include/net/sch_generic.h

   1 #ifndef __NET_SCHED_GENERIC_H
   2 #define __NET_SCHED_GENERIC_H
   3
   4 #include <linux/netdevice.h>
   5 #include <linux/types.h>
   6 #include <linux/rcupdate.h>
   7 #include <linux/module.h>
   8 #include <linux/pkt_sched.h>
   9 #include <linux/pkt_cls.h>
  10 #include <net/gen_stats.h>
  11 #include <net/rtnetlink.h>
  12
  13 struct Qdisc_ops;
  14 struct qdisc_walker;
  15 struct tcf_walker;
  16 struct module;
  17
  18 struct qdisc_rate_table
  19 {
  20         struct tc_ratespec rate;
  21         u32             data[256];
  22         struct qdisc_rate_table *next;
  23         int             refcnt;
  24 };
  25
  26 enum qdisc_state_t
  27 {
  28         __QDISC_STATE_RUNNING,
  29         __QDISC_STATE_SCHED,
  30         __QDISC_STATE_DEACTIVATED,
  31 };
  32
  33 struct qdisc_size_table {
  34         struct list_head        list;
  35         struct tc_sizespec      szopts;
  36         int                     refcnt;
  37         u16                     data[];
  38 };
  39
  40 struct Qdisc
  41 {
  42         int                     (*enqueue)(struct sk_buff *skb, struct Qdisc *dev);
  43         struct sk_buff *        (*dequeue)(struct Qdisc *dev);
  44         unsigned                flags;
  45 #define TCQ_F_BUILTIN   1
  46 #define TCQ_F_THROTTLED 2
  47 #define TCQ_F_INGRESS   4
  48         int                     padded;
  49         struct Qdisc_ops        *ops;
  50         struct qdisc_size_table *stab;
  51         u32                     handle;
  52         u32                     parent;
  53         atomic_t                refcnt;
  54         unsigned long           state;
  55         struct sk_buff          *gso_skb;
  56         struct sk_buff_head     q;
  57         struct netdev_queue     *dev_queue;
  58         struct Qdisc            *next_sched;
  59         struct list_head        list;
  60
  61         struct gnet_stats_basic bstats;
  62         struct gnet_stats_queue qstats;
  63         struct gnet_stats_rate_est      rate_est;
  64         int                     (*reshape_fail)(struct sk_buff *skb,
  65                                         struct Qdisc *q);
  66
  67         void                    *u32_node;
  68
  69         /* This field is deprecated, but it is still used by CBQ
  70          * and it will live until better solution will be invented.
  71          */
  72         struct Qdisc            *__parent;
  73 };
  74
  75 struct Qdisc_class_ops
  76 {
  77         /* Child qdisc manipulation */
  78         int                     (*graft)(struct Qdisc *, unsigned long cl,
  79                                         struct Qdisc *, struct Qdisc **);
  80         struct Qdisc *          (*leaf)(struct Qdisc *, unsigned long cl);
  81         void                    (*qlen_notify)(struct Qdisc *, unsigned long);
  82
  83         /* Class manipulation routines */
  84         unsigned long           (*get)(struct Qdisc *, u32 classid);
  85         void                    (*put)(struct Qdisc *, unsigned long);
  86         int                     (*change)(struct Qdisc *, u32, u32,
  87                                         struct nlattr **, unsigned long *);
  88         int                     (*delete)(struct Qdisc *, unsigned long);
  89         void                    (*walk)(struct Qdisc *, struct qdisc_walker * arg);
  90
  91         /* Filter manipulation */
  92         struct tcf_proto **     (*tcf_chain)(struct Qdisc *, unsigned long);
  93         unsigned long           (*bind_tcf)(struct Qdisc *, unsigned long,
  94                                         u32 classid);
  95         void                    (*unbind_tcf)(struct Qdisc *, unsigned long);
  96
  97         /* rtnetlink specific */
  98         int                     (*dump)(struct Qdisc *, unsigned long,
  99                                         struct sk_buff *skb, struct tcmsg*);
 100         int                     (*dump_stats)(struct Qdisc *, unsigned long,
 101                                         struct gnet_dump *);
 102 };
 103
 104 struct Qdisc_ops
 105 {
 106         struct Qdisc_ops        *next;
 107         const struct Qdisc_class_ops    *cl_ops;
 108         char                    id[IFNAMSIZ];
 109         int                     priv_size;
 110
 111         int                     (*enqueue)(struct sk_buff *, struct Qdisc *);
 112         struct sk_buff *        (*dequeue)(struct Qdisc *);
 113         struct sk_buff *        (*peek)(struct Qdisc *);
 114         unsigned int            (*drop)(struct Qdisc *);
 115
 116         int                     (*init)(struct Qdisc *, struct nlattr *arg);
 117         void                    (*reset)(struct Qdisc *);
 118         void                    (*destroy)(struct Qdisc *);
 119         int                     (*change)(struct Qdisc *, struct nlattr *arg);
 120
 121         int                     (*dump)(struct Qdisc *, struct sk_buff *);
 122         int                     (*dump_stats)(struct Qdisc *, struct gnet_dump *);
 123
 124         struct module           *owner;
 125 };
 126
 127
 128 struct tcf_result
 129 {
 130         unsigned long   class;
 131         u32             classid;
 132 };
 133
 134 struct tcf_proto_ops
 135 {
 136         struct tcf_proto_ops    *next;
 137         char                    kind[IFNAMSIZ];
 138
 139         int                     (*classify)(struct sk_buff*, struct tcf_proto*,
 140                                         struct tcf_result *);
 141         int                     (*init)(struct tcf_proto*);
 142         void                    (*destroy)(struct tcf_proto*);
 143
 144         unsigned long           (*get)(struct tcf_proto*, u32 handle);
 145         void                    (*put)(struct tcf_proto*, unsigned long);
 146         int                     (*change)(struct tcf_proto*, unsigned long,
 147                                         u32 handle, struct nlattr **,
 148                                         unsigned long *);
 149         int                     (*delete)(struct tcf_proto*, unsigned long);
 150         void                    (*walk)(struct tcf_proto*, struct tcf_walker *arg);
 151
 152         /* rtnetlink specific */
 153         int                     (*dump)(struct tcf_proto*, unsigned long,
 154                                         struct sk_buff *skb, struct tcmsg*);
 155
 156         struct module           *owner;
 157 };
 158
 159 struct tcf_proto
 160 {
 161         /* Fast access part */
 162         struct tcf_proto        *next;
 163         void                    *root;
 164         int                     (*classify)(struct sk_buff*, struct tcf_proto*,
 165                                         struct tcf_result *);
 166         __be16                  protocol;
 167
 168         /* All the rest */
 169         u32                     prio;
 170         u32                     classid;
 171         struct Qdisc            *q;
 172         void                    *data;
 173         struct tcf_proto_ops    *ops;
 174 };
 175
 176 struct qdisc_skb_cb {
 177         unsigned int            pkt_len;
 178         char                    data[];
 179 };
 180
 181 static inline struct qdisc_skb_cb *qdisc_skb_cb(struct sk_buff *skb)
 182 {
 183         return (struct qdisc_skb_cb *)skb->cb;
 184 }
 185
 186 static inline spinlock_t *qdisc_lock(struct Qdisc *qdisc)
 187 {
 188         return &qdisc->q.lock;
 189 }
 190
 191 static inline struct Qdisc *qdisc_root(struct Qdisc *qdisc)
 192 {
 193         return qdisc->dev_queue->qdisc;
 194 }
 195
 196 static inline struct Qdisc *qdisc_root_sleeping(struct Qdisc *qdisc)
 197 {
 198         return qdisc->dev_queue->qdisc_sleeping;
 199 }
 200
 201 /* The qdisc root lock is a mechanism by which to top level
 202  * of a qdisc tree can be locked from any qdisc node in the
 203  * forest.  This allows changing the configuration of some
 204  * aspect of the qdisc tree while blocking out asynchronous
 205  * qdisc access in the packet processing paths.
 206  *
 207  * It is only legal to do this when the root will not change
 208  * on us.  Otherwise we'll potentially lock the wrong qdisc
 209  * root.  This is enforced by holding the RTNL semaphore, which
 210  * all users of this lock accessor must do.
 211  */
 212 static inline spinlock_t *qdisc_root_lock(struct Qdisc *qdisc)
 213 {
 214         struct Qdisc *root = qdisc_root(qdisc);
 215
 216         ASSERT_RTNL();
 217         return qdisc_lock(root);
 218 }
 219
 220 static inline spinlock_t *qdisc_root_sleeping_lock(struct Qdisc *qdisc)
 221 {
 222         struct Qdisc *root = qdisc_root_sleeping(qdisc);
 223
 224         ASSERT_RTNL();
 225         return qdisc_lock(root);
 226 }
 227
 228 static inline struct net_device *qdisc_dev(struct Qdisc *qdisc)
 229 {
 230         return qdisc->dev_queue->dev;
 231 }
 232
 233 static inline void sch_tree_lock(struct Qdisc *q)
 234 {
 235         spin_lock_bh(qdisc_root_sleeping_lock(q));
 236 }
 237
 238 static inline void sch_tree_unlock(struct Qdisc *q)
 239 {
 240         spin_unlock_bh(qdisc_root_sleeping_lock(q));
 241 }
 242
 243 #define tcf_tree_lock(tp)       sch_tree_lock((tp)->q)
 244 #define tcf_tree_unlock(tp)     sch_tree_unlock((tp)->q)
 245
 246 extern struct Qdisc noop_qdisc;
 247 extern struct Qdisc_ops noop_qdisc_ops;
 248
 249 struct Qdisc_class_common
 250 {
 251         u32                     classid;
 252         struct hlist_node       hnode;
 253 };
 254
 255 struct Qdisc_class_hash
 256 {
 257         struct hlist_head       *hash;
 258         unsigned int            hashsize;
 259         unsigned int            hashmask;
 260         unsigned int            hashelems;
 261 };
 262
 263 static inline unsigned int qdisc_class_hash(u32 id, u32 mask)
 264 {
 265         id ^= id >> 8;
 266         id ^= id >> 4;
 267         return id & mask;
 268 }
 269
 270 static inline struct Qdisc_class_common *
 271 qdisc_class_find(struct Qdisc_class_hash *hash, u32 id)
 272 {
 273         struct Qdisc_class_common *cl;
 274         struct hlist_node *n;
 275         unsigned int h;
 276
 277         h = qdisc_class_hash(id, hash->hashmask);
 278         hlist_for_each_entry(cl, n, &hash->hash[h], hnode) {
 279                 if (cl->classid == id)
 280                         return cl;
 281         }
 282         return NULL;
 283 }
 284
 285 extern int qdisc_class_hash_init(struct Qdisc_class_hash *);
 286 extern void qdisc_class_hash_insert(struct Qdisc_class_hash *, struct Qdisc_class_common *);
 287 extern void qdisc_class_hash_remove(struct Qdisc_class_hash *, struct Qdisc_class_common *);
 288 extern void qdisc_class_hash_grow(struct Qdisc *, struct Qdisc_class_hash *);
 289 extern void qdisc_class_hash_destroy(struct Qdisc_class_hash *);
 290
 291 extern void dev_init_scheduler(struct net_device *dev);
 292 extern void dev_shutdown(struct net_device *dev);
 293 extern void dev_activate(struct net_device *dev);
 294 extern void dev_deactivate(struct net_device *dev);
 295 extern void qdisc_reset(struct Qdisc *qdisc);
 296 extern void qdisc_destroy(struct Qdisc *qdisc);
 297 extern void qdisc_tree_decrease_qlen(struct Qdisc *qdisc, unsigned int n);
 298 extern struct Qdisc *qdisc_alloc(struct netdev_queue *dev_queue,
 299                                  struct Qdisc_ops *ops);
 300 extern struct Qdisc *qdisc_create_dflt(struct net_device *dev,
 301                                        struct netdev_queue *dev_queue,
 302                                        struct Qdisc_ops *ops, u32 parentid);
 303 extern void qdisc_calculate_pkt_len(struct sk_buff *skb,
 304                                    struct qdisc_size_table *stab);
 305 extern void tcf_destroy(struct tcf_proto *tp);
 306 extern void tcf_destroy_chain(struct tcf_proto **fl);
 307
 308 /* Reset all TX qdiscs of a device.  */
 309 static inline void qdisc_reset_all_tx(struct net_device *dev)
 310 {
 311         unsigned int i;
 312         for (i = 0; i < dev->num_tx_queues; i++)
 313                 qdisc_reset(netdev_get_tx_queue(dev, i)->qdisc);
 314 }
 315
 316 /* Are all TX queues of the device empty?  */
 317 static inline bool qdisc_all_tx_empty(const struct net_device *dev)
 318 {
 319         unsigned int i;
 320         for (i = 0; i < dev->num_tx_queues; i++) {
 321                 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 322                 const struct Qdisc *q = txq->qdisc;
 323
 324                 if (q->q.qlen)
 325                         return false;
 326         }
 327         return true;
 328 }
 329
 330 /* Are any of the TX qdiscs changing?  */
 331 static inline bool qdisc_tx_changing(struct net_device *dev)
 332 {
 333         unsigned int i;
 334         for (i = 0; i < dev->num_tx_queues; i++) {
 335                 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 336                 if (txq->qdisc != txq->qdisc_sleeping)
 337                         return true;
 338         }
 339         return false;
 340 }
 341
 342 /* Is the device using the noop qdisc on all queues?  */
 343 static inline bool qdisc_tx_is_noop(const struct net_device *dev)
 344 {
 345         unsigned int i;
 346         for (i = 0; i < dev->num_tx_queues; i++) {
 347                 struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
 348                 if (txq->qdisc != &noop_qdisc)
 349                         return false;
 350         }
 351         return true;
 352 }
 353
 354 static inline unsigned int qdisc_pkt_len(struct sk_buff *skb)
 355 {
 356         return qdisc_skb_cb(skb)->pkt_len;
 357 }
 358
 359 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
 360 enum net_xmit_qdisc_t {
 361         __NET_XMIT_STOLEN = 0x00010000,
 362         __NET_XMIT_BYPASS = 0x00020000,
 363 };
 364
 365 #ifdef CONFIG_NET_CLS_ACT
 366 #define net_xmit_drop_count(e)  ((e) & __NET_XMIT_STOLEN ? 0 : 1)
 367 #else
 368 #define net_xmit_drop_count(e)  (1)
 369 #endif
 370
 371 static inline int qdisc_enqueue(struct sk_buff *skb, struct Qdisc *sch)
 372 {
 373 #ifdef CONFIG_NET_SCHED
 374         if (sch->stab)
 375                 qdisc_calculate_pkt_len(skb, sch->stab);
 376 #endif
 377         return sch->enqueue(skb, sch);
 378 }
 379
 380 static inline int qdisc_enqueue_root(struct sk_buff *skb, struct Qdisc *sch)
 381 {
 382         qdisc_skb_cb(skb)->pkt_len = skb->len;
 383         return qdisc_enqueue(skb, sch) & NET_XMIT_MASK;
 384 }
 385
 386 static inline int __qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch,
 387                                        struct sk_buff_head *list)
 388 {
 389         __skb_queue_tail(list, skb);
 390         sch->qstats.backlog += qdisc_pkt_len(skb);
 391         sch->bstats.bytes += qdisc_pkt_len(skb);
 392         sch->bstats.packets++;
 393
 394         return NET_XMIT_SUCCESS;
 395 }
 396
 397 static inline int qdisc_enqueue_tail(struct sk_buff *skb, struct Qdisc *sch)
 398 {
 399         return __qdisc_enqueue_tail(skb, sch, &sch->q);
 400 }
 401
 402 static inline struct sk_buff *__qdisc_dequeue_head(struct Qdisc *sch,
 403                                                    struct sk_buff_head *list)
 404 {
 405         struct sk_buff *skb = __skb_dequeue(list);
 406
 407         if (likely(skb != NULL))
 408                 sch->qstats.backlog -= qdisc_pkt_len(skb);
 409
 410         return skb;
 411 }
 412
 413 static inline struct sk_buff *qdisc_dequeue_head(struct Qdisc *sch)
 414 {
 415         return __qdisc_dequeue_head(sch, &sch->q);
 416 }
 417
 418 static inline struct sk_buff *__qdisc_dequeue_tail(struct Qdisc *sch,
 419                                                    struct sk_buff_head *list)
 420 {
 421         struct sk_buff *skb = __skb_dequeue_tail(list);
 422
 423         if (likely(skb != NULL))
 424                 sch->qstats.backlog -= qdisc_pkt_len(skb);
 425
 426         return skb;
 427 }
 428
 429 static inline struct sk_buff *qdisc_dequeue_tail(struct Qdisc *sch)
 430 {
 431         return __qdisc_dequeue_tail(sch, &sch->q);
 432 }
 433
 434 static inline struct sk_buff *qdisc_peek_head(struct Qdisc *sch)
 435 {
 436         return skb_peek(&sch->q);
 437 }
 438
 439 /* generic pseudo peek method for non-work-conserving qdisc */
 440 static inline struct sk_buff *qdisc_peek_dequeued(struct Qdisc *sch)
 441 {
 442         /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
 443         if (!sch->gso_skb) {
 444                 sch->gso_skb = sch->dequeue(sch);
 445                 if (sch->gso_skb)
 446                         /* it's still part of the queue */
 447                         sch->q.qlen++;
 448         }
 449
 450         return sch->gso_skb;
 451 }
 452
 453 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
 454 static inline struct sk_buff *qdisc_dequeue_peeked(struct Qdisc *sch)
 455 {
 456         struct sk_buff *skb = sch->gso_skb;
 457
 458         if (skb) {
 459                 sch->gso_skb = NULL;
 460                 sch->q.qlen--;
 461         } else {
 462                 skb = sch->dequeue(sch);
 463         }
 464
 465         return skb;
 466 }
 467
 468 static inline void __qdisc_reset_queue(struct Qdisc *sch,
 469                                        struct sk_buff_head *list)
 470 {
 471         /*
 472          * We do not know the backlog in bytes of this list, it
 473          * is up to the caller to correct it
 474          */
 475         __skb_queue_purge(list);
 476 }
 477
 478 static inline void qdisc_reset_queue(struct Qdisc *sch)
 479 {
 480         __qdisc_reset_queue(sch, &sch->q);
 481         sch->qstats.backlog = 0;
 482 }
 483
 484 static inline unsigned int __qdisc_queue_drop(struct Qdisc *sch,
 485                                               struct sk_buff_head *list)
 486 {
 487         struct sk_buff *skb = __qdisc_dequeue_tail(sch, list);
 488
 489         if (likely(skb != NULL)) {
 490                 unsigned int len = qdisc_pkt_len(skb);
 491                 kfree_skb(skb);
 492                 return len;
 493         }
 494
 495         return 0;
 496 }
 497
 498 static inline unsigned int qdisc_queue_drop(struct Qdisc *sch)
 499 {
 500         return __qdisc_queue_drop(sch, &sch->q);
 501 }
 502
 503 static inline int qdisc_drop(struct sk_buff *skb, struct Qdisc *sch)
 504 {
 505         kfree_skb(skb);
 506         sch->qstats.drops++;
 507
 508         return NET_XMIT_DROP;
 509 }
 510
 511 static inline int qdisc_reshape_fail(struct sk_buff *skb, struct Qdisc *sch)
 512 {
 513         sch->qstats.drops++;
 514
 515 #ifdef CONFIG_NET_CLS_ACT
 516         if (sch->reshape_fail == NULL || sch->reshape_fail(skb, sch))
 517                 goto drop;
 518
 519         return NET_XMIT_SUCCESS;
 520
 521 drop:
 522 #endif
 523         kfree_skb(skb);
 524         return NET_XMIT_DROP;
 525 }
 526
 527 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
 528    long it will take to send a packet given its size.
 529  */
 530 static inline u32 qdisc_l2t(struct qdisc_rate_table* rtab, unsigned int pktlen)
 531 {
 532         int slot = pktlen + rtab->rate.cell_align + rtab->rate.overhead;
 533         if (slot < 0)
 534                 slot = 0;
 535         slot >>= rtab->rate.cell_log;
 536         if (slot > 255)
 537                 return (rtab->data[255]*(slot >> 8) + rtab->data[slot & 0xFF]);
 538         return rtab->data[slot];
 539 }
 540
 541 #ifdef CONFIG_NET_CLS_ACT
 542 static inline struct sk_buff *skb_act_clone(struct sk_buff *skb, gfp_t gfp_mask)
 543 {
 544         struct sk_buff *n = skb_clone(skb, gfp_mask);
 545
 546         if (n) {
 547                 n->tc_verd = SET_TC_VERD(n->tc_verd, 0);
 548                 n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
 549                 n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
 550         }
 551         return n;
 552 }
 553 #endif
 554
 555 #endif