Eric W. Biederman | 5f256be | 2007-09-12 11:50:50 +0200 | [diff] [blame] | 1 | #include <linux/workqueue.h> |
| 2 | #include <linux/rtnetlink.h> |
| 3 | #include <linux/cache.h> |
| 4 | #include <linux/slab.h> |
| 5 | #include <linux/list.h> |
| 6 | #include <linux/delay.h> |
| 7 | #include <net/net_namespace.h> |
| 8 | |
| 9 | /* |
| 10 | * Our network namespace constructor/destructor lists |
| 11 | */ |
| 12 | |
| 13 | static LIST_HEAD(pernet_list); |
| 14 | static struct list_head *first_device = &pernet_list; |
| 15 | static DEFINE_MUTEX(net_mutex); |
| 16 | |
| 17 | static DEFINE_MUTEX(net_list_mutex); |
| 18 | LIST_HEAD(net_namespace_list); |
| 19 | |
| 20 | static struct kmem_cache *net_cachep; |
| 21 | |
| 22 | struct net init_net; |
| 23 | EXPORT_SYMBOL_GPL(init_net); |
| 24 | |
| 25 | void net_lock(void) |
| 26 | { |
| 27 | mutex_lock(&net_list_mutex); |
| 28 | } |
| 29 | |
| 30 | void net_unlock(void) |
| 31 | { |
| 32 | mutex_unlock(&net_list_mutex); |
| 33 | } |
| 34 | |
| 35 | static struct net *net_alloc(void) |
| 36 | { |
| 37 | return kmem_cache_alloc(net_cachep, GFP_KERNEL); |
| 38 | } |
| 39 | |
| 40 | static void net_free(struct net *net) |
| 41 | { |
| 42 | if (!net) |
| 43 | return; |
| 44 | |
| 45 | if (unlikely(atomic_read(&net->use_count) != 0)) { |
| 46 | printk(KERN_EMERG "network namespace not free! Usage: %d\n", |
| 47 | atomic_read(&net->use_count)); |
| 48 | return; |
| 49 | } |
| 50 | |
| 51 | kmem_cache_free(net_cachep, net); |
| 52 | } |
| 53 | |
| 54 | static void cleanup_net(struct work_struct *work) |
| 55 | { |
| 56 | struct pernet_operations *ops; |
| 57 | struct list_head *ptr; |
| 58 | struct net *net; |
| 59 | |
| 60 | net = container_of(work, struct net, work); |
| 61 | |
| 62 | mutex_lock(&net_mutex); |
| 63 | |
| 64 | /* Don't let anyone else find us. */ |
| 65 | net_lock(); |
| 66 | list_del(&net->list); |
| 67 | net_unlock(); |
| 68 | |
| 69 | /* Run all of the network namespace exit methods */ |
| 70 | list_for_each_prev(ptr, &pernet_list) { |
| 71 | ops = list_entry(ptr, struct pernet_operations, list); |
| 72 | if (ops->exit) |
| 73 | ops->exit(net); |
| 74 | } |
| 75 | |
| 76 | mutex_unlock(&net_mutex); |
| 77 | |
| 78 | /* Ensure there are no outstanding rcu callbacks using this |
| 79 | * network namespace. |
| 80 | */ |
| 81 | rcu_barrier(); |
| 82 | |
| 83 | /* Finally it is safe to free my network namespace structure */ |
| 84 | net_free(net); |
| 85 | } |
| 86 | |
| 87 | |
| 88 | void __put_net(struct net *net) |
| 89 | { |
| 90 | /* Cleanup the network namespace in process context */ |
| 91 | INIT_WORK(&net->work, cleanup_net); |
| 92 | schedule_work(&net->work); |
| 93 | } |
| 94 | EXPORT_SYMBOL_GPL(__put_net); |
| 95 | |
| 96 | /* |
| 97 | * setup_net runs the initializers for the network namespace object. |
| 98 | */ |
| 99 | static int setup_net(struct net *net) |
| 100 | { |
| 101 | /* Must be called with net_mutex held */ |
| 102 | struct pernet_operations *ops; |
| 103 | struct list_head *ptr; |
| 104 | int error; |
| 105 | |
| 106 | memset(net, 0, sizeof(struct net)); |
| 107 | atomic_set(&net->count, 1); |
| 108 | atomic_set(&net->use_count, 0); |
| 109 | |
| 110 | error = 0; |
| 111 | list_for_each(ptr, &pernet_list) { |
| 112 | ops = list_entry(ptr, struct pernet_operations, list); |
| 113 | if (ops->init) { |
| 114 | error = ops->init(net); |
| 115 | if (error < 0) |
| 116 | goto out_undo; |
| 117 | } |
| 118 | } |
| 119 | out: |
| 120 | return error; |
| 121 | out_undo: |
| 122 | /* Walk through the list backwards calling the exit functions |
| 123 | * for the pernet modules whose init functions did not fail. |
| 124 | */ |
| 125 | for (ptr = ptr->prev; ptr != &pernet_list; ptr = ptr->prev) { |
| 126 | ops = list_entry(ptr, struct pernet_operations, list); |
| 127 | if (ops->exit) |
| 128 | ops->exit(net); |
| 129 | } |
| 130 | goto out; |
| 131 | } |
| 132 | |
| 133 | static int __init net_ns_init(void) |
| 134 | { |
| 135 | int err; |
| 136 | |
| 137 | printk(KERN_INFO "net_namespace: %zd bytes\n", sizeof(struct net)); |
| 138 | net_cachep = kmem_cache_create("net_namespace", sizeof(struct net), |
| 139 | SMP_CACHE_BYTES, |
| 140 | SLAB_PANIC, NULL); |
| 141 | mutex_lock(&net_mutex); |
| 142 | err = setup_net(&init_net); |
| 143 | |
| 144 | net_lock(); |
| 145 | list_add_tail(&init_net.list, &net_namespace_list); |
| 146 | net_unlock(); |
| 147 | |
| 148 | mutex_unlock(&net_mutex); |
| 149 | if (err) |
| 150 | panic("Could not setup the initial network namespace"); |
| 151 | |
| 152 | return 0; |
| 153 | } |
| 154 | |
| 155 | pure_initcall(net_ns_init); |
| 156 | |
| 157 | static int register_pernet_operations(struct list_head *list, |
| 158 | struct pernet_operations *ops) |
| 159 | { |
| 160 | struct net *net, *undo_net; |
| 161 | int error; |
| 162 | |
| 163 | error = 0; |
| 164 | list_add_tail(&ops->list, list); |
| 165 | for_each_net(net) { |
| 166 | if (ops->init) { |
| 167 | error = ops->init(net); |
| 168 | if (error) |
| 169 | goto out_undo; |
| 170 | } |
| 171 | } |
| 172 | out: |
| 173 | return error; |
| 174 | |
| 175 | out_undo: |
| 176 | /* If I have an error cleanup all namespaces I initialized */ |
| 177 | list_del(&ops->list); |
| 178 | for_each_net(undo_net) { |
| 179 | if (undo_net == net) |
| 180 | goto undone; |
| 181 | if (ops->exit) |
| 182 | ops->exit(undo_net); |
| 183 | } |
| 184 | undone: |
| 185 | goto out; |
| 186 | } |
| 187 | |
| 188 | static void unregister_pernet_operations(struct pernet_operations *ops) |
| 189 | { |
| 190 | struct net *net; |
| 191 | |
| 192 | list_del(&ops->list); |
| 193 | for_each_net(net) |
| 194 | if (ops->exit) |
| 195 | ops->exit(net); |
| 196 | } |
| 197 | |
| 198 | /** |
| 199 | * register_pernet_subsys - register a network namespace subsystem |
| 200 | * @ops: pernet operations structure for the subsystem |
| 201 | * |
| 202 | * Register a subsystem which has init and exit functions |
| 203 | * that are called when network namespaces are created and |
| 204 | * destroyed respectively. |
| 205 | * |
| 206 | * When registered all network namespace init functions are |
| 207 | * called for every existing network namespace. Allowing kernel |
| 208 | * modules to have a race free view of the set of network namespaces. |
| 209 | * |
| 210 | * When a new network namespace is created all of the init |
| 211 | * methods are called in the order in which they were registered. |
| 212 | * |
| 213 | * When a network namespace is destroyed all of the exit methods |
| 214 | * are called in the reverse of the order with which they were |
| 215 | * registered. |
| 216 | */ |
| 217 | int register_pernet_subsys(struct pernet_operations *ops) |
| 218 | { |
| 219 | int error; |
| 220 | mutex_lock(&net_mutex); |
| 221 | error = register_pernet_operations(first_device, ops); |
| 222 | mutex_unlock(&net_mutex); |
| 223 | return error; |
| 224 | } |
| 225 | EXPORT_SYMBOL_GPL(register_pernet_subsys); |
| 226 | |
| 227 | /** |
| 228 | * unregister_pernet_subsys - unregister a network namespace subsystem |
| 229 | * @ops: pernet operations structure to manipulate |
| 230 | * |
| 231 | * Remove the pernet operations structure from the list to be |
| 232 | * used when network namespaces are created or destoryed. In |
| 233 | * addition run the exit method for all existing network |
| 234 | * namespaces. |
| 235 | */ |
| 236 | void unregister_pernet_subsys(struct pernet_operations *module) |
| 237 | { |
| 238 | mutex_lock(&net_mutex); |
| 239 | unregister_pernet_operations(module); |
| 240 | mutex_unlock(&net_mutex); |
| 241 | } |
| 242 | EXPORT_SYMBOL_GPL(unregister_pernet_subsys); |
| 243 | |
| 244 | /** |
| 245 | * register_pernet_device - register a network namespace device |
| 246 | * @ops: pernet operations structure for the subsystem |
| 247 | * |
| 248 | * Register a device which has init and exit functions |
| 249 | * that are called when network namespaces are created and |
| 250 | * destroyed respectively. |
| 251 | * |
| 252 | * When registered all network namespace init functions are |
| 253 | * called for every existing network namespace. Allowing kernel |
| 254 | * modules to have a race free view of the set of network namespaces. |
| 255 | * |
| 256 | * When a new network namespace is created all of the init |
| 257 | * methods are called in the order in which they were registered. |
| 258 | * |
| 259 | * When a network namespace is destroyed all of the exit methods |
| 260 | * are called in the reverse of the order with which they were |
| 261 | * registered. |
| 262 | */ |
| 263 | int register_pernet_device(struct pernet_operations *ops) |
| 264 | { |
| 265 | int error; |
| 266 | mutex_lock(&net_mutex); |
| 267 | error = register_pernet_operations(&pernet_list, ops); |
| 268 | if (!error && (first_device == &pernet_list)) |
| 269 | first_device = &ops->list; |
| 270 | mutex_unlock(&net_mutex); |
| 271 | return error; |
| 272 | } |
| 273 | EXPORT_SYMBOL_GPL(register_pernet_device); |
| 274 | |
| 275 | /** |
| 276 | * unregister_pernet_device - unregister a network namespace netdevice |
| 277 | * @ops: pernet operations structure to manipulate |
| 278 | * |
| 279 | * Remove the pernet operations structure from the list to be |
| 280 | * used when network namespaces are created or destoryed. In |
| 281 | * addition run the exit method for all existing network |
| 282 | * namespaces. |
| 283 | */ |
| 284 | void unregister_pernet_device(struct pernet_operations *ops) |
| 285 | { |
| 286 | mutex_lock(&net_mutex); |
| 287 | if (&ops->list == first_device) |
| 288 | first_device = first_device->next; |
| 289 | unregister_pernet_operations(ops); |
| 290 | mutex_unlock(&net_mutex); |
| 291 | } |
| 292 | EXPORT_SYMBOL_GPL(unregister_pernet_device); |