diff -urN ../v2.2.21-ds9/linux/Documentation/Configure.help linux/Documentation/Configure.help --- ../v2.2.21-ds9/linux/Documentation/Configure.help Tue May 21 23:26:01 2002 +++ linux/Documentation/Configure.help Sat Aug 10 10:48:07 2002 @@ -6118,6 +6118,20 @@ whenever you want). If you want to compile it as a module, say M here and read Documentation/modules.txt. +CONFIG_NET_SCH_HTB + Say Y here if you want to use the Hierarchical Token Buckets (HTB) + packet scheduling algorithm for some of your network devices. See + URL http://luxik.cdi.cz/~devik/qos/htb/ for complete manual and + in-depth articles. + + HTB is very similar to the CBQ regarding its goals however is has + different properties and different algorithm. + + This code is also available as a module called sch_htb.o ( = code + which can be inserted in and removed from the running kernel + whenever you want). If you want to compile it as a module, say M + here and read Documentation/modules.txt. + CSZ packet scheduler CONFIG_NET_SCH_CSZ Say Y here if you want to use the Clark-Shenker-Zhang (CSZ) packet diff -urN ../v2.2.21-ds9/linux/include/linux/pkt_sched.h linux/include/linux/pkt_sched.h --- ../v2.2.21-ds9/linux/include/linux/pkt_sched.h Sun Aug 4 14:54:40 2002 +++ linux/include/linux/pkt_sched.h Sat Aug 10 10:31:11 2002 @@ -246,6 +246,48 @@ }; +/* HTB section */ +#define TC_HTB_NUMPRIO 8 +#define TC_HTB_MAXDEPTH 8 +#define TC_HTB_PROTOVER 3 /* the same as HTB and TC's major */ + +struct tc_htb_opt +{ + struct tc_ratespec rate; + struct tc_ratespec ceil; + __u32 buffer; + __u32 cbuffer; + __u32 quantum; + __u32 level; /* out only */ + __u32 prio; +}; +struct tc_htb_glob +{ + __u32 version; /* to match HTB/TC */ + __u32 rate2quantum; /* bps->quantum divisor */ + __u32 defcls; /* default class number */ + __u32 debug; /* debug flags */ + + /* stats */ + __u32 direct_pkts; /* count of non shapped packets */ +}; +enum +{ + TCA_HTB_UNSPEC, + TCA_HTB_PARMS, + TCA_HTB_INIT, + TCA_HTB_CTAB, + TCA_HTB_RTAB, +}; +struct tc_htb_xstats +{ + __u32 lends; + __u32 borrows; + __u32 giants; /* too big packets (rate will not be accurate) */ + __u32 tokens; + __u32 ctokens; +}; + /* CBQ section */ #define TC_CBQ_MAXPRIO 8 diff -urN ../v2.2.21-ds9/linux/net/sched/Config.in linux/net/sched/Config.in --- ../v2.2.21-ds9/linux/net/sched/Config.in Wed Jul 31 00:00:59 2002 +++ linux/net/sched/Config.in Sat Aug 10 10:31:11 2002 @@ -4,6 +4,7 @@ define_bool CONFIG_NETLINK y define_bool CONFIG_RTNETLINK y tristate 'CBQ packet scheduler' CONFIG_NET_SCH_CBQ +tristate 'HTB packet scheduler' CONFIG_NET_SCH_HTB tristate 'CSZ packet scheduler' CONFIG_NET_SCH_CSZ #tristate 'H-PFQ packet scheduler' CONFIG_NET_SCH_HPFQ #tristate 'H-FSC packet scheduler' CONFIG_NET_SCH_HFCS diff -urN ../v2.2.21-ds9/linux/net/sched/Makefile linux/net/sched/Makefile --- ../v2.2.21-ds9/linux/net/sched/Makefile Wed Jul 31 00:00:59 2002 +++ linux/net/sched/Makefile Sat Aug 10 10:31:11 2002 @@ -68,6 +68,14 @@ endif endif +ifeq ($(CONFIG_NET_SCH_HTB), y) +O_OBJS += sch_htb.o +else + ifeq ($(CONFIG_NET_SCH_HTB), m) + M_OBJS += sch_htb.o + endif +endif + ifeq ($(CONFIG_NET_SCH_SFQ), y) O_OBJS += sch_sfq.o diff -urN ../v2.2.21-ds9/linux/net/sched/sch_htb.c linux/net/sched/sch_htb.c --- ../v2.2.21-ds9/linux/net/sched/sch_htb.c Thu Jan 1 00:00:00 1970 +++ linux/net/sched/sch_htb.c Thu Oct 24 00:45:35 2002 @@ -1,5 +1,5 @@ -/* vim: ts=8 sw=4 - * net/sched/sch_htb.c Hierarchical token bucket +/* vim: ts=8 sw=8 + * net/sched/sch_htb.c Hierarchical token bucket, feed tree version * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public License @@ -8,18 +8,19 @@ * * Authors: Martin Devera, * - * Credits (in time order): + * Credits (in time order) for older HTB versions: * Ondrej Kraus, * found missing INIT_QDISC(htb) * Vladimir Smelhaus, Aamer Akhter, Bert Hubert * helped a lot to locate nasty class stall bug * Andi Kleen, Jamal Hadi, Bert Hubert * code review and helpful comments on shaping + * Tomasz Wrona, + * created test case so that I was able to fix nasty bug * and many others. thanks. * - * $Id$ + * $Id$ */ - #include #include #include @@ -44,15 +45,17 @@ #include #include #include +#include #include #include +#include /* HTB algorithm. Author: devik@cdi.cz - ======================================= + ======================================================================== HTB is like TBF with multiple classes. It is also similar to CBQ because it allows to assign priority to each class in hierarchy. - In fact it is another implementation os Floyd's formal sharing. + In fact it is another implementation of Floyd's formal sharing. Levels: Each class is assigned level. Leaf has ALWAYS level 0 and root @@ -63,37 +66,19 @@ #define HTB_HSIZE 16 /* classid hash size */ #define HTB_EWMAC 2 /* rate average over HTB_EWMAC*HTB_HSIZE sec */ #define HTB_DEBUG 1 /* compile debugging support (activated by tc tool) */ -#define HTB_QLOCK(S) spin_lock_bh(&(S)->dev->queue_lock) -#define HTB_QUNLOCK(S) spin_unlock_bh(&(S)->dev->queue_lock) - -/* ======== Begin of part to be deleted for 2.4 merged one ========= */ -#if LINUX_VERSION_CODE < 0x20300 -#define MODULE_LICENSE(X) - -#define NET_XMIT_SUCCESS 1 -#define NET_XMIT_DROP 0 - -static inline void __skb_queue_purge(struct sk_buff_head *list) -{ - struct sk_buff *skb; - while ((skb=__skb_dequeue(list))!=NULL) - kfree_skb(skb); -} -#define del_timer_sync(t) del_timer(t) +#define HTB_RATECM 1 /* whether to use rate computer */ +#define HTB_HYSTERESIS 1/* whether to use mode hysteresis for speedup */ +#define HTB_QLOCK(S) sch_dev_queue_lock((S)->dev) +#define HTB_QUNLOCK(S) sch_dev_queue_unlock((S)->dev) +#define HTB_VER 0x30007 /* major must be matched with number suplied by TC as version */ -#define netif_schedule qdisc_wakeup -#define netif_queue_stopped(D) (D->tbusy) -#define sch_tree_lock(S) start_bh_atomic() -#define sch_tree_unlock(S) end_bh_atomic() -#undef HTB_QLOCK -#undef HTB_QUNLOCK -#define HTB_QLOCK(S) -#define HTB_QUNLOCK(S) -#ifndef BUG_TRAP -#define BUG_TRAP(x) if (!(x)) { printk("Assertion (" #x ") failed at " __FILE__ "(%d):" __FUNCTION__ "\n", __LINE__); } +#if HTB_VER >> 16 != TC_HTB_PROTOVER +#error "Mismatched sch_htb.c and pkt_sch.h" #endif -#endif -/* ======== End of part to be deleted for 2.4 merged one =========== */ + +/* temporary debug defines to be removed after beta stage */ +#define DEVIK_MEND(N) +#define DEVIK_MSTART(N) /* debugging support; S is subsystem, these are defined: 0 - netlink messages @@ -102,7 +87,9 @@ 3 - dequeue main 4 - dequeue one prio DRR part 5 - dequeue class accounting - 6 - dequeue rcache (ready level computation) + 6 - class overlimit status computation + 7 - hint tree + 8 - event queue 10 - rate estimator 11 - classifier 12 - fast dequeue cache @@ -111,80 +98,88 @@ q->debug uint32 contains 16 2-bit fields one for subsystem starting from LSB */ -#if HTB_DEBUG +#ifdef HTB_DEBUG #define HTB_DBG(S,L,FMT,ARG...) if (((q->debug>>(2*S))&3) >= L) \ printk(KERN_DEBUG FMT,##ARG) +#define HTB_CHCL(cl) BUG_TRAP((cl)->magic == HTB_CMAGIC) +#define HTB_PASSQ q, +#define HTB_ARGQ struct htb_sched *q, +#define static +#define __inline__ +#define inline +#define HTB_CMAGIC 0xFEFAFEF1 +#define htb_safe_rb_erase(N,R) do { BUG_TRAP((N)->rb_color != -1); \ + if ((N)->rb_color == -1) break; \ + rb_erase(N,R); \ + (N)->rb_color = -1; } while (0) #else #define HTB_DBG(S,L,FMT,ARG...) +#define HTB_PASSQ +#define HTB_ARGQ +#define HTB_CHCL(cl) +#define htb_safe_rb_erase(N,R) rb_erase(N,R) #endif -/* used internaly to pass status of single class */ +/* used internaly to keep status of single class */ enum htb_cmode { HTB_CANT_SEND, /* class can't send and can't borrow */ HTB_MAY_BORROW, /* class can't send but may borrow */ HTB_CAN_SEND /* class can send */ }; -#define HTB_F_INJ 0x10000 /* to mark dequeue level as injected one */ - -/* often used circular list of classes; I didn't use generic linux - double linked list to avoid casts and before I rely on some behaviour - of insert and delete functions; item not bound to list is guaranted - to have prev member NULL (we don't mangle next pointer as we often - need it */ -struct htb_litem { - struct htb_class *prev, *next; -}; -/* circular list insert and delete macros; these also maintain - correct value of pointer to the list; insert adds 'new' class - before 'cl' class using prev/next member 'list' */ -#define HTB_INSERTB(list,cl,new) \ -do { if (!cl) new->list.prev = cl = new; \ - new->list.next = cl; new->list.prev = cl->list.prev; \ - cl->list.prev->list.next = cl->list.prev = new; } while(0) - -/* remove 'cl' class from 'list' repairing 'ptr' if not null */ -#define HTB_DELETE(list,cl,ptr) do { \ - if (cl->list.prev) { cl->list.prev->list.next = cl->list.next; \ - cl->list.next->list.prev = cl->list.prev; \ - if (ptr == cl) ptr = cl->list.next; \ - if (ptr == cl) ptr = NULL; cl->list.prev = NULL; } \ - else printk(KERN_ERR "htb: DELETE BUG [" #list "," #cl "," #ptr "]\n"); \ - } while (0) /* interior & leaf nodes; props specific to leaves are marked L: */ struct htb_class { +#ifdef HTB_DEBUG + unsigned magic; +#endif /* general class parameters */ u32 classid; struct tc_stats stats; /* generic stats */ struct tc_htb_xstats xstats;/* our special stats */ int refcnt; /* usage count of this class */ - struct Qdisc *q; /* L: elem. qdisc */ +#ifdef HTB_RATECM /* rate measurement counters */ unsigned long rate_bytes,sum_bytes; unsigned long rate_packets,sum_packets; +#endif - /* DRR scheduler parameters */ - int quantum; /* L: round quantum computed from rate */ - int deficit[TC_HTB_MAXDEPTH]; /* L: deficit for class at level */ - char prio; /* L: priority of the class; 0 is the highest */ - char aprio; /* L: prio at which we were last adding to active list - it is used to change priority at runtime */ /* topology */ - char level; /* our level (see above) */ - char injectd; /* distance from injected parent */ + int level; /* our level (see above) */ struct htb_class *parent; /* parent class */ - struct htb_class *children; /* pointer to children list */ - struct htb_litem hlist; /* classid hash list */ - struct htb_litem active; /* L: prio level active DRR list */ - struct htb_litem sibling; /* sibling list */ + struct list_head hlist; /* classid hash list item */ + struct list_head sibling; /* sibling list item */ + struct list_head children; /* children list */ + + union { + struct htb_class_leaf { + struct Qdisc *q; + int prio; + int aprio; + int quantum; + int deficit[TC_HTB_MAXDEPTH]; + struct list_head drop_list; + } leaf; + struct htb_class_inner { + rb_root_t feed[TC_HTB_NUMPRIO]; /* feed trees */ + rb_node_t *ptr[TC_HTB_NUMPRIO]; /* current class ptr */ + } inner; + } un; + rb_node_t node[TC_HTB_NUMPRIO]; /* node for self or feed tree */ + rb_node_t pq_node; /* node for event queue */ + unsigned long pq_key; /* the same type as jiffies global */ + int prio_activity; /* for which prios are we active */ + enum htb_cmode cmode; /* current mode of the class */ + /* class attached filters */ struct tcf_proto *filter_list; int filter_cnt; + int warned; /* only one warning about non work conserving .. */ + /* token bucket parameters */ struct qdisc_rate_table *rate; /* rate table of the class itself */ struct qdisc_rate_table *ceil; /* ceiling rate (limits borrows too) */ @@ -192,10 +187,6 @@ long mbuffer; /* max wait time */ long tokens,ctokens; /* current number of tokens */ psched_time_t t_c; /* checkpoint time */ - - /* walk result cache for leaves */ - unsigned long rcache_sn; /* SN of cache validity */ - unsigned rc_level; /* victim's level */ }; /* TODO: maybe compute rate when size is too large .. or drop ? */ @@ -212,19 +203,23 @@ struct htb_sched { - struct htb_class *root; /* root classes circular list */ - struct htb_class *hash[HTB_HSIZE]; /* hashed by classid */ + struct list_head root; /* root classes list */ + struct list_head hash[HTB_HSIZE]; /* hashed by classid */ + struct list_head drops[TC_HTB_NUMPRIO]; /* active leaves (for drops) */ - /* active classes table; this needs explanation. This table contains - one set of pointers per priority, it is obvious. The set contains - one pointer per class level in the same way as cl->deficit is - independent for each level. This allows us to maintain correct - DRR position independent of borrowing level. - If we used single active/deficit items then DRR fairness'd suffer - from frequent class level changes. - Note that htb_[de]activate must be used to update this item - because it needs to keep all pointers in set coherent. */ - struct htb_class *active[TC_HTB_NUMPRIO][TC_HTB_MAXDEPTH]; + /* self list - roots of self generating tree */ + rb_root_t row[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; + int row_mask[TC_HTB_MAXDEPTH]; + rb_node_t *ptr[TC_HTB_MAXDEPTH][TC_HTB_NUMPRIO]; + + /* self wait list - roots of wait PQs per row */ + rb_root_t wait_pq[TC_HTB_MAXDEPTH]; + + /* time of nearest event per level (row) */ + unsigned long near_ev_cache[TC_HTB_MAXDEPTH]; + + /* whether we hit non-work conserving class during this dequeue; we use */ + int nwc_hit; /* this to disable mindelay complaint in dequeue */ int defcls; /* class where unclassified flows go to */ u32 debug; /* subsystem debug levels */ @@ -233,27 +228,18 @@ struct tcf_proto *filter_list; int filter_cnt; - unsigned long sn; /* result cache serial number */ int rate2quantum; /* quant = rate / rate2quantum */ psched_time_t now; /* cached dequeue time */ - long delay; /* how long to deactivate for */ struct timer_list timer; /* send delay timer */ +#ifdef HTB_RATECM struct timer_list rttim; /* rate computer timer */ int recmp_bucket; /* which hash bucket to recompute next */ - - /* cache of last dequeued class */ - struct htb_class *last_tx; - int use_dcache; - - /* non shapped skbs; let them go directly thru */ +#endif + + /* non shaped skbs; let them go directly thru */ struct sk_buff_head direct_queue; int direct_qlen; /* max qlen of above */ - /* statistics (see tc_htb_glob for explanation) */ - long deq_rate,deq_rate_c; - long utilz,utilz_c; - long trials,trials_c; - long dcache_hits; long direct_pkts; }; @@ -271,117 +257,477 @@ /* find class in global hash table using given handle */ static __inline__ struct htb_class *htb_find(u32 handle, struct Qdisc *sch) { - struct htb_sched *q = (struct htb_sched *)sch->data; - int h = htb_hash(handle); - struct htb_class *cl; - if (TC_H_MAJ(handle) != sch->handle) return NULL; - cl = q->hash[h]; - if (cl) do { - if (cl->classid == handle) return cl; - - } while ((cl = cl->hlist.next) != q->hash[h]); - return NULL; + struct htb_sched *q = (struct htb_sched *)sch->data; + struct list_head *p; + if (TC_H_MAJ(handle) != sch->handle) + return NULL; + + list_for_each (p,q->hash+htb_hash(handle)) { + struct htb_class *cl = list_entry(p,struct htb_class,hlist); + if (cl->classid == handle) + return cl; + } + return NULL; } -/* classify packet into class TODO: use inner filters & marks here */ -static struct htb_class *htb_clasify(struct sk_buff *skb, struct Qdisc *sch) +/** + * htb_classify - classify a packet into class + * + * It returns NULL if the packet should be dropped or -1 if the packet + * should be passed directly thru. In all other cases leaf class is returned. + * We allow direct class selection by classid in priority. The we examine + * filters in qdisc and in inner nodes (if higher filter points to the inner + * node). If we end up with classid MAJOR:0 we enqueue the skb into special + * internal fifo (direct). These packets then go directly thru. If we still + * have no valid leaf we try to use MAJOR:default leaf. It still unsuccessfull + * then finish and return direct queue. + */ +#define HTB_DIRECT (struct htb_class*)-1 +static struct htb_class *htb_classify(struct sk_buff *skb, struct Qdisc *sch) { - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl; - struct tcf_result res; - struct tcf_proto *tcf; - - /* allow to select class by setting skb->priority to valid classid; - note that nfmark can be used too by attaching filter fw with no - rules in it */ - if (skb->priority == sch->handle) - return NULL; /* X:0 (direct flow) selected */ - if ((cl = htb_find(skb->priority,sch)) != NULL) + struct htb_sched *q = (struct htb_sched *)sch->data; + struct htb_class *cl; + struct tcf_result res; + struct tcf_proto *tcf; + int result; + + /* allow to select class by setting skb->priority to valid classid; + note that nfmark can be used too by attaching filter fw with no + rules in it */ + if (skb->priority == sch->handle) + return HTB_DIRECT; /* X:0 (direct flow) selected */ + if ((cl = htb_find(skb->priority,sch)) != NULL) + return cl; + + tcf = q->filter_list; + while (tcf && (result = tc_classify(skb, tcf, &res)) >= 0) { +#ifdef CONFIG_NET_CLS_POLICE + if (result == TC_POLICE_SHOT) + return NULL; +#endif + if ((cl = (void*)res.class) == NULL) { + if (res.classid == sch->handle) + return HTB_DIRECT; /* X:0 (direct flow) */ + if ((cl = htb_find(res.classid,sch)) == NULL) + break; /* filter selected invalid classid */ + } + if (!cl->level) + return cl; /* we hit leaf; return it */ + + /* we have got inner class; apply inner filter chain */ + tcf = cl->filter_list; + } + /* classification failed; try to use default class */ + cl = htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle),q->defcls),sch); + if (!cl || cl->level) + return HTB_DIRECT; /* bad default .. this is safe bet */ return cl; +} + +#ifdef HTB_DEBUG +static void htb_next_rb_node(rb_node_t **n); +#define HTB_DUMTREE(root,memb) if(root) { \ + rb_node_t *n = (root)->rb_node; \ + while (n->rb_left) n = n->rb_left; \ + while (n) { \ + struct htb_class *cl = rb_entry(n, struct htb_class, memb); \ + printk(" %x",cl->classid); htb_next_rb_node (&n); \ + } } + +static void htb_debug_dump (struct htb_sched *q) +{ + int i,p; + printk(KERN_DEBUG "htb*g j=%lu\n",jiffies); + /* rows */ + for (i=TC_HTB_MAXDEPTH-1;i>=0;i--) { + printk(KERN_DEBUG "htb*r%d m=%x",i,q->row_mask[i]); + for (p=0;prow[i][p].rb_node) continue; + printk(" p%d:",p); + HTB_DUMTREE(q->row[i]+p,node[p]); + } + printk("\n"); + } + /* classes */ + for (i = 0; i < HTB_HSIZE; i++) { + struct list_head *l; + list_for_each (l,q->hash+i) { + struct htb_class *cl = list_entry(l,struct htb_class,hlist); + long diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32)cl->mbuffer, 0); + printk(KERN_DEBUG "htb*c%x m=%d t=%ld c=%ld pq=%lu df=%ld ql=%d " + "pa=%x f:", + cl->classid,cl->cmode,cl->tokens,cl->ctokens, + cl->pq_node.rb_color==-1?0:cl->pq_key,diff, + cl->level?0:cl->un.leaf.q->q.qlen,cl->prio_activity); + if (cl->level) + for (p=0;pun.inner.feed[p].rb_node) continue; + printk(" p%d a=%x:",p,cl->un.inner.ptr[p]?rb_entry(cl->un.inner.ptr[p], struct htb_class,node[p])->classid:0); + HTB_DUMTREE(cl->un.inner.feed+p,node[p]); + } + printk("\n"); + } + } +} +#endif +/** + * htb_add_to_id_tree - adds class to the round robin list + * + * Routine adds class to the list (actually tree) sorted by classid. + * Make sure that class is not already on such list for given prio. + */ +static void htb_add_to_id_tree (HTB_ARGQ rb_root_t *root, + struct htb_class *cl,int prio) +{ + rb_node_t **p = &root->rb_node, *parent = NULL; + HTB_DBG(7,3,"htb_add_id_tree cl=%X prio=%d\n",cl->classid,prio); +#ifdef HTB_DEBUG + if (cl->node[prio].rb_color != -1) { BUG_TRAP(0); return; } + HTB_CHCL(cl); + if (*p) { + struct htb_class *x = rb_entry(*p,struct htb_class,node[prio]); + HTB_CHCL(x); + } +#endif + while (*p) { + struct htb_class *c; parent = *p; + c = rb_entry(parent, struct htb_class, node[prio]); + HTB_CHCL(c); + if (cl->classid > c->classid) + p = &parent->rb_right; + else + p = &parent->rb_left; + } + rb_link_node(&cl->node[prio], parent, p); + rb_insert_color(&cl->node[prio], root); +} + +/** + * htb_add_to_wait_tree - adds class to the event queue with delay + * + * The class is added to priority event queue to indicate that class will + * change its mode in cl->pq_key microseconds. Make sure that class is not + * already in the queue. + */ +static void htb_add_to_wait_tree (struct htb_sched *q, + struct htb_class *cl,long delay,int debug_hint) +{ + rb_node_t **p = &q->wait_pq[cl->level].rb_node, *parent = NULL; + HTB_DBG(7,3,"htb_add_wt cl=%X key=%lu\n",cl->classid,cl->pq_key); +#ifdef HTB_DEBUG + if (cl->pq_node.rb_color != -1) { BUG_TRAP(0); return; } + HTB_CHCL(cl); + if ((delay <= 0 || delay > cl->mbuffer) && net_ratelimit()) + printk(KERN_ERR "HTB: suspicious delay in wait_tree d=%ld cl=%X h=%d\n",delay,cl->classid,debug_hint); +#endif + DEVIK_MSTART(9); + cl->pq_key = jiffies + PSCHED_US2JIFFIE(delay); + if (cl->pq_key == jiffies) + cl->pq_key++; + + /* update the nearest event cache */ + if (q->near_ev_cache[cl->level] - cl->pq_key < 0x80000000) + q->near_ev_cache[cl->level] = cl->pq_key; + + while (*p) { + struct htb_class *c; parent = *p; + c = rb_entry(parent, struct htb_class, pq_node); + if (cl->pq_key - c->pq_key < 0x80000000) + p = &parent->rb_right; + else + p = &parent->rb_left; + } + rb_link_node(&cl->pq_node, parent, p); + rb_insert_color(&cl->pq_node, &q->wait_pq[cl->level]); + DEVIK_MEND(9); +} - tcf = q->filter_list; - while (tcf && !tc_classify(skb, tcf, &res)) { - if (res.classid == sch->handle) - return NULL; /* X:0 (direct flow) selected */ - if ((cl = htb_find(res.classid,sch)) == NULL) - break; /* filter selected invalid classid */ - if (!cl->level) - return cl; /* we hit leaf; return it */ +/** + * htb_next_rb_node - finds next node in binary tree + * + * When we are past last key we return NULL. + * Average complexity is 2 steps per call. + */ +static void htb_next_rb_node(rb_node_t **n) +{ + rb_node_t *p; + if ((*n)->rb_right) { + *n = (*n)->rb_right; + while ((*n)->rb_left) + *n = (*n)->rb_left; + return; + } + while ((p = (*n)->rb_parent) != NULL) { + if (p->rb_left == *n) break; + *n = p; + } + *n = p; +} - /* we have got inner class; apply inner filter chain */ - tcf = cl->filter_list; +/** + * htb_add_class_to_row - add class to its row + * + * The class is added to row at priorities marked in mask. + * It does nothing if mask == 0. + */ +static inline void htb_add_class_to_row(struct htb_sched *q, + struct htb_class *cl,int mask) +{ + HTB_DBG(7,2,"htb_addrow cl=%X mask=%X rmask=%X\n", + cl->classid,mask,q->row_mask[cl->level]); + HTB_CHCL(cl); + q->row_mask[cl->level] |= mask; + while (mask) { + int prio = ffz(~mask); + mask &= ~(1 << prio); + htb_add_to_id_tree(HTB_PASSQ q->row[cl->level]+prio,cl,prio); + } +} + +/** + * htb_remove_class_from_row - removes class from its row + * + * The class is removed from row at priorities marked in mask. + * It does nothing if mask == 0. + */ +static __inline__ void htb_remove_class_from_row(struct htb_sched *q, + struct htb_class *cl,int mask) +{ + int m = 0; + HTB_CHCL(cl); + while (mask) { + int prio = ffz(~mask); + mask &= ~(1 << prio); + if (q->ptr[cl->level][prio] == cl->node+prio) + htb_next_rb_node(q->ptr[cl->level]+prio); + htb_safe_rb_erase(cl->node + prio,q->row[cl->level]+prio); + if (!q->row[cl->level][prio].rb_node) + m |= 1 << prio; + } + HTB_DBG(7,2,"htb_delrow cl=%X mask=%X rmask=%X maskdel=%X\n", + cl->classid,mask,q->row_mask[cl->level],m); + q->row_mask[cl->level] &= ~m; +} + +/** + * htb_activate_prios - creates active classe's feed chain + * + * The class is connected to ancestors and/or appropriate rows + * for priorities it is participating on. cl->cmode must be new + * (activated) mode. It does nothing if cl->prio_activity == 0. + */ +static void htb_activate_prios(struct htb_sched *q,struct htb_class *cl) +{ + struct htb_class *p = cl->parent; + long m,mask = cl->prio_activity; + HTB_DBG(7,2,"htb_act_prios cl=%X mask=%lX cmode=%d\n",cl->classid,mask,cl->cmode); + HTB_CHCL(cl); + + while (cl->cmode == HTB_MAY_BORROW && p && mask) { + HTB_CHCL(p); + m = mask; while (m) { + int prio = ffz(~m); + m &= ~(1 << prio); + + if (p->un.inner.feed[prio].rb_node) + /* parent already has its feed in use so that + reset bit in mask as parent is already ok */ + mask &= ~(1 << prio); + + htb_add_to_id_tree(HTB_PASSQ p->un.inner.feed+prio,cl,prio); + } + HTB_DBG(7,3,"htb_act_pr_aft p=%X pact=%X mask=%lX pmode=%d\n", + p->classid,p->prio_activity,mask,p->cmode); + p->prio_activity |= mask; + cl = p; p = cl->parent; + HTB_CHCL(cl); + } + if (cl->cmode == HTB_CAN_SEND && mask) + htb_add_class_to_row(q,cl,mask); +} + +/** + * htb_deactivate_prios - remove class from feed chain + * + * cl->cmode must represent old mode (before deactivation). It does + * nothing if cl->prio_activity == 0. Class is removed from all feed + * chains and rows. + */ +static void htb_deactivate_prios(struct htb_sched *q, struct htb_class *cl) +{ + struct htb_class *p = cl->parent; + long m,mask = cl->prio_activity; + HTB_DBG(7,2,"htb_deact_prios cl=%X mask=%lX cmode=%d\n",cl->classid,mask,cl->cmode); + HTB_CHCL(cl); + + while (cl->cmode == HTB_MAY_BORROW && p && mask) { + m = mask; mask = 0; + while (m) { + int prio = ffz(~m); + m &= ~(1 << prio); + + if (p->un.inner.ptr[prio] == cl->node+prio) + htb_next_rb_node(p->un.inner.ptr + prio); + + htb_safe_rb_erase(cl->node + prio,p->un.inner.feed + prio); + + if (!p->un.inner.feed[prio].rb_node) + mask |= 1 << prio; + } + HTB_DBG(7,3,"htb_deact_pr_aft p=%X pact=%X mask=%lX pmode=%d\n", + p->classid,p->prio_activity,mask,p->cmode); + p->prio_activity &= ~mask; + cl = p; p = cl->parent; + HTB_CHCL(cl); + } + if (cl->cmode == HTB_CAN_SEND && mask) + htb_remove_class_from_row(q,cl,mask); +} + +/** + * htb_class_mode - computes and returns current class mode + * + * It computes cl's mode at time cl->t_c+diff and returns it. If mode + * is not HTB_CAN_SEND then cl->pq_key is updated to time difference + * from now to time when cl will change its state. + * Also it is worth to note that class mode doesn't change simply + * at cl->{c,}tokens == 0 but there can rather be hysteresis of + * 0 .. -cl->{c,}buffer range. It is meant to limit number of + * mode transitions per time unit. The speed gain is about 1/6. + */ +static __inline__ enum htb_cmode +htb_class_mode(struct htb_class *cl,long *diff) +{ + long toks; + + if ((toks = (cl->ctokens + *diff)) < ( +#ifdef HTB_HYSTERESIS + cl->cmode != HTB_CANT_SEND ? -cl->cbuffer : +#endif + 0)) { + *diff = -toks; + return HTB_CANT_SEND; } - /* classification failed; try to use default class */ - return htb_find(TC_H_MAKE(TC_H_MAJ(sch->handle),q->defcls),sch); + if ((toks = (cl->tokens + *diff)) >= ( +#ifdef HTB_HYSTERESIS + cl->cmode == HTB_CAN_SEND ? -cl->buffer : +#endif + 0)) + return HTB_CAN_SEND; + + *diff = -toks; + return HTB_MAY_BORROW; } -/* inserts cl into appropriate active lists (for all levels) */ +/** + * htb_change_class_mode - changes classe's mode + * + * This should be the only way how to change classe's mode under normal + * cirsumstances. Routine will update feed lists linkage, change mode + * and add class to the wait event queue if appropriate. New mode should + * be different from old one and cl->pq_key has to be valid if changing + * to mode other than HTB_CAN_SEND (see htb_add_to_wait_tree). + */ +static void +htb_change_class_mode(struct htb_sched *q, struct htb_class *cl, long *diff) +{ + enum htb_cmode new_mode = htb_class_mode(cl,diff); + + HTB_CHCL(cl); + HTB_DBG(7,1,"htb_chging_clmode %d->%d cl=%X\n",cl->cmode,new_mode,cl->classid); + + if (new_mode == cl->cmode) + return; + + if (cl->prio_activity) { /* not neccessary: speed optimization */ + if (cl->cmode != HTB_CANT_SEND) + htb_deactivate_prios(q,cl); + cl->cmode = new_mode; + if (new_mode != HTB_CANT_SEND) + htb_activate_prios(q,cl); + } else + cl->cmode = new_mode; +} + +/** + * htb_activate - inserts leaf cl into appropriate active feeds + * + * Routine learns (new) priority of leaf and activates feed chain + * for the prio. It can be called on already active leaf safely. + * It also adds leaf into droplist. + */ static __inline__ void htb_activate(struct htb_sched *q,struct htb_class *cl) { - if (!cl->active.prev) { - struct htb_class **ap = q->active[(int)(cl->aprio=cl->prio)]; - int i = !ap[0]; - HTB_INSERTB(active,ap[0],cl); - if (i) /* set also all level pointers */ - for (i = 1; i < TC_HTB_MAXDEPTH; i++) ap[i] = ap[0]; - } + BUG_TRAP(!cl->level && cl->un.leaf.q && cl->un.leaf.q->q.qlen); + HTB_CHCL(cl); + if (!cl->prio_activity) { + cl->prio_activity = 1 << (cl->un.leaf.aprio = cl->un.leaf.prio); + htb_activate_prios(q,cl); + list_add_tail(&cl->un.leaf.drop_list,q->drops+cl->un.leaf.aprio); + } } -/* remove cl from active lists; lev is level at which we dequeued - so that we know that active[prio][lev] points to cl */ +/** + * htb_deactivate - remove leaf cl from active feeds + * + * Make sure that leaf is active. In the other words it can't be called + * with non-active leaf. It also removes class from the drop list. + */ static __inline__ void -htb_deactivate(struct htb_sched *q,struct htb_class *cl,int lev) +htb_deactivate(struct htb_sched *q,struct htb_class *cl) { - int i; - struct htb_class **ap = q->active[(int)cl->aprio]; - HTB_DELETE(active,cl,ap[lev]); - if (ap[lev]) { - /* repair other level pointers if they've pointed - to the deleted class */ - for (i = 0; i < TC_HTB_MAXDEPTH; i++) - if (ap[i] == cl) ap[i] = ap[lev]; - } else - memset(ap,0,sizeof(*ap)*TC_HTB_MAXDEPTH); + BUG_TRAP(cl->prio_activity); + HTB_CHCL(cl); + htb_deactivate_prios(q,cl); + cl->prio_activity = 0; + list_del_init(&cl->un.leaf.drop_list); } static int htb_enqueue(struct sk_buff *skb, struct Qdisc *sch) { struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl = htb_clasify(skb,sch); + struct htb_class *cl = htb_classify(skb,sch); - if (!cl || !cl->q) { - /* bad class; enqueue to helper queue */ - if (q->direct_queue.qlen < q->direct_qlen) { + DEVIK_MSTART(0); + if (cl == HTB_DIRECT || !cl) { + /* enqueue to helper queue */ + if (q->direct_queue.qlen < q->direct_qlen && cl) { __skb_queue_tail(&q->direct_queue, skb); q->direct_pkts++; } else { kfree_skb (skb); sch->stats.drops++; + DEVIK_MEND(0); return NET_XMIT_DROP; } - } else if (cl->q->enqueue(skb, cl->q) != NET_XMIT_SUCCESS) { + } else if (cl->un.leaf.q->enqueue(skb, cl->un.leaf.q) != NET_XMIT_SUCCESS) { sch->stats.drops++; cl->stats.drops++; + DEVIK_MEND(0); return NET_XMIT_DROP; } else { cl->stats.packets++; cl->stats.bytes += skb->len; + DEVIK_MSTART(1); htb_activate (q,cl); + DEVIK_MEND(1); } sch->q.qlen++; sch->stats.packets++; sch->stats.bytes += skb->len; HTB_DBG(1,1,"htb_enq_ok cl=%X skb=%p\n",cl?cl->classid:0,skb); + DEVIK_MEND(0); return NET_XMIT_SUCCESS; } +/* TODO: requeuing packet charges it to policers again !! */ static int htb_requeue(struct sk_buff *skb, struct Qdisc *sch) { struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl = htb_clasify(skb,sch); + struct htb_class *cl = htb_classify(skb,sch); - if (!cl || !cl->q) { - /* bad class; enqueue to helper queue */ - if (q->direct_queue.qlen < q->direct_qlen) { + if (cl == HTB_DIRECT || !cl) { + /* enqueue to helper queue */ + if (q->direct_queue.qlen < q->direct_qlen && cl) { __skb_queue_tail(&q->direct_queue, skb); q->direct_pkts++; } else { @@ -389,12 +735,12 @@ sch->stats.drops++; return NET_XMIT_DROP; } - } else if (cl->q->ops->requeue(skb, cl->q) != NET_XMIT_SUCCESS) { + } else if (cl->un.leaf.q->ops->requeue(skb, cl->un.leaf.q) != NET_XMIT_SUCCESS) { sch->stats.drops++; cl->stats.drops++; return NET_XMIT_DROP; } else - htb_activate (q,cl); + htb_activate (q,cl); sch->q.qlen++; HTB_DBG(1,1,"htb_req_ok cl=%X skb=%p\n",cl?cl->classid:0,skb); @@ -409,315 +755,251 @@ netif_schedule(sch->dev); } +#ifdef HTB_RATECM #define RT_GEN(D,R) R+=D-(R/HTB_EWMAC);D=0 static void htb_rate_timer(unsigned long arg) { - struct Qdisc *sch = (struct Qdisc*)arg; - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl; - - /* lock queue so that we can muck with it */ - HTB_QLOCK(sch); - HTB_DBG(10,1,"htb_rttmr j=%ld\n",jiffies); - - q->rttim.expires = jiffies + HZ; - add_timer(&q->rttim); - - /* scan and recompute one bucket at time */ - if (++q->recmp_bucket >= HTB_HSIZE) q->recmp_bucket = 0; - if ((cl = q->hash[q->recmp_bucket]) != NULL) do { - HTB_DBG(10,2,"htb_rttmr_cl cl=%X sbyte=%lu spkt=%lu\n",cl->classid,cl->sum_bytes,cl->sum_packets); - RT_GEN (cl->sum_bytes,cl->rate_bytes); - RT_GEN (cl->sum_packets,cl->rate_packets); - } while ((cl = cl->hlist.next) != q->hash[q->recmp_bucket]); - - /* global stats */ - RT_GEN (q->trials_c,q->trials); - RT_GEN (q->utilz_c,q->utilz); - RT_GEN (q->deq_rate_c,q->deq_rate); - - HTB_QUNLOCK(sch); -} - -/* test whether class can send or borrow packet */ -static enum htb_cmode -htb_class_mode(struct htb_sched *q, struct htb_class *cl) -{ - long toks,diff; - diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32)cl->mbuffer, 0); - HTB_DBG(6,3,"htb_cm diff=%ld\n",diff); - - /* check whether we are over ceil */ - if ((toks = (cl->ctokens + diff)) < 0) { - if (q->delay > -toks || !q->delay) q->delay = -toks; - return HTB_CANT_SEND; - } - - /* our regular rate */ - if ((toks = (cl->tokens + diff)) >= 0) - return HTB_CAN_SEND; - - /* record time when we can transmit */ - if (q->delay > -toks || !q->delay) q->delay = -toks; - - return HTB_MAY_BORROW; -} - -/* computes (possibly ancestor) class ready to send; cl is leaf; - cl's rc_level is then filled with level we are borrowing at; - it is set to TC_HTB_MAXDEPTH if we can't borrow at all and can be - ORed with HTB_F_INJ if bw was injected. */ -static void htb_ready_level(struct htb_sched *q,struct htb_class *cl) -{ - struct htb_class *stack[TC_HTB_MAXDEPTH],**sp = stack; - int level = TC_HTB_MAXDEPTH, injdist = cl->injectd; - enum htb_cmode mode; - HTB_DBG(6,1,"htb_rl cl=%X tok=%ld ctok=%ld buf=%ld cbuf=%ld\n",cl->classid,cl->tokens,cl->ctokens,cl->buffer,cl->cbuffer); - - /* traverse tree upward looking for ready class */ - for (;;) { - *sp++ = cl; /* push at stack */ - - /* test mode */ - mode = htb_class_mode(q,cl); - HTB_DBG(6,2,"htb_clmod cl=%X m=%d tok=%ld ctok=%ld buf=%ld cbuf=%ld\n",cl->classid,mode,cl->tokens,cl->ctokens,cl->buffer,cl->cbuffer); - if (mode != HTB_MAY_BORROW) { - if (mode == HTB_CAN_SEND) level = cl->level; - break; - } - /* update injdist from current node */ - if (injdist > cl->injectd) injdist = cl->injectd; - - /* if this is leaf's injector then resolve borrow positively */ - if (!injdist--) { - /* don't cache this result in interior nodes */ - stack[0]->rc_level = cl->level|HTB_F_INJ; - stack[0]->rcache_sn = q->sn; - return; - } - if ((cl = cl->parent) == NULL) break; - if (q->sn == cl->rcache_sn) { - /* the node has already computed result; use it */ - level = cl->rc_level; break; + struct Qdisc *sch = (struct Qdisc*)arg; + struct htb_sched *q = (struct htb_sched *)sch->data; + struct list_head *p; + + /* lock queue so that we can muck with it */ + HTB_QLOCK(sch); + HTB_DBG(10,1,"htb_rttmr j=%ld\n",jiffies); + + q->rttim.expires = jiffies + HZ; + add_timer(&q->rttim); + + /* scan and recompute one bucket at time */ + if (++q->recmp_bucket >= HTB_HSIZE) + q->recmp_bucket = 0; + list_for_each (p,q->hash+q->recmp_bucket) { + struct htb_class *cl = list_entry(p,struct htb_class,hlist); + HTB_DBG(10,2,"htb_rttmr_cl cl=%X sbyte=%lu spkt=%lu\n", + cl->classid,cl->sum_bytes,cl->sum_packets); + RT_GEN (cl->sum_bytes,cl->rate_bytes); + RT_GEN (cl->sum_packets,cl->rate_packets); } - } - while (--sp >= stack) { /* update mode cache */ - (*sp)->rcache_sn = q->sn; - (*sp)->rc_level = level; - } + HTB_QUNLOCK(sch); } +#endif -/* pull packet from class and charge to ancestors */ -static struct sk_buff * -htb_dequeue_class(struct Qdisc *sch, struct htb_class *cl) +/** + * htb_charge_class - charges ammount "bytes" to leaf and ancestors + * + * Routine assumes that packet "bytes" long was dequeued from leaf cl + * borrowing from "level". It accounts bytes to ceil leaky bucket for + * leaf and all ancestors and to rate bucket for ancestors at levels + * "level" and higher. It also handles possible change of mode resulting + * from the update. Note that mode can also increase here (MAY_BORROW to + * CAN_SEND) because we can use more precise clock that event queue here. + * In such case we remove class from event queue first. + */ +static void htb_charge_class(struct htb_sched *q,struct htb_class *cl, + int level,int bytes) { - struct htb_sched *q = (struct htb_sched *)sch->data; - long toks,diff; - int injecting = cl->rc_level & HTB_F_INJ, injdist = cl->injectd; - int level = cl->rc_level & 0xff; - struct sk_buff *skb = cl->q->dequeue(cl->q); - HTB_DBG(5,1,"htb_deq_cl cl=%X skb=%p lev=%d inj=%d\n",cl->classid,skb,level,injecting); - if (!skb) return NULL; - - /* we have got skb, account it to victim and its parents - and also to all ceil estimators under victim */ - while (cl) { - diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32)cl->mbuffer, 0); + long toks,diff; + enum htb_cmode old_mode; + HTB_DBG(5,1,"htb_chrg_cl cl=%X lev=%d len=%d\n",cl->classid,level,bytes); #define HTB_ACCNT(T,B,R) toks = diff + cl->T; \ if (toks > cl->B) toks = cl->B; \ - toks -= L2T(cl, cl->R, skb->len); \ + toks -= L2T(cl, cl->R, bytes); \ if (toks <= -cl->mbuffer) toks = 1-cl->mbuffer; \ - cl->T = toks - - HTB_ACCNT (ctokens,cbuffer,ceil); - if (cl->level >= level) { - if (cl->level == level) cl->xstats.lends++; - HTB_ACCNT (tokens,buffer,rate); - } else { - cl->xstats.borrows++; - cl->tokens += diff; /* we moved t_c; update tokens */ - } - cl->t_c = q->now; - HTB_DBG(5,2,"htb_deq_clp cl=%X clev=%d diff=%ld\n",cl->classid,cl->level,diff); + cl->T = toks - /* update rate counters */ - cl->sum_bytes += skb->len; cl->sum_packets++; + while (cl) { + HTB_CHCL(cl); + diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32)cl->mbuffer, 0); +#ifdef HTB_DEBUG + if (diff > cl->mbuffer || diff < 0 || PSCHED_TLESS(q->now, cl->t_c)) { + if (net_ratelimit()) + printk(KERN_ERR "HTB: bad diff in charge, cl=%X diff=%lX now=%Lu then=%Lu j=%lu\n", + cl->classid, diff, + (unsigned long long) q->now, + (unsigned long long) cl->t_c, + jiffies); + diff = 1000; + } +#endif + if (cl->level >= level) { + if (cl->level == level) cl->xstats.lends++; + HTB_ACCNT (tokens,buffer,rate); + } else { + cl->xstats.borrows++; + cl->tokens += diff; /* we moved t_c; update tokens */ + } + HTB_ACCNT (ctokens,cbuffer,ceil); + cl->t_c = q->now; + HTB_DBG(5,2,"htb_chrg_clp cl=%X diff=%ld tok=%ld ctok=%ld\n",cl->classid,diff,cl->tokens,cl->ctokens); + + old_mode = cl->cmode; diff = 0; + htb_change_class_mode(q,cl,&diff); + if (old_mode != cl->cmode) { + if (old_mode != HTB_CAN_SEND) + htb_safe_rb_erase(&cl->pq_node,q->wait_pq+cl->level); + if (cl->cmode != HTB_CAN_SEND) + htb_add_to_wait_tree (q,cl,diff,1); + } + +#ifdef HTB_RATECM + /* update rate counters */ + cl->sum_bytes += bytes; cl->sum_packets++; +#endif - /* update byte stats except for leaves which are already updated */ - if (cl->level) { - cl->stats.bytes += skb->len; - cl->stats.packets++; - } - /* finish if we hit stop-class and we are injecting */ - if (injecting) { - if (injdist > cl->injectd) injdist = cl->injectd; - if (!injdist--) { - cl->xstats.injects++; break; - } + /* update byte stats except for leaves which are already updated */ + if (cl->level) { + cl->stats.bytes += bytes; + cl->stats.packets++; + } + cl = cl->parent; } - cl = cl->parent; - } - return skb; } -/* dequeues packet at given priority borrowing from given level; - if unsuccessfull then it returns level at which someone can - dequeue. If it sets level to TC_HTB_MAXDEPTH then no one can. */ -static struct sk_buff *htb_dequeue_prio(struct Qdisc *sch,int prio,int *level) +/** + * htb_do_events - make mode changes to classes at the level + * + * Scans event queue for pending events and applies them. Returns jiffies to + * next pending event (0 for no event in pq). + */ +static long htb_do_events(struct htb_sched *q,int level) { - struct sk_buff *skb = NULL; - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class **ap = q->active[prio], *cl = ap[*level]; - int done,top = TC_HTB_MAXDEPTH,rclev; - - HTB_DBG(4,1,"htb_deq_pr pr=%d lev=%d cl=%X\n",prio,*level,cl->classid); - /* this is DRR algorithm */ - do { - done = 1; - do { - /* catch empty classes here; note that we don't remove them - immediately after dequeue but rather delay remove to next - DRR round because if packet arrive for just emptied class - then we don't need to remove and again add it */ - if (!cl->q->q.qlen) { - ap[*level] = cl; /* needed for HTB_DELETE in deactivate */ - htb_deactivate (q,cl,*level); - - HTB_DBG(4,2,"htb_deq_deact cl=%X ncl=%X\n",cl->classid,ap[*level]?ap[*level]->classid:0); - if (ap[*level]) continue; - *level = TC_HTB_MAXDEPTH; - return NULL; /* NO class remains active */ - } - /* test whether class can send at all borrowing from level */ - if (cl->rcache_sn != q->sn) htb_ready_level(q,cl); - rclev = cl->rc_level & 0xff; /* filter injecting flag out */ - - HTB_DBG(4,2,"htb_deq_rd cl=%X rc_lev=0x%x dfct=%d qnt=%d\n", - cl->classid,cl->rc_level,cl->deficit[*level],cl->quantum); - - if (rclev == TC_HTB_MAXDEPTH) { - /* TODO: overlimit increment here is not proven correct */ - if (cl->deficit[*level] > 0) cl->stats.overlimits++; - continue; /* can't send or borrow */ - } - /* if we can't send at this level, remember where we can */ - if (rclev > *level) { - if (rclev < top) /* keep lowest level */ - top = rclev; - - HTB_DBG(4,2,"htb_deq_badlev top=%d\n",top); - continue; - } - if (cl->deficit[*level] <= 0) { - /* haven't allotment, increase and try again */ - done = 0; cl->deficit[*level] += cl->quantum; - continue; - } - if ((skb = htb_dequeue_class(sch,cl)) == NULL) { - /* nonempty class can't dequeue so that mark it as such; - note that rcache_sn is already set and thus this remarking - will be valid only for rest of this dequeue; this is - possible if child class is non work conserving */ - cl->rc_level = TC_HTB_MAXDEPTH; - - HTB_DBG(4,2,"htb_deq_noskb cl=%X len=%d\n",cl->classid,cl->q->q.qlen); - continue; - } - sch->q.qlen--; - /* prepare next class if we can't stay valid */ - if ((cl->deficit[*level] -= skb->len) <= 0) cl = cl->active.next; - else if (q->use_dcache) - q->last_tx = cl; /* cache cl if it still can transmit */ - ap[*level] = cl; - - HTB_DBG(4,1,"htb_deq_haspkt ncl=%X sqlen=%d\n",cl->classid,sch->q.qlen); - return skb; - - } while ((cl = cl->active.next) != ap[*level]); - - } while (!done); - *level = top; - HTB_DBG(4,1,"htb_deq_quit top=%d\n",top); - return NULL; + int i; + HTB_DBG(8,1,"htb_do_events l=%d root=%p rmask=%X\n", + level,q->wait_pq[level].rb_node,q->row_mask[level]); + for (i = 0; i < 500; i++) { + struct htb_class *cl; + long diff; + rb_node_t *p = q->wait_pq[level].rb_node; + if (!p) return 0; + while (p->rb_left) p = p->rb_left; + + cl = rb_entry(p, struct htb_class, pq_node); + if (cl->pq_key - (jiffies+1) < 0x80000000) { + HTB_DBG(8,3,"htb_do_ev_ret delay=%ld\n",cl->pq_key - jiffies); + return cl->pq_key - jiffies; + } + htb_safe_rb_erase(p,q->wait_pq+level); + diff = PSCHED_TDIFF_SAFE(q->now, cl->t_c, (u32)cl->mbuffer, 0); +#ifdef HTB_DEBUG + if (diff > cl->mbuffer || diff < 0 || PSCHED_TLESS(q->now, cl->t_c)) { + if (net_ratelimit()) + printk(KERN_ERR "HTB: bad diff in events, cl=%X diff=%lX now=%Lu then=%Lu j=%lu\n", + cl->classid, diff, + (unsigned long long) q->now, + (unsigned long long) cl->t_c, + jiffies); + diff = 1000; + } +#endif + htb_change_class_mode(q,cl,&diff); + if (cl->cmode != HTB_CAN_SEND) + htb_add_to_wait_tree (q,cl,diff,2); + } + if (net_ratelimit()) + printk(KERN_WARNING "htb: too many events !\n"); + return HZ/10; } -static struct sk_buff *htb_dequeue(struct Qdisc *sch) +/** + * htb_lookup_leaf - returns next leaf class in DRR order + * + * Find leaf where current feed pointers points to. + */ +static struct htb_class * +htb_lookup_leaf(rb_root_t *tree,int prio,rb_node_t **pptr) { - struct sk_buff *skb = NULL; - struct htb_sched *q = (struct htb_sched *)sch->data; - int prio,oklev,okprio = 0 /* avoid unused warning */,lev,i; - struct htb_class *cl; - psched_time_t endt; - - HTB_DBG(3,1,"htb_deq dircnt=%d ltx=%X\n",skb_queue_len(&q->direct_queue), - q->last_tx?q->last_tx->classid:0); - - /* try to dequeue direct packets as high prio (!) to minimize cpu work */ - if ((skb = __skb_dequeue(&q->direct_queue)) != NULL) { - sch->flags &= ~TCQ_F_THROTTLED; - sch->q.qlen--; - return skb; - } - - PSCHED_GET_TIME(q->now); /* htb_dequeue_class needs it too */ - q->delay = 0; q->sn++; + int i; + struct { + rb_node_t *root; + rb_node_t **pptr; + } stk[TC_HTB_MAXDEPTH],*sp = stk; + + sp->root = tree->rb_node; + sp->pptr = pptr; + + for (i = 0; i < 65535; i++) { + if (!*sp->pptr) { /* we are at right end; rewind & go up */ + *sp->pptr = sp->root; + while ((*sp->pptr)->rb_left) + *sp->pptr = (*sp->pptr)->rb_left; + if (sp > stk) { + sp--; + BUG_TRAP(*sp->pptr); if(!*sp->pptr) return NULL; + htb_next_rb_node (sp->pptr); + } + } else { + struct htb_class *cl; + cl = rb_entry(*sp->pptr,struct htb_class,node[prio]); + HTB_CHCL(cl); + if (!cl->level) + return cl; + (++sp)->root = cl->un.inner.feed[prio].rb_node; + sp->pptr = cl->un.inner.ptr+prio; + } + } + BUG_TRAP(0); + return NULL; +} - /* well here I bite CBQ's speed :-) if last dequeued class is - still active and is not deficit then we can dequeue it again */ - if ((cl = q->last_tx) != NULL && cl->q->q.qlen > 0 && - cl->deficit[cl->rc_level & 0xff] > 0 && - (skb = htb_dequeue_class(sch,cl)) != NULL) { - sch->q.qlen--; - cl->deficit[cl->rc_level & 0xff] -= skb->len; - sch->flags &= ~TCQ_F_THROTTLED; - q->dcache_hits++; - HTB_DBG(3,1,"htb_deq_hit skb=%p\n",skb); +/* dequeues packet at given priority and level; call only if + you are sure that there is active class at prio/level */ +static struct sk_buff * +htb_dequeue_tree(struct htb_sched *q,int prio,int level) +{ + struct sk_buff *skb = NULL; + //struct htb_sched *q = (struct htb_sched *)sch->data; + struct htb_class *cl,*start; + /* look initial class up in the row */ + DEVIK_MSTART(6); + start = cl = htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio); + + do { + BUG_TRAP(cl && cl->un.leaf.q->q.qlen); if (!cl) return NULL; + HTB_DBG(4,1,"htb_deq_tr prio=%d lev=%d cl=%X defic=%d\n", + prio,level,cl->classid,cl->un.leaf.deficit[level]); + + if (likely((skb = cl->un.leaf.q->dequeue(cl->un.leaf.q)) != NULL)) + break; + if (!cl->warned) { + printk(KERN_WARNING "htb: class %X isn't work conserving ?!\n",cl->classid); + cl->warned = 1; + } + q->nwc_hit++; + htb_next_rb_node((level?cl->parent->un.inner.ptr:q->ptr[0])+prio); + cl = htb_lookup_leaf (q->row[level]+prio,prio,q->ptr[level]+prio); + } while (cl != start); + + DEVIK_MEND(6); + DEVIK_MSTART(7); + if (likely(skb != NULL)) { + if ((cl->un.leaf.deficit[level] -= skb->len) < 0) { + HTB_DBG(4,2,"htb_next_cl oldptr=%p quant_add=%d\n", + level?cl->parent->un.inner.ptr[prio]:q->ptr[0][prio],cl->un.leaf.quantum); + cl->un.leaf.deficit[level] += cl->un.leaf.quantum; + htb_next_rb_node((level?cl->parent->un.inner.ptr:q->ptr[0])+prio); + } + /* this used to be after charge_class but this constelation + gives us slightly better performance */ + if (!cl->un.leaf.q->q.qlen) + htb_deactivate (q,cl); + DEVIK_MSTART(8); + htb_charge_class (q,cl,level,skb->len); + DEVIK_MEND(8); + } + DEVIK_MEND(7); return skb; - } - q->last_tx = NULL; /* can't use cache ? invalidate */ - - for (i = 0; i < TC_HTB_MAXDEPTH; i++) { - /* first try: dequeue leaves (level 0) */ - oklev = TC_HTB_MAXDEPTH; - q->trials_c++; - for (prio = 0; prio < TC_HTB_NUMPRIO; prio++) { - if (!q->active[prio][0]) continue; - lev = 0; skb = htb_dequeue_prio(sch, prio, &lev); - HTB_DBG(3,2,"htb_deq_1 i=%d p=%d skb=%p blev=%d\n",i,prio,skb,lev); - if (skb) { - sch->flags &= ~TCQ_F_THROTTLED; - goto fin; - } - if (lev < oklev) { - oklev = lev; okprio = prio; - } - } - if (oklev >= TC_HTB_MAXDEPTH) break; - /* second try: use ok level we learned in first try; - it really should succeed */ - q->trials_c++; - skb = htb_dequeue_prio(sch, okprio, &oklev); - HTB_DBG(3,2,"htb_deq_2 p=%d lev=%d skb=%p\n",okprio,oklev,skb); - if (skb) { - sch->flags &= ~TCQ_F_THROTTLED; - goto fin; - } - /* probably qdisc at oklev can't transmit - it is not good - idea to have TBF as HTB's child ! retry with that node - disabled */ - } - if (i >= TC_HTB_MAXDEPTH) - printk(KERN_ERR "htb: too many dequeue trials\n"); +} - /* no-one gave us packet, setup timer if someone wants it */ - if (sch->q.qlen && !netif_queue_stopped(sch->dev) && q->delay) { - long delay = PSCHED_US2JIFFIE(q->delay); - if (delay == 0) delay = 1; - if (delay > 5*HZ) { - if (net_ratelimit()) - printk(KERN_INFO "HTB delay %ld > 5sec\n", delay); - delay = 5*HZ; +static void htb_delay_by(struct Qdisc *sch,long delay) +{ + struct htb_sched *q = (struct htb_sched *)sch->data; + if (netif_queue_stopped(sch->dev)) return; + if (delay <= 0) delay = 1; + if (unlikely(delay > 5*HZ)) { + if (net_ratelimit()) + printk(KERN_INFO "HTB delay %ld > 5sec\n", delay); + delay = 5*HZ; } del_timer(&q->timer); q->timer.expires = jiffies + delay; @@ -725,439 +1007,605 @@ sch->flags |= TCQ_F_THROTTLED; sch->stats.overlimits++; HTB_DBG(3,1,"htb_deq t_delay=%ld\n",delay); - } -fin: - do { - static unsigned util = 0; unsigned d; - PSCHED_GET_TIME(endt); q->deq_rate_c++; - d = PSCHED_TDIFF(endt,q->now); - q->utilz_c += d; util += d; -#if 0 - /* special debug hack */ - if (skb) { - memcpy (skb->data+28,_dbg,sizeof(_dbg)); - memset (_dbg,0,sizeof(_dbg)); +} + +static struct sk_buff *htb_dequeue(struct Qdisc *sch) +{ + struct sk_buff *skb = NULL; + struct htb_sched *q = (struct htb_sched *)sch->data; + int level; + long min_delay; + + HTB_DBG(3,1,"htb_deq dircnt=%d qlen=%d\n",skb_queue_len(&q->direct_queue), + sch->q.qlen); + + /* try to dequeue direct packets as high prio (!) to minimize cpu work */ + if ((skb = __skb_dequeue(&q->direct_queue)) != NULL) { + sch->flags &= ~TCQ_F_THROTTLED; + sch->q.qlen--; + return skb; + } + + DEVIK_MSTART(2); + if (!sch->q.qlen) goto fin; + PSCHED_GET_TIME(q->now); + + min_delay = HZ*5; + q->nwc_hit = 0; + for (level = 0; level < TC_HTB_MAXDEPTH; level++) { + /* common case optimization - skip event handler quickly */ + int m; + long delay; + DEVIK_MSTART(3); + if (jiffies - q->near_ev_cache[level] < 0x80000000 || 0) { + delay = htb_do_events(q,level); + q->near_ev_cache[level] += delay ? delay : HZ; + } else + delay = q->near_ev_cache[level] - jiffies; + + if (delay && min_delay > delay) + min_delay = delay; + DEVIK_MEND(3); + DEVIK_MSTART(5); + m = ~q->row_mask[level]; + while (m != (int)(-1)) { + int prio = ffz (m); + m |= 1 << prio; + skb = htb_dequeue_tree(q,prio,level); + if (likely(skb != NULL)) { + sch->q.qlen--; + sch->flags &= ~TCQ_F_THROTTLED; + DEVIK_MEND(5); + goto fin; + } + } + DEVIK_MEND(5); + } + DEVIK_MSTART(4); +#ifdef HTB_DEBUG + if (!q->nwc_hit && min_delay >= 5*HZ && net_ratelimit()) { + printk(KERN_ERR "HTB: mindelay=%ld, report it please !\n",min_delay); + htb_debug_dump(q); } #endif - } while (0); - HTB_DBG(3,1,"htb_deq_end %s j=%lu skb=%p (28/11)\n",sch->dev->name,jiffies,skb); - return skb; + htb_delay_by (sch,min_delay); + DEVIK_MEND(4); +fin: + HTB_DBG(3,1,"htb_deq_end %s j=%lu skb=%p\n",sch->dev->name,jiffies,skb); + DEVIK_MEND(2); + return skb; } /* try to drop from each class (by prio) until one succeed */ static int htb_drop(struct Qdisc* sch) { - struct htb_sched *q = (struct htb_sched *)sch->data; - int prio; + struct htb_sched *q = (struct htb_sched *)sch->data; + int prio; - for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) { - struct htb_class *cl = q->active[prio][0]; - if (cl) do { - if (cl->q->ops->drop && cl->q->ops->drop(cl->q)) { - sch->q.qlen--; - return 1; - } - } while ((cl = cl->active.next) != q->active[prio][0]); - } - return 0; + for (prio = TC_HTB_NUMPRIO - 1; prio >= 0; prio--) { + struct list_head *p; + list_for_each (p,q->drops+prio) { + struct htb_class *cl = list_entry(p,struct htb_class, + un.leaf.drop_list); + if (cl->un.leaf.q->ops->drop && + cl->un.leaf.q->ops->drop(cl->un.leaf.q)) { + sch->q.qlen--; + if (!cl->un.leaf.q->q.qlen) + htb_deactivate (q,cl); + return 1; + } + } + } + return 0; } /* reset all classes */ /* always caled under BH & queue lock */ static void htb_reset(struct Qdisc* sch) { - struct htb_sched *q = (struct htb_sched *)sch->data; - int i; - HTB_DBG(0,1,"htb_reset sch=%X, handle=%X\n",(int)sch,sch->handle); - - for (i = 0; i < HTB_HSIZE; i++) { - struct htb_class *cl = q->hash[i]; - if (cl) do { - if (cl->q) qdisc_reset(cl->q); + struct htb_sched *q = (struct htb_sched *)sch->data; + int i; + HTB_DBG(0,1,"htb_reset sch=%p, handle=%X\n",sch,sch->handle); + + for (i = 0; i < HTB_HSIZE; i++) { + struct list_head *p; + list_for_each (p,q->hash+i) { + struct htb_class *cl = list_entry(p,struct htb_class,hlist); + if (cl->level) + memset(&cl->un.inner,0,sizeof(cl->un.inner)); + else { + if (cl->un.leaf.q) + qdisc_reset(cl->un.leaf.q); + INIT_LIST_HEAD(&cl->un.leaf.drop_list); + } + cl->prio_activity = 0; + cl->cmode = HTB_CAN_SEND; +#ifdef HTB_DEBUG + cl->pq_node.rb_color = -1; + memset(cl->node,255,sizeof(cl->node)); +#endif - } while ((cl = cl->hlist.next) != q->hash[i]); - } - sch->flags &= ~TCQ_F_THROTTLED; - del_timer(&q->timer); - __skb_queue_purge(&q->direct_queue); - sch->q.qlen = 0; q->last_tx = NULL; + } + } + sch->flags &= ~TCQ_F_THROTTLED; + del_timer(&q->timer); + __skb_queue_purge(&q->direct_queue); + sch->q.qlen = 0; + memset(q->row,0,sizeof(q->row)); + memset(q->row_mask,0,sizeof(q->row_mask)); + memset(q->wait_pq,0,sizeof(q->wait_pq)); + memset(q->ptr,0,sizeof(q->ptr)); + for (i = 0; i < TC_HTB_NUMPRIO; i++) + INIT_LIST_HEAD(q->drops+i); } static int htb_init(struct Qdisc *sch, struct rtattr *opt) { - struct htb_sched *q = (struct htb_sched*)sch->data; - struct rtattr *tb[TCA_HTB_INIT]; - struct tc_htb_glob *gopt; - - if (!opt || - rtattr_parse(tb, TCA_HTB_INIT, RTA_DATA(opt), RTA_PAYLOAD(opt)) || - tb[TCA_HTB_INIT-1] == NULL || - RTA_PAYLOAD(tb[TCA_HTB_INIT-1]) < sizeof(*gopt)) - return -EINVAL; - - gopt = RTA_DATA(tb[TCA_HTB_INIT-1]); - memset(q,0,sizeof(*q)); - q->debug = gopt->debug; - HTB_DBG(0,1,"htb_init sch=%p handle=%X r2q=%d\n",sch,sch->handle,gopt->rate2quantum); - init_timer(&q->timer); - init_timer(&q->rttim); - skb_queue_head_init(&q->direct_queue); - q->direct_qlen = sch->dev->tx_queue_len; - q->timer.function = htb_timer; - q->timer.data = (unsigned long)sch; - q->rttim.function = htb_rate_timer; - q->rttim.data = (unsigned long)sch; - q->rttim.expires = jiffies + HZ; - add_timer(&q->rttim); - if ((q->rate2quantum = gopt->rate2quantum) < 1) - q->rate2quantum = 1; - q->defcls = gopt->defcls; - q->use_dcache = gopt->use_dcache; + struct htb_sched *q = (struct htb_sched*)sch->data; + struct rtattr *tb[TCA_HTB_INIT]; + struct tc_htb_glob *gopt; + int i; +#ifdef HTB_DEBUG + printk(KERN_INFO "HTB init, kernel part version %d.%d\n", + HTB_VER >> 16,HTB_VER & 0xffff); +#endif + if (!opt || rtattr_parse(tb, TCA_HTB_INIT, RTA_DATA(opt), RTA_PAYLOAD(opt)) || + tb[TCA_HTB_INIT-1] == NULL || + RTA_PAYLOAD(tb[TCA_HTB_INIT-1]) < sizeof(*gopt)) { + printk(KERN_ERR "HTB: hey probably you have bad tc tool ?\n"); + return -EINVAL; + } + gopt = RTA_DATA(tb[TCA_HTB_INIT-1]); + if (gopt->version != HTB_VER >> 16) { + printk(KERN_ERR "HTB: need tc/htb version %d (minor is %d), you have %d\n", + HTB_VER >> 16,HTB_VER & 0xffff,gopt->version); + return -EINVAL; + } + memset(q,0,sizeof(*q)); + q->debug = gopt->debug; + HTB_DBG(0,1,"htb_init sch=%p handle=%X r2q=%d\n",sch,sch->handle,gopt->rate2quantum); + + INIT_LIST_HEAD(&q->root); + for (i = 0; i < HTB_HSIZE; i++) + INIT_LIST_HEAD(q->hash+i); + for (i = 0; i < TC_HTB_NUMPRIO; i++) + INIT_LIST_HEAD(q->drops+i); + + init_timer(&q->timer); + skb_queue_head_init(&q->direct_queue); + + q->direct_qlen = sch->dev->tx_queue_len; + if (q->direct_qlen < 2) /* some devices have zero tx_queue_len */ + q->direct_qlen = 2; + q->timer.function = htb_timer; + q->timer.data = (unsigned long)sch; + +#ifdef HTB_RATECM + init_timer(&q->rttim); + q->rttim.function = htb_rate_timer; + q->rttim.data = (unsigned long)sch; + q->rttim.expires = jiffies + HZ; + add_timer(&q->rttim); +#endif + if ((q->rate2quantum = gopt->rate2quantum) < 1) + q->rate2quantum = 1; + q->defcls = gopt->defcls; - MOD_INC_USE_COUNT; - return 0; + MOD_INC_USE_COUNT; + return 0; } -#ifdef CONFIG_RTNETLINK static int htb_dump(struct Qdisc *sch, struct sk_buff *skb) { - struct htb_sched *q = (struct htb_sched*)sch->data; - unsigned char *b = skb->tail; - struct rtattr *rta; - struct tc_htb_glob gopt; - HTB_DBG(0,1,"htb_dump sch=%p, handle=%X\n",sch,sch->handle); - /* stats */ - HTB_QLOCK(sch); - gopt.deq_rate = q->deq_rate/HTB_EWMAC; - gopt.utilz = q->utilz/HTB_EWMAC; - gopt.trials = q->trials/HTB_EWMAC; - gopt.dcache_hits = q->dcache_hits; - gopt.direct_pkts = q->direct_pkts; - - gopt.use_dcache = q->use_dcache; - gopt.rate2quantum = q->rate2quantum; - gopt.defcls = q->defcls; - gopt.debug = q->debug; - rta = (struct rtattr*)b; - RTA_PUT(skb, TCA_OPTIONS, 0, NULL); - RTA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt); - rta->rta_len = skb->tail - b; - sch->stats.qlen = sch->q.qlen; - RTA_PUT(skb, TCA_STATS, sizeof(sch->stats), &sch->stats); - HTB_QUNLOCK(sch); - return skb->len; + struct htb_sched *q = (struct htb_sched*)sch->data; + unsigned char *b = skb->tail; + struct rtattr *rta; + struct tc_htb_glob gopt; + HTB_DBG(0,1,"htb_dump sch=%p, handle=%X\n",sch,sch->handle); + /* stats */ + HTB_QLOCK(sch); + gopt.direct_pkts = q->direct_pkts; + +#ifdef HTB_DEBUG + htb_debug_dump(q); +#endif + gopt.version = HTB_VER; + gopt.rate2quantum = q->rate2quantum; + gopt.defcls = q->defcls; + gopt.debug = q->debug; + rta = (struct rtattr*)b; + RTA_PUT(skb, TCA_OPTIONS, 0, NULL); + RTA_PUT(skb, TCA_HTB_INIT, sizeof(gopt), &gopt); + rta->rta_len = skb->tail - b; + sch->stats.qlen = sch->q.qlen; + RTA_PUT(skb, TCA_STATS, sizeof(sch->stats), &sch->stats); + HTB_QUNLOCK(sch); + return skb->len; rtattr_failure: - HTB_QUNLOCK(sch); - skb_trim(skb, skb->tail - skb->data); - return -1; + HTB_QUNLOCK(sch); + skb_trim(skb, skb->tail - skb->data); + return -1; } -static int -htb_dump_class(struct Qdisc *sch, unsigned long arg, +static int htb_dump_class(struct Qdisc *sch, unsigned long arg, struct sk_buff *skb, struct tcmsg *tcm) { - struct htb_sched *q = (struct htb_sched*)sch->data; - struct htb_class *cl = (struct htb_class*)arg; - unsigned char *b = skb->tail; - struct rtattr *rta; - struct tc_htb_opt opt; - - HTB_DBG(0,1,"htb_dump_class handle=%X clid=%X\n",sch->handle,cl->classid); - - HTB_QLOCK(sch); - tcm->tcm_parent = cl->parent ? cl->parent->classid : TC_H_ROOT; - tcm->tcm_handle = cl->classid; - if (cl->q) { - tcm->tcm_info = cl->q->handle; - cl->stats.qlen = cl->q->q.qlen; - } +#ifdef HTB_DEBUG + struct htb_sched *q = (struct htb_sched*)sch->data; +#endif + struct htb_class *cl = (struct htb_class*)arg; + unsigned char *b = skb->tail; + struct rtattr *rta; + struct tc_htb_opt opt; + + HTB_DBG(0,1,"htb_dump_class handle=%X clid=%X\n",sch->handle,cl->classid); + + HTB_QLOCK(sch); + tcm->tcm_parent = cl->parent ? cl->parent->classid : TC_H_ROOT; + tcm->tcm_handle = cl->classid; + if (!cl->level && cl->un.leaf.q) { + tcm->tcm_info = cl->un.leaf.q->handle; + cl->stats.qlen = cl->un.leaf.q->q.qlen; + } + + rta = (struct rtattr*)b; + RTA_PUT(skb, TCA_OPTIONS, 0, NULL); - rta = (struct rtattr*)b; - RTA_PUT(skb, TCA_OPTIONS, 0, NULL); + memset (&opt,0,sizeof(opt)); - memset (&opt,0,sizeof(opt)); + opt.rate = cl->rate->rate; opt.buffer = cl->buffer; + opt.ceil = cl->ceil->rate; opt.cbuffer = cl->cbuffer; + opt.quantum = cl->un.leaf.quantum; opt.prio = cl->un.leaf.prio; + opt.level = cl->level; + RTA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt); + rta->rta_len = skb->tail - b; + +#ifdef HTB_RATECM + cl->stats.bps = cl->rate_bytes/(HTB_EWMAC*HTB_HSIZE); + cl->stats.pps = cl->rate_packets/(HTB_EWMAC*HTB_HSIZE); +#endif - opt.rate = cl->rate->rate; opt.buffer = cl->buffer; - opt.ceil = cl->ceil->rate; opt.cbuffer = cl->cbuffer; - opt.quantum = cl->quantum; opt.prio = cl->prio; - opt.level = cl->level; opt.injectd = cl->injectd; - RTA_PUT(skb, TCA_HTB_PARMS, sizeof(opt), &opt); - rta->rta_len = skb->tail - b; - - cl->stats.bps = cl->rate_bytes/(HTB_EWMAC*HTB_HSIZE); - cl->stats.pps = cl->rate_packets/(HTB_EWMAC*HTB_HSIZE); - - cl->xstats.tokens = cl->tokens; - cl->xstats.ctokens = cl->ctokens; - RTA_PUT(skb, TCA_STATS, sizeof(cl->stats), &cl->stats); - RTA_PUT(skb, TCA_XSTATS, sizeof(cl->xstats), &cl->xstats); - HTB_QUNLOCK(sch); - return skb->len; + cl->xstats.tokens = cl->tokens; + cl->xstats.ctokens = cl->ctokens; + RTA_PUT(skb, TCA_STATS, sizeof(cl->stats), &cl->stats); + RTA_PUT(skb, TCA_XSTATS, sizeof(cl->xstats), &cl->xstats); + HTB_QUNLOCK(sch); + return skb->len; rtattr_failure: - HTB_QUNLOCK(sch); - skb_trim(skb, b - skb->data); - return -1; + HTB_QUNLOCK(sch); + skb_trim(skb, b - skb->data); + return -1; } -#endif static int htb_graft(struct Qdisc *sch, unsigned long arg, struct Qdisc *new, struct Qdisc **old) { - struct htb_class *cl = (struct htb_class*)arg; + struct htb_class *cl = (struct htb_class*)arg; - if (cl && !cl->level) { - if (new == NULL) { - if ((new = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops)) == NULL) - return -ENOBUFS; - } - if ((*old = xchg(&cl->q, new)) != NULL) /* xchg is atomical :-) */ - qdisc_reset(*old); - return 0; - } - return -ENOENT; + if (cl && !cl->level) { + if (new == NULL && (new = qdisc_create_dflt(sch->dev, + &pfifo_qdisc_ops)) == NULL) + return -ENOBUFS; + sch_tree_lock(sch); + if ((*old = xchg(&cl->un.leaf.q, new)) != NULL) { + /* TODO: is it correct ? Why CBQ doesn't do it ? */ + sch->q.qlen -= (*old)->q.qlen; + qdisc_reset(*old); + } + sch_tree_unlock(sch); + return 0; + } + return -ENOENT; } -static struct Qdisc * -htb_leaf(struct Qdisc *sch, unsigned long arg) +static struct Qdisc * htb_leaf(struct Qdisc *sch, unsigned long arg) { - struct htb_class *cl = (struct htb_class*)arg; - return cl ? cl->q : NULL; + struct htb_class *cl = (struct htb_class*)arg; + return (cl && !cl->level) ? cl->un.leaf.q : NULL; } static unsigned long htb_get(struct Qdisc *sch, u32 classid) { - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl = htb_find(classid,sch); - HTB_DBG(0,1,"htb_get clid=%X q=%p cl=%p ref=%d\n",classid,q,cl,cl?cl->refcnt:0); - if (cl) cl->refcnt++; - return (unsigned long)cl; +#ifdef HTB_DEBUG + struct htb_sched *q = (struct htb_sched *)sch->data; +#endif + struct htb_class *cl = htb_find(classid,sch); + HTB_DBG(0,1,"htb_get clid=%X q=%p cl=%p ref=%d\n",classid,q,cl,cl?cl->refcnt:0); + if (cl) + cl->refcnt++; + return (unsigned long)cl; } static void htb_destroy_filters(struct tcf_proto **fl) { - struct tcf_proto *tp; + struct tcf_proto *tp; - while ((tp = *fl) != NULL) { - *fl = tp->next; - tp->ops->destroy(tp); - } + while ((tp = *fl) != NULL) { + *fl = tp->next; + tp->ops->destroy(tp); + } } static void htb_destroy_class(struct Qdisc* sch,struct htb_class *cl) { - struct htb_sched *q = (struct htb_sched *)sch->data; - HTB_DBG(0,1,"htb_destrycls clid=%X q=%p ref=%d\n", cl?cl->classid:0,cl->q,cl?cl->refcnt:0); - if (cl->q) qdisc_destroy(cl->q); - qdisc_put_rtab(cl->rate); - qdisc_put_rtab(cl->ceil); + struct htb_sched *q = (struct htb_sched *)sch->data; + HTB_DBG(0,1,"htb_destrycls clid=%X ref=%d\n", cl?cl->classid:0,cl?cl->refcnt:0); + if (!cl->level) { + BUG_TRAP(cl->un.leaf.q); + sch->q.qlen -= cl->un.leaf.q->q.qlen; + qdisc_destroy(cl->un.leaf.q); + } + qdisc_put_rtab(cl->rate); + qdisc_put_rtab(cl->ceil); + #ifdef CONFIG_NET_ESTIMATOR - qdisc_kill_estimator(&cl->stats); + qdisc_kill_estimator(&cl->stats); #endif - htb_destroy_filters (&cl->filter_list); - /* remove children */ - while (cl->children) htb_destroy_class (sch,cl->children); - - /* remove class from all lists it is on */ - q->last_tx = NULL; - if (cl->hlist.prev) - HTB_DELETE(hlist,cl,q->hash[htb_hash(cl->classid)]); - if (cl->active.prev) - htb_deactivate (q,cl,0); - if (cl->parent) - HTB_DELETE(sibling,cl,cl->parent->children); - else - HTB_DELETE(sibling,cl,q->root); - - kfree(cl); + htb_destroy_filters (&cl->filter_list); + + while (!list_empty(&cl->children)) + htb_destroy_class (sch,list_entry(cl->children.next, + struct htb_class,sibling)); + + /* note: this delete may happen twice (see htb_delete) */ + list_del(&cl->hlist); + list_del(&cl->sibling); + + if (cl->prio_activity) + htb_deactivate (q,cl); + + if (cl->cmode != HTB_CAN_SEND) + htb_safe_rb_erase(&cl->pq_node,q->wait_pq+cl->level); + + kfree(cl); } /* always caled under BH & queue lock */ static void htb_destroy(struct Qdisc* sch) { - struct htb_sched *q = (struct htb_sched *)sch->data; - HTB_DBG(0,1,"htb_destroy q=%p\n",q); + struct htb_sched *q = (struct htb_sched *)sch->data; + HTB_DBG(0,1,"htb_destroy q=%p\n",q); - del_timer_sync (&q->timer); - del_timer_sync (&q->rttim); - while (q->root) htb_destroy_class(sch,q->root); - htb_destroy_filters(&q->filter_list); - __skb_queue_purge(&q->direct_queue); - MOD_DEC_USE_COUNT; + del_timer_sync (&q->timer); +#ifdef HTB_RATECM + del_timer_sync (&q->rttim); +#endif + while (!list_empty(&q->root)) + htb_destroy_class (sch,list_entry(q->root.next, + struct htb_class,sibling)); + + htb_destroy_filters(&q->filter_list); + __skb_queue_purge(&q->direct_queue); + MOD_DEC_USE_COUNT; } -static void htb_put(struct Qdisc *sch, unsigned long arg) +static int htb_delete(struct Qdisc *sch, unsigned long arg) { - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl = (struct htb_class*)arg; - HTB_DBG(0,1,"htb_put q=%p cl=%X ref=%d\n",q,cl?cl->classid:0,cl?cl->refcnt:0); + struct htb_sched *q = (struct htb_sched *)sch->data; + struct htb_class *cl = (struct htb_class*)arg; + HTB_DBG(0,1,"htb_delete q=%p cl=%X ref=%d\n",q,cl?cl->classid:0,cl?cl->refcnt:0); + + // TODO: why don't allow to delete subtree ? references ? does + // tc subsys quarantee us that in htb_destroy it holds no class + // refs so that we can remove children safely there ? + if (!list_empty(&cl->children) || cl->filter_cnt) + return -EBUSY; + + sch_tree_lock(sch); + + /* delete from hash and active; remainder in destroy_class */ + list_del_init(&cl->hlist); + if (cl->prio_activity) + htb_deactivate (q,cl); + + if (--cl->refcnt == 0) + htb_destroy_class(sch,cl); - if (--cl->refcnt == 0) - htb_destroy_class(sch,cl); + sch_tree_unlock(sch); + return 0; } -static int -htb_change_class(struct Qdisc *sch, u32 classid, u32 parentid, struct rtattr **tca, - unsigned long *arg) +static void htb_put(struct Qdisc *sch, unsigned long arg) { - int err = -EINVAL,h; - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl = (struct htb_class*)*arg,*parent; - struct rtattr *opt = tca[TCA_OPTIONS-1]; - struct qdisc_rate_table *rtab = NULL, *ctab = NULL; - struct rtattr *tb[TCA_HTB_RTAB]; - struct tc_htb_opt *hopt; - - if (parentid == TC_H_ROOT) parent = NULL; - else parent = htb_find (parentid,sch); - - /* extract all subattrs from opt attr */ - if (!opt || - rtattr_parse(tb, TCA_HTB_RTAB, RTA_DATA(opt), RTA_PAYLOAD(opt)) || - tb[TCA_HTB_PARMS-1] == NULL || - RTA_PAYLOAD(tb[TCA_HTB_PARMS-1]) < sizeof(*hopt)) - goto failure; - - hopt = RTA_DATA(tb[TCA_HTB_PARMS-1]); - HTB_DBG(0,1,"htb_chg cl=%p, clid=%X, opt/prio=%d, rate=%u, buff=%d, quant=%d\n", cl,cl?cl->classid:0,(int)hopt->prio,hopt->rate.rate,hopt->buffer,hopt->quantum); - rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB-1]); - ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB-1]); - if (!rtab || !ctab) goto failure; - - if (!cl) { /* new class */ - /* check maximal depth */ - if (parent && parent->parent && parent->parent->level < 2) { - printk(KERN_ERR "htb: tree is too deep\n"); - goto failure; - } - err = -ENOBUFS; - cl = kmalloc(sizeof(*cl), GFP_KERNEL); - if (cl == NULL) goto failure; - memset(cl, 0, sizeof(*cl)); - cl->refcnt = 1; cl->level = 0; /* assume leaf */ - - if (parent && !parent->level) { - /* turn parent into inner node */ - qdisc_destroy (parent->q); parent->q = &noop_qdisc; - parent->level = (parent->parent ? - parent->parent->level : TC_HTB_MAXDEPTH) - 1; - } - /* leaf (we) needs elementary qdisc */ - if (!(cl->q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops))) - cl->q = &noop_qdisc; - - cl->classid = classid; cl->parent = parent; - cl->tokens = hopt->buffer; - cl->ctokens = hopt->cbuffer; - cl->mbuffer = 60000000; /* 1min */ - PSCHED_GET_TIME(cl->t_c); +#ifdef HTB_DEBUG + struct htb_sched *q = (struct htb_sched *)sch->data; +#endif + struct htb_class *cl = (struct htb_class*)arg; + HTB_DBG(0,1,"htb_put q=%p cl=%X ref=%d\n",q,cl?cl->classid:0,cl?cl->refcnt:0); - /* attach to the hash list and parent's family */ - sch_tree_lock(sch); - h = htb_hash(classid); - if (!cl->hlist.prev) - HTB_INSERTB(hlist,q->hash[h],cl); - if (parent) - HTB_INSERTB(sibling,parent->children,cl); - else HTB_INSERTB(sibling,q->root,cl); - - } else sch_tree_lock(sch); - - q->last_tx = NULL; - cl->quantum = rtab->rate.rate / q->rate2quantum; - cl->injectd = hopt->injectd; - if (cl->quantum < 100) cl->quantum = 100; - if (cl->quantum > 60000) cl->quantum = 60000; - if ((cl->prio = hopt->prio) >= TC_HTB_NUMPRIO) - cl->prio = TC_HTB_NUMPRIO - 1; - - cl->buffer = hopt->buffer; - cl->cbuffer = hopt->cbuffer; - if (cl->rate) qdisc_put_rtab(cl->rate); cl->rate = rtab; - if (cl->ceil) qdisc_put_rtab(cl->ceil); cl->ceil = ctab; - sch_tree_unlock(sch); + if (--cl->refcnt == 0) + htb_destroy_class(sch,cl); +} - *arg = (unsigned long)cl; - return 0; +static int htb_change_class(struct Qdisc *sch, u32 classid, + u32 parentid, struct rtattr **tca, unsigned long *arg) +{ + int err = -EINVAL; + struct htb_sched *q = (struct htb_sched *)sch->data; + struct htb_class *cl = (struct htb_class*)*arg,*parent; + struct rtattr *opt = tca[TCA_OPTIONS-1]; + struct qdisc_rate_table *rtab = NULL, *ctab = NULL; + struct rtattr *tb[TCA_HTB_RTAB]; + struct tc_htb_opt *hopt; + + /* extract all subattrs from opt attr */ + if (!opt || rtattr_parse(tb, TCA_HTB_RTAB, RTA_DATA(opt), RTA_PAYLOAD(opt)) || + tb[TCA_HTB_PARMS-1] == NULL || + RTA_PAYLOAD(tb[TCA_HTB_PARMS-1]) < sizeof(*hopt)) + goto failure; + + parent = parentid == TC_H_ROOT ? NULL : htb_find (parentid,sch); + + hopt = RTA_DATA(tb[TCA_HTB_PARMS-1]); + HTB_DBG(0,1,"htb_chg cl=%p, clid=%X, opt/prio=%d, rate=%u, buff=%d, quant=%d\n", cl,cl?cl->classid:0,(int)hopt->prio,hopt->rate.rate,hopt->buffer,hopt->quantum); + rtab = qdisc_get_rtab(&hopt->rate, tb[TCA_HTB_RTAB-1]); + ctab = qdisc_get_rtab(&hopt->ceil, tb[TCA_HTB_CTAB-1]); + if (!rtab || !ctab) goto failure; + + if (!cl) { /* new class */ + /* check for valid classid */ + if (!classid || TC_H_MAJ(classid^sch->handle) || htb_find(classid,sch)) + goto failure; + + /* check maximal depth */ + if (parent && parent->parent && parent->parent->level < 2) { + printk(KERN_ERR "htb: tree is too deep\n"); + goto failure; + } + err = -ENOBUFS; + if ((cl = kmalloc(sizeof(*cl), GFP_KERNEL)) == NULL) + goto failure; + + memset(cl, 0, sizeof(*cl)); + cl->refcnt = 1; + INIT_LIST_HEAD(&cl->sibling); + INIT_LIST_HEAD(&cl->hlist); + INIT_LIST_HEAD(&cl->children); + INIT_LIST_HEAD(&cl->un.leaf.drop_list); +#ifdef HTB_DEBUG + cl->magic = HTB_CMAGIC; +#endif -failure: - if (rtab) qdisc_put_rtab(rtab); - if (ctab) qdisc_put_rtab(ctab); - return err; -} + sch_tree_lock(sch); + if (parent && !parent->level) { + /* turn parent into inner node */ + sch->q.qlen -= parent->un.leaf.q->q.qlen; + qdisc_destroy (parent->un.leaf.q); + if (parent->prio_activity) + htb_deactivate (q,parent); + + /* remove from evt list because of level change */ + if (parent->cmode != HTB_CAN_SEND) { + htb_safe_rb_erase(&parent->pq_node,q->wait_pq /*+0*/); + parent->cmode = HTB_CAN_SEND; + } + parent->level = (parent->parent ? parent->parent->level + : TC_HTB_MAXDEPTH) - 1; + memset (&parent->un.inner,0,sizeof(parent->un.inner)); + } + /* leaf (we) needs elementary qdisc */ + if (!(cl->un.leaf.q = qdisc_create_dflt(sch->dev, &pfifo_qdisc_ops))) + cl->un.leaf.q = &noop_qdisc; + + cl->classid = classid; cl->parent = parent; + + /* set class to be in HTB_CAN_SEND state */ + cl->tokens = hopt->buffer; + cl->ctokens = hopt->cbuffer; + cl->mbuffer = 60000000; /* 1min */ + PSCHED_GET_TIME(cl->t_c); + cl->cmode = HTB_CAN_SEND; + + /* attach to the hash list and parent's family */ + list_add_tail(&cl->hlist, q->hash+htb_hash(classid)); + list_add_tail(&cl->sibling, parent ? &parent->children : &q->root); +#ifdef HTB_DEBUG + { + int i; + for (i = 0; i < TC_HTB_NUMPRIO; i++) cl->node[i].rb_color = -1; + cl->pq_node.rb_color = -1; + } +#endif + } else sch_tree_lock(sch); -static int htb_delete(struct Qdisc *sch, unsigned long arg) -{ - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl = (struct htb_class*)arg; - HTB_DBG(0,1,"htb_delete q=%p cl=%X ref=%d\n",q,cl?cl->classid:0,cl?cl->refcnt:0); + /* it used to be a nasty bug here, we have to check that node + is really leaf before changing cl->un.leaf ! */ + if (!cl->level) { + cl->un.leaf.quantum = rtab->rate.rate / q->rate2quantum; + if (!hopt->quantum && cl->un.leaf.quantum < 1000) { + printk(KERN_WARNING "HTB: quantum of class %X is small. Consider r2q change.", cl->classid); + cl->un.leaf.quantum = 1000; + } + if (!hopt->quantum && cl->un.leaf.quantum > 200000) { + printk(KERN_WARNING "HTB: quantum of class %X is big. Consider r2q change.", cl->classid); + cl->un.leaf.quantum = 200000; + } + if (hopt->quantum) + cl->un.leaf.quantum = hopt->quantum; + if ((cl->un.leaf.prio = hopt->prio) >= TC_HTB_NUMPRIO) + cl->un.leaf.prio = TC_HTB_NUMPRIO - 1; + } - if (cl->children || cl->filter_cnt) return -EBUSY; - sch_tree_lock(sch); - /* delete from hash and active; remainder in destroy_class */ - if (cl->hlist.prev) - HTB_DELETE(hlist,cl,q->hash[htb_hash(cl->classid)]); - if (cl->active.prev) - htb_deactivate (q,cl,0); - q->last_tx = NULL; + cl->buffer = hopt->buffer; + cl->cbuffer = hopt->cbuffer; + if (cl->rate) qdisc_put_rtab(cl->rate); cl->rate = rtab; + if (cl->ceil) qdisc_put_rtab(cl->ceil); cl->ceil = ctab; + sch_tree_unlock(sch); - if (--cl->refcnt == 0) - htb_destroy_class(sch,cl); + *arg = (unsigned long)cl; + return 0; - sch_tree_unlock(sch); - return 0; +failure: + if (rtab) qdisc_put_rtab(rtab); + if (ctab) qdisc_put_rtab(ctab); + return err; } static struct tcf_proto **htb_find_tcf(struct Qdisc *sch, unsigned long arg) { - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl = (struct htb_class *)arg; - struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list; - HTB_DBG(0,2,"htb_tcf q=%p clid=%X fref=%d fl=%p\n",q,cl?cl->classid:0,cl?cl->filter_cnt:q->filter_cnt,*fl); - return fl; + struct htb_sched *q = (struct htb_sched *)sch->data; + struct htb_class *cl = (struct htb_class *)arg; + struct tcf_proto **fl = cl ? &cl->filter_list : &q->filter_list; + HTB_DBG(0,2,"htb_tcf q=%p clid=%X fref=%d fl=%p\n",q,cl?cl->classid:0,cl?cl->filter_cnt:q->filter_cnt,*fl); + return fl; } static unsigned long htb_bind_filter(struct Qdisc *sch, unsigned long parent, u32 classid) { - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl = htb_find (classid,sch); - HTB_DBG(0,2,"htb_bind q=%p clid=%X cl=%p fref=%d\n",q,classid,cl,cl?cl->filter_cnt:q->filter_cnt); - if (cl && !cl->level) return 0; - if (cl) cl->filter_cnt++; else q->filter_cnt++; - return (unsigned long)cl; + struct htb_sched *q = (struct htb_sched *)sch->data; + struct htb_class *cl = htb_find (classid,sch); + HTB_DBG(0,2,"htb_bind q=%p clid=%X cl=%p fref=%d\n",q,classid,cl,cl?cl->filter_cnt:q->filter_cnt); + /*if (cl && !cl->level) return 0; + The line above used to be there to prevent attaching filters to + leaves. But at least tc_index filter uses this just to get class + for other reasons so that we have to allow for it. + ---- + 19.6.2002 As Werner explained it is ok - bind filter is just + another way to "lock" the class - unlike "get" this lock can + be broken by class during destroy IIUC. + */ + if (cl) + cl->filter_cnt++; + else + q->filter_cnt++; + return (unsigned long)cl; } static void htb_unbind_filter(struct Qdisc *sch, unsigned long arg) { - struct htb_sched *q = (struct htb_sched *)sch->data; - struct htb_class *cl = (struct htb_class *)arg; - HTB_DBG(0,2,"htb_unbind q=%p cl=%p fref=%d\n",q,cl,cl?cl->filter_cnt:q->filter_cnt); - if (cl) cl->filter_cnt--; else q->filter_cnt--; + struct htb_sched *q = (struct htb_sched *)sch->data; + struct htb_class *cl = (struct htb_class *)arg; + HTB_DBG(0,2,"htb_unbind q=%p cl=%p fref=%d\n",q,cl,cl?cl->filter_cnt:q->filter_cnt); + if (cl) + cl->filter_cnt--; + else + q->filter_cnt--; } static void htb_walk(struct Qdisc *sch, struct qdisc_walker *arg) { - struct htb_sched *q = (struct htb_sched *)sch->data; - int i; + struct htb_sched *q = (struct htb_sched *)sch->data; + int i; - if (arg->stop) - return; - - for (i = 0; i < HTB_HSIZE; i++) { - struct htb_class *cl = q->hash[i]; - if (cl) do { - if (arg->count < arg->skip) { - arg->count++; - continue; - } - if (arg->fn(sch, (unsigned long)cl, arg) < 0) { - arg->stop = 1; + if (arg->stop) return; - } - arg->count++; - } while ((cl = cl->hlist.next) != q->hash[i]); - } + for (i = 0; i < HTB_HSIZE; i++) { + struct list_head *p; + list_for_each (p,q->hash+i) { + struct htb_class *cl = list_entry(p,struct htb_class,hlist); + if (arg->count < arg->skip) { + arg->count++; + continue; + } + if (arg->fn(sch, (unsigned long)cl, arg) < 0) { + arg->stop = 1; + return; + } + arg->count++; + } + } } static struct Qdisc_class_ops htb_class_ops = @@ -1174,9 +1622,7 @@ htb_bind_filter, htb_unbind_filter, -#ifdef CONFIG_RTNETLINK htb_dump_class, -#endif }; struct Qdisc_ops htb_qdisc_ops = @@ -1196,9 +1642,7 @@ htb_destroy, NULL /* htb_change */, -#ifdef CONFIG_RTNETLINK htb_dump, -#endif }; #ifdef MODULE