i; netdev_initialize(); - netdev_obj = shash_find_data(&netdev_obj_shash, name); - if (netdev_obj) { - netdev_obj->ref_cnt++; - error = netdev_obj->netdev_class->open(name, ethertype, &netdev); - } else { - /* Default to "system". */ - error = EAFNOSUPPORT; - for (i = 0; i < n_netdev_classes; i++) { - const struct netdev_class *class = netdev_classes[i]; - if (!strcmp(class->type, "system")) { - struct shash empty_args = SHASH_INITIALIZER(&empty_args); - - /* Dynamically create the netdev object, but indicate - * that it should be destroyed when the the last user - * closes its handle. */ - error = class->create(name, "system", &empty_args, false); + 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) { - netdev_obj = shash_find_data(&netdev_obj_shash, name); - netdev_obj->ref_cnt++; - error = class->open(name, ethertype, &netdev); + 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); } - break; + } else { + error = ENOMEM; } + } else { + VLOG_WARN("could not create netdev %s of unknown type %s", + name, type); + error = EAFNOSUPPORT; } + } else { + error = 0; } - *netdevp = error ? NULL : netdev; + ovs_mutex_unlock(&netdev_mutex); + ovs_mutex_unlock(&netdev_class_mutex); + + if (!error) { + netdev->ref_cnt++; + *netdevp = netdev; + } else { + *netdevp = NULL; + } return error; } -/* Closes and destroys 'netdev'. */ -void -netdev_close(struct netdev *netdev) +/* 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) { - struct netdev_obj *netdev_obj; - char *name = netdev->name; + 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 smap *args) + OVS_EXCLUDED(netdev_mutex) +{ + if (netdev->netdev_class->set_config) { + const struct smap no_args = SMAP_INITIALIZER(&no_args); int error; - netdev_obj = shash_find_data(&netdev_obj_shash, name); - assert(netdev_obj); - if (netdev_obj->ref_cnt > 0) { - netdev_obj->ref_cnt--; - } else { - VLOG_WARN("netdev %s closed too many times", name); + 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; +} - /* If the reference count for the netdev object is zero, and it - * was dynamically created by netdev_open(), destroy it. */ - if (!netdev_obj->ref_cnt && !netdev_obj->created) { - netdev_destroy(name); - } +/* Returns the current configuration for 'netdev' in 'args'. The caller must + * 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 + * smap_destroy(). */ +int +netdev_get_config(const struct netdev *netdev, struct smap *args) + OVS_EXCLUDED(netdev_mutex) +{ + int error; - /* Restore flags that we changed, if any. */ - fatal_signal_block(); - error = restore_flags(netdev); - list_remove(&netdev->node); - fatal_signal_unblock(); + smap_clear(args); + if (netdev->netdev_class->get_config) { + error = netdev->netdev_class->get_config(netdev, args); if (error) { - VLOG_WARN("failed to restore network device flags on %s: %s", - name, strerror(error)); + smap_clear(args); } + } else { + error = 0; + } + + return error; +} - /* Free. */ - netdev->netdev_class->close(netdev); - free(name); +const struct netdev_tunnel_config * +netdev_get_tunnel_config(const struct netdev *netdev) + OVS_EXCLUDED(netdev_mutex) +{ + 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) +static void +netdev_unref(struct netdev *dev) + OVS_RELEASES(netdev_mutex) { - struct netdev *netdev; - int error; + 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 { + ovs_mutex_unlock(&netdev_mutex); + } +} - error = netdev_open(name, NETDEV_ETH_TYPE_NONE, &netdev); - if (!error) { - netdev_close(netdev); - return true; +/* Closes and destroys 'netdev'. */ +void +netdev_close(struct netdev *netdev) + OVS_EXCLUDED(netdev_mutex) +{ + if (netdev) { + ovs_mutex_lock(&netdev_mutex); + netdev_unref(netdev); + } +} + +/* Parses 'netdev_name_', which is of the form [type@]name into its component + * pieces. 'name' and 'type' must be freed by the caller. */ +void +netdev_parse_name(const char *netdev_name_, char **name, char **type) +{ + char *netdev_name = xstrdup(netdev_name_); + char *separator; + + separator = strchr(netdev_name, '@'); + if (separator) { + *separator = '\0'; + *type = netdev_name; + *name = xstrdup(separator + 1); } else { - if (error != ENODEV) { - VLOG_WARN("failed to open network device %s: %s", - name, strerror(error)); - } - return false; + *name = netdev_name; + *type = xstrdup("system"); } } -/* Initializes 'svec' with a list of the names of all known network devices. */ +/* 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_enumerate(struct svec *svec) +netdev_rxq_open(struct netdev *netdev, struct netdev_rxq **rxp, int id) + OVS_EXCLUDED(netdev_mutex) { int error; - int i; - svec_init(svec); - - netdev_initialize(); - - error = 0; - for (i = 0; i < n_netdev_classes; i++) { - const struct netdev_class *class = netdev_classes[i]; - if (class->enumerate) { - int retval = class->enumerate(svec); - if (retval) { - VLOG_WARN("failed to enumerate %s network devices: %s", - class->type, strerror(retval)); - if (!error) { - error = retval; - } + 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; } + + *rxp = NULL; return error; } -/* 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.) +/* 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 batch of packets from 'rx'. * - * 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 EAGAIN immediately if no packet is ready to be received. + * + * Returns EMSGSIZE, and discards the packet, if the received packet is longer + * than 'ofpbuf_tailroom(buffer)'. + * + * 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 retval; - assert(buffer->size == 0); - assert(ofpbuf_tailroom(buffer) >= ETH_TOTAL_MIN); - - retval = netdev->netdev_class->recv(netdev, - buffer->data, ofpbuf_tailroom(buffer)); - 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) { - netdev->netdev_class->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) { - return netdev->netdev_class->drain(netdev); + 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 @@ -405,14 +634,21 @@ 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. */ + * expected to do additional queuing of packets. + * + * 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 error = netdev->netdev_class->send(netdev, buffer->data, buffer->size); + int error; + + error = (netdev->netdev_class->send + ? netdev->netdev_class->send(netdev, buffer, may_steal) + : EOPNOTSUPP); if (!error) { COVERAGE_INC(netdev_sent); } @@ -429,7 +665,9 @@ netdev_send(struct netdev *netdev, const struct ofpbuf *buffer) void netdev_send_wait(struct netdev *netdev) { - return netdev->netdev_class->send_wait(netdev); + if (netdev->netdev_class->send_wait) { + netdev->netdev_class->send_wait(netdev); + } } /* Attempts to set 'netdev''s MAC address to 'mac'. Returns 0 if successful, @@ -461,18 +699,45 @@ netdev_get_name(const struct netdev *netdev) * (and received) packets, in bytes, not including the hardware header; thus, * this is typically 1500 bytes for Ethernet devices. * - * If successful, returns 0 and stores the MTU size in '*mtup'. On failure, - * returns a positive errno value and stores ETH_PAYLOAD_MAX (1500) in - * '*mtup'. */ + * If successful, returns 0 and stores the MTU size in '*mtup'. Returns + * EOPNOTSUPP if 'netdev' does not have an MTU (as e.g. some tunnels do not). + * On other failure, returns a positive errno value. On failure, sets '*mtup' + * to 0. */ int netdev_get_mtu(const struct netdev *netdev, int *mtup) { - int error = netdev->netdev_class->get_mtu(netdev, mtup); + const struct netdev_class *class = netdev->netdev_class; + int error; + + error = class->get_mtu ? class->get_mtu(netdev, mtup) : EOPNOTSUPP; if (error) { - VLOG_WARN_RL(&rl, "failed to retrieve MTU for network device %s: %s", - netdev_get_name(netdev), strerror(error)); - *mtup = ETH_PAYLOAD_MAX; + *mtup = 0; + if (error != EOPNOTSUPP) { + VLOG_DBG_RL(&rl, "failed to retrieve MTU for network device %s: " + "%s", netdev_get_name(netdev), ovs_strerror(error)); + } + } + return error; +} + +/* Sets the MTU of 'netdev'. The MTU is the maximum size of transmitted + * (and received) packets, in bytes. + * + * If successful, returns 0. Returns EOPNOTSUPP if 'netdev' does not have an + * MTU (as e.g. some tunnels do not). On other failure, returns a positive + * errno value. */ +int +netdev_set_mtu(const struct netdev *netdev, int mtu) +{ + 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_DBG_RL(&rl, "failed to set MTU for network device %s: %s", + netdev_get_name(netdev), ovs_strerror(error)); } + return error; } @@ -484,24 +749,41 @@ netdev_get_mtu(const struct netdev *netdev, int *mtup) * value should be unique within a host and remain stable at least until * reboot. SNMP says an ifindex "ranges between 1 and the value of ifNumber" * but many systems do not follow this rule anyhow. + * + * Some network devices may not implement support for this function. In such + * cases this function will always return -EOPNOTSUPP. */ int netdev_get_ifindex(const struct netdev *netdev) { - return netdev->netdev_class->get_ifindex(netdev); + int (*get_ifindex)(const struct netdev *); + + get_ifindex = netdev->netdev_class->get_ifindex; + + return get_ifindex ? get_ifindex(netdev) : -EOPNOTSUPP; } /* Stores the features supported by 'netdev' into each of '*current', * '*advertised', '*supported', and '*peer' that are non-null. Each value is a * bitmap of "enum ofp_port_features" bits, in host byte order. Returns 0 if * successful, otherwise a positive errno value. On failure, all of the - * passed-in values are set to 0. */ + * passed-in values are set to 0. + * + * Some network devices may not implement support for this function. In such + * cases this function will always return EOPNOTSUPP. */ int -netdev_get_features(struct netdev *netdev, - uint32_t *current, uint32_t *advertised, - uint32_t *supported, uint32_t *peer) +netdev_get_features(const struct netdev *netdev, + enum netdev_features *current, + enum netdev_features *advertised, + enum netdev_features *supported, + enum netdev_features *peer) { - uint32_t dummy[4]; + int (*get_features)(const struct netdev *netdev, + enum netdev_features *current, + enum netdev_features *advertised, + enum netdev_features *supported, + enum netdev_features *peer); + enum netdev_features dummy[4]; int error; if (!current) { @@ -517,50 +799,63 @@ netdev_get_features(struct netdev *netdev, peer = &dummy[3]; } - error = netdev->netdev_class->get_features(netdev, current, advertised, - supported, peer); + get_features = netdev->netdev_class->get_features; + error = get_features + ? get_features(netdev, current, advertised, supported, + peer) + : EOPNOTSUPP; if (error) { *current = *advertised = *supported = *peer = 0; } return error; } -/* Returns the maximum speed of a network connection that has the "enum - * ofp_port_features" bits in 'features', in bits per second. If no bits that - * indicate a speed are set in 'features', assumes 100Mbps. */ +/* 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', returns 'default_bps'. */ uint64_t -netdev_features_to_bps(uint32_t features) +netdev_features_to_bps(enum netdev_features features, + uint64_t default_bps) { enum { - F_10000MB = OFPPF_10GB_FD, - F_1000MB = OFPPF_1GB_HD | OFPPF_1GB_FD, - F_100MB = OFPPF_100MB_HD | OFPPF_100MB_FD, - F_10MB = OFPPF_10MB_HD | OFPPF_10MB_FD + F_1000000MB = NETDEV_F_1TB_FD, + F_100000MB = NETDEV_F_100GB_FD, + F_40000MB = NETDEV_F_40GB_FD, + F_10000MB = NETDEV_F_10GB_FD, + F_1000MB = NETDEV_F_1GB_HD | NETDEV_F_1GB_FD, + F_100MB = NETDEV_F_100MB_HD | NETDEV_F_100MB_FD, + F_10MB = NETDEV_F_10MB_HD | NETDEV_F_10MB_FD }; - return ( features & F_10000MB ? UINT64_C(10000000000) - : features & F_1000MB ? UINT64_C(1000000000) - : features & F_100MB ? UINT64_C(100000000) - : features & F_10MB ? UINT64_C(10000000) - : UINT64_C(100000000)); + return ( features & F_1000000MB ? UINT64_C(1000000000000) + : features & F_100000MB ? UINT64_C(100000000000) + : features & F_40000MB ? UINT64_C(40000000000) + : features & F_10000MB ? UINT64_C(10000000000) + : features & F_1000MB ? UINT64_C(1000000000) + : features & F_100MB ? UINT64_C(100000000) + : features & F_10MB ? UINT64_C(10000000) + : default_bps); } -/* Returns true if any of the "enum ofp_port_features" bits that indicate a - * full-duplex link are set in 'features', otherwise false. */ +/* Returns true if any of the NETDEV_F_* bits that indicate a full-duplex link + * are set in 'features', otherwise false. */ bool -netdev_features_is_full_duplex(uint32_t features) +netdev_features_is_full_duplex(enum netdev_features features) { - return (features & (OFPPF_10MB_FD | OFPPF_100MB_FD | OFPPF_1GB_FD - | OFPPF_10GB_FD)) != 0; + return (features & (NETDEV_F_10MB_FD | NETDEV_F_100MB_FD | NETDEV_F_1GB_FD + | NETDEV_F_10GB_FD | NETDEV_F_40GB_FD + | NETDEV_F_100GB_FD | NETDEV_F_1TB_FD)) != 0; } /* Set the features advertised by 'netdev' to 'advertise'. Returns 0 if * successful, otherwise a positive errno value. */ int -netdev_set_advertisements(struct netdev *netdev, uint32_t advertise) +netdev_set_advertisements(struct netdev *netdev, + enum netdev_features advertise) { return (netdev->netdev_class->set_advertisements - ? netdev->netdev_class->set_advertisements(netdev, advertise) + ? netdev->netdev_class->set_advertisements( + netdev, advertise) : EOPNOTSUPP); } @@ -574,8 +869,8 @@ netdev_set_advertisements(struct netdev *netdev, uint32_t advertise) * * - EOPNOTSUPP: No IPv4 network stack attached to 'netdev'. * - * 'address' or 'netmask' or both may be null, in which case the address or netmask - * is not reported. */ + * 'address' or 'netmask' or both may be null, in which case the address or + * netmask is not reported. */ int netdev_get_in4(const struct netdev *netdev, struct in_addr *address_, struct in_addr *netmask_) @@ -585,7 +880,8 @@ netdev_get_in4(const struct netdev *netdev, int error; error = (netdev->netdev_class->get_in4 - ? netdev->netdev_class->get_in4(netdev, &address, &netmask) + ? netdev->netdev_class->get_in4(netdev, + &address, &netmask) : EOPNOTSUPP); if (address_) { address_->s_addr = error ? 0 : address.s_addr; @@ -607,6 +903,26 @@ netdev_set_in4(struct netdev *netdev, struct in_addr addr, struct in_addr mask) : EOPNOTSUPP); } +/* Obtains ad IPv4 address from device name and save the address in + * in4. Returns 0 if successful, otherwise a positive errno value. + */ +int +netdev_get_in4_by_name(const char *device_name, struct in_addr *in4) +{ + struct netdev *netdev; + int error; + + error = netdev_open(device_name, "system", &netdev); + if (error) { + in4->s_addr = htonl(0); + return error; + } + + error = netdev_get_in4(netdev, in4, NULL); + netdev_close(netdev); + return error; +} + /* Adds 'router' as a default IP gateway for the TCP/IP stack that corresponds * to 'netdev'. */ int @@ -631,8 +947,8 @@ netdev_get_next_hop(const struct netdev *netdev, char **netdev_name) { int error = (netdev->netdev_class->get_next_hop - ? netdev->netdev_class->get_next_hop(host, next_hop, - netdev_name) + ? netdev->netdev_class->get_next_hop( + host, next_hop, netdev_name) : EOPNOTSUPP); if (error) { next_hop->s_addr = 0; @@ -641,6 +957,19 @@ netdev_get_next_hop(const struct netdev *netdev, return error; } +/* Populates 'smap' with status information. + * + * 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 smap *smap) +{ + return (netdev->netdev_class->get_status + ? netdev->netdev_class->get_status(netdev, smap) + : EOPNOTSUPP); +} + /* If 'netdev' has an assigned IPv6 address, sets '*in6' to that address and * returns 0. Otherwise, returns a positive errno value and sets '*in6' to * all-zero-bits (in6addr_any). @@ -659,7 +988,8 @@ netdev_get_in6(const struct netdev *netdev, struct in6_addr *in6) int error; error = (netdev->netdev_class->get_in6 - ? netdev->netdev_class->get_in6(netdev, in6 ? in6 : &dummy) + ? netdev->netdev_class->get_in6(netdev, + in6 ? in6 : &dummy) : EOPNOTSUPP); if (error && in6) { memset(in6, 0, sizeof *in6); @@ -668,37 +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->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; } @@ -708,41 +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, permanent); + 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) +{ + 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 @@ -751,7 +1116,7 @@ netdev_turn_flags_off(struct netdev *netdev, enum netdev_flags flags, * ENXIO indicates that there is no ARP table entry for 'ip' on 'netdev'. */ int netdev_arp_lookup(const struct netdev *netdev, - uint32_t ip, uint8_t mac[ETH_ADDR_LEN]) + ovs_be32 ip, uint8_t mac[ETH_ADDR_LEN]) { int error = (netdev->netdev_class->arp_lookup ? netdev->netdev_class->arp_lookup(netdev, ip, mac) @@ -762,18 +1127,56 @@ netdev_arp_lookup(const struct netdev *netdev, return error; } -/* Sets 'carrier' to true if carrier is active (link light is on) on - * 'netdev'. */ -int -netdev_get_carrier(const struct netdev *netdev, bool *carrier) +/* Returns true if carrier is active (link light is on) on 'netdev'. */ +bool +netdev_get_carrier(const struct netdev *netdev) { - int error = (netdev->netdev_class->get_carrier - ? netdev->netdev_class->get_carrier(netdev, carrier) - : EOPNOTSUPP); + int error; + enum netdev_flags flags; + bool carrier; + + netdev_get_flags(netdev, &flags); + if (!(flags & NETDEV_UP)) { + return false; + } + + if (!netdev->netdev_class->get_carrier) { + return true; + } + + error = netdev->netdev_class->get_carrier(netdev, &carrier); if (error) { - *carrier = false; + VLOG_DBG("%s: failed to get network device carrier status, assuming " + "down: %s", netdev_get_name(netdev), ovs_strerror(error)); + carrier = false; } - return error; + + return carrier; +} + +/* Returns the number of times 'netdev''s carrier has changed. */ +long long int +netdev_get_carrier_resets(const struct netdev *netdev) +{ + return (netdev->netdev_class->get_carrier_resets + ? netdev->netdev_class->get_carrier_resets(netdev) + : 0); +} + +/* Attempts to force netdev_get_carrier() to poll 'netdev''s MII registers for + * link status instead of checking 'netdev''s carrier. 'netdev''s MII + * registers will be polled once ever 'interval' milliseconds. If 'netdev' + * does not support MII, another method may be used as a fallback. If + * 'interval' is less than or equal to zero, reverts netdev_get_carrier() to + * its normal behavior. + * + * Returns 0 if successful, otherwise a positive errno value. */ +int +netdev_set_miimon_interval(struct netdev *netdev, long long int interval) +{ + return (netdev->netdev_class->set_miimon_interval + ? netdev->netdev_class->set_miimon_interval(netdev, interval) + : EOPNOTSUPP); } /* Retrieves current device stats for 'netdev'. */ @@ -792,6 +1195,19 @@ netdev_get_stats(const struct netdev *netdev, struct netdev_stats *stats) return error; } +/* Attempts to change the stats for 'netdev' to those provided in 'stats'. + * Returns 0 if successful, otherwise a positive errno value. + * + * This will probably fail for most network devices. Some devices might only + * allow setting their stats to 0. */ +int +netdev_set_stats(struct netdev *netdev, const struct netdev_stats *stats) +{ + return (netdev->netdev_class->set_stats + ? netdev->netdev_class->set_stats(netdev, stats) + : EOPNOTSUPP); +} + /* Attempts to set input rate limiting (policing) policy, such that up to * 'kbits_rate' kbps of traffic is accepted, with a maximum accumulative burst * size of 'kbits' kb. */ @@ -801,263 +1217,489 @@ netdev_set_policing(struct netdev *netdev, uint32_t kbits_rate, { return (netdev->netdev_class->set_policing ? netdev->netdev_class->set_policing(netdev, - kbits_rate, kbits_burst) + kbits_rate, kbits_burst) : EOPNOTSUPP); } -/* If 'netdev' is a VLAN network device (e.g. one created with vconfig(8)), - * sets '*vlan_vid' to the VLAN VID associated with that device and returns 0. - * Otherwise returns a errno value (specifically ENOENT if 'netdev_name' is the - * name of a network device that is not a VLAN device) and sets '*vlan_vid' to - * -1. */ +/* Adds to 'types' all of the forms of QoS supported by 'netdev', or leaves it + * empty if 'netdev' does not support QoS. Any names added to 'types' should + * be documented as valid for the "type" column in the "QoS" table in + * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). + * + * Every network device supports disabling QoS with a type of "", but this type + * will not be added to 'types'. + * + * The caller must initialize 'types' (e.g. with sset_init()) before calling + * this function. The caller is responsible for destroying 'types' (e.g. with + * sset_destroy()) when it is no longer needed. + * + * Returns 0 if successful, otherwise a positive errno value. */ int -netdev_get_vlan_vid(const struct netdev *netdev, int *vlan_vid) +netdev_get_qos_types(const struct netdev *netdev, struct sset *types) { - int error = (netdev->netdev_class->get_vlan_vid - ? netdev->netdev_class->get_vlan_vid(netdev, vlan_vid) - : ENOENT); - if (error) { - *vlan_vid = 0; - } - return error; + const struct netdev_class *class = netdev->netdev_class; + return (class->get_qos_types + ? class->get_qos_types(netdev, types) + : 0); } -/* Returns a network device that has 'in4' as its IP address, if one exists, - * otherwise a null pointer. */ -struct netdev * -netdev_find_dev_by_in4(const struct in_addr *in4) +/* Queries 'netdev' for its capabilities regarding the specified 'type' of QoS, + * which should be "" or one of the types returned by netdev_get_qos_types() + * for 'netdev'. Returns 0 if successful, otherwise a positive errno value. + * On success, initializes 'caps' with the QoS capabilities; on failure, clears + * 'caps' to all zeros. */ +int +netdev_get_qos_capabilities(const struct netdev *netdev, const char *type, + struct netdev_qos_capabilities *caps) { - struct netdev *netdev; - struct svec dev_list; - size_t i; - - netdev_enumerate(&dev_list); - for (i = 0; i < dev_list.n; i++) { - const char *name = dev_list.names[i]; - struct in_addr dev_in4; - - if (!netdev_open(name, NETDEV_ETH_TYPE_NONE, &netdev) - && !netdev_get_in4(netdev, &dev_in4, NULL) - && dev_in4.s_addr == in4->s_addr) { - goto exit; + const struct netdev_class *class = netdev->netdev_class; + + if (*type) { + int retval = (class->get_qos_capabilities + ? class->get_qos_capabilities(netdev, type, caps) + : EOPNOTSUPP); + if (retval) { + memset(caps, 0, sizeof *caps); } - netdev_close(netdev); + return retval; + } else { + /* Every netdev supports turning off QoS. */ + memset(caps, 0, sizeof *caps); + return 0; } - netdev = NULL; +} -exit: - svec_destroy(&dev_list); - return netdev; +/* Obtains the number of queues supported by 'netdev' for the specified 'type' + * of QoS. Returns 0 if successful, otherwise a positive errno value. Stores + * the number of queues (zero on failure) in '*n_queuesp'. + * + * This is just a simple wrapper around netdev_get_qos_capabilities(). */ +int +netdev_get_n_queues(const struct netdev *netdev, + const char *type, unsigned int *n_queuesp) +{ + struct netdev_qos_capabilities caps; + int retval; + + retval = netdev_get_qos_capabilities(netdev, type, &caps); + *n_queuesp = caps.n_queues; + return retval; } - -/* Initializes 'netdev_obj' as a netdev object named 'name' of the - * specified 'netdev_class'. + +/* Queries 'netdev' about its currently configured form of QoS. If successful, + * stores the name of the current form of QoS into '*typep', stores any details + * of configuration as string key-value pairs in 'details', and returns 0. On + * failure, sets '*typep' to NULL and returns a positive errno value. * - * This function adds 'netdev_obj' to a netdev-owned shash, so it is - * very important that 'netdev_obj' only be freed after calling - * netdev_destroy(). */ -void -netdev_obj_init(struct netdev_obj *netdev_obj, const char *name, - const struct netdev_class *netdev_class, bool created) + * A '*typep' of "" indicates that QoS is currently disabled on 'netdev'. + * + * 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'. + * + * '*typep' will be one of the types returned by netdev_get_qos_types() for + * 'netdev'. The contents of 'details' should be documented as valid for + * '*typep' in the "other_config" column in the "QoS" table in + * vswitchd/vswitch.xml (which is built as ovs-vswitchd.conf.db(8)). */ +int +netdev_get_qos(const struct netdev *netdev, + const char **typep, struct smap *details) { - assert(!shash_find(&netdev_obj_shash, name)); + const struct netdev_class *class = netdev->netdev_class; + int retval; - netdev_obj->netdev_class = netdev_class; - netdev_obj->ref_cnt = 0; - netdev_obj->created = created; - shash_add(&netdev_obj_shash, name, netdev_obj); + if (class->get_qos) { + retval = class->get_qos(netdev, typep, details); + if (retval) { + *typep = NULL; + smap_clear(details); + } + return retval; + } else { + /* 'netdev' doesn't support QoS, so report that QoS is disabled. */ + *typep = ""; + return 0; + } } -/* Initializes 'netdev' as a netdev named 'name' of the specified - * 'netdev_class'. +/* Attempts to reconfigure QoS on 'netdev', changing the form of QoS to 'type' + * with details of configuration from 'details'. Returns 0 if successful, + * otherwise a positive errno value. On error, the previous QoS configuration + * is retained. * - * 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, const char *name, - const struct netdev_class *netdev_class) + * When this function changes the type of QoS (not just 'details'), this also + * resets all queue configuration for 'netdev' to their defaults (which depend + * on the specific type of QoS). Otherwise, the queue configuration for + * 'netdev' is unchanged. + * + * 'type' should be "" (to disable QoS) or one of the types returned by + * netdev_get_qos_types() for 'netdev'. The contents of 'details' should be + * documented as valid for the given 'type' in the "other_config" column in the + * "QoS" table in vswitchd/vswitch.xml (which is built as + * ovs-vswitchd.conf.db(8)). + * + * NULL may be specified for 'details' if there are no configuration + * details. */ +int +netdev_set_qos(struct netdev *netdev, + const char *type, const struct smap *details) { - netdev->netdev_class = netdev_class; - netdev->name = xstrdup(name); - netdev->save_flags = 0; - netdev->changed_flags = 0; - list_push_back(&netdev_list, &netdev->node); + const struct netdev_class *class = netdev->netdev_class; + + if (!type) { + type = ""; + } + + if (class->set_qos) { + if (!details) { + static const struct smap empty = SMAP_INITIALIZER(&empty); + details = ∅ + } + return class->set_qos(netdev, type, details); + } else { + return *type ? EOPNOTSUPP : 0; + } } -/* Returns the class type of 'netdev'. +/* Queries 'netdev' for information about the queue numbered 'queue_id'. If + * successful, adds that information as string key-value pairs to 'details'. + * Returns 0 if successful, otherwise a positive errno value. * - * The caller must not free the returned value. */ -const char *netdev_get_type(const struct netdev *netdev) + * 'queue_id' must be less than the number of queues supported by 'netdev' for + * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)). + * + * 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. 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 smap *details) { - return netdev->netdev_class->type; + const struct netdev_class *class = netdev->netdev_class; + int retval; + + retval = (class->get_queue + ? class->get_queue(netdev, queue_id, details) + : EOPNOTSUPP); + if (retval) { + smap_clear(details); + } + return retval; } -/* Initializes 'notifier' as a netdev notifier for 'netdev', for which - * notification will consist of calling 'cb', with auxiliary data 'aux'. */ -void -netdev_notifier_init(struct netdev_notifier *notifier, struct netdev *netdev, - void (*cb)(struct netdev_notifier *), void *aux) +/* Configures the queue numbered 'queue_id' on 'netdev' with the key-value + * string pairs in 'details'. The 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)). + * Returns 0 if successful, otherwise a positive errno value. On failure, the + * given queue's configuration should be unmodified. + * + * 'queue_id' must be less than the number of queues supported by 'netdev' for + * the current form of QoS (e.g. as returned by netdev_get_n_queues(netdev)). + * + * This function does not modify 'details', and the caller retains ownership of + * it. */ +int +netdev_set_queue(struct netdev *netdev, + unsigned int queue_id, const struct smap *details) { - notifier->netdev = netdev; - notifier->cb = cb; - notifier->aux = aux; + const struct netdev_class *class = netdev->netdev_class; + return (class->set_queue + ? class->set_queue(netdev, queue_id, details) + : EOPNOTSUPP); } - -/* Tracks changes in the status of a set of network devices. */ -struct netdev_monitor { - struct shash polled_netdevs; - struct shash changed_netdevs; -}; -/* Creates and returns a new structure for monitor changes in the status of - * network devices. */ -struct netdev_monitor * -netdev_monitor_create(void) +/* Attempts to delete the queue numbered 'queue_id' from 'netdev'. Some kinds + * of QoS may have a fixed set of queues, in which case attempts to delete them + * will fail with EOPNOTSUPP. + * + * Returns 0 if successful, otherwise a positive errno value. On failure, the + * given queue will be unmodified. + * + * 'queue_id' must be less than the number of queues supported by 'netdev' for + * the current form of QoS (e.g. as returned by + * netdev_get_n_queues(netdev)). */ +int +netdev_delete_queue(struct netdev *netdev, unsigned int queue_id) { - struct netdev_monitor *monitor = xmalloc(sizeof *monitor); - shash_init(&monitor->polled_netdevs); - shash_init(&monitor->changed_netdevs); - return monitor; + const struct netdev_class *class = netdev->netdev_class; + return (class->delete_queue + ? class->delete_queue(netdev, queue_id) + : EOPNOTSUPP); } -/* Destroys 'monitor'. */ +/* 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 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->netdev_class; + int retval; + + retval = (class->get_queue_stats + ? class->get_queue_stats(netdev, queue_id, stats) + : EOPNOTSUPP); + if (retval) { + stats->tx_bytes = UINT64_MAX; + stats->tx_packets = UINT64_MAX; + stats->tx_errors = UINT64_MAX; + stats->created = LLONG_MIN; + } + return retval; +} + +/* 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_monitor_destroy(struct netdev_monitor *monitor) +netdev_queue_dump_start(struct netdev_queue_dump *dump, + const struct netdev *netdev) { - if (monitor) { - struct shash_node *node; + 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; + } +} - SHASH_FOR_EACH (node, &monitor->polled_netdevs) { - struct netdev_notifier *notifier = node->data; - notifier->netdev->netdev_class->poll_remove(notifier); - } +/* 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. + * + * Queues are not necessarily dumped in increasing order of queue ID (or any + * other predictable order). + * + * 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(). + * + * 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; - shash_destroy(&monitor->polled_netdevs); - shash_destroy(&monitor->changed_netdevs); - free(monitor); + 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; } -static void -netdev_monitor_cb(struct netdev_notifier *notifier) +/* 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_queue_dump_done(struct netdev_queue_dump *dump) { - struct netdev_monitor *monitor = notifier->aux; - const char *name = netdev_get_name(notifier->netdev); - if (!shash_find(&monitor->changed_netdevs, name)) { - shash_add(&monitor->changed_netdevs, name, NULL); + 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; } -/* Attempts to add 'netdev' as a netdev monitored by 'monitor'. Returns 0 if - * successful, otherwise a positive errno value. +/* Iterates over all of 'netdev''s queues, calling 'cb' with the queue's ID, + * its statistics, and the 'aux' specified by the caller. The order of + * iteration is unspecified, but (when successful) each queue is visited + * exactly once. + * + * Calling this function may be more efficient than calling + * netdev_get_queue_stats() for every queue. * - * Adding a given 'netdev' to a monitor multiple times is equivalent to adding - * it once. */ + * 'cb' must not modify or free the statistics passed in. + * + * Returns 0 if successful, otherwise a positive errno value. On error, some + * configured queues may not have been included in the iteration. */ int -netdev_monitor_add(struct netdev_monitor *monitor, struct netdev *netdev) -{ - const char *netdev_name = netdev_get_name(netdev); - int error = 0; - if (!shash_find(&monitor->polled_netdevs, netdev_name) - && netdev->netdev_class->poll_add) - { - struct netdev_notifier *notifier; - error = netdev->netdev_class->poll_add(netdev, netdev_monitor_cb, - monitor, ¬ifier); - if (!error) { - assert(notifier->netdev == netdev); - shash_add(&monitor->polled_netdevs, netdev_name, notifier); - } +netdev_dump_queue_stats(const struct netdev *netdev, + netdev_dump_queue_stats_cb *cb, void *aux) +{ + const struct netdev_class *class = netdev->netdev_class; + return (class->dump_queue_stats + ? class->dump_queue_stats(netdev, cb, aux) + : EOPNOTSUPP); +} + + +/* Returns the class type of 'netdev'. + * + * The caller must not free the returned value. */ +const char * +netdev_get_type(const struct netdev *netdev) +{ + return netdev->netdev_class->type; +} + +/* Returns the class associated with 'netdev'. */ +const struct netdev_class * +netdev_get_class(const struct netdev *netdev) +{ + return netdev->netdev_class; +} + +/* Returns the netdev with 'name' or NULL if there is none. + * + * The caller must free the returned netdev with netdev_close(). */ +struct netdev * +netdev_from_name(const char *name) + OVS_EXCLUDED(netdev_mutex) +{ + struct netdev *netdev; + + ovs_mutex_lock(&netdev_mutex); + netdev = shash_find_data(&netdev_shash, name); + if (netdev) { + netdev->ref_cnt++; } - return error; + ovs_mutex_unlock(&netdev_mutex); + + return netdev; } -/* Removes 'netdev' from the set of netdevs monitored by 'monitor'. (This has - * no effect if 'netdev' is not in the set of devices monitored by - * 'monitor'.) */ +/* Fills 'device_list' with devices that match 'netdev_class'. + * + * The caller is responsible for initializing and destroying 'device_list' and + * must close each device on the list. */ void -netdev_monitor_remove(struct netdev_monitor *monitor, struct netdev *netdev) +netdev_get_devices(const struct netdev_class *netdev_class, + struct shash *device_list) + OVS_EXCLUDED(netdev_mutex) { - const char *netdev_name = netdev_get_name(netdev); struct shash_node *node; - node = shash_find(&monitor->polled_netdevs, netdev_name); - if (node) { - /* Cancel future notifications. */ - struct netdev_notifier *notifier = node->data; - netdev->netdev_class->poll_remove(notifier); - shash_delete(&monitor->polled_netdevs, node); - - /* Drop any pending notification. */ - node = shash_find(&monitor->changed_netdevs, netdev_name); - if (node) { - shash_delete(&monitor->changed_netdevs, node); + 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); } -/* Checks for changes to netdevs in the set monitored by 'monitor'. If any of - * the attributes (Ethernet address, carrier status, speed or peer-advertised - * speed, flags, etc.) of a network device monitored by 'monitor' has changed, - * sets '*devnamep' to the name of a device that has changed and returns 0. - * The caller is responsible for freeing '*devnamep' (with free()). +/* Extracts pointers to all 'netdev-vports' into an array 'vports' + * and returns it. Stores the size of the array into '*size'. * - * If no devices have changed, sets '*devnamep' to NULL and returns EAGAIN. - */ -int -netdev_monitor_poll(struct netdev_monitor *monitor, char **devnamep) + * 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) { - struct shash_node *node = shash_first(&monitor->changed_netdevs); - if (!node) { - *devnamep = NULL; - return EAGAIN; - } else { - *devnamep = xstrdup(node->name); - shash_delete(&monitor->changed_netdevs, node); - return 0; + struct netdev **vports; + struct shash_node *node; + size_t n = 0; + + if (!size) { + return NULL; + } + + /* 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; } -/* Registers with the poll loop to wake up from the next call to poll_block() - * when netdev_monitor_poll(monitor) would indicate that a device has - * changed. */ -void -netdev_monitor_poll_wait(const struct netdev_monitor *monitor) +const char * +netdev_get_type_from_name(const char *name) { - if (!shash_is_empty(&monitor->changed_netdevs)) { - poll_immediate_wake(); - } else { - /* XXX Nothing needed here for netdev_linux, but maybe other netdev - * classes need help. */ - } + struct netdev *dev = netdev_from_name(name); + const char *type = dev ? netdev_get_type(dev) : NULL; + netdev_close(dev); + return type; } -/* 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) +struct netdev * +netdev_rxq_get_netdev(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->netdev_class->update_flags(netdev, - netdev->changed_flags - & ~restore, - restore, &old_flags); - } - return 0; + ovs_assert(rx->netdev->ref_cnt > 0); + return rx->netdev; +} + +const char * +netdev_rxq_get_name(const struct netdev_rxq *rx) +{ + return netdev_get_name(netdev_rxq_get_netdev(rx)); } -/* Retores all the flags on all network devices that we modified. Called from - * a signal handler, so it does not attempt to report error conditions. */ static void restore_all_flags(void *aux OVS_UNUSED) { - struct netdev *netdev; - LIST_FOR_EACH (netdev, struct netdev, node, &netdev_list) { - restore_flags(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; +}