diff options
author | Luiz Otavio O Souza <luiz@netgate.com> | 2015-11-11 13:59:04 -0600 |
---|---|---|
committer | Luiz Otavio O Souza <luiz@netgate.com> | 2015-11-11 13:59:04 -0600 |
commit | c548a571ceaeae698d02b862b4170d015a265ba7 (patch) | |
tree | 537578451fed3a14cc2be8fbd1560ab95ed77a8d | |
parent | b24875947f184e461a05c24a653b99071580feef (diff) | |
download | FreeBSD-src-c548a571ceaeae698d02b862b4170d015a265ba7.zip FreeBSD-src-c548a571ceaeae698d02b862b4170d015a265ba7.tar.gz |
MFC r284777:
ALTQ FAIRQ discipline import from DragonFLY
Differential Revision: https://reviews.freebsd.org/D2847
Reviewed by: glebius, wblock(manpage)
Approved by: gnn(mentor)
Obtained from: pfSense
Sponsored by: Netgate
TAG: FAIRQ
-rw-r--r-- | sbin/pfctl/parse.y | 75 | ||||
-rw-r--r-- | sbin/pfctl/pfctl_altq.c | 221 | ||||
-rw-r--r-- | sbin/pfctl/pfctl_parser.h | 15 | ||||
-rw-r--r-- | sbin/pfctl/pfctl_qstats.c | 30 | ||||
-rw-r--r-- | share/man/man4/altq.4 | 7 | ||||
-rw-r--r-- | sys/conf/NOTES | 1 | ||||
-rw-r--r-- | sys/conf/files | 1 | ||||
-rw-r--r-- | sys/conf/options | 1 | ||||
-rw-r--r-- | sys/contrib/altq/altq/altq.h | 3 | ||||
-rw-r--r-- | sys/contrib/altq/altq/altq_fairq.c | 889 | ||||
-rw-r--r-- | sys/contrib/altq/altq/altq_fairq.h | 137 | ||||
-rw-r--r-- | sys/contrib/altq/altq/altq_subr.c | 30 | ||||
-rw-r--r-- | sys/contrib/altq/altq/altq_var.h | 7 | ||||
-rw-r--r-- | sys/netpfil/pf/pf.c | 18 | ||||
-rw-r--r-- | sys/netpfil/pf/pf_altq.h | 15 | ||||
-rw-r--r-- | sys/netpfil/pf/pf_mtag.h | 1 |
16 files changed, 1445 insertions, 6 deletions
diff --git a/sbin/pfctl/parse.y b/sbin/pfctl/parse.y index 3edb634..2f98692 100644 --- a/sbin/pfctl/parse.y +++ b/sbin/pfctl/parse.y @@ -50,6 +50,7 @@ __FBSDID("$FreeBSD$"); #include <altq/altq_cbq.h> #include <altq/altq_priq.h> #include <altq/altq_hfsc.h> +#include <altq/altq_fairq.h> #include <stdio.h> #include <unistd.h> @@ -317,6 +318,7 @@ struct pool_opts { struct node_hfsc_opts hfsc_opts; +struct node_fairq_opts fairq_opts; struct node_state_opt *keep_state_defaults = NULL; int disallow_table(struct node_host *, const char *); @@ -440,6 +442,7 @@ typedef struct { struct table_opts table_opts; struct pool_opts pool_opts; struct node_hfsc_opts hfsc_opts; + struct node_fairq_opts fairq_opts; } v; int lineno; } YYSTYPE; @@ -464,8 +467,8 @@ int parseport(char *, struct range *r, int); %token REQUIREORDER SYNPROXY FINGERPRINTS NOSYNC DEBUG SKIP HOSTID %token ANTISPOOF FOR INCLUDE %token BITMASK RANDOM SOURCEHASH ROUNDROBIN STATICPORT PROBABILITY -%token ALTQ CBQ PRIQ HFSC BANDWIDTH TBRSIZE LINKSHARE REALTIME UPPERLIMIT -%token QUEUE PRIORITY QLIMIT RTABLE +%token ALTQ CBQ PRIQ HFSC FAIRQ BANDWIDTH TBRSIZE LINKSHARE REALTIME UPPERLIMIT +%token QUEUE PRIORITY QLIMIT HOGS BUCKETS RTABLE %token DNPIPE DNQUEUE %token LOAD RULESET_OPTIMIZATION %token STICKYADDRESS MAXSRCSTATES MAXSRCNODES SOURCETRACK GLOBAL RULE @@ -515,6 +518,7 @@ int parseport(char *, struct range *r, int); %type <v.number> cbqflags_list cbqflags_item %type <v.number> priqflags_list priqflags_item %type <v.hfsc_opts> hfscopts_list hfscopts_item hfsc_opts +%type <v.fairq_opts> fairqopts_list fairqopts_item fairq_opts %type <v.queue_bwspec> bandwidth %type <v.filter_opts> filter_opts filter_opt filter_opts_l %type <v.antispoof_opts> antispoof_opts antispoof_opt antispoof_opts_l @@ -1697,6 +1701,15 @@ scheduler : CBQ { $$.qtype = ALTQT_HFSC; $$.data.hfsc_opts = $3; } + | FAIRQ { + $$.qtype = ALTQT_FAIRQ; + bzero(&$$.data.fairq_opts, + sizeof(struct node_fairq_opts)); + } + | FAIRQ '(' fairq_opts ')' { + $$.qtype = ALTQT_FAIRQ; + $$.data.fairq_opts = $3; + } ; cbqflags_list : cbqflags_item { $$ |= $1; } @@ -1845,6 +1858,61 @@ hfscopts_item : LINKSHARE bandwidth { } ; +fairq_opts : { + bzero(&fairq_opts, + sizeof(struct node_fairq_opts)); + } + fairqopts_list { + $$ = fairq_opts; + } + ; + +fairqopts_list : fairqopts_item + | fairqopts_list comma fairqopts_item + ; + +fairqopts_item : LINKSHARE bandwidth { + if (fairq_opts.linkshare.used) { + yyerror("linkshare already specified"); + YYERROR; + } + fairq_opts.linkshare.m2 = $2; + fairq_opts.linkshare.used = 1; + } + | LINKSHARE '(' bandwidth number bandwidth ')' { + if (fairq_opts.linkshare.used) { + yyerror("linkshare already specified"); + YYERROR; + } + fairq_opts.linkshare.m1 = $3; + fairq_opts.linkshare.d = $4; + fairq_opts.linkshare.m2 = $5; + fairq_opts.linkshare.used = 1; + } + | HOGS bandwidth { + fairq_opts.hogs_bw = $2; + } + | BUCKETS number { + fairq_opts.nbuckets = $2; + } + | STRING { + if (!strcmp($1, "default")) + fairq_opts.flags |= FARF_DEFAULTCLASS; + else if (!strcmp($1, "red")) + fairq_opts.flags |= FARF_RED; + else if (!strcmp($1, "ecn")) + fairq_opts.flags |= FARF_RED|FARF_ECN; + else if (!strcmp($1, "rio")) + fairq_opts.flags |= FARF_RIO; + else { + yyerror("unknown fairq flag \"%s\"", $1); + free($1); + YYERROR; + } + free($1); + } + ; + qassign : /* empty */ { $$ = NULL; } | qassign_item { $$ = $1; } | '{' optnl qassign_list '}' { $$ = $3; } @@ -5516,6 +5584,7 @@ lookup(char *s) { "bitmask", BITMASK}, { "block", BLOCK}, { "block-policy", BLOCKPOLICY}, + { "buckets", BUCKETS}, { "cbq", CBQ}, { "code", CODE}, { "crop", FRAGCROP}, @@ -5528,6 +5597,7 @@ lookup(char *s) { "drop-ovl", FRAGDROP}, { "dscp", DSCP}, { "dup-to", DUPTO}, + { "fairq", FAIRQ}, { "fastroute", FASTROUTE}, { "file", FILENAME}, { "fingerprints", FINGERPRINTS}, @@ -5540,6 +5610,7 @@ lookup(char *s) { "global", GLOBAL}, { "group", GROUP}, { "hfsc", HFSC}, + { "hogs", HOGS}, { "hostid", HOSTID}, { "icmp-type", ICMPTYPE}, { "icmp6-type", ICMP6TYPE}, diff --git a/sbin/pfctl/pfctl_altq.c b/sbin/pfctl/pfctl_altq.c index be2776c..adba457 100644 --- a/sbin/pfctl/pfctl_altq.c +++ b/sbin/pfctl/pfctl_altq.c @@ -43,6 +43,7 @@ __FBSDID("$FreeBSD$"); #include <altq/altq_cbq.h> #include <altq/altq_priq.h> #include <altq/altq_hfsc.h> +#include <altq/altq_fairq.h> #include "pfctl_parser.h" #include "pfctl.h" @@ -69,6 +70,11 @@ static int check_commit_hfsc(int, int, struct pf_altq *); static int print_hfsc_opts(const struct pf_altq *, const struct node_queue_opt *); +static int eval_pfqueue_fairq(struct pfctl *, struct pf_altq *); +static int print_fairq_opts(const struct pf_altq *, + const struct node_queue_opt *); +static int check_commit_fairq(int, int, struct pf_altq *); + static void gsc_add_sc(struct gen_sc *, struct service_curve *); static int is_gsc_under_sc(struct gen_sc *, struct service_curve *); @@ -89,6 +95,8 @@ int eval_queue_opts(struct pf_altq *, struct node_queue_opt *, u_int32_t eval_bwspec(struct node_queue_bw *, u_int32_t); void print_hfsc_sc(const char *, u_int, u_int, u_int, const struct node_hfsc_sc *); +void print_fairq_sc(const char *, u_int, u_int, u_int, + const struct node_fairq_sc *); void pfaltq_store(struct pf_altq *a) @@ -174,6 +182,10 @@ print_altq(const struct pf_altq *a, unsigned int level, if (!print_hfsc_opts(a, qopts)) printf("hfsc "); break; + case ALTQT_FAIRQ: + if (!print_fairq_opts(a, qopts)) + printf("fairq "); + break; } if (bw != NULL && bw->bw_percent > 0) { @@ -204,7 +216,8 @@ print_queue(const struct pf_altq *a, unsigned int level, printf("%s ", a->qname); if (print_interface) printf("on %s ", a->ifname); - if (a->scheduler == ALTQT_CBQ || a->scheduler == ALTQT_HFSC) { + if (a->scheduler == ALTQT_CBQ || a->scheduler == ALTQT_HFSC || + a->scheduler == ALTQT_FAIRQ) { if (bw != NULL && bw->bw_percent > 0) { if (bw->bw_percent < 100) printf("bandwidth %u%% ", bw->bw_percent); @@ -225,6 +238,9 @@ print_queue(const struct pf_altq *a, unsigned int level, case ALTQT_HFSC: print_hfsc_opts(a, qopts); break; + case ALTQT_FAIRQ: + print_fairq_opts(a, qopts); + break; } } @@ -291,6 +307,9 @@ check_commit_altq(int dev, int opts) case ALTQT_HFSC: error = check_commit_hfsc(dev, opts, altq); break; + case ALTQT_FAIRQ: + error = check_commit_fairq(dev, opts, altq); + break; default: break; } @@ -339,7 +358,8 @@ eval_pfqueue(struct pfctl *pf, struct pf_altq *pa, struct node_queue_bw *bw, if (pa->qlimit == 0) pa->qlimit = DEFAULT_QLIMIT; - if (pa->scheduler == ALTQT_CBQ || pa->scheduler == ALTQT_HFSC) { + if (pa->scheduler == ALTQT_CBQ || pa->scheduler == ALTQT_HFSC || + pa->scheduler == ALTQT_FAIRQ) { pa->bandwidth = eval_bwspec(bw, parent == NULL ? 0 : parent->bandwidth); @@ -385,6 +405,9 @@ eval_pfqueue(struct pfctl *pf, struct pf_altq *pa, struct node_queue_bw *bw, case ALTQT_HFSC: error = eval_pfqueue_hfsc(pf, pa); break; + case ALTQT_FAIRQ: + error = eval_pfqueue_fairq(pf, pa); + break; default: break; } @@ -797,6 +820,85 @@ err_ret: return (-1); } +/* + * FAIRQ support functions + */ +static int +eval_pfqueue_fairq(struct pfctl *pf __unused, struct pf_altq *pa) +{ + struct pf_altq *altq, *parent; + struct fairq_opts *opts; + struct service_curve sc; + + opts = &pa->pq_u.fairq_opts; + + if (pa->parent[0] == 0) { + /* root queue */ + opts->lssc_m1 = pa->ifbandwidth; + opts->lssc_m2 = pa->ifbandwidth; + opts->lssc_d = 0; + return (0); + } + + LIST_INIT(&lssc); + + /* if link_share is not specified, use bandwidth */ + if (opts->lssc_m2 == 0) + opts->lssc_m2 = pa->bandwidth; + + /* + * admission control: + * for the real-time service curve, the sum of the service curves + * should not exceed 80% of the interface bandwidth. 20% is reserved + * not to over-commit the actual interface bandwidth. + * for the link-sharing service curve, the sum of the child service + * curve should not exceed the parent service curve. + * for the upper-limit service curve, the assigned bandwidth should + * be smaller than the interface bandwidth, and the upper-limit should + * be larger than the real-time service curve when both are defined. + */ + parent = qname_to_pfaltq(pa->parent, pa->ifname); + if (parent == NULL) + errx(1, "parent %s not found for %s", pa->parent, pa->qname); + + TAILQ_FOREACH(altq, &altqs, entries) { + if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0) + continue; + if (altq->qname[0] == 0) /* this is for interface */ + continue; + + if (strncmp(altq->parent, pa->parent, PF_QNAME_SIZE) != 0) + continue; + + /* if the class has a link-sharing service curve, add it. */ + if (opts->lssc_m2 != 0 && altq->pq_u.fairq_opts.lssc_m2 != 0) { + sc.m1 = altq->pq_u.fairq_opts.lssc_m1; + sc.d = altq->pq_u.fairq_opts.lssc_d; + sc.m2 = altq->pq_u.fairq_opts.lssc_m2; + gsc_add_sc(&lssc, &sc); + } + } + + /* check the link-sharing service curve. */ + if (opts->lssc_m2 != 0) { + sc.m1 = parent->pq_u.fairq_opts.lssc_m1; + sc.d = parent->pq_u.fairq_opts.lssc_d; + sc.m2 = parent->pq_u.fairq_opts.lssc_m2; + if (!is_gsc_under_sc(&lssc, &sc)) { + warnx("link-sharing sc exceeds parent's sc"); + goto err_ret; + } + } + + gsc_destroy(&lssc); + + return (0); + +err_ret: + gsc_destroy(&lssc); + return (-1); +} + static int check_commit_hfsc(int dev, int opts, struct pf_altq *pa) { @@ -837,6 +939,43 @@ check_commit_hfsc(int dev, int opts, struct pf_altq *pa) } static int +check_commit_fairq(int dev __unused, int opts __unused, struct pf_altq *pa) +{ + struct pf_altq *altq, *def = NULL; + int default_class; + int error = 0; + + /* check if fairq has one default queue for this interface */ + default_class = 0; + TAILQ_FOREACH(altq, &altqs, entries) { + if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0) + continue; + if (altq->qname[0] == 0) /* this is for interface */ + continue; + if (altq->pq_u.fairq_opts.flags & FARF_DEFAULTCLASS) { + default_class++; + def = altq; + } + } + if (default_class != 1) { + warnx("should have one default queue on %s", pa->ifname); + return (1); + } + /* make sure the default queue is a leaf */ + TAILQ_FOREACH(altq, &altqs, entries) { + if (strncmp(altq->ifname, pa->ifname, IFNAMSIZ) != 0) + continue; + if (altq->qname[0] == 0) /* this is for interface */ + continue; + if (strncmp(altq->parent, def->qname, PF_QNAME_SIZE) == 0) { + warnx("default queue is not a leaf"); + error++; + } + } + return (error); +} + +static int print_hfsc_opts(const struct pf_altq *a, const struct node_queue_opt *qopts) { const struct hfsc_opts *opts; @@ -882,6 +1021,43 @@ print_hfsc_opts(const struct pf_altq *a, const struct node_queue_opt *qopts) return (0); } +static int +print_fairq_opts(const struct pf_altq *a, const struct node_queue_opt *qopts) +{ + const struct fairq_opts *opts; + const struct node_fairq_sc *loc_lssc; + + opts = &a->pq_u.fairq_opts; + if (qopts == NULL) + loc_lssc = NULL; + else + loc_lssc = &qopts->data.fairq_opts.linkshare; + + if (opts->flags || + (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth || + opts->lssc_d != 0))) { + printf("fairq("); + if (opts->flags & FARF_RED) + printf(" red"); + if (opts->flags & FARF_ECN) + printf(" ecn"); + if (opts->flags & FARF_RIO) + printf(" rio"); + if (opts->flags & FARF_CLEARDSCP) + printf(" cleardscp"); + if (opts->flags & FARF_DEFAULTCLASS) + printf(" default"); + if (opts->lssc_m2 != 0 && (opts->lssc_m2 != a->bandwidth || + opts->lssc_d != 0)) + print_fairq_sc("linkshare", opts->lssc_m1, opts->lssc_d, + opts->lssc_m2, loc_lssc); + printf(" ) "); + + return (1); + } else + return (0); +} + /* * admission control using generalized service curve */ @@ -1201,6 +1377,23 @@ eval_queue_opts(struct pf_altq *pa, struct node_queue_opt *opts, opts->data.hfsc_opts.upperlimit.d; } break; + case ALTQT_FAIRQ: + pa->pq_u.fairq_opts.flags = opts->data.fairq_opts.flags; + pa->pq_u.fairq_opts.nbuckets = opts->data.fairq_opts.nbuckets; + pa->pq_u.fairq_opts.hogs_m1 = + eval_bwspec(&opts->data.fairq_opts.hogs_bw, ref_bw); + + if (opts->data.fairq_opts.linkshare.used) { + pa->pq_u.fairq_opts.lssc_m1 = + eval_bwspec(&opts->data.fairq_opts.linkshare.m1, + ref_bw); + pa->pq_u.fairq_opts.lssc_m2 = + eval_bwspec(&opts->data.fairq_opts.linkshare.m2, + ref_bw); + pa->pq_u.fairq_opts.lssc_d = + opts->data.fairq_opts.linkshare.d; + } + break; default: warnx("eval_queue_opts: unknown scheduler type %u", opts->qtype); @@ -1246,3 +1439,27 @@ print_hfsc_sc(const char *scname, u_int m1, u_int d, u_int m2, if (d != 0) printf(")"); } + +void +print_fairq_sc(const char *scname, u_int m1, u_int d, u_int m2, + const struct node_fairq_sc *sc) +{ + printf(" %s", scname); + + if (d != 0) { + printf("("); + if (sc != NULL && sc->m1.bw_percent > 0) + printf("%u%%", sc->m1.bw_percent); + else + printf("%s", rate2str((double)m1)); + printf(" %u", d); + } + + if (sc != NULL && sc->m2.bw_percent > 0) + printf(" %u%%", sc->m2.bw_percent); + else + printf(" %s", rate2str((double)m2)); + + if (d != 0) + printf(")"); +} diff --git a/sbin/pfctl/pfctl_parser.h b/sbin/pfctl/pfctl_parser.h index 5c909e9..8a4e84e 100644 --- a/sbin/pfctl/pfctl_parser.h +++ b/sbin/pfctl/pfctl_parser.h @@ -150,12 +150,27 @@ struct node_hfsc_opts { int flags; }; +struct node_fairq_sc { + struct node_queue_bw m1; /* slope of 1st segment; bps */ + u_int d; /* x-projection of m1; msec */ + struct node_queue_bw m2; /* slope of 2nd segment; bps */ + u_int8_t used; +}; + +struct node_fairq_opts { + struct node_fairq_sc linkshare; + struct node_queue_bw hogs_bw; + u_int nbuckets; + int flags; +}; + struct node_queue_opt { int qtype; union { struct cbq_opts cbq_opts; struct priq_opts priq_opts; struct node_hfsc_opts hfsc_opts; + struct node_fairq_opts fairq_opts; } data; }; diff --git a/sbin/pfctl/pfctl_qstats.c b/sbin/pfctl/pfctl_qstats.c index 95371e4..4087d71 100644 --- a/sbin/pfctl/pfctl_qstats.c +++ b/sbin/pfctl/pfctl_qstats.c @@ -38,6 +38,7 @@ __FBSDID("$FreeBSD$"); #include <altq/altq_cbq.h> #include <altq/altq_priq.h> #include <altq/altq_hfsc.h> +#include <altq/altq_fairq.h> #include "pfctl.h" #include "pfctl_parser.h" @@ -46,6 +47,7 @@ union class_stats { class_stats_t cbq_stats; struct priq_classstats priq_stats; struct hfsc_classstats hfsc_stats; + struct fairq_classstats fairq_stats; }; #define AVGN_MAX 8 @@ -77,6 +79,7 @@ void pfctl_print_altq_node(int, const struct pf_altq_node *, void print_cbqstats(struct queue_stats); void print_priqstats(struct queue_stats); void print_hfscstats(struct queue_stats); +void print_fairqstats(struct queue_stats); void pfctl_free_altq_node(struct pf_altq_node *); void pfctl_print_altq_nodestat(int, const struct pf_altq_node *); @@ -317,6 +320,9 @@ pfctl_print_altq_nodestat(int dev, const struct pf_altq_node *a) case ALTQT_HFSC: print_hfscstats(a->qstats); break; + case ALTQT_FAIRQ: + print_fairqstats(a->qstats); + break; } } @@ -382,6 +388,26 @@ print_hfscstats(struct queue_stats cur) } void +print_fairqstats(struct queue_stats cur) +{ + printf(" [ pkts: %10llu bytes: %10llu " + "dropped pkts: %6llu bytes: %6llu ]\n", + (unsigned long long)cur.data.fairq_stats.xmit_cnt.packets, + (unsigned long long)cur.data.fairq_stats.xmit_cnt.bytes, + (unsigned long long)cur.data.fairq_stats.drop_cnt.packets, + (unsigned long long)cur.data.fairq_stats.drop_cnt.bytes); + printf(" [ qlength: %3d/%3d ]\n", + cur.data.fairq_stats.qlength, cur.data.fairq_stats.qlimit); + + if (cur.avgn < 2) + return; + + printf(" [ measured: %7.1f packets/s, %s/s ]\n", + cur.avg_packets / STAT_INTERVAL, + rate2str((8 * cur.avg_bytes) / STAT_INTERVAL)); +} + +void pfctl_free_altq_node(struct pf_altq_node *node) { while (node != NULL) { @@ -421,6 +447,10 @@ update_avg(struct pf_altq_node *a) b = qs->data.hfsc_stats.xmit_cnt.bytes; p = qs->data.hfsc_stats.xmit_cnt.packets; break; + case ALTQT_FAIRQ: + b = qs->data.fairq_stats.xmit_cnt.bytes; + p = qs->data.fairq_stats.xmit_cnt.packets; + break; default: b = 0; p = 0; diff --git a/share/man/man4/altq.4 b/share/man/man4/altq.4 index 1effdb2..5191f6d 100644 --- a/share/man/man4/altq.4 +++ b/share/man/man4/altq.4 @@ -25,7 +25,7 @@ .\" .\" $FreeBSD$ .\" -.Dd December 9, 2011 +.Dd June 24, 2015 .Dt ALTQ 4 .Os .Sh NAME @@ -40,6 +40,7 @@ .Cd options ALTQ_HFSC .Cd options ALTQ_CDNR .Cd options ALTQ_PRIQ +.Cd options ALTQ_FAIRQ .Sh DESCRIPTION The .Nm @@ -93,6 +94,10 @@ any of the available disciplines or consumers. Build the .Dq "Priority Queuing" discipline. +.It Dv ALTQ_FAIRQ +Build the +.Dq "Fair Queuing" +discipline. .It Dv ALTQ_NOPCC Required if the TSC is unusable. .It Dv ALTQ_DEBUG diff --git a/sys/conf/NOTES b/sys/conf/NOTES index 0792681..99c3d0b 100644 --- a/sys/conf/NOTES +++ b/sys/conf/NOTES @@ -697,6 +697,7 @@ options ALTQ_CBQ # Class Based Queueing options ALTQ_RED # Random Early Detection options ALTQ_RIO # RED In/Out options ALTQ_HFSC # Hierarchical Packet Scheduler +options ALTQ_FAIRQ # Fair Packet Scheduler options ALTQ_CDNR # Traffic conditioner options ALTQ_PRIQ # Priority Queueing options ALTQ_NOPCC # Required if the TSC is unusable diff --git a/sys/conf/files b/sys/conf/files index 6e3ca1c..3e6ca83 100644 --- a/sys/conf/files +++ b/sys/conf/files @@ -253,6 +253,7 @@ compat/freebsd32/freebsd32_sysent.c optional compat_freebsd32 contrib/altq/altq/altq_cbq.c optional altq contrib/altq/altq/altq_cdnr.c optional altq contrib/altq/altq/altq_hfsc.c optional altq +contrib/altq/altq/altq_fairq.c optional altq contrib/altq/altq/altq_priq.c optional altq contrib/altq/altq/altq_red.c optional altq contrib/altq/altq/altq_rio.c optional altq diff --git a/sys/conf/options b/sys/conf/options index a154b4c..45aeb51 100644 --- a/sys/conf/options +++ b/sys/conf/options @@ -385,6 +385,7 @@ ALTQ_CBQ opt_altq.h ALTQ_CDNR opt_altq.h ALTQ_DEBUG opt_altq.h ALTQ_HFSC opt_altq.h +ALTQ_FAIRQ opt_altq.h ALTQ_NOPCC opt_altq.h ALTQ_PRIQ opt_altq.h ALTQ_RED opt_altq.h diff --git a/sys/contrib/altq/altq/altq.h b/sys/contrib/altq/altq/altq.h index c740ed3..1600ba8 100644 --- a/sys/contrib/altq/altq/altq.h +++ b/sys/contrib/altq/altq/altq.h @@ -63,7 +63,8 @@ #define ALTQT_BLUE 10 /* blue */ #define ALTQT_PRIQ 11 /* priority queue */ #define ALTQT_JOBS 12 /* JoBS */ -#define ALTQT_MAX 13 /* should be max discipline type + 1 */ +#define ALTQT_FAIRQ 13 /* fairq */ +#define ALTQT_MAX 14 /* should be max discipline type + 1 */ #ifdef ALTQ3_COMPAT struct altqreq { diff --git a/sys/contrib/altq/altq/altq_fairq.c b/sys/contrib/altq/altq/altq_fairq.c new file mode 100644 index 0000000..038e6e1 --- /dev/null +++ b/sys/contrib/altq/altq/altq_fairq.c @@ -0,0 +1,889 @@ +/* + * Copyright (c) 2008 The DragonFly Project. All rights reserved. + * + * This code is derived from software contributed to The DragonFly Project + * by Matthew Dillon <dillon@backplane.com> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * 3. Neither the name of The DragonFly Project nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific, prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE + * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, + * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, + * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED + * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, + * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT + * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $DragonFly: src/sys/net/altq/altq_fairq.c,v 1.1 2008/04/06 18:58:15 dillon Exp $ + * $FreeBSD$ + */ +/* + * Matt: I gutted altq_priq.c and used it as a skeleton on which to build + * fairq. The fairq algorithm is completely different then priq, of course, + * but because I used priq's skeleton I believe I should include priq's + * copyright. + * + * Copyright (C) 2000-2003 + * Sony Computer Science Laboratories Inc. All rights reserved. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * + * THIS SOFTWARE IS PROVIDED BY SONY CSL AND CONTRIBUTORS ``AS IS'' AND + * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE + * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE + * ARE DISCLAIMED. IN NO EVENT SHALL SONY CSL OR CONTRIBUTORS BE LIABLE + * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL + * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS + * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) + * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT + * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY + * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + */ + +/* + * FAIRQ - take traffic classified by keep state (hashed into + * mbuf->m_pkthdr.altq_state_hash) and bucketize it. Fairly extract + * the first packet from each bucket in a round-robin fashion. + * + * TODO - better overall qlimit support (right now it is per-bucket). + * - NOTE: red etc is per bucket, not overall. + * - better service curve support. + * + * EXAMPLE: + * + * altq on em0 fairq bandwidth 650Kb queue { std, bulk } + * queue std priority 3 bandwidth 400Kb \ + * fairq (buckets 64, default, hogs 1Kb) qlimit 50 + * queue bulk priority 2 bandwidth 100Kb \ + * fairq (buckets 64, hogs 1Kb) qlimit 50 + * + * pass out on em0 from any to any keep state queue std + * pass out on em0 inet proto tcp ..... port ... keep state queue bulk + */ +#include "opt_altq.h" +#include "opt_inet.h" +#include "opt_inet6.h" + +#ifdef ALTQ_FAIRQ /* fairq is enabled in the kernel conf */ + +#include <sys/param.h> +#include <sys/malloc.h> +#include <sys/mbuf.h> +#include <sys/socket.h> +#include <sys/sockio.h> +#include <sys/systm.h> +#include <sys/proc.h> +#include <sys/errno.h> +#include <sys/kernel.h> +#include <sys/queue.h> + +#include <net/if.h> +#include <netinet/in.h> + +#include <net/pfvar.h> +#include <altq/altq.h> +#include <altq/altq_fairq.h> + +/* + * function prototypes + */ +static int fairq_clear_interface(struct fairq_if *); +static int fairq_request(struct ifaltq *, int, void *); +static void fairq_purge(struct fairq_if *); +static struct fairq_class *fairq_class_create(struct fairq_if *, int, int, u_int, struct fairq_opts *, int); +static int fairq_class_destroy(struct fairq_class *); +static int fairq_enqueue(struct ifaltq *, struct mbuf *, struct altq_pktattr *); +static struct mbuf *fairq_dequeue(struct ifaltq *, int); + +static int fairq_addq(struct fairq_class *, struct mbuf *, u_int32_t); +static struct mbuf *fairq_getq(struct fairq_class *, uint64_t); +static struct mbuf *fairq_pollq(struct fairq_class *, uint64_t, int *); +static fairq_bucket_t *fairq_selectq(struct fairq_class *, int); +static void fairq_purgeq(struct fairq_class *); + +static void get_class_stats(struct fairq_classstats *, struct fairq_class *); +static struct fairq_class *clh_to_clp(struct fairq_if *, uint32_t); + +int +fairq_pfattach(struct pf_altq *a) +{ + struct ifnet *ifp; + int error; + + if ((ifp = ifunit(a->ifname)) == NULL || a->altq_disc == NULL) + return (EINVAL); + + error = altq_attach(&ifp->if_snd, ALTQT_FAIRQ, a->altq_disc, + fairq_enqueue, fairq_dequeue, fairq_request, NULL, NULL); + + return (error); +} + +int +fairq_add_altq(struct pf_altq *a) +{ + struct fairq_if *pif; + struct ifnet *ifp; + + if ((ifp = ifunit(a->ifname)) == NULL) + return (EINVAL); + if (!ALTQ_IS_READY(&ifp->if_snd)) + return (ENODEV); + + + pif = malloc(sizeof(struct fairq_if), + M_DEVBUF, M_WAITOK | M_ZERO); + pif->pif_bandwidth = a->ifbandwidth; + pif->pif_maxpri = -1; + pif->pif_ifq = &ifp->if_snd; + + /* keep the state in pf_altq */ + a->altq_disc = pif; + + return (0); +} + +int +fairq_remove_altq(struct pf_altq *a) +{ + struct fairq_if *pif; + + if ((pif = a->altq_disc) == NULL) + return (EINVAL); + a->altq_disc = NULL; + + fairq_clear_interface(pif); + + free(pif, M_DEVBUF); + return (0); +} + +int +fairq_add_queue(struct pf_altq *a) +{ + struct fairq_if *pif; + struct fairq_class *cl; + + if ((pif = a->altq_disc) == NULL) + return (EINVAL); + + /* check parameters */ + if (a->priority >= FAIRQ_MAXPRI) + return (EINVAL); + if (a->qid == 0) + return (EINVAL); + if (pif->pif_classes[a->priority] != NULL) + return (EBUSY); + if (clh_to_clp(pif, a->qid) != NULL) + return (EBUSY); + + cl = fairq_class_create(pif, a->priority, a->qlimit, a->bandwidth, + &a->pq_u.fairq_opts, a->qid); + if (cl == NULL) + return (ENOMEM); + + return (0); +} + +int +fairq_remove_queue(struct pf_altq *a) +{ + struct fairq_if *pif; + struct fairq_class *cl; + + if ((pif = a->altq_disc) == NULL) + return (EINVAL); + + if ((cl = clh_to_clp(pif, a->qid)) == NULL) + return (EINVAL); + + return (fairq_class_destroy(cl)); +} + +int +fairq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes) +{ + struct fairq_if *pif; + struct fairq_class *cl; + struct fairq_classstats stats; + int error = 0; + + if ((pif = altq_lookup(a->ifname, ALTQT_FAIRQ)) == NULL) + return (EBADF); + + if ((cl = clh_to_clp(pif, a->qid)) == NULL) + return (EINVAL); + + if (*nbytes < sizeof(stats)) + return (EINVAL); + + get_class_stats(&stats, cl); + + if ((error = copyout((caddr_t)&stats, ubuf, sizeof(stats))) != 0) + return (error); + *nbytes = sizeof(stats); + return (0); +} + +/* + * bring the interface back to the initial state by discarding + * all the filters and classes. + */ +static int +fairq_clear_interface(struct fairq_if *pif) +{ + struct fairq_class *cl; + int pri; + + /* clear out the classes */ + for (pri = 0; pri <= pif->pif_maxpri; pri++) { + if ((cl = pif->pif_classes[pri]) != NULL) + fairq_class_destroy(cl); + } + + return (0); +} + +static int +fairq_request(struct ifaltq *ifq, int req, void *arg) +{ + struct fairq_if *pif = (struct fairq_if *)ifq->altq_disc; + + IFQ_LOCK_ASSERT(ifq); + + switch (req) { + case ALTRQ_PURGE: + fairq_purge(pif); + break; + } + return (0); +} + +/* discard all the queued packets on the interface */ +static void +fairq_purge(struct fairq_if *pif) +{ + struct fairq_class *cl; + int pri; + + for (pri = 0; pri <= pif->pif_maxpri; pri++) { + if ((cl = pif->pif_classes[pri]) != NULL && cl->cl_head) + fairq_purgeq(cl); + } + if (ALTQ_IS_ENABLED(pif->pif_ifq)) + pif->pif_ifq->ifq_len = 0; +} + +static struct fairq_class * +fairq_class_create(struct fairq_if *pif, int pri, int qlimit, + u_int bandwidth, struct fairq_opts *opts, int qid) +{ + struct fairq_class *cl; + int flags = opts->flags; + u_int nbuckets = opts->nbuckets; + int i; + +#ifndef ALTQ_RED + if (flags & FARF_RED) { +#ifdef ALTQ_DEBUG + printf("fairq_class_create: RED not configured for FAIRQ!\n"); +#endif + return (NULL); + } +#endif + if (nbuckets == 0) + nbuckets = 256; + if (nbuckets > FAIRQ_MAX_BUCKETS) + nbuckets = FAIRQ_MAX_BUCKETS; + /* enforce power-of-2 size */ + while ((nbuckets ^ (nbuckets - 1)) != ((nbuckets << 1) - 1)) + ++nbuckets; + + if ((cl = pif->pif_classes[pri]) != NULL) { + /* modify the class instead of creating a new one */ + IFQ_LOCK(cl->cl_pif->pif_ifq); + if (cl->cl_head) + fairq_purgeq(cl); + IFQ_UNLOCK(cl->cl_pif->pif_ifq); +#ifdef ALTQ_RIO + if (cl->cl_qtype == Q_RIO) + rio_destroy((rio_t *)cl->cl_red); +#endif +#ifdef ALTQ_RED + if (cl->cl_qtype == Q_RED) + red_destroy(cl->cl_red); +#endif + } else { + cl = malloc(sizeof(struct fairq_class), + M_DEVBUF, M_WAITOK | M_ZERO); + cl->cl_nbuckets = nbuckets; + cl->cl_nbucket_mask = nbuckets - 1; + + cl->cl_buckets = malloc( + sizeof(struct fairq_bucket) * cl->cl_nbuckets, + M_DEVBUF, M_WAITOK | M_ZERO); + cl->cl_head = NULL; + } + + pif->pif_classes[pri] = cl; + if (flags & FARF_DEFAULTCLASS) + pif->pif_default = cl; + if (qlimit == 0) + qlimit = 50; /* use default */ + cl->cl_qlimit = qlimit; + for (i = 0; i < cl->cl_nbuckets; ++i) { + qlimit(&cl->cl_buckets[i].queue) = qlimit; + } + cl->cl_bandwidth = bandwidth / 8; + cl->cl_qtype = Q_DROPTAIL; + cl->cl_flags = flags & FARF_USERFLAGS; + cl->cl_pri = pri; + if (pri > pif->pif_maxpri) + pif->pif_maxpri = pri; + cl->cl_pif = pif; + cl->cl_handle = qid; + cl->cl_hogs_m1 = opts->hogs_m1 / 8; + cl->cl_lssc_m1 = opts->lssc_m1 / 8; /* NOT YET USED */ + +#ifdef ALTQ_RED + if (flags & (FARF_RED|FARF_RIO)) { + int red_flags, red_pkttime; + + red_flags = 0; + if (flags & FARF_ECN) + red_flags |= REDF_ECN; +#ifdef ALTQ_RIO + if (flags & FARF_CLEARDSCP) + red_flags |= RIOF_CLEARDSCP; +#endif + if (pif->pif_bandwidth < 8) + red_pkttime = 1000 * 1000 * 1000; /* 1 sec */ + else + red_pkttime = (int64_t)pif->pif_ifq->altq_ifp->if_mtu + * 1000 * 1000 * 1000 / (pif->pif_bandwidth / 8); +#ifdef ALTQ_RIO + if (flags & FARF_RIO) { + cl->cl_red = (red_t *)rio_alloc(0, NULL, + red_flags, red_pkttime); + if (cl->cl_red != NULL) + cl->cl_qtype = Q_RIO; + } else +#endif + if (flags & FARF_RED) { + cl->cl_red = red_alloc(0, 0, + cl->cl_qlimit * 10/100, + cl->cl_qlimit * 30/100, + red_flags, red_pkttime); + if (cl->cl_red != NULL) + cl->cl_qtype = Q_RED; + } + } +#endif /* ALTQ_RED */ + + return (cl); + +err_buckets: + if (cl->cl_buckets != NULL) + free(cl->cl_buckets, M_DEVBUF); +err_ret: + if (cl->cl_red != NULL) { +#ifdef ALTQ_RIO + if (cl->cl_qtype == Q_RIO) + rio_destroy((rio_t *)cl->cl_red); +#endif +#ifdef ALTQ_RED + if (cl->cl_qtype == Q_RED) + red_destroy(cl->cl_red); +#endif + } + if (cl != NULL) + free(cl, M_DEVBUF); + return (NULL); +} + +static int +fairq_class_destroy(struct fairq_class *cl) +{ + struct fairq_if *pif; + int pri; + + IFQ_LOCK(cl->cl_pif->pif_ifq); + + if (cl->cl_head) + fairq_purgeq(cl); + + pif = cl->cl_pif; + pif->pif_classes[cl->cl_pri] = NULL; + if (pif->pif_poll_cache == cl) + pif->pif_poll_cache = NULL; + if (pif->pif_maxpri == cl->cl_pri) { + for (pri = cl->cl_pri; pri >= 0; pri--) + if (pif->pif_classes[pri] != NULL) { + pif->pif_maxpri = pri; + break; + } + if (pri < 0) + pif->pif_maxpri = -1; + } + IFQ_UNLOCK(cl->cl_pif->pif_ifq); + + if (cl->cl_red != NULL) { +#ifdef ALTQ_RIO + if (cl->cl_qtype == Q_RIO) + rio_destroy((rio_t *)cl->cl_red); +#endif +#ifdef ALTQ_RED + if (cl->cl_qtype == Q_RED) + red_destroy(cl->cl_red); +#endif + } + free(cl->cl_buckets, M_DEVBUF); + free(cl, M_DEVBUF); + + return (0); +} + +/* + * fairq_enqueue is an enqueue function to be registered to + * (*altq_enqueue) in struct ifaltq. + */ +static int +fairq_enqueue(struct ifaltq *ifq, struct mbuf *m, struct altq_pktattr *pktattr) +{ + struct fairq_if *pif = (struct fairq_if *)ifq->altq_disc; + struct fairq_class *cl = NULL; /* Make compiler happy */ + struct pf_mtag *t; + u_int32_t qid_hash = 0; + int len; + + IFQ_LOCK_ASSERT(ifq); + + /* grab class set by classifier */ + if ((m->m_flags & M_PKTHDR) == 0) { + /* should not happen */ + printf("altq: packet for %s does not have pkthdr\n", + ifq->altq_ifp->if_xname); + m_freem(m); + return (ENOBUFS); + } + + if ((t = pf_find_mtag(m)) != NULL) { + cl = clh_to_clp(pif, t->qid); + qid_hash = t->qid_hash; + } + if (cl == NULL) { + cl = pif->pif_default; + if (cl == NULL) { + m_freem(m); + return (ENOBUFS); + } + } + cl->cl_flags |= FARF_HAS_PACKETS; + cl->cl_pktattr = NULL; + len = m_pktlen(m); + if (fairq_addq(cl, m, qid_hash) != 0) { + /* drop occurred. mbuf was freed in fairq_addq. */ + PKTCNTR_ADD(&cl->cl_dropcnt, len); + return (ENOBUFS); + } + IFQ_INC_LEN(ifq); + + return (0); +} + +/* + * fairq_dequeue is a dequeue function to be registered to + * (*altq_dequeue) in struct ifaltq. + * + * note: ALTDQ_POLL returns the next packet without removing the packet + * from the queue. ALTDQ_REMOVE is a normal dequeue operation. + * ALTDQ_REMOVE must return the same packet if called immediately + * after ALTDQ_POLL. + */ +static struct mbuf * +fairq_dequeue(struct ifaltq *ifq, int op) +{ + struct fairq_if *pif = (struct fairq_if *)ifq->altq_disc; + struct fairq_class *cl; + struct fairq_class *best_cl; + struct mbuf *best_m; + struct mbuf *m = NULL; + uint64_t cur_time = read_machclk(); + int pri; + int hit_limit; + + IFQ_LOCK_ASSERT(ifq); + + if (IFQ_IS_EMPTY(ifq)) { + return (NULL); + } + + if (pif->pif_poll_cache && op == ALTDQ_REMOVE) { + best_cl = pif->pif_poll_cache; + m = fairq_getq(best_cl, cur_time); + pif->pif_poll_cache = NULL; + if (m) { + IFQ_DEC_LEN(ifq); + PKTCNTR_ADD(&best_cl->cl_xmitcnt, m_pktlen(m)); + return (m); + } + } else { + best_cl = NULL; + best_m = NULL; + + for (pri = pif->pif_maxpri; pri >= 0; pri--) { + if ((cl = pif->pif_classes[pri]) == NULL) + continue; + if ((cl->cl_flags & FARF_HAS_PACKETS) == 0) + continue; + m = fairq_pollq(cl, cur_time, &hit_limit); + if (m == NULL) { + cl->cl_flags &= ~FARF_HAS_PACKETS; + continue; + } + + /* + * Only override the best choice if we are under + * the BW limit. + */ + if (hit_limit == 0 || best_cl == NULL) { + best_cl = cl; + best_m = m; + } + + /* + * Remember the highest priority mbuf in case we + * do not find any lower priority mbufs. + */ + if (hit_limit) + continue; + break; + } + if (op == ALTDQ_POLL) { + pif->pif_poll_cache = best_cl; + m = best_m; + } else if (best_cl) { + m = fairq_getq(best_cl, cur_time); + if (m != NULL) { + IFQ_DEC_LEN(ifq); + PKTCNTR_ADD(&best_cl->cl_xmitcnt, m_pktlen(m)); + } + } + return (m); + } + return (NULL); +} + +static int +fairq_addq(struct fairq_class *cl, struct mbuf *m, u_int32_t bucketid) +{ + fairq_bucket_t *b; + u_int hindex; + uint64_t bw; + + /* + * If the packet doesn't have any keep state put it on the end of + * our queue. XXX this can result in out of order delivery. + */ + if (bucketid == 0) { + if (cl->cl_head) + b = cl->cl_head->prev; + else + b = &cl->cl_buckets[0]; + } else { + hindex = bucketid & cl->cl_nbucket_mask; + b = &cl->cl_buckets[hindex]; + } + + /* + * Add the bucket to the end of the circular list of active buckets. + * + * As a special case we add the bucket to the beginning of the list + * instead of the end if it was not previously on the list and if + * its traffic is less then the hog level. + */ + if (b->in_use == 0) { + b->in_use = 1; + if (cl->cl_head == NULL) { + cl->cl_head = b; + b->next = b; + b->prev = b; + } else { + b->next = cl->cl_head; + b->prev = cl->cl_head->prev; + b->prev->next = b; + b->next->prev = b; + + if (b->bw_delta && cl->cl_hogs_m1) { + bw = b->bw_bytes * machclk_freq / b->bw_delta; + if (bw < cl->cl_hogs_m1) + cl->cl_head = b; + } + } + } + +#ifdef ALTQ_RIO + if (cl->cl_qtype == Q_RIO) + return rio_addq((rio_t *)cl->cl_red, &b->queue, m, cl->cl_pktattr); +#endif +#ifdef ALTQ_RED + if (cl->cl_qtype == Q_RED) + return red_addq(cl->cl_red, &b->queue, m, cl->cl_pktattr); +#endif + if (qlen(&b->queue) >= qlimit(&b->queue)) { + m_freem(m); + return (-1); + } + + if (cl->cl_flags & FARF_CLEARDSCP) + write_dsfield(m, cl->cl_pktattr, 0); + + _addq(&b->queue, m); + + return (0); +} + +static struct mbuf * +fairq_getq(struct fairq_class *cl, uint64_t cur_time) +{ + fairq_bucket_t *b; + struct mbuf *m; + + b = fairq_selectq(cl, 0); + if (b == NULL) + m = NULL; +#ifdef ALTQ_RIO + else if (cl->cl_qtype == Q_RIO) + m = rio_getq((rio_t *)cl->cl_red, &b->queue); +#endif +#ifdef ALTQ_RED + else if (cl->cl_qtype == Q_RED) + m = red_getq(cl->cl_red, &b->queue); +#endif + else + m = _getq(&b->queue); + + /* + * Calculate the BW change + */ + if (m != NULL) { + uint64_t delta; + + /* + * Per-class bandwidth calculation + */ + delta = (cur_time - cl->cl_last_time); + if (delta > machclk_freq * 8) + delta = machclk_freq * 8; + cl->cl_bw_delta += delta; + cl->cl_bw_bytes += m->m_pkthdr.len; + cl->cl_last_time = cur_time; + cl->cl_bw_delta -= cl->cl_bw_delta >> 3; + cl->cl_bw_bytes -= cl->cl_bw_bytes >> 3; + + /* + * Per-bucket bandwidth calculation + */ + delta = (cur_time - b->last_time); + if (delta > machclk_freq * 8) + delta = machclk_freq * 8; + b->bw_delta += delta; + b->bw_bytes += m->m_pkthdr.len; + b->last_time = cur_time; + b->bw_delta -= b->bw_delta >> 3; + b->bw_bytes -= b->bw_bytes >> 3; + } + return(m); +} + +/* + * Figure out what the next packet would be if there were no limits. If + * this class hits its bandwidth limit *hit_limit is set to no-zero, otherwise + * it is set to 0. A non-NULL mbuf is returned either way. + */ +static struct mbuf * +fairq_pollq(struct fairq_class *cl, uint64_t cur_time, int *hit_limit) +{ + fairq_bucket_t *b; + struct mbuf *m; + uint64_t delta; + uint64_t bw; + + *hit_limit = 0; + b = fairq_selectq(cl, 1); + if (b == NULL) + return(NULL); + m = qhead(&b->queue); + + /* + * Did this packet exceed the class bandwidth? Calculate the + * bandwidth component of the packet. + * + * - Calculate bytes per second + */ + delta = cur_time - cl->cl_last_time; + if (delta > machclk_freq * 8) + delta = machclk_freq * 8; + cl->cl_bw_delta += delta; + cl->cl_last_time = cur_time; + if (cl->cl_bw_delta) { + bw = cl->cl_bw_bytes * machclk_freq / cl->cl_bw_delta; + + if (bw > cl->cl_bandwidth) + *hit_limit = 1; +#ifdef ALTQ_DEBUG + printf("BW %6lld relative to %6u %d queue %p\n", + bw, cl->cl_bandwidth, *hit_limit, b); +#endif + } + return(m); +} + +/* + * Locate the next queue we want to pull a packet out of. This code + * is also responsible for removing empty buckets from the circular list. + */ +static +fairq_bucket_t * +fairq_selectq(struct fairq_class *cl, int ispoll) +{ + fairq_bucket_t *b; + uint64_t bw; + + if (ispoll == 0 && cl->cl_polled) { + b = cl->cl_polled; + cl->cl_polled = NULL; + return(b); + } + + while ((b = cl->cl_head) != NULL) { + /* + * Remove empty queues from consideration + */ + if (qempty(&b->queue)) { + b->in_use = 0; + cl->cl_head = b->next; + if (cl->cl_head == b) { + cl->cl_head = NULL; + } else { + b->next->prev = b->prev; + b->prev->next = b->next; + } + continue; + } + + /* + * Advance the round robin. Queues with bandwidths less + * then the hog bandwidth are allowed to burst. + */ + if (cl->cl_hogs_m1 == 0) { + cl->cl_head = b->next; + } else if (b->bw_delta) { + bw = b->bw_bytes * machclk_freq / b->bw_delta; + if (bw >= cl->cl_hogs_m1) { + cl->cl_head = b->next; + } + /* + * XXX TODO - + */ + } + + /* + * Return bucket b. + */ + break; + } + if (ispoll) + cl->cl_polled = b; + return(b); +} + +static void +fairq_purgeq(struct fairq_class *cl) +{ + fairq_bucket_t *b; + struct mbuf *m; + + while ((b = fairq_selectq(cl, 0)) != NULL) { + while ((m = _getq(&b->queue)) != NULL) { + PKTCNTR_ADD(&cl->cl_dropcnt, m_pktlen(m)); + m_freem(m); + } + ASSERT(qlen(&b->queue) == 0); + } +} + +static void +get_class_stats(struct fairq_classstats *sp, struct fairq_class *cl) +{ + fairq_bucket_t *b; + + sp->class_handle = cl->cl_handle; + sp->qlimit = cl->cl_qlimit; + sp->xmit_cnt = cl->cl_xmitcnt; + sp->drop_cnt = cl->cl_dropcnt; + sp->qtype = cl->cl_qtype; + sp->qlength = 0; + + if (cl->cl_head) { + b = cl->cl_head; + do { + sp->qlength += qlen(&b->queue); + b = b->next; + } while (b != cl->cl_head); + } + +#ifdef ALTQ_RED + if (cl->cl_qtype == Q_RED) + red_getstats(cl->cl_red, &sp->red[0]); +#endif +#ifdef ALTQ_RIO + if (cl->cl_qtype == Q_RIO) + rio_getstats((rio_t *)cl->cl_red, &sp->red[0]); +#endif +} + +/* convert a class handle to the corresponding class pointer */ +static struct fairq_class * +clh_to_clp(struct fairq_if *pif, uint32_t chandle) +{ + struct fairq_class *cl; + int idx; + + if (chandle == 0) + return (NULL); + + for (idx = pif->pif_maxpri; idx >= 0; idx--) + if ((cl = pif->pif_classes[idx]) != NULL && + cl->cl_handle == chandle) + return (cl); + + return (NULL); +} + +#endif /* ALTQ_FAIRQ */ diff --git a/sys/contrib/altq/altq/altq_fairq.h b/sys/contrib/altq/altq/altq_fairq.h new file mode 100644 index 0000000..7ff6913 --- /dev/null +++ b/sys/contrib/altq/altq/altq_fairq.h @@ -0,0 +1,137 @@ +/* + * Copyright (c) 2008 The DragonFly Project. All rights reserved. + * + * This code is derived from software contributed to The DragonFly Project + * by Matthew Dillon <dillon@backplane.com> + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions + * are met: + * + * 1. Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * 2. Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in + * the documentation and/or other materials provided with the + * distribution. + * 3. Neither the name of The DragonFly Project nor the names of its + * contributors may be used to endorse or promote products derived + * from this software without specific, prior written permission. + * + * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS + * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT + * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS + * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE + * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, + * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, + * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED + * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, + * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT + * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF + * SUCH DAMAGE. + * + * $DragonFly: src/sys/net/altq/altq_fairq.h,v 1.1 2008/04/06 18:58:15 dillon Exp $ + * $FreeBSD$ + */ + +#ifndef _ALTQ_ALTQ_FAIRQ_H_ +#define _ALTQ_ALTQ_FAIRQ_H_ + +#include <altq/altq.h> +#include <altq/altq_classq.h> +#include <altq/altq_red.h> +#include <altq/altq_rio.h> +#include <altq/altq_rmclass.h> + +#define FAIRQ_MAX_BUCKETS 2048 /* maximum number of sorting buckets */ +#define FAIRQ_MAXPRI RM_MAXPRIO +#define FAIRQ_BITMAP_WIDTH (sizeof(fairq_bitmap_t)*8) +#define FAIRQ_BITMAP_MASK (FAIRQ_BITMAP_WIDTH - 1) + +/* fairq class flags */ +#define FARF_RED 0x0001 /* use RED */ +#define FARF_ECN 0x0002 /* use RED/ECN */ +#define FARF_RIO 0x0004 /* use RIO */ +#define FARF_CLEARDSCP 0x0010 /* clear diffserv codepoint */ +#define FARF_DEFAULTCLASS 0x1000 /* default class */ + +#define FARF_HAS_PACKETS 0x2000 /* might have queued packets */ + +#define FARF_USERFLAGS (FARF_RED|FARF_ECN|FARF_RIO|FARF_CLEARDSCP| \ + FARF_DEFAULTCLASS) + +/* special class handles */ +#define FAIRQ_NULLCLASS_HANDLE 0 + +typedef u_int fairq_bitmap_t; + +struct fairq_classstats { + uint32_t class_handle; + + u_int qlength; + u_int qlimit; + struct pktcntr xmit_cnt; /* transmitted packet counter */ + struct pktcntr drop_cnt; /* dropped packet counter */ + + /* red and rio related info */ + int qtype; + struct redstats red[3]; /* rio has 3 red stats */ +}; + +#ifdef _KERNEL + +typedef struct fairq_bucket { + struct fairq_bucket *next; /* circular list */ + struct fairq_bucket *prev; /* circular list */ + class_queue_t queue; /* the actual queue */ + uint64_t bw_bytes; /* statistics used to calculate bw */ + uint64_t bw_delta; /* statistics used to calculate bw */ + uint64_t last_time; + int in_use; +} fairq_bucket_t; + +struct fairq_class { + uint32_t cl_handle; /* class handle */ + u_int cl_nbuckets; /* (power of 2) */ + u_int cl_nbucket_mask; /* bucket mask */ + fairq_bucket_t *cl_buckets; + fairq_bucket_t *cl_head; /* head of circular bucket list */ + fairq_bucket_t *cl_polled; + struct red *cl_red; /* RED state */ + u_int cl_hogs_m1; + u_int cl_lssc_m1; + u_int cl_bandwidth; + uint64_t cl_bw_bytes; + uint64_t cl_bw_delta; + uint64_t cl_last_time; + int cl_qtype; /* rollup */ + int cl_qlimit; + int cl_pri; /* priority */ + int cl_flags; /* class flags */ + struct fairq_if *cl_pif; /* back pointer to pif */ + struct altq_pktattr *cl_pktattr; /* saved header used by ECN */ + + /* round robin index */ + + /* statistics */ + struct pktcntr cl_xmitcnt; /* transmitted packet counter */ + struct pktcntr cl_dropcnt; /* dropped packet counter */ +}; + +/* + * fairq interface state + */ +struct fairq_if { + struct fairq_if *pif_next; /* interface state list */ + struct ifaltq *pif_ifq; /* backpointer to ifaltq */ + u_int pif_bandwidth; /* link bandwidth in bps */ + int pif_maxpri; /* max priority in use */ + struct fairq_class *pif_poll_cache;/* cached poll */ + struct fairq_class *pif_default; /* default class */ + struct fairq_class *pif_classes[FAIRQ_MAXPRI]; /* classes */ +}; + +#endif /* _KERNEL */ + +#endif /* _ALTQ_ALTQ_FAIRQ_H_ */ diff --git a/sys/contrib/altq/altq/altq_subr.c b/sys/contrib/altq/altq/altq_subr.c index 16b796a..d59751a 100644 --- a/sys/contrib/altq/altq/altq_subr.c +++ b/sys/contrib/altq/altq/altq_subr.c @@ -537,6 +537,11 @@ altq_pfattach(struct pf_altq *a) error = hfsc_pfattach(a); break; #endif +#ifdef ALTQ_FAIRQ + case ALTQT_FAIRQ: + error = fairq_pfattach(a); + break; +#endif default: error = ENXIO; } @@ -612,6 +617,11 @@ altq_add(struct pf_altq *a) error = hfsc_add_altq(a); break; #endif +#ifdef ALTQ_FAIRQ + case ALTQT_FAIRQ: + error = fairq_add_altq(a); + break; +#endif default: error = ENXIO; } @@ -648,6 +658,11 @@ altq_remove(struct pf_altq *a) error = hfsc_remove_altq(a); break; #endif +#ifdef ALTQ_FAIRQ + case ALTQT_FAIRQ: + error = fairq_remove_altq(a); + break; +#endif default: error = ENXIO; } @@ -681,6 +696,11 @@ altq_add_queue(struct pf_altq *a) error = hfsc_add_queue(a); break; #endif +#ifdef ALTQ_FAIRQ + case ALTQT_FAIRQ: + error = fairq_add_queue(a); + break; +#endif default: error = ENXIO; } @@ -714,6 +734,11 @@ altq_remove_queue(struct pf_altq *a) error = hfsc_remove_queue(a); break; #endif +#ifdef ALTQ_FAIRQ + case ALTQT_FAIRQ: + error = fairq_remove_queue(a); + break; +#endif default: error = ENXIO; } @@ -747,6 +772,11 @@ altq_getqstats(struct pf_altq *a, void *ubuf, int *nbytes) error = hfsc_getqstats(a, ubuf, nbytes); break; #endif +#ifdef ALTQ_FAIRQ + case ALTQT_FAIRQ: + error = fairq_getqstats(a, ubuf, nbytes); + break; +#endif default: error = ENXIO; } diff --git a/sys/contrib/altq/altq/altq_var.h b/sys/contrib/altq/altq/altq_var.h index 956ee16..eb603ea 100644 --- a/sys/contrib/altq/altq/altq_var.h +++ b/sys/contrib/altq/altq/altq_var.h @@ -257,5 +257,12 @@ int hfsc_add_queue(struct pf_altq *); int hfsc_remove_queue(struct pf_altq *); int hfsc_getqstats(struct pf_altq *, void *, int *); +int fairq_pfattach(struct pf_altq *); +int fairq_add_altq(struct pf_altq *); +int fairq_remove_altq(struct pf_altq *); +int fairq_add_queue(struct pf_altq *); +int fairq_remove_queue(struct pf_altq *); +int fairq_getqstats(struct pf_altq *, void *, int *); + #endif /* _KERNEL */ #endif /* _ALTQ_ALTQ_VAR_H_ */ diff --git a/sys/netpfil/pf/pf.c b/sys/netpfil/pf/pf.c index 91ca4fe..a11f7f8 100644 --- a/sys/netpfil/pf/pf.c +++ b/sys/netpfil/pf/pf.c @@ -614,6 +614,20 @@ pf_hashsrc(struct pf_addr *addr, sa_family_t af) return (h & pf_srchashmask); } +#ifdef ALTQ +static int +pf_state_hash(struct pf_state *s) +{ + u_int32_t hv = (intptr_t)s / sizeof(*s); + + hv ^= crc32(&s->src, sizeof(s->src)); + hv ^= crc32(&s->dst, sizeof(s->dst)); + if (hv == 0) + hv = 1; + return (hv); +} +#endif + #ifdef INET6 void pf_addrcpy(struct pf_addr *dst, struct pf_addr *src, sa_family_t af) @@ -6430,6 +6444,8 @@ done: pd.act.qid = r->qid; } if (action == PF_PASS && pd.act.qid) { + if (s != NULL) + pd.pf_mtag->qid_hash = pf_state_hash(s); if (pqid || (pd.tos & IPTOS_LOWDELAY)) pd.pf_mtag->qid = pd.act.pqid; else @@ -7004,6 +7020,8 @@ done: pd.act.qid = r->qid; } if (action == PF_PASS && pd.act.qid) { + if (s != NULL) + pd.pf_mtag->qid_hash = pf_state_hash(s); if (pd.tos & IPTOS_LOWDELAY) pd.pf_mtag->qid = pd.act.pqid; else diff --git a/sys/netpfil/pf/pf_altq.h b/sys/netpfil/pf/pf_altq.h index eda0965..db681fb 100644 --- a/sys/netpfil/pf/pf_altq.h +++ b/sys/netpfil/pf/pf_altq.h @@ -65,6 +65,20 @@ struct hfsc_opts { int flags; }; +/* + * XXX this needs some work + */ +struct fairq_opts { + u_int nbuckets; + u_int hogs_m1; + int flags; + + /* link sharing service curve */ + u_int lssc_m1; + u_int lssc_d; + u_int lssc_m2; +}; + struct pf_altq { char ifname[IFNAMSIZ]; @@ -91,6 +105,7 @@ struct pf_altq { struct cbq_opts cbq_opts; struct priq_opts priq_opts; struct hfsc_opts hfsc_opts; + struct fairq_opts fairq_opts; } pq_u; uint32_t qid; /* return value */ diff --git a/sys/netpfil/pf/pf_mtag.h b/sys/netpfil/pf/pf_mtag.h index 3aacb2e..fd8554a 100644 --- a/sys/netpfil/pf/pf_mtag.h +++ b/sys/netpfil/pf/pf_mtag.h @@ -44,6 +44,7 @@ struct pf_mtag { void *hdr; /* saved hdr pos in mbuf, for ECN */ u_int32_t qid; /* queue id */ + u_int32_t qid_hash; /* queue hashid used by WFQ like algos */ u_int16_t tag; /* tag id */ u_int8_t flags; u_int8_t routed; |