1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __NET_SCHED_GENERIC_H
3 #define __NET_SCHED_GENERIC_H
5 #include <linux/netdevice.h>
6 #include <linux/types.h>
7 #include <linux/rcupdate.h>
8 #include <linux/pkt_sched.h>
9 #include <linux/pkt_cls.h>
10 #include <linux/percpu.h>
11 #include <linux/dynamic_queue_limits.h>
12 #include <linux/list.h>
13 #include <linux/refcount.h>
14 #include <linux/workqueue.h>
15 #include <linux/mutex.h>
16 #include <linux/rwsem.h>
17 #include <linux/atomic.h>
18 #include <linux/hashtable.h>
19 #include <net/gen_stats.h>
20 #include <net/rtnetlink.h>
21 #include <net/flow_offload.h>
29 struct qdisc_rate_table
{
30 struct tc_ratespec rate
;
32 struct qdisc_rate_table
*next
;
38 __QDISC_STATE_DEACTIVATED
,
41 struct qdisc_size_table
{
43 struct list_head list
;
44 struct tc_sizespec szopts
;
49 /* similar to sk_buff_head, but skb->prev pointer is undefined. */
50 struct qdisc_skb_head
{
58 int (*enqueue
)(struct sk_buff
*skb
,
60 struct sk_buff
**to_free
);
61 struct sk_buff
* (*dequeue
)(struct Qdisc
*sch
);
63 #define TCQ_F_BUILTIN 1
64 #define TCQ_F_INGRESS 2
65 #define TCQ_F_CAN_BYPASS 4
66 #define TCQ_F_MQROOT 8
67 #define TCQ_F_ONETXQUEUE 0x10 /* dequeue_skb() can assume all skbs are for
68 * q->dev_queue : It can test
69 * netif_xmit_frozen_or_stopped() before
70 * dequeueing next packet.
71 * Its true for MQ/MQPRIO slaves, or non
74 #define TCQ_F_WARN_NONWC (1 << 16)
75 #define TCQ_F_CPUSTATS 0x20 /* run using percpu statistics */
76 #define TCQ_F_NOPARENT 0x40 /* root of its hierarchy :
77 * qdisc_tree_decrease_qlen() should stop.
79 #define TCQ_F_INVISIBLE 0x80 /* invisible by default in dump */
80 #define TCQ_F_NOLOCK 0x100 /* qdisc does not require locking */
81 #define TCQ_F_OFFLOADED 0x200 /* qdisc is offloaded to HW */
83 const struct Qdisc_ops
*ops
;
84 struct qdisc_size_table __rcu
*stab
;
85 struct hlist_node hash
;
89 struct netdev_queue
*dev_queue
;
91 struct net_rate_estimator __rcu
*rate_est
;
92 struct gnet_stats_basic_cpu __percpu
*cpu_bstats
;
93 struct gnet_stats_queue __percpu
*cpu_qstats
;
98 * For performance sake on SMP, we put highly modified fields at the end
100 struct sk_buff_head gso_skb ____cacheline_aligned_in_smp
;
101 struct qdisc_skb_head q
;
102 struct gnet_stats_basic_packed bstats
;
104 struct gnet_stats_queue qstats
;
106 struct Qdisc
*next_sched
;
107 struct sk_buff_head skb_bad_txq
;
109 spinlock_t busylock ____cacheline_aligned_in_smp
;
112 /* for NOLOCK qdisc, true if there are no enqueued skbs */
117 static inline void qdisc_refcount_inc(struct Qdisc
*qdisc
)
119 if (qdisc
->flags
& TCQ_F_BUILTIN
)
121 refcount_inc(&qdisc
->refcnt
);
124 /* Intended to be used by unlocked users, when concurrent qdisc release is
128 static inline struct Qdisc
*qdisc_refcount_inc_nz(struct Qdisc
*qdisc
)
130 if (qdisc
->flags
& TCQ_F_BUILTIN
)
132 if (refcount_inc_not_zero(&qdisc
->refcnt
))
137 static inline bool qdisc_is_running(struct Qdisc
*qdisc
)
139 if (qdisc
->flags
& TCQ_F_NOLOCK
)
140 return spin_is_locked(&qdisc
->seqlock
);
141 return (raw_read_seqcount(&qdisc
->running
) & 1) ? true : false;
144 static inline bool qdisc_is_percpu_stats(const struct Qdisc
*q
)
146 return q
->flags
& TCQ_F_CPUSTATS
;
149 static inline bool qdisc_is_empty(const struct Qdisc
*qdisc
)
151 if (qdisc_is_percpu_stats(qdisc
))
152 return READ_ONCE(qdisc
->empty
);
153 return !READ_ONCE(qdisc
->q
.qlen
);
156 static inline bool qdisc_run_begin(struct Qdisc
*qdisc
)
158 if (qdisc
->flags
& TCQ_F_NOLOCK
) {
159 if (!spin_trylock(&qdisc
->seqlock
))
161 WRITE_ONCE(qdisc
->empty
, false);
162 } else if (qdisc_is_running(qdisc
)) {
165 /* Variant of write_seqcount_begin() telling lockdep a trylock
168 raw_write_seqcount_begin(&qdisc
->running
);
169 seqcount_acquire(&qdisc
->running
.dep_map
, 0, 1, _RET_IP_
);
173 static inline void qdisc_run_end(struct Qdisc
*qdisc
)
175 write_seqcount_end(&qdisc
->running
);
176 if (qdisc
->flags
& TCQ_F_NOLOCK
)
177 spin_unlock(&qdisc
->seqlock
);
180 static inline bool qdisc_may_bulk(const struct Qdisc
*qdisc
)
182 return qdisc
->flags
& TCQ_F_ONETXQUEUE
;
185 static inline int qdisc_avail_bulklimit(const struct netdev_queue
*txq
)
188 /* Non-BQL migrated drivers will return 0, too. */
189 return dql_avail(&txq
->dql
);
195 struct Qdisc_class_ops
{
197 /* Child qdisc manipulation */
198 struct netdev_queue
* (*select_queue
)(struct Qdisc
*, struct tcmsg
*);
199 int (*graft
)(struct Qdisc
*, unsigned long cl
,
200 struct Qdisc
*, struct Qdisc
**,
201 struct netlink_ext_ack
*extack
);
202 struct Qdisc
* (*leaf
)(struct Qdisc
*, unsigned long cl
);
203 void (*qlen_notify
)(struct Qdisc
*, unsigned long);
205 /* Class manipulation routines */
206 unsigned long (*find
)(struct Qdisc
*, u32 classid
);
207 int (*change
)(struct Qdisc
*, u32
, u32
,
208 struct nlattr
**, unsigned long *,
209 struct netlink_ext_ack
*);
210 int (*delete)(struct Qdisc
*, unsigned long);
211 void (*walk
)(struct Qdisc
*, struct qdisc_walker
* arg
);
213 /* Filter manipulation */
214 struct tcf_block
* (*tcf_block
)(struct Qdisc
*sch
,
216 struct netlink_ext_ack
*extack
);
217 unsigned long (*bind_tcf
)(struct Qdisc
*, unsigned long,
219 void (*unbind_tcf
)(struct Qdisc
*, unsigned long);
221 /* rtnetlink specific */
222 int (*dump
)(struct Qdisc
*, unsigned long,
223 struct sk_buff
*skb
, struct tcmsg
*);
224 int (*dump_stats
)(struct Qdisc
*, unsigned long,
228 /* Qdisc_class_ops flag values */
230 /* Implements API that doesn't require rtnl lock */
231 enum qdisc_class_ops_flags
{
232 QDISC_CLASS_OPS_DOIT_UNLOCKED
= 1,
236 struct Qdisc_ops
*next
;
237 const struct Qdisc_class_ops
*cl_ops
;
240 unsigned int static_flags
;
242 int (*enqueue
)(struct sk_buff
*skb
,
244 struct sk_buff
**to_free
);
245 struct sk_buff
* (*dequeue
)(struct Qdisc
*);
246 struct sk_buff
* (*peek
)(struct Qdisc
*);
248 int (*init
)(struct Qdisc
*sch
, struct nlattr
*arg
,
249 struct netlink_ext_ack
*extack
);
250 void (*reset
)(struct Qdisc
*);
251 void (*destroy
)(struct Qdisc
*);
252 int (*change
)(struct Qdisc
*sch
,
254 struct netlink_ext_ack
*extack
);
255 void (*attach
)(struct Qdisc
*sch
);
256 int (*change_tx_queue_len
)(struct Qdisc
*, unsigned int);
258 int (*dump
)(struct Qdisc
*, struct sk_buff
*);
259 int (*dump_stats
)(struct Qdisc
*, struct gnet_dump
*);
261 void (*ingress_block_set
)(struct Qdisc
*sch
,
263 void (*egress_block_set
)(struct Qdisc
*sch
,
265 u32 (*ingress_block_get
)(struct Qdisc
*sch
);
266 u32 (*egress_block_get
)(struct Qdisc
*sch
);
268 struct module
*owner
;
278 const struct tcf_proto
*goto_tp
;
280 /* used in the skb_tc_reinsert function */
283 struct gnet_stats_queue
*qstats
;
290 struct tcf_proto_ops
{
291 struct list_head head
;
294 int (*classify
)(struct sk_buff
*,
295 const struct tcf_proto
*,
296 struct tcf_result
*);
297 int (*init
)(struct tcf_proto
*);
298 void (*destroy
)(struct tcf_proto
*tp
, bool rtnl_held
,
299 struct netlink_ext_ack
*extack
);
301 void* (*get
)(struct tcf_proto
*, u32 handle
);
302 void (*put
)(struct tcf_proto
*tp
, void *f
);
303 int (*change
)(struct net
*net
, struct sk_buff
*,
304 struct tcf_proto
*, unsigned long,
305 u32 handle
, struct nlattr
**,
307 struct netlink_ext_ack
*);
308 int (*delete)(struct tcf_proto
*tp
, void *arg
,
309 bool *last
, bool rtnl_held
,
310 struct netlink_ext_ack
*);
311 void (*walk
)(struct tcf_proto
*tp
,
312 struct tcf_walker
*arg
, bool rtnl_held
);
313 int (*reoffload
)(struct tcf_proto
*tp
, bool add
,
314 flow_setup_cb_t
*cb
, void *cb_priv
,
315 struct netlink_ext_ack
*extack
);
316 void (*hw_add
)(struct tcf_proto
*tp
,
318 void (*hw_del
)(struct tcf_proto
*tp
,
320 void (*bind_class
)(void *, u32
, unsigned long);
321 void * (*tmplt_create
)(struct net
*net
,
322 struct tcf_chain
*chain
,
324 struct netlink_ext_ack
*extack
);
325 void (*tmplt_destroy
)(void *tmplt_priv
);
327 /* rtnetlink specific */
328 int (*dump
)(struct net
*, struct tcf_proto
*, void *,
329 struct sk_buff
*skb
, struct tcmsg
*,
331 int (*tmplt_dump
)(struct sk_buff
*skb
,
335 struct module
*owner
;
339 enum tcf_proto_ops_flags
{
340 TCF_PROTO_OPS_DOIT_UNLOCKED
= 1,
344 /* Fast access part */
345 struct tcf_proto __rcu
*next
;
348 /* called under RCU BH lock*/
349 int (*classify
)(struct sk_buff
*,
350 const struct tcf_proto
*,
351 struct tcf_result
*);
357 const struct tcf_proto_ops
*ops
;
358 struct tcf_chain
*chain
;
359 /* Lock protects tcf_proto shared state and can be used by unlocked
360 * classifiers to protect their private data.
366 struct hlist_node destroy_ht_node
;
369 struct qdisc_skb_cb
{
371 unsigned int pkt_len
;
372 u16 slave_dev_queue_mapping
;
375 #define QDISC_CB_PRIV_LEN 20
376 unsigned char data
[QDISC_CB_PRIV_LEN
];
379 typedef void tcf_chain_head_change_t(struct tcf_proto
*tp_head
, void *priv
);
382 /* Protects filter_chain. */
383 struct mutex filter_chain_lock
;
384 struct tcf_proto __rcu
*filter_chain
;
385 struct list_head list
;
386 struct tcf_block
*block
;
387 u32 index
; /* chain index */
389 unsigned int action_refcnt
;
390 bool explicitly_created
;
392 const struct tcf_proto_ops
*tmplt_ops
;
398 /* Lock protects tcf_block and lifetime-management data of chains
399 * attached to the block (refcnt, action_refcnt, explicitly_created).
402 struct list_head chain_list
;
403 u32 index
; /* block index for shared blocks */
407 struct rw_semaphore cb_lock
; /* protects cb_list and offload counters */
408 struct flow_block flow_block
;
409 struct list_head owner_list
;
411 atomic_t offloadcnt
; /* Number of oddloaded filters */
412 unsigned int nooffloaddevcnt
; /* Number of devs unable to do offload */
413 unsigned int lockeddevcnt
; /* Number of devs that require rtnl lock. */
415 struct tcf_chain
*chain
;
416 struct list_head filter_chain_list
;
419 DECLARE_HASHTABLE(proto_destroy_ht
, 7);
420 struct mutex proto_destroy_lock
; /* Lock for proto_destroy hashtable. */
423 #ifdef CONFIG_PROVE_LOCKING
424 static inline bool lockdep_tcf_chain_is_locked(struct tcf_chain
*chain
)
426 return lockdep_is_held(&chain
->filter_chain_lock
);
429 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto
*tp
)
431 return lockdep_is_held(&tp
->lock
);
434 static inline bool lockdep_tcf_chain_is_locked(struct tcf_block
*chain
)
439 static inline bool lockdep_tcf_proto_is_locked(struct tcf_proto
*tp
)
443 #endif /* #ifdef CONFIG_PROVE_LOCKING */
445 #define tcf_chain_dereference(p, chain) \
446 rcu_dereference_protected(p, lockdep_tcf_chain_is_locked(chain))
448 #define tcf_proto_dereference(p, tp) \
449 rcu_dereference_protected(p, lockdep_tcf_proto_is_locked(tp))
451 static inline void qdisc_cb_private_validate(const struct sk_buff
*skb
, int sz
)
453 struct qdisc_skb_cb
*qcb
;
455 BUILD_BUG_ON(sizeof(skb
->cb
) < offsetof(struct qdisc_skb_cb
, data
) + sz
);
456 BUILD_BUG_ON(sizeof(qcb
->data
) < sz
);
459 static inline int qdisc_qlen_cpu(const struct Qdisc
*q
)
461 return this_cpu_ptr(q
->cpu_qstats
)->qlen
;
464 static inline int qdisc_qlen(const struct Qdisc
*q
)
469 static inline int qdisc_qlen_sum(const struct Qdisc
*q
)
471 __u32 qlen
= q
->qstats
.qlen
;
474 if (qdisc_is_percpu_stats(q
)) {
475 for_each_possible_cpu(i
)
476 qlen
+= per_cpu_ptr(q
->cpu_qstats
, i
)->qlen
;
484 static inline struct qdisc_skb_cb
*qdisc_skb_cb(const struct sk_buff
*skb
)
486 return (struct qdisc_skb_cb
*)skb
->cb
;
489 static inline spinlock_t
*qdisc_lock(struct Qdisc
*qdisc
)
491 return &qdisc
->q
.lock
;
494 static inline struct Qdisc
*qdisc_root(const struct Qdisc
*qdisc
)
496 struct Qdisc
*q
= rcu_dereference_rtnl(qdisc
->dev_queue
->qdisc
);
501 static inline struct Qdisc
*qdisc_root_bh(const struct Qdisc
*qdisc
)
503 return rcu_dereference_bh(qdisc
->dev_queue
->qdisc
);
506 static inline struct Qdisc
*qdisc_root_sleeping(const struct Qdisc
*qdisc
)
508 return qdisc
->dev_queue
->qdisc_sleeping
;
511 /* The qdisc root lock is a mechanism by which to top level
512 * of a qdisc tree can be locked from any qdisc node in the
513 * forest. This allows changing the configuration of some
514 * aspect of the qdisc tree while blocking out asynchronous
515 * qdisc access in the packet processing paths.
517 * It is only legal to do this when the root will not change
518 * on us. Otherwise we'll potentially lock the wrong qdisc
519 * root. This is enforced by holding the RTNL semaphore, which
520 * all users of this lock accessor must do.
522 static inline spinlock_t
*qdisc_root_lock(const struct Qdisc
*qdisc
)
524 struct Qdisc
*root
= qdisc_root(qdisc
);
527 return qdisc_lock(root
);
530 static inline spinlock_t
*qdisc_root_sleeping_lock(const struct Qdisc
*qdisc
)
532 struct Qdisc
*root
= qdisc_root_sleeping(qdisc
);
535 return qdisc_lock(root
);
538 static inline seqcount_t
*qdisc_root_sleeping_running(const struct Qdisc
*qdisc
)
540 struct Qdisc
*root
= qdisc_root_sleeping(qdisc
);
543 return &root
->running
;
546 static inline struct net_device
*qdisc_dev(const struct Qdisc
*qdisc
)
548 return qdisc
->dev_queue
->dev
;
551 static inline void sch_tree_lock(const struct Qdisc
*q
)
553 spin_lock_bh(qdisc_root_sleeping_lock(q
));
556 static inline void sch_tree_unlock(const struct Qdisc
*q
)
558 spin_unlock_bh(qdisc_root_sleeping_lock(q
));
561 extern struct Qdisc noop_qdisc
;
562 extern struct Qdisc_ops noop_qdisc_ops
;
563 extern struct Qdisc_ops pfifo_fast_ops
;
564 extern struct Qdisc_ops mq_qdisc_ops
;
565 extern struct Qdisc_ops noqueue_qdisc_ops
;
566 extern const struct Qdisc_ops
*default_qdisc_ops
;
567 static inline const struct Qdisc_ops
*
568 get_default_qdisc_ops(const struct net_device
*dev
, int ntx
)
570 return ntx
< dev
->real_num_tx_queues
?
571 default_qdisc_ops
: &pfifo_fast_ops
;
574 struct Qdisc_class_common
{
576 struct hlist_node hnode
;
579 struct Qdisc_class_hash
{
580 struct hlist_head
*hash
;
581 unsigned int hashsize
;
582 unsigned int hashmask
;
583 unsigned int hashelems
;
586 static inline unsigned int qdisc_class_hash(u32 id
, u32 mask
)
593 static inline struct Qdisc_class_common
*
594 qdisc_class_find(const struct Qdisc_class_hash
*hash
, u32 id
)
596 struct Qdisc_class_common
*cl
;
602 h
= qdisc_class_hash(id
, hash
->hashmask
);
603 hlist_for_each_entry(cl
, &hash
->hash
[h
], hnode
) {
604 if (cl
->classid
== id
)
610 static inline int tc_classid_to_hwtc(struct net_device
*dev
, u32 classid
)
612 u32 hwtc
= TC_H_MIN(classid
) - TC_H_MIN_PRIORITY
;
614 return (hwtc
< netdev_get_num_tc(dev
)) ? hwtc
: -EINVAL
;
617 int qdisc_class_hash_init(struct Qdisc_class_hash
*);
618 void qdisc_class_hash_insert(struct Qdisc_class_hash
*,
619 struct Qdisc_class_common
*);
620 void qdisc_class_hash_remove(struct Qdisc_class_hash
*,
621 struct Qdisc_class_common
*);
622 void qdisc_class_hash_grow(struct Qdisc
*, struct Qdisc_class_hash
*);
623 void qdisc_class_hash_destroy(struct Qdisc_class_hash
*);
625 int dev_qdisc_change_tx_queue_len(struct net_device
*dev
);
626 void dev_init_scheduler(struct net_device
*dev
);
627 void dev_shutdown(struct net_device
*dev
);
628 void dev_activate(struct net_device
*dev
);
629 void dev_deactivate(struct net_device
*dev
);
630 void dev_deactivate_many(struct list_head
*head
);
631 struct Qdisc
*dev_graft_qdisc(struct netdev_queue
*dev_queue
,
632 struct Qdisc
*qdisc
);
633 void qdisc_reset(struct Qdisc
*qdisc
);
634 void qdisc_put(struct Qdisc
*qdisc
);
635 void qdisc_put_unlocked(struct Qdisc
*qdisc
);
636 void qdisc_tree_reduce_backlog(struct Qdisc
*qdisc
, int n
, int len
);
637 #ifdef CONFIG_NET_SCHED
638 int qdisc_offload_dump_helper(struct Qdisc
*q
, enum tc_setup_type type
,
640 void qdisc_offload_graft_helper(struct net_device
*dev
, struct Qdisc
*sch
,
641 struct Qdisc
*new, struct Qdisc
*old
,
642 enum tc_setup_type type
, void *type_data
,
643 struct netlink_ext_ack
*extack
);
646 qdisc_offload_dump_helper(struct Qdisc
*q
, enum tc_setup_type type
,
649 q
->flags
&= ~TCQ_F_OFFLOADED
;
654 qdisc_offload_graft_helper(struct net_device
*dev
, struct Qdisc
*sch
,
655 struct Qdisc
*new, struct Qdisc
*old
,
656 enum tc_setup_type type
, void *type_data
,
657 struct netlink_ext_ack
*extack
)
661 struct Qdisc
*qdisc_alloc(struct netdev_queue
*dev_queue
,
662 const struct Qdisc_ops
*ops
,
663 struct netlink_ext_ack
*extack
);
664 void qdisc_free(struct Qdisc
*qdisc
);
665 struct Qdisc
*qdisc_create_dflt(struct netdev_queue
*dev_queue
,
666 const struct Qdisc_ops
*ops
, u32 parentid
,
667 struct netlink_ext_ack
*extack
);
668 void __qdisc_calculate_pkt_len(struct sk_buff
*skb
,
669 const struct qdisc_size_table
*stab
);
670 int skb_do_redirect(struct sk_buff
*);
672 static inline void skb_reset_tc(struct sk_buff
*skb
)
674 #ifdef CONFIG_NET_CLS_ACT
675 skb
->tc_redirected
= 0;
679 static inline bool skb_is_tc_redirected(const struct sk_buff
*skb
)
681 #ifdef CONFIG_NET_CLS_ACT
682 return skb
->tc_redirected
;
688 static inline bool skb_at_tc_ingress(const struct sk_buff
*skb
)
690 #ifdef CONFIG_NET_CLS_ACT
691 return skb
->tc_at_ingress
;
697 static inline bool skb_skip_tc_classify(struct sk_buff
*skb
)
699 #ifdef CONFIG_NET_CLS_ACT
700 if (skb
->tc_skip_classify
) {
701 skb
->tc_skip_classify
= 0;
708 /* Reset all TX qdiscs greater than index of a device. */
709 static inline void qdisc_reset_all_tx_gt(struct net_device
*dev
, unsigned int i
)
713 for (; i
< dev
->num_tx_queues
; i
++) {
714 qdisc
= rtnl_dereference(netdev_get_tx_queue(dev
, i
)->qdisc
);
716 spin_lock_bh(qdisc_lock(qdisc
));
718 spin_unlock_bh(qdisc_lock(qdisc
));
723 static inline void qdisc_reset_all_tx(struct net_device
*dev
)
725 qdisc_reset_all_tx_gt(dev
, 0);
728 /* Are all TX queues of the device empty? */
729 static inline bool qdisc_all_tx_empty(const struct net_device
*dev
)
734 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
735 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
736 const struct Qdisc
*q
= rcu_dereference(txq
->qdisc
);
738 if (!qdisc_is_empty(q
)) {
747 /* Are any of the TX qdiscs changing? */
748 static inline bool qdisc_tx_changing(const struct net_device
*dev
)
752 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
753 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
754 if (rcu_access_pointer(txq
->qdisc
) != txq
->qdisc_sleeping
)
760 /* Is the device using the noop qdisc on all queues? */
761 static inline bool qdisc_tx_is_noop(const struct net_device
*dev
)
765 for (i
= 0; i
< dev
->num_tx_queues
; i
++) {
766 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, i
);
767 if (rcu_access_pointer(txq
->qdisc
) != &noop_qdisc
)
773 static inline unsigned int qdisc_pkt_len(const struct sk_buff
*skb
)
775 return qdisc_skb_cb(skb
)->pkt_len
;
778 /* additional qdisc xmit flags (NET_XMIT_MASK in linux/netdevice.h) */
779 enum net_xmit_qdisc_t
{
780 __NET_XMIT_STOLEN
= 0x00010000,
781 __NET_XMIT_BYPASS
= 0x00020000,
784 #ifdef CONFIG_NET_CLS_ACT
785 #define net_xmit_drop_count(e) ((e) & __NET_XMIT_STOLEN ? 0 : 1)
787 #define net_xmit_drop_count(e) (1)
790 static inline void qdisc_calculate_pkt_len(struct sk_buff
*skb
,
791 const struct Qdisc
*sch
)
793 #ifdef CONFIG_NET_SCHED
794 struct qdisc_size_table
*stab
= rcu_dereference_bh(sch
->stab
);
797 __qdisc_calculate_pkt_len(skb
, stab
);
801 static inline int qdisc_enqueue(struct sk_buff
*skb
, struct Qdisc
*sch
,
802 struct sk_buff
**to_free
)
804 qdisc_calculate_pkt_len(skb
, sch
);
805 return sch
->enqueue(skb
, sch
, to_free
);
808 static inline void _bstats_update(struct gnet_stats_basic_packed
*bstats
,
809 __u64 bytes
, __u32 packets
)
811 bstats
->bytes
+= bytes
;
812 bstats
->packets
+= packets
;
815 static inline void bstats_update(struct gnet_stats_basic_packed
*bstats
,
816 const struct sk_buff
*skb
)
818 _bstats_update(bstats
,
820 skb_is_gso(skb
) ? skb_shinfo(skb
)->gso_segs
: 1);
823 static inline void _bstats_cpu_update(struct gnet_stats_basic_cpu
*bstats
,
824 __u64 bytes
, __u32 packets
)
826 u64_stats_update_begin(&bstats
->syncp
);
827 _bstats_update(&bstats
->bstats
, bytes
, packets
);
828 u64_stats_update_end(&bstats
->syncp
);
831 static inline void bstats_cpu_update(struct gnet_stats_basic_cpu
*bstats
,
832 const struct sk_buff
*skb
)
834 u64_stats_update_begin(&bstats
->syncp
);
835 bstats_update(&bstats
->bstats
, skb
);
836 u64_stats_update_end(&bstats
->syncp
);
839 static inline void qdisc_bstats_cpu_update(struct Qdisc
*sch
,
840 const struct sk_buff
*skb
)
842 bstats_cpu_update(this_cpu_ptr(sch
->cpu_bstats
), skb
);
845 static inline void qdisc_bstats_update(struct Qdisc
*sch
,
846 const struct sk_buff
*skb
)
848 bstats_update(&sch
->bstats
, skb
);
851 static inline void qdisc_qstats_backlog_dec(struct Qdisc
*sch
,
852 const struct sk_buff
*skb
)
854 sch
->qstats
.backlog
-= qdisc_pkt_len(skb
);
857 static inline void qdisc_qstats_cpu_backlog_dec(struct Qdisc
*sch
,
858 const struct sk_buff
*skb
)
860 this_cpu_sub(sch
->cpu_qstats
->backlog
, qdisc_pkt_len(skb
));
863 static inline void qdisc_qstats_backlog_inc(struct Qdisc
*sch
,
864 const struct sk_buff
*skb
)
866 sch
->qstats
.backlog
+= qdisc_pkt_len(skb
);
869 static inline void qdisc_qstats_cpu_backlog_inc(struct Qdisc
*sch
,
870 const struct sk_buff
*skb
)
872 this_cpu_add(sch
->cpu_qstats
->backlog
, qdisc_pkt_len(skb
));
875 static inline void qdisc_qstats_cpu_qlen_inc(struct Qdisc
*sch
)
877 this_cpu_inc(sch
->cpu_qstats
->qlen
);
880 static inline void qdisc_qstats_cpu_qlen_dec(struct Qdisc
*sch
)
882 this_cpu_dec(sch
->cpu_qstats
->qlen
);
885 static inline void qdisc_qstats_cpu_requeues_inc(struct Qdisc
*sch
)
887 this_cpu_inc(sch
->cpu_qstats
->requeues
);
890 static inline void __qdisc_qstats_drop(struct Qdisc
*sch
, int count
)
892 sch
->qstats
.drops
+= count
;
895 static inline void qstats_drop_inc(struct gnet_stats_queue
*qstats
)
900 static inline void qstats_overlimit_inc(struct gnet_stats_queue
*qstats
)
902 qstats
->overlimits
++;
905 static inline void qdisc_qstats_drop(struct Qdisc
*sch
)
907 qstats_drop_inc(&sch
->qstats
);
910 static inline void qdisc_qstats_cpu_drop(struct Qdisc
*sch
)
912 this_cpu_inc(sch
->cpu_qstats
->drops
);
915 static inline void qdisc_qstats_overlimit(struct Qdisc
*sch
)
917 sch
->qstats
.overlimits
++;
920 static inline int qdisc_qstats_copy(struct gnet_dump
*d
, struct Qdisc
*sch
)
922 __u32 qlen
= qdisc_qlen_sum(sch
);
924 return gnet_stats_copy_queue(d
, sch
->cpu_qstats
, &sch
->qstats
, qlen
);
927 static inline void qdisc_qstats_qlen_backlog(struct Qdisc
*sch
, __u32
*qlen
,
930 struct gnet_stats_queue qstats
= { 0 };
931 __u32 len
= qdisc_qlen_sum(sch
);
933 __gnet_stats_copy_queue(&qstats
, sch
->cpu_qstats
, &sch
->qstats
, len
);
935 *backlog
= qstats
.backlog
;
938 static inline void qdisc_tree_flush_backlog(struct Qdisc
*sch
)
942 qdisc_qstats_qlen_backlog(sch
, &qlen
, &backlog
);
943 qdisc_tree_reduce_backlog(sch
, qlen
, backlog
);
946 static inline void qdisc_purge_queue(struct Qdisc
*sch
)
950 qdisc_qstats_qlen_backlog(sch
, &qlen
, &backlog
);
952 qdisc_tree_reduce_backlog(sch
, qlen
, backlog
);
955 static inline void qdisc_skb_head_init(struct qdisc_skb_head
*qh
)
962 static inline void __qdisc_enqueue_tail(struct sk_buff
*skb
,
963 struct qdisc_skb_head
*qh
)
965 struct sk_buff
*last
= qh
->tail
;
978 static inline int qdisc_enqueue_tail(struct sk_buff
*skb
, struct Qdisc
*sch
)
980 __qdisc_enqueue_tail(skb
, &sch
->q
);
981 qdisc_qstats_backlog_inc(sch
, skb
);
982 return NET_XMIT_SUCCESS
;
985 static inline void __qdisc_enqueue_head(struct sk_buff
*skb
,
986 struct qdisc_skb_head
*qh
)
988 skb
->next
= qh
->head
;
996 static inline struct sk_buff
*__qdisc_dequeue_head(struct qdisc_skb_head
*qh
)
998 struct sk_buff
*skb
= qh
->head
;
1000 if (likely(skb
!= NULL
)) {
1001 qh
->head
= skb
->next
;
1003 if (qh
->head
== NULL
)
1011 static inline struct sk_buff
*qdisc_dequeue_head(struct Qdisc
*sch
)
1013 struct sk_buff
*skb
= __qdisc_dequeue_head(&sch
->q
);
1015 if (likely(skb
!= NULL
)) {
1016 qdisc_qstats_backlog_dec(sch
, skb
);
1017 qdisc_bstats_update(sch
, skb
);
1023 /* Instead of calling kfree_skb() while root qdisc lock is held,
1024 * queue the skb for future freeing at end of __dev_xmit_skb()
1026 static inline void __qdisc_drop(struct sk_buff
*skb
, struct sk_buff
**to_free
)
1028 skb
->next
= *to_free
;
1032 static inline void __qdisc_drop_all(struct sk_buff
*skb
,
1033 struct sk_buff
**to_free
)
1036 skb
->prev
->next
= *to_free
;
1038 skb
->next
= *to_free
;
1042 static inline unsigned int __qdisc_queue_drop_head(struct Qdisc
*sch
,
1043 struct qdisc_skb_head
*qh
,
1044 struct sk_buff
**to_free
)
1046 struct sk_buff
*skb
= __qdisc_dequeue_head(qh
);
1048 if (likely(skb
!= NULL
)) {
1049 unsigned int len
= qdisc_pkt_len(skb
);
1051 qdisc_qstats_backlog_dec(sch
, skb
);
1052 __qdisc_drop(skb
, to_free
);
1059 static inline unsigned int qdisc_queue_drop_head(struct Qdisc
*sch
,
1060 struct sk_buff
**to_free
)
1062 return __qdisc_queue_drop_head(sch
, &sch
->q
, to_free
);
1065 static inline struct sk_buff
*qdisc_peek_head(struct Qdisc
*sch
)
1067 const struct qdisc_skb_head
*qh
= &sch
->q
;
1072 /* generic pseudo peek method for non-work-conserving qdisc */
1073 static inline struct sk_buff
*qdisc_peek_dequeued(struct Qdisc
*sch
)
1075 struct sk_buff
*skb
= skb_peek(&sch
->gso_skb
);
1077 /* we can reuse ->gso_skb because peek isn't called for root qdiscs */
1079 skb
= sch
->dequeue(sch
);
1082 __skb_queue_head(&sch
->gso_skb
, skb
);
1083 /* it's still part of the queue */
1084 qdisc_qstats_backlog_inc(sch
, skb
);
1092 static inline void qdisc_update_stats_at_dequeue(struct Qdisc
*sch
,
1093 struct sk_buff
*skb
)
1095 if (qdisc_is_percpu_stats(sch
)) {
1096 qdisc_qstats_cpu_backlog_dec(sch
, skb
);
1097 qdisc_bstats_cpu_update(sch
, skb
);
1098 qdisc_qstats_cpu_qlen_dec(sch
);
1100 qdisc_qstats_backlog_dec(sch
, skb
);
1101 qdisc_bstats_update(sch
, skb
);
1106 static inline void qdisc_update_stats_at_enqueue(struct Qdisc
*sch
,
1107 unsigned int pkt_len
)
1109 if (qdisc_is_percpu_stats(sch
)) {
1110 qdisc_qstats_cpu_qlen_inc(sch
);
1111 this_cpu_add(sch
->cpu_qstats
->backlog
, pkt_len
);
1113 sch
->qstats
.backlog
+= pkt_len
;
1118 /* use instead of qdisc->dequeue() for all qdiscs queried with ->peek() */
1119 static inline struct sk_buff
*qdisc_dequeue_peeked(struct Qdisc
*sch
)
1121 struct sk_buff
*skb
= skb_peek(&sch
->gso_skb
);
1124 skb
= __skb_dequeue(&sch
->gso_skb
);
1125 if (qdisc_is_percpu_stats(sch
)) {
1126 qdisc_qstats_cpu_backlog_dec(sch
, skb
);
1127 qdisc_qstats_cpu_qlen_dec(sch
);
1129 qdisc_qstats_backlog_dec(sch
, skb
);
1133 skb
= sch
->dequeue(sch
);
1139 static inline void __qdisc_reset_queue(struct qdisc_skb_head
*qh
)
1142 * We do not know the backlog in bytes of this list, it
1143 * is up to the caller to correct it
1147 rtnl_kfree_skbs(qh
->head
, qh
->tail
);
1155 static inline void qdisc_reset_queue(struct Qdisc
*sch
)
1157 __qdisc_reset_queue(&sch
->q
);
1158 sch
->qstats
.backlog
= 0;
1161 static inline struct Qdisc
*qdisc_replace(struct Qdisc
*sch
, struct Qdisc
*new,
1162 struct Qdisc
**pold
)
1170 qdisc_tree_flush_backlog(old
);
1171 sch_tree_unlock(sch
);
1176 static inline void rtnl_qdisc_drop(struct sk_buff
*skb
, struct Qdisc
*sch
)
1178 rtnl_kfree_skbs(skb
, skb
);
1179 qdisc_qstats_drop(sch
);
1182 static inline int qdisc_drop_cpu(struct sk_buff
*skb
, struct Qdisc
*sch
,
1183 struct sk_buff
**to_free
)
1185 __qdisc_drop(skb
, to_free
);
1186 qdisc_qstats_cpu_drop(sch
);
1188 return NET_XMIT_DROP
;
1191 static inline int qdisc_drop(struct sk_buff
*skb
, struct Qdisc
*sch
,
1192 struct sk_buff
**to_free
)
1194 __qdisc_drop(skb
, to_free
);
1195 qdisc_qstats_drop(sch
);
1197 return NET_XMIT_DROP
;
1200 static inline int qdisc_drop_all(struct sk_buff
*skb
, struct Qdisc
*sch
,
1201 struct sk_buff
**to_free
)
1203 __qdisc_drop_all(skb
, to_free
);
1204 qdisc_qstats_drop(sch
);
1206 return NET_XMIT_DROP
;
1209 /* Length to Time (L2T) lookup in a qdisc_rate_table, to determine how
1210 long it will take to send a packet given its size.
1212 static inline u32
qdisc_l2t(struct qdisc_rate_table
* rtab
, unsigned int pktlen
)
1214 int slot
= pktlen
+ rtab
->rate
.cell_align
+ rtab
->rate
.overhead
;
1217 slot
>>= rtab
->rate
.cell_log
;
1219 return rtab
->data
[255]*(slot
>> 8) + rtab
->data
[slot
& 0xFF];
1220 return rtab
->data
[slot
];
1223 struct psched_ratecfg
{
1224 u64 rate_bytes_ps
; /* bytes per second */
1231 static inline u64
psched_l2t_ns(const struct psched_ratecfg
*r
,
1236 if (unlikely(r
->linklayer
== TC_LINKLAYER_ATM
))
1237 return ((u64
)(DIV_ROUND_UP(len
,48)*53) * r
->mult
) >> r
->shift
;
1239 return ((u64
)len
* r
->mult
) >> r
->shift
;
1242 void psched_ratecfg_precompute(struct psched_ratecfg
*r
,
1243 const struct tc_ratespec
*conf
,
1246 static inline void psched_ratecfg_getrate(struct tc_ratespec
*res
,
1247 const struct psched_ratecfg
*r
)
1249 memset(res
, 0, sizeof(*res
));
1251 /* legacy struct tc_ratespec has a 32bit @rate field
1252 * Qdisc using 64bit rate should add new attributes
1253 * in order to maintain compatibility.
1255 res
->rate
= min_t(u64
, r
->rate_bytes_ps
, ~0U);
1257 res
->overhead
= r
->overhead
;
1258 res
->linklayer
= (r
->linklayer
& TC_LINKLAYER_MASK
);
1261 /* Mini Qdisc serves for specific needs of ingress/clsact Qdisc.
1262 * The fast path only needs to access filter list and to update stats
1265 struct tcf_proto
*filter_list
;
1266 struct gnet_stats_basic_cpu __percpu
*cpu_bstats
;
1267 struct gnet_stats_queue __percpu
*cpu_qstats
;
1268 struct rcu_head rcu
;
1271 static inline void mini_qdisc_bstats_cpu_update(struct mini_Qdisc
*miniq
,
1272 const struct sk_buff
*skb
)
1274 bstats_cpu_update(this_cpu_ptr(miniq
->cpu_bstats
), skb
);
1277 static inline void mini_qdisc_qstats_cpu_drop(struct mini_Qdisc
*miniq
)
1279 this_cpu_inc(miniq
->cpu_qstats
->drops
);
1282 struct mini_Qdisc_pair
{
1283 struct mini_Qdisc miniq1
;
1284 struct mini_Qdisc miniq2
;
1285 struct mini_Qdisc __rcu
**p_miniq
;
1288 void mini_qdisc_pair_swap(struct mini_Qdisc_pair
*miniqp
,
1289 struct tcf_proto
*tp_head
);
1290 void mini_qdisc_pair_init(struct mini_Qdisc_pair
*miniqp
, struct Qdisc
*qdisc
,
1291 struct mini_Qdisc __rcu
**p_miniq
);
1293 static inline int skb_tc_reinsert(struct sk_buff
*skb
, struct tcf_result
*res
)
1295 return res
->ingress
? netif_receive_skb(skb
) : dev_queue_xmit(skb
);