Date: Sun, 31 Jan 2021 14:49:59 +0100 From: Gordon Bergling <gbe@freebsd.org> To: Michael Tuexen <Michael.Tuexen@macmic.franken.de> Cc: Mateusz Guzik <mjguzik@gmail.com>, Randall Stewart <rrs@netflix.com>, Randall Stewart <rrs@freebsd.org>, src-committers <src-committers@freebsd.org>, dev-commits-src-all@freebsd.org, dev-commits-src-main@freebsd.org Subject: Re: git: 1a714ff20419 - main - This pulls over all the changes that are in the netflix tree that fix the ratelimit code. There were several bugs in tcp_ratelimit itself and we needed further work to support the multiple tag format coming for the joint TLS and Ratelimit dances. Message-ID: <YBa1h6uc1uPaEOCt@lion.0xfce3.net> In-Reply-To: <E0ABCA13-DDCC-45E7-A12A-AB1805782152@macmic.franken.de> References: <202101281653.10SGrJWf002301@gitrepo.freebsd.org> <CAGudoHFh9HMQNUOKpfrj3tBnO2%2BhUyhnnLFVF5c1NnPD=31OFA@mail.gmail.com> <CAGudoHH=Eu7zRLWezj9mOtXH827A4k4rFsVhRn45aK589mDWvQ@mail.gmail.com> <E0ABCA13-DDCC-45E7-A12A-AB1805782152@macmic.franken.de>
next in thread | previous in thread | raw e-mail | index | archive | help
--EB6sWAVXHCNbq2EJ Content-Type: text/plain; charset=us-ascii Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Sun, Jan 31, 2021 at 02:30:24PM +0100, Michael Tuexen wrote: > > On 31. Jan 2021, at 12:09, Mateusz Guzik <mjguzik@gmail.com> wrote: > >=20 > > ping? I'm going to revert this on Tuesday to unbreak the build > >=20 > > On 1/29/21, Mateusz Guzik <mjguzik@gmail.com> wrote: > >> This breaks tinderbox: > >> arm LINT kernel failed, check _.arm.LINT for details > >> i386 LINT-NOINET kernel failed, check _.i386.LINT-NOINET for details > >> i386 LINT-NOINET6 kernel failed, check _.i386.LINT-NOINET6 for details > >> i386 LINT kernel failed, check _.i386.LINT for details > >> amd64 LINT-NOINET kernel failed, check _.amd64.LINT-NOINET for details > Hi Mateusz, >=20 > where can the logfiles be accessed? > I don't see anything red at https://ci.freebsd.org/tinderbox/ >=20 > Best regards > Michael Besides of the reported build issues, this should be an excellent MFC candidate for the upcoming 13.0-RELEASE. --Gordon > >> On 1/28/21, Randall Stewart <rrs@freebsd.org> wrote: > >>> The branch main has been updated by rrs: > >>>=20 > >>> URL: > >>> https://cgit.FreeBSD.org/src/commit/?id=3D1a714ff204193b9eb810426048e= 03f5d76e9730e > >>>=20 > >>> commit 1a714ff204193b9eb810426048e03f5d76e9730e > >>> Author: Randall Stewart <rrs@FreeBSD.org> > >>> AuthorDate: 2021-01-26 16:54:42 +0000 > >>> Commit: Randall Stewart <rrs@FreeBSD.org> > >>> CommitDate: 2021-01-28 16:53:05 +0000 > >>>=20 > >>> This pulls over all the changes that are in the netflix > >>> tree that fix the ratelimit code. There were several bugs > >>> in tcp_ratelimit itself and we needed further work to support > >>> the multiple tag format coming for the joint TLS and Ratelimit > >>> dances. > >>>=20 > >>> Sponsored by: Netflix Inc. > >>> Differential Revision: https://reviews.freebsd.org/D28357 > >>> --- > >>> sys/net/if_lagg.c | 11 + > >>> sys/net/if_var.h | 2 + > >>> sys/net/if_vlan.c | 30 +++ > >>> sys/netinet/in_pcb.c | 9 + > >>> sys/netinet/tcp_log_buf.h | 2 +- > >>> sys/netinet/tcp_ratelimit.c | 584 > >>> +++++++++++++++++++++++++++--------------- > >>> sys/netinet/tcp_ratelimit.h | 24 +- > >>> sys/netinet/tcp_stacks/bbr.c | 4 +- > >>> sys/netinet/tcp_stacks/rack.c | 14 +- > >>> 9 files changed, 464 insertions(+), 216 deletions(-) > >>>=20 > >>> diff --git a/sys/net/if_lagg.c b/sys/net/if_lagg.c > >>> index 3144f86901b2..9a3c22789fa5 100644 > >>> --- a/sys/net/if_lagg.c > >>> +++ b/sys/net/if_lagg.c > >>> @@ -151,6 +151,7 @@ static int lagg_snd_tag_modify(struct m_snd_tag *, > >>> static int lagg_snd_tag_query(struct m_snd_tag *, > >>> union if_snd_tag_query_params *); > >>> static void lagg_snd_tag_free(struct m_snd_tag *); > >>> +static struct m_snd_tag *lagg_next_snd_tag(struct m_snd_tag *); > >>> static void lagg_ratelimit_query(struct ifnet *, > >>> struct if_ratelimit_query_results *); > >>> #endif > >>> @@ -585,6 +586,7 @@ lagg_clone_create(struct if_clone *ifc, int unit, > >>> caddr_t params) > >>> ifp->if_snd_tag_modify =3D lagg_snd_tag_modify; > >>> ifp->if_snd_tag_query =3D lagg_snd_tag_query; > >>> ifp->if_snd_tag_free =3D lagg_snd_tag_free; > >>> + ifp->if_next_snd_tag =3D lagg_next_snd_tag; > >>> ifp->if_ratelimit_query =3D lagg_ratelimit_query; > >>> #endif > >>> ifp->if_capenable =3D ifp->if_capabilities =3D IFCAP_HWSTATS; > >>> @@ -1834,6 +1836,15 @@ lagg_snd_tag_alloc(struct ifnet *ifp, > >>> return (0); > >>> } > >>>=20 > >>> +static struct m_snd_tag * > >>> +lagg_next_snd_tag(struct m_snd_tag *mst) > >>> +{ > >>> + struct lagg_snd_tag *lst; > >>> + > >>> + lst =3D mst_to_lst(mst); > >>> + return (lst->tag); > >>> +} > >>> + > >>> static int > >>> lagg_snd_tag_modify(struct m_snd_tag *mst, > >>> union if_snd_tag_modify_params *params) > >>> diff --git a/sys/net/if_var.h b/sys/net/if_var.h > >>> index beb9596895ee..9ecdfb684296 100644 > >>> --- a/sys/net/if_var.h > >>> +++ b/sys/net/if_var.h > >>> @@ -278,6 +278,7 @@ typedef int (if_snd_tag_alloc_t)(struct ifnet *, > >>> union > >>> if_snd_tag_alloc_params * > >>> typedef int (if_snd_tag_modify_t)(struct m_snd_tag *, union > >>> if_snd_tag_modify_params *); > >>> typedef int (if_snd_tag_query_t)(struct m_snd_tag *, union > >>> if_snd_tag_query_params *); > >>> typedef void (if_snd_tag_free_t)(struct m_snd_tag *); > >>> +typedef struct m_snd_tag *(if_next_send_tag_t)(struct m_snd_tag *); > >>> typedef void (if_ratelimit_query_t)(struct ifnet *, > >>> struct if_ratelimit_query_results *); > >>> typedef int (if_ratelimit_setup_t)(struct ifnet *, uint64_t, uint32_t= ); > >>> @@ -422,6 +423,7 @@ struct ifnet { > >>> if_snd_tag_modify_t *if_snd_tag_modify; > >>> if_snd_tag_query_t *if_snd_tag_query; > >>> if_snd_tag_free_t *if_snd_tag_free; > >>> + if_next_send_tag_t *if_next_snd_tag; > >>> if_ratelimit_query_t *if_ratelimit_query; > >>> if_ratelimit_setup_t *if_ratelimit_setup; > >>>=20 > >>> diff --git a/sys/net/if_vlan.c b/sys/net/if_vlan.c > >>> index 9358724844cb..9703e5e7753f 100644 > >>> --- a/sys/net/if_vlan.c > >>> +++ b/sys/net/if_vlan.c > >>> @@ -295,6 +295,9 @@ static int vlan_snd_tag_modify(struct m_snd_tag *, > >>> static int vlan_snd_tag_query(struct m_snd_tag *, > >>> union if_snd_tag_query_params *); > >>> static void vlan_snd_tag_free(struct m_snd_tag *); > >>> +static struct m_snd_tag *vlan_next_snd_tag(struct m_snd_tag *); > >>> +static void vlan_ratelimit_query(struct ifnet *, > >>> + struct if_ratelimit_query_results *); > >>> #endif > >>> static void vlan_qflush(struct ifnet *ifp); > >>> static int vlan_setflag(struct ifnet *ifp, int flag, int status, > >>> @@ -1071,6 +1074,8 @@ vlan_clone_create(struct if_clone *ifc, char *n= ame, > >>> size_t len, caddr_t params) > >>> ifp->if_snd_tag_modify =3D vlan_snd_tag_modify; > >>> ifp->if_snd_tag_query =3D vlan_snd_tag_query; > >>> ifp->if_snd_tag_free =3D vlan_snd_tag_free; > >>> + ifp->if_next_snd_tag =3D vlan_next_snd_tag; > >>> + ifp->if_ratelimit_query =3D vlan_ratelimit_query; > >>> #endif > >>> ifp->if_flags =3D VLAN_IFFLAGS; > >>> ether_ifattach(ifp, eaddr); > >>> @@ -2073,6 +2078,15 @@ vlan_snd_tag_alloc(struct ifnet *ifp, > >>> return (0); > >>> } > >>>=20 > >>> +static struct m_snd_tag * > >>> +vlan_next_snd_tag(struct m_snd_tag *mst) > >>> +{ > >>> + struct vlan_snd_tag *vst; > >>> + > >>> + vst =3D mst_to_vst(mst); > >>> + return (vst->tag); > >>> +} > >>> + > >>> static int > >>> vlan_snd_tag_modify(struct m_snd_tag *mst, > >>> union if_snd_tag_modify_params *params) > >>> @@ -2102,4 +2116,20 @@ vlan_snd_tag_free(struct m_snd_tag *mst) > >>> m_snd_tag_rele(vst->tag); > >>> free(vst, M_VLAN); > >>> } > >>> + > >>> +static void > >>> +vlan_ratelimit_query(struct ifnet *ifp __unused, struct > >>> if_ratelimit_query_results *q) > >>> +{ > >>> + /* > >>> + * For vlan, we have an indirect > >>> + * interface. The caller needs to > >>> + * get a ratelimit tag on the actual > >>> + * interface the flow will go on. > >>> + */ > >>> + q->rate_table =3D NULL; > >>> + q->flags =3D RT_IS_INDIRECT; > >>> + q->max_flows =3D 0; > >>> + q->number_of_rates =3D 0; > >>> +} > >>> + > >>> #endif > >>> diff --git a/sys/netinet/in_pcb.c b/sys/netinet/in_pcb.c > >>> index 59695ee3715d..40a0b4c0676e 100644 > >>> --- a/sys/netinet/in_pcb.c > >>> +++ b/sys/netinet/in_pcb.c > >>> @@ -224,6 +224,8 @@ SYSCTL_INT(_net_inet_ip_portrange, OID_AUTO, > >>> randomtime, > >>> "allocation before switching to a random one"); > >>>=20 > >>> #ifdef RATELIMIT > >>> +counter_u64_t rate_limit_new; > >>> +counter_u64_t rate_limit_chg; > >>> counter_u64_t rate_limit_active; > >>> counter_u64_t rate_limit_alloc_fail; > >>> counter_u64_t rate_limit_set_ok; > >>> @@ -236,6 +238,11 @@ SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, > >>> alloc_fail, CTLFLAG_RD, > >>> &rate_limit_alloc_fail, "Rate limited connection failures"); > >>> SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, set_ok, CTLFLAG_RD, > >>> &rate_limit_set_ok, "Rate limited setting succeeded"); > >>> +SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, newrl, CTLFLAG_RD, > >>> + &rate_limit_new, "Total Rate limit new attempts"); > >>> +SYSCTL_COUNTER_U64(_net_inet_ip_rl, OID_AUTO, chgrl, CTLFLAG_RD, > >>> + &rate_limit_chg, "Total Rate limited change attempts"); > >>> + > >>> #endif /* RATELIMIT */ > >>>=20 > >>> #endif /* INET */ > >>> @@ -3591,6 +3598,8 @@ in_pcboutput_eagain(struct inpcb *inp) > >>> static void > >>> rl_init(void *st) > >>> { > >>> + rate_limit_new =3D counter_u64_alloc(M_WAITOK); > >>> + rate_limit_chg =3D counter_u64_alloc(M_WAITOK); > >>> rate_limit_active =3D counter_u64_alloc(M_WAITOK); > >>> rate_limit_alloc_fail =3D counter_u64_alloc(M_WAITOK); > >>> rate_limit_set_ok =3D counter_u64_alloc(M_WAITOK); > >>> diff --git a/sys/netinet/tcp_log_buf.h b/sys/netinet/tcp_log_buf.h > >>> index 0d9b14a08f53..436383124dce 100644 > >>> --- a/sys/netinet/tcp_log_buf.h > >>> +++ b/sys/netinet/tcp_log_buf.h > >>> @@ -221,7 +221,7 @@ enum tcp_log_events { > >>> BBR_LOG_SETTINGS_CHG, /* Settings changed for loss response 48 */ > >>> BBR_LOG_SRTT_GAIN_EVENT, /* SRTT gaining -- now not used 49 */ > >>> TCP_LOG_REASS, /* Reassembly buffer logging 50 */ > >>> - TCP_HDWR_TLS, /* TCP Hardware TLS logs 51 */ > >>> + TCP_HDWR_PACE_SIZE, /* TCP pacing size set (rl and rack uses this)= 51 > >>> */ > >>> BBR_LOG_HDWR_PACE, /* TCP Hardware pacing log 52 */ > >>> BBR_LOG_TSTMP_VAL, /* Temp debug timestamp validation 53 */ > >>> TCP_LOG_CONNEND, /* End of connection 54 */ > >>> diff --git a/sys/netinet/tcp_ratelimit.c b/sys/netinet/tcp_ratelimit.c > >>> index 97f066be69e0..aeb1ed6906b5 100644 > >>> --- a/sys/netinet/tcp_ratelimit.c > >>> +++ b/sys/netinet/tcp_ratelimit.c > >>> @@ -57,6 +57,8 @@ __FBSDID("$FreeBSD$"); > >>> #ifdef INET6 > >>> #include <netinet6/tcp6_var.h> > >>> #endif > >>> +#include <netinet/tcp_hpts.h> > >>> +#include <netinet/tcp_log_buf.h> > >>> #include <netinet/tcp_ratelimit.h> > >>> #ifndef USECS_IN_SECOND > >>> #define USECS_IN_SECOND 1000000 > >>> @@ -154,108 +156,77 @@ const uint64_t desired_rates[] =3D { > >>> 180500, /* 1.44Mpbs - rate 2 common rate */ > >>> 375000, /* 3Mbps - rate 3 */ > >>> 625000, /* 5Mbps - rate 4 */ > >>> - 875000, /* 7Mbps - rate 5 */ > >>> - 1125000, /* 9Mbps - rate 6 */ > >>> - 1375000, /* 11Mbps - rate 7 */ > >>> - 1625000, /* 13Mbps - rate 8 */ > >>> - 2625000, /* 21Mbps - rate 9 */ > >>> - 3875000, /* 31Mbps - rate 10 */ > >>> - 5125000, /* 41Meg - rate 11 */ > >>> - 12500000, /* 100Mbps - rate 12 */ > >>> - 25000000, /* 200Mbps - rate 13 */ > >>> - 50000000, /* 400Mbps - rate 14 */ > >>> - 63750000, /* 51Mbps - rate 15 */ > >>> + 1250000, /* 10Mbps - rate 5 */ > >>> + 1875000, /* 15Mbps - rate 6 */ > >>> + 2500000, /* 20Mbps - rate 7 */ > >>> + 3125000, /* 25Mbps - rate 8 */ > >>> + 3750000, /* 30Mbps - rate 9 */ > >>> + 4375000, /* 35Mbps - rate 10 */ > >>> + 5000000, /* 40Meg - rate 11 */ > >>> + 6250000, /* 50Mbps - rate 12 */ > >>> + 12500000, /* 100Mbps - rate 13 */ > >>> + 25000000, /* 200Mbps - rate 14 */ > >>> + 50000000, /* 400Mbps - rate 15 */ > >>> 100000000, /* 800Mbps - rate 16 */ > >>> - 1875000, /* 15Mbps - rate 17 */ > >>> - 2125000, /* 17Mbps - rate 18 */ > >>> - 2375000, /* 19Mbps - rate 19 */ > >>> - 2875000, /* 23Mbps - rate 20 */ > >>> - 3125000, /* 25Mbps - rate 21 */ > >>> - 3375000, /* 27Mbps - rate 22 */ > >>> - 3625000, /* 29Mbps - rate 23 */ > >>> - 4125000, /* 33Mbps - rate 24 */ > >>> - 4375000, /* 35Mbps - rate 25 */ > >>> - 4625000, /* 37Mbps - rate 26 */ > >>> - 4875000, /* 39Mbps - rate 27 */ > >>> - 5375000, /* 43Mbps - rate 28 */ > >>> - 5625000, /* 45Mbps - rate 29 */ > >>> - 5875000, /* 47Mbps - rate 30 */ > >>> - 6125000, /* 49Mbps - rate 31 */ > >>> - 6625000, /* 53Mbps - rate 32 */ > >>> - 6875000, /* 55Mbps - rate 33 */ > >>> - 7125000, /* 57Mbps - rate 34 */ > >>> - 7375000, /* 59Mbps - rate 35 */ > >>> - 7625000, /* 61Mbps - rate 36 */ > >>> - 7875000, /* 63Mbps - rate 37 */ > >>> - 8125000, /* 65Mbps - rate 38 */ > >>> - 8375000, /* 67Mbps - rate 39 */ > >>> - 8625000, /* 69Mbps - rate 40 */ > >>> - 8875000, /* 71Mbps - rate 41 */ > >>> - 9125000, /* 73Mbps - rate 42 */ > >>> - 9375000, /* 75Mbps - rate 43 */ > >>> - 9625000, /* 77Mbps - rate 44 */ > >>> - 9875000, /* 79Mbps - rate 45 */ > >>> - 10125000, /* 81Mbps - rate 46 */ > >>> - 10375000, /* 83Mbps - rate 47 */ > >>> - 10625000, /* 85Mbps - rate 48 */ > >>> - 10875000, /* 87Mbps - rate 49 */ > >>> - 11125000, /* 89Mbps - rate 50 */ > >>> - 11375000, /* 91Mbps - rate 51 */ > >>> - 11625000, /* 93Mbps - rate 52 */ > >>> - 11875000, /* 95Mbps - rate 53 */ > >>> - 13125000, /* 105Mbps - rate 54 */ > >>> - 13750000, /* 110Mbps - rate 55 */ > >>> - 14375000, /* 115Mbps - rate 56 */ > >>> - 15000000, /* 120Mbps - rate 57 */ > >>> - 15625000, /* 125Mbps - rate 58 */ > >>> - 16250000, /* 130Mbps - rate 59 */ > >>> - 16875000, /* 135Mbps - rate 60 */ > >>> - 17500000, /* 140Mbps - rate 61 */ > >>> - 18125000, /* 145Mbps - rate 62 */ > >>> - 18750000, /* 150Mbps - rate 64 */ > >>> - 20000000, /* 160Mbps - rate 65 */ > >>> - 21250000, /* 170Mbps - rate 66 */ > >>> - 22500000, /* 180Mbps - rate 67 */ > >>> - 23750000, /* 190Mbps - rate 68 */ > >>> - 26250000, /* 210Mbps - rate 69 */ > >>> - 27500000, /* 220Mbps - rate 70 */ > >>> - 28750000, /* 230Mbps - rate 71 */ > >>> - 30000000, /* 240Mbps - rate 72 */ > >>> - 31250000, /* 250Mbps - rate 73 */ > >>> - 34375000, /* 275Mbps - rate 74 */ > >>> - 37500000, /* 300Mbps - rate 75 */ > >>> - 40625000, /* 325Mbps - rate 76 */ > >>> - 43750000, /* 350Mbps - rate 77 */ > >>> - 46875000, /* 375Mbps - rate 78 */ > >>> - 53125000, /* 425Mbps - rate 79 */ > >>> - 56250000, /* 450Mbps - rate 80 */ > >>> - 59375000, /* 475Mbps - rate 81 */ > >>> - 62500000, /* 500Mbps - rate 82 */ > >>> - 68750000, /* 550Mbps - rate 83 */ > >>> - 75000000, /* 600Mbps - rate 84 */ > >>> - 81250000, /* 650Mbps - rate 85 */ > >>> - 87500000, /* 700Mbps - rate 86 */ > >>> - 93750000, /* 750Mbps - rate 87 */ > >>> - 106250000, /* 850Mbps - rate 88 */ > >>> - 112500000, /* 900Mbps - rate 89 */ > >>> - 125000000, /* 1Gbps - rate 90 */ > >>> - 156250000, /* 1.25Gps - rate 91 */ > >>> - 187500000, /* 1.5Gps - rate 92 */ > >>> - 218750000, /* 1.75Gps - rate 93 */ > >>> - 250000000, /* 2Gbps - rate 94 */ > >>> - 281250000, /* 2.25Gps - rate 95 */ > >>> - 312500000, /* 2.5Gbps - rate 96 */ > >>> - 343750000, /* 2.75Gbps - rate 97 */ > >>> - 375000000, /* 3Gbps - rate 98 */ > >>> - 500000000, /* 4Gbps - rate 99 */ > >>> - 625000000, /* 5Gbps - rate 100 */ > >>> - 750000000, /* 6Gbps - rate 101 */ > >>> - 875000000, /* 7Gbps - rate 102 */ > >>> - 1000000000, /* 8Gbps - rate 103 */ > >>> - 1125000000, /* 9Gbps - rate 104 */ > >>> - 1250000000, /* 10Gbps - rate 105 */ > >>> - 1875000000, /* 15Gbps - rate 106 */ > >>> - 2500000000 /* 20Gbps - rate 107 */ > >>> + 5625000, /* 45Mbps - rate 17 */ > >>> + 6875000, /* 55Mbps - rate 19 */ > >>> + 7500000, /* 60Mbps - rate 20 */ > >>> + 8125000, /* 65Mbps - rate 21 */ > >>> + 8750000, /* 70Mbps - rate 22 */ > >>> + 9375000, /* 75Mbps - rate 23 */ > >>> + 10000000, /* 80Mbps - rate 24 */ > >>> + 10625000, /* 85Mbps - rate 25 */ > >>> + 11250000, /* 90Mbps - rate 26 */ > >>> + 11875000, /* 95Mbps - rate 27 */ > >>> + 12500000, /* 100Mbps - rate 28 */ > >>> + 13750000, /* 110Mbps - rate 29 */ > >>> + 15000000, /* 120Mbps - rate 30 */ > >>> + 16250000, /* 130Mbps - rate 31 */ > >>> + 17500000, /* 140Mbps - rate 32 */ > >>> + 18750000, /* 150Mbps - rate 33 */ > >>> + 20000000, /* 160Mbps - rate 34 */ > >>> + 21250000, /* 170Mbps - rate 35 */ > >>> + 22500000, /* 180Mbps - rate 36 */ > >>> + 23750000, /* 190Mbps - rate 37 */ > >>> + 26250000, /* 210Mbps - rate 38 */ > >>> + 27500000, /* 220Mbps - rate 39 */ > >>> + 28750000, /* 230Mbps - rate 40 */ > >>> + 30000000, /* 240Mbps - rate 41 */ > >>> + 31250000, /* 250Mbps - rate 42 */ > >>> + 34375000, /* 275Mbps - rate 43 */ > >>> + 37500000, /* 300Mbps - rate 44 */ > >>> + 40625000, /* 325Mbps - rate 45 */ > >>> + 43750000, /* 350Mbps - rate 46 */ > >>> + 46875000, /* 375Mbps - rate 47 */ > >>> + 53125000, /* 425Mbps - rate 48 */ > >>> + 56250000, /* 450Mbps - rate 49 */ > >>> + 59375000, /* 475Mbps - rate 50 */ > >>> + 62500000, /* 500Mbps - rate 51 */ > >>> + 68750000, /* 550Mbps - rate 52 */ > >>> + 75000000, /* 600Mbps - rate 53 */ > >>> + 81250000, /* 650Mbps - rate 54 */ > >>> + 87500000, /* 700Mbps - rate 55 */ > >>> + 93750000, /* 750Mbps - rate 56 */ > >>> + 106250000, /* 850Mbps - rate 57 */ > >>> + 112500000, /* 900Mbps - rate 58 */ > >>> + 125000000, /* 1Gbps - rate 59 */ > >>> + 156250000, /* 1.25Gps - rate 60 */ > >>> + 187500000, /* 1.5Gps - rate 61 */ > >>> + 218750000, /* 1.75Gps - rate 62 */ > >>> + 250000000, /* 2Gbps - rate 63 */ > >>> + 281250000, /* 2.25Gps - rate 64 */ > >>> + 312500000, /* 2.5Gbps - rate 65 */ > >>> + 343750000, /* 2.75Gbps - rate 66 */ > >>> + 375000000, /* 3Gbps - rate 67 */ > >>> + 500000000, /* 4Gbps - rate 68 */ > >>> + 625000000, /* 5Gbps - rate 69 */ > >>> + 750000000, /* 6Gbps - rate 70 */ > >>> + 875000000, /* 7Gbps - rate 71 */ > >>> + 1000000000, /* 8Gbps - rate 72 */ > >>> + 1125000000, /* 9Gbps - rate 73 */ > >>> + 1250000000, /* 10Gbps - rate 74 */ > >>> + 1875000000, /* 15Gbps - rate 75 */ > >>> + 2500000000 /* 20Gbps - rate 76 */ > >>> }; > >>>=20 > >>> #define MAX_HDWR_RATES (sizeof(desired_rates)/sizeof(uint64_t)) > >>> @@ -283,6 +254,10 @@ static struct head_tcp_rate_set int_rs; > >>> static struct mtx rs_mtx; > >>> uint32_t rs_number_alive; > >>> uint32_t rs_number_dead; > >>> +static uint32_t rs_floor_mss =3D 0; > >>> +static uint32_t wait_time_floor =3D 8000; /* 8 ms */ > >>> +static uint32_t rs_hw_floor_mss =3D 16; > >>> +static uint32_t num_of_waits_allowed =3D 1; /* How many time blocks = are we > >>> willing to wait */ > >>>=20 > >>> SYSCTL_NODE(_net_inet_tcp, OID_AUTO, rl, CTLFLAG_RW | CTLFLAG_MPSAFE,= 0, > >>> "TCP Ratelimit stats"); > >>> @@ -292,6 +267,20 @@ SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, alive, > >>> CTLFLAG_RW, > >>> SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, dead, CTLFLAG_RW, > >>> &rs_number_dead, 0, > >>> "Number of interfaces departing from ratelimiting"); > >>> +SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, floor_mss, CTLFLAG_RW, > >>> + &rs_floor_mss, 0, > >>> + "Number of MSS that will override the normal minimums (0 means d= on't > >>> enforce)"); > >>> +SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, wait_floor, CTLFLAG_RW, > >>> + &wait_time_floor, 2000, > >>> + "Has b/w increases what is the wait floor we are willing to wait= at > >>> the > >>> end?"); > >>> +SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, time_blocks, CTLFLAG_RW, > >>> + &num_of_waits_allowed, 1, > >>> + "How many time blocks on the end should software pacing be willi= ng > >>> to > >>> wait?"); > >>> + > >>> +SYSCTL_UINT(_net_inet_tcp_rl, OID_AUTO, hw_floor_mss, CTLFLAG_RW, > >>> + &rs_hw_floor_mss, 16, > >>> + "Number of mss that are a minum for hardware pacing?"); > >>> + > >>>=20 > >>> static void > >>> rl_add_syctl_entries(struct sysctl_oid *rl_sysctl_root, struct > >>> tcp_rate_set > >>> *rs) > >>> @@ -383,6 +372,17 @@ rl_add_syctl_entries(struct sysctl_oid > >>> *rl_sysctl_root, > >>> struct tcp_rate_set *rs) > >>> OID_AUTO, "rate", CTLFLAG_RD, > >>> &rs->rs_rlt[i].rate, 0, > >>> "Rate in bytes per second"); > >>> + SYSCTL_ADD_U64(&rs->sysctl_ctx, > >>> + SYSCTL_CHILDREN(rl_rate_num), > >>> + OID_AUTO, "using", CTLFLAG_RD, > >>> + &rs->rs_rlt[i].using, 0, > >>> + "Number of flows using"); > >>> + SYSCTL_ADD_U64(&rs->sysctl_ctx, > >>> + SYSCTL_CHILDREN(rl_rate_num), > >>> + OID_AUTO, "enobufs", CTLFLAG_RD, > >>> + &rs->rs_rlt[i].rs_num_enobufs, 0, > >>> + "Number of enobufs logged on this rate"); > >>> + > >>> } > >>> } > >>> #endif > >>> @@ -443,6 +443,8 @@ rs_defer_destroy(struct tcp_rate_set *rs) > >>> } > >>>=20 > >>> #ifdef INET > >>> +extern counter_u64_t rate_limit_new; > >>> +extern counter_u64_t rate_limit_chg; > >>> extern counter_u64_t rate_limit_set_ok; > >>> extern counter_u64_t rate_limit_active; > >>> extern counter_u64_t rate_limit_alloc_fail; > >>> @@ -519,6 +521,7 @@ rt_setup_new_rs(struct ifnet *ifp, int *error) > >>> int i; > >>> struct if_ratelimit_query_results rl; > >>> struct sysctl_oid *rl_sysctl_root; > >>> + struct epoch_tracker et; > >>> /* > >>> * We expect to enter with the > >>> * mutex locked. > >>> @@ -562,9 +565,11 @@ rt_setup_new_rs(struct ifnet *ifp, int *error) > >>> CTLFLAG_RW | CTLFLAG_MPSAFE, 0, > >>> ""); > >>> rl_add_syctl_entries(rl_sysctl_root, rs); > >>> + NET_EPOCH_ENTER(et); > >>> mtx_lock(&rs_mtx); > >>> CK_LIST_INSERT_HEAD(&int_rs, rs, next); > >>> mtx_unlock(&rs_mtx); > >>> + NET_EPOCH_EXIT(et); > >>> return (rs); > >>> } else if ((rl.flags & RT_IS_INDIRECT) =3D=3D RT_IS_INDIRECT) { > >>> memset(rs, 0, sizeof(struct tcp_rate_set)); > >>> @@ -580,9 +585,11 @@ rt_setup_new_rs(struct ifnet *ifp, int *error) > >>> CTLFLAG_RW | CTLFLAG_MPSAFE, 0, > >>> ""); > >>> rl_add_syctl_entries(rl_sysctl_root, rs); > >>> + NET_EPOCH_ENTER(et); > >>> mtx_lock(&rs_mtx); > >>> CK_LIST_INSERT_HEAD(&int_rs, rs, next); > >>> mtx_unlock(&rs_mtx); > >>> + NET_EPOCH_EXIT(et); > >>> return (rs); > >>> } else if ((rl.flags & RT_IS_FIXED_TABLE) =3D=3D RT_IS_FIXED_TABLE) { > >>> /* Mellanox C4 likely */ > >>> @@ -671,6 +678,8 @@ bail: > >>> */ > >>> rs->rs_rlt[i].ptbl =3D rs; > >>> rs->rs_rlt[i].tag =3D NULL; > >>> + rs->rs_rlt[i].using =3D 0; > >>> + rs->rs_rlt[i].rs_num_enobufs =3D 0; > >>> /* > >>> * Calculate the time between. > >>> */ > >>> @@ -741,18 +750,24 @@ handle_err: > >>> CTLFLAG_RW | CTLFLAG_MPSAFE, 0, > >>> ""); > >>> rl_add_syctl_entries(rl_sysctl_root, rs); > >>> + NET_EPOCH_ENTER(et); > >>> mtx_lock(&rs_mtx); > >>> CK_LIST_INSERT_HEAD(&int_rs, rs, next); > >>> mtx_unlock(&rs_mtx); > >>> + NET_EPOCH_EXIT(et); > >>> return (rs); > >>> } > >>>=20 > >>> +/* > >>> + * For an explanation of why the argument is volatile please > >>> + * look at the comments around rt_setup_rate(). > >>> + */ > >>> static const struct tcp_hwrate_limit_table * > >>> -tcp_int_find_suitable_rate(const struct tcp_rate_set *rs, > >>> - uint64_t bytes_per_sec, uint32_t flags) > >>> +tcp_int_find_suitable_rate(const volatile struct tcp_rate_set *rs, > >>> + uint64_t bytes_per_sec, uint32_t flags, uint64_t *lower_rate) > >>> { > >>> struct tcp_hwrate_limit_table *arte =3D NULL, *rte =3D NULL; > >>> - uint64_t mbits_per_sec, ind_calc; > >>> + uint64_t mbits_per_sec, ind_calc, previous_rate =3D 0; > >>> int i; > >>>=20 > >>> mbits_per_sec =3D (bytes_per_sec * 8); > >>> @@ -763,6 +778,7 @@ tcp_int_find_suitable_rate(const struct tcp_rate_= set > >>> *rs, > >>> * Smaller than 1Meg, only > >>> * 3 entries can match it. > >>> */ > >>> + previous_rate =3D 0; > >>> for(i =3D rs->rs_lowest_valid; i < 3; i++) { > >>> if (bytes_per_sec <=3D rs->rs_rlt[i].rate) { > >>> rte =3D &rs->rs_rlt[i]; > >>> @@ -770,6 +786,7 @@ tcp_int_find_suitable_rate(const struct tcp_rate_= set > >>> *rs, > >>> } else if (rs->rs_rlt[i].flags & HDWRPACE_INITED) { > >>> arte =3D &rs->rs_rlt[i]; > >>> } > >>> + previous_rate =3D rs->rs_rlt[i].rate; > >>> } > >>> goto done; > >>> } else if ((mbits_per_sec > RS_ONE_GIGABIT_PERSEC) && > >>> @@ -782,6 +799,7 @@ tcp_int_find_suitable_rate(const struct tcp_rate_= set > >>> *rs, > >>> rte =3D &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; > >>> else > >>> arte =3D &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; > >>> + previous_rate =3D rs->rs_rlt[(ALL_HARDWARE_RATES-2)].rate; > >>> goto done; > >>> } > >>> /* > >>> @@ -800,8 +818,11 @@ tcp_int_find_suitable_rate(const struct tcp_rate= _set > >>> *rs, > >>> ind_calc =3D ALL_HARDWARE_RATES-1; > >>> } > >>> if ((ind_calc >=3D rs->rs_lowest_valid) && > >>> - (ind_calc <=3D rs->rs_highest_valid)) > >>> - rte =3D &rs->rs_rlt[ind_calc]; > >>> + (ind_calc <=3D rs->rs_highest_valid)) { > >>> + rte =3D &rs->rs_rlt[ind_calc]; > >>> + if (ind_calc >=3D 1) > >>> + previous_rate =3D rs->rs_rlt[(ind_calc-1)].rate; > >>> + } > >>> } else if (flags & RS_PACING_EXACT_MATCH) { > >>> if ((mbits_per_sec < RS_ONE_MEGABIT_PERSEC) && > >>> (rs->rs_lowest_valid <=3D 2)){ > >>> @@ -840,10 +861,16 @@ tcp_int_find_suitable_rate(const struct > >>> tcp_rate_set > >>> *rs, > >>> for (i=3D2; i>=3Drs->rs_lowest_valid; i--) { > >>> if (bytes_per_sec < rs->rs_rlt[i].rate) { > >>> rte =3D &rs->rs_rlt[i]; > >>> + if (i >=3D 1) { > >>> + previous_rate =3D rs->rs_rlt[(i-1)].rate; > >>> + } > >>> break; > >>> } else if ((flags & RS_PACING_GEQ) && > >>> (bytes_per_sec =3D=3D rs->rs_rlt[i].rate)) { > >>> rte =3D &rs->rs_rlt[i]; > >>> + if (i >=3D 1) { > >>> + previous_rate =3D rs->rs_rlt[(i-1)].rate; > >>> + } > >>> break; > >>> } else { > >>> arte =3D &rs->rs_rlt[i]; /* new alternate */ > >>> @@ -863,6 +890,7 @@ tcp_int_find_suitable_rate(const struct tcp_rate_= set > >>> *rs, > >>> /* The top rate is an alternative */ > >>> arte =3D &rs->rs_rlt[(ALL_HARDWARE_RATES-1)]; > >>> } > >>> + previous_rate =3D rs->rs_rlt[(ALL_HARDWARE_RATES-2)].rate; > >>> } else { > >>> /* Its in our range 1Meg - 1Gig */ > >>> if (flags & RS_PACING_GEQ) { > >>> @@ -873,6 +901,8 @@ tcp_int_find_suitable_rate(const struct tcp_rate_= set > >>> *rs, > >>> ind_calc =3D (ALL_HARDWARE_RATES-1); > >>> } > >>> rte =3D &rs->rs_rlt[ind_calc]; > >>> + if (ind_calc >=3D 1) > >>> + previous_rate =3D rs->rs_rlt[(ind_calc-1)].rate; > >>> } > >>> goto done; > >>> } > >>> @@ -882,8 +912,11 @@ tcp_int_find_suitable_rate(const struct tcp_rate= _set > >>> *rs, > >>> /* This should not happen */ > >>> ind_calc =3D ALL_HARDWARE_RATES-1; > >>> } > >>> - if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) > >>> + if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) { > >>> rte =3D &rs->rs_rlt[ind_calc]; > >>> + if (ind_calc >=3D 1) > >>> + previous_rate =3D rs->rs_rlt[(ind_calc-1)].rate; > >>> + } > >>> } > >>> } > >>> done: > >>> @@ -893,11 +926,17 @@ done: > >>> /* We can use the substitute */ > >>> rte =3D arte; > >>> } > >>> + if (lower_rate) > >>> + *lower_rate =3D previous_rate; > >>> return (rte); > >>> } > >>>=20 > >>> +/* > >>> + * For an explanation of why the argument is volatile please > >>> + * look at the comments around rt_setup_rate(). > >>> + */ > >>> static const struct tcp_hwrate_limit_table * > >>> -tcp_find_suitable_rate(const struct tcp_rate_set *rs, uint64_t > >>> bytes_per_sec, uint32_t flags) > >>> +tcp_find_suitable_rate(const volatile struct tcp_rate_set *rs, uint6= 4_t > >>> bytes_per_sec, uint32_t flags, uint64_t *lower_rate) > >>> { > >>> /** > >>> * Hunt the rate table with the restrictions in flags and find a > >>> @@ -911,6 +950,7 @@ tcp_find_suitable_rate(const struct tcp_rate_set = *rs, > >>> uint64_t bytes_per_sec, ui > >>> */ > >>> int i, matched; > >>> struct tcp_hwrate_limit_table *rte =3D NULL; > >>> + uint64_t previous_rate =3D 0; > >>>=20 > >>> if ((rs->rs_flags & RS_INT_TBL) && > >>> (rs->rs_rate_cnt >=3D ALL_HARDWARE_RATES)) { > >>> @@ -920,7 +960,7 @@ tcp_find_suitable_rate(const struct tcp_rate_set = *rs, > >>> uint64_t bytes_per_sec, ui > >>> * from 1Meg - 1000Meg in 1Meg increments. > >>> * Use an alternate method to "lookup". > >>> */ > >>> - return (tcp_int_find_suitable_rate(rs, bytes_per_sec, flags)); > >>> + return (tcp_int_find_suitable_rate(rs, bytes_per_sec, flags, > >>> lower_rate)); > >>> } > >>> if ((flags & RS_PACING_LT) || > >>> (flags & RS_PACING_EXACT_MATCH)) { > >>> @@ -934,13 +974,18 @@ tcp_find_suitable_rate(const struct tcp_rate_set > >>> *rs, > >>> uint64_t bytes_per_sec, ui > >>> (bytes_per_sec =3D=3D rs->rs_rlt[i].rate)) { > >>> rte =3D &rs->rs_rlt[i]; > >>> matched =3D 1; > >>> + if (lower_rate !=3D NULL) > >>> + *lower_rate =3D previous_rate; > >>> break; > >>> } else if ((flags & RS_PACING_LT) && > >>> (bytes_per_sec <=3D rs->rs_rlt[i].rate)) { > >>> rte =3D &rs->rs_rlt[i]; > >>> matched =3D 1; > >>> + if (lower_rate !=3D NULL) > >>> + *lower_rate =3D previous_rate; > >>> break; > >>> } > >>> + previous_rate =3D rs->rs_rlt[i].rate; > >>> if (bytes_per_sec > rs->rs_rlt[i].rate) > >>> break; > >>> } > >>> @@ -979,6 +1024,8 @@ tcp_find_suitable_rate(const struct tcp_rate_set > >>> *rs, > >>> uint64_t bytes_per_sec, ui > >>> * We found a table entry that is smaller, > >>> * stop there will be none greater or equal. > >>> */ > >>> + if (lower_rate !=3D NULL) > >>> + *lower_rate =3D rs->rs_rlt[i].rate; > >>> break; > >>> } > >>> } > >>> @@ -995,10 +1042,10 @@ static struct ifnet * > >>> rt_find_real_interface(struct ifnet *ifp, struct inpcb *inp, int *err= or) > >>> { > >>> struct ifnet *tifp; > >>> - struct m_snd_tag *tag; > >>> + struct m_snd_tag *tag, *ntag; > >>> union if_snd_tag_alloc_params params =3D { > >>> .rate_limit.hdr.type =3D IF_SND_TAG_TYPE_RATE_LIMIT, > >>> - .rate_limit.hdr.flowid =3D 1, > >>> + .rate_limit.hdr.flowid =3D inp->inp_flowid, > >>> .rate_limit.hdr.numa_domain =3D inp->inp_numa_domain, > >>> .rate_limit.max_rate =3D COMMON_RATE, > >>> .rate_limit.flags =3D M_NOWAIT, > >>> @@ -1017,38 +1064,92 @@ rt_find_real_interface(struct ifnet *ifp, str= uct > >>> inpcb *inp, int *error) > >>> *error =3D err; > >>> return (NULL); > >>> } > >>> - tifp =3D tag->ifp; > >>> + ntag =3D tag; > >>> + while(ntag->ifp->if_next_snd_tag !=3D NULL) { > >>> + ntag =3D ntag->ifp->if_next_snd_tag(ntag); > >>> + } > >>> + tifp =3D ntag->ifp; > >>> m_snd_tag_rele(tag); > >>> return (tifp); > >>> } > >>>=20 > >>> +static void > >>> +rl_increment_using(const struct tcp_hwrate_limit_table *rte) > >>> +{ > >>> + struct tcp_hwrate_limit_table *decon_rte; > >>> + > >>> + decon_rte =3D __DECONST(struct tcp_hwrate_limit_table *, rte); > >>> + atomic_add_long(&decon_rte->using, 1); > >>> +} > >>> + > >>> +static void > >>> +rl_decrement_using(const struct tcp_hwrate_limit_table *rte) > >>> +{ > >>> + struct tcp_hwrate_limit_table *decon_rte; > >>> + > >>> + decon_rte =3D __DECONST(struct tcp_hwrate_limit_table *, rte); > >>> + atomic_subtract_long(&decon_rte->using, 1); > >>> +} > >>> + > >>> +void > >>> +tcp_rl_log_enobuf(const struct tcp_hwrate_limit_table *rte) > >>> +{ > >>> + struct tcp_hwrate_limit_table *decon_rte; > >>> + > >>> + decon_rte =3D __DECONST(struct tcp_hwrate_limit_table *, rte); > >>> + atomic_add_long(&decon_rte->rs_num_enobufs, 1); > >>> +} > >>> + > >>> +/* > >>> + * Do NOT take the __noinline out of the > >>> + * find_rs_for_ifp() function. If you do the inline > >>> + * of it for the rt_setup_rate() will show you a > >>> + * compiler bug. For some reason the compiler thinks > >>> + * the list can never be empty. The consequence of > >>> + * this will be a crash when we dereference NULL > >>> + * if an ifp is removed just has a hw rate limit > >>> + * is attempted. If you are working on the compiler > >>> + * and want to "test" this go ahead and take the noinline > >>> + * out otherwise let sleeping dogs ly until such time > >>> + * as we get a compiler fix 10/2/20 -- RRS > >>> + */ > >>> +static __noinline struct tcp_rate_set * > >>> +find_rs_for_ifp(struct ifnet *ifp) > >>> +{ > >>> + struct tcp_rate_set *rs; > >>> + > >>> + CK_LIST_FOREACH(rs, &int_rs, next) { > >>> + if ((rs->rs_ifp =3D=3D ifp) && > >>> + (rs->rs_if_dunit =3D=3D ifp->if_dunit)) { > >>> + /* Ok we found it */ > >>> + return (rs); > >>> + } > >>> + } > >>> + return (NULL); > >>> +} > >>> + > >>> + > >>> static const struct tcp_hwrate_limit_table * > >>> rt_setup_rate(struct inpcb *inp, struct ifnet *ifp, uint64_t > >>> bytes_per_sec, > >>> - uint32_t flags, int *error) > >>> + uint32_t flags, int *error, uint64_t *lower_rate) > >>> { > >>> /* First lets find the interface if it exists */ > >>> const struct tcp_hwrate_limit_table *rte; > >>> - struct tcp_rate_set *rs; > >>> + /* > >>> + * So why is rs volatile? This is to defeat a > >>> + * compiler bug where in the compiler is convinced > >>> + * that rs can never be NULL (which is not true). Because > >>> + * of its conviction it nicely optimizes out the if ((rs =3D=3D NULL > >>> + * below which means if you get a NULL back you dereference it. > >>> + */ > >>> + volatile struct tcp_rate_set *rs; > >>> struct epoch_tracker et; > >>> + struct ifnet *oifp =3D ifp; > >>> int err; > >>>=20 > >>> NET_EPOCH_ENTER(et); > >>> use_real_interface: > >>> - CK_LIST_FOREACH(rs, &int_rs, next) { > >>> - /* > >>> - * Note we don't look with the lock since we either see a > >>> - * new entry or will get one when we try to add it. > >>> - */ > >>> - if (rs->rs_flags & RS_IS_DEAD) { > >>> - /* The dead are not looked at */ > >>> - continue; > >>> - } > >>> - if ((rs->rs_ifp =3D=3D ifp) && > >>> - (rs->rs_if_dunit =3D=3D ifp->if_dunit)) { > >>> - /* Ok we found it */ > >>> - break; > >>> - } > >>> - } > >>> + rs =3D find_rs_for_ifp(ifp); > >>> if ((rs =3D=3D NULL) || > >>> (rs->rs_flags & RS_INTF_NO_SUP) || > >>> (rs->rs_flags & RS_IS_DEAD)) { > >>> @@ -1063,14 +1164,14 @@ use_real_interface: > >>> * might be arguable, but its impossible > >>> * to tell from the departing case. > >>> */ > >>> - if (rs->rs_disable && error) > >>> + if (error) > >>> *error =3D ENODEV; > >>> NET_EPOCH_EXIT(et); > >>> return (NULL); > >>> } > >>>=20 > >>> if ((rs =3D=3D NULL) || (rs->rs_disable !=3D 0)) { > >>> - if (rs->rs_disable && error) > >>> + if (error) > >>> *error =3D ENOSPC; > >>> NET_EPOCH_EXIT(et); > >>> return (NULL); > >>> @@ -1086,6 +1187,10 @@ use_real_interface: > >>> NET_EPOCH_EXIT(et); > >>> return (NULL); > >>> } > >>> + KASSERT((tifp !=3D ifp), > >>> + ("Lookup failure ifp:%p inp:%p rt_find_real_interface() returns t= he > >>> same > >>> interface tifp:%p?\n", > >>> + ifp, inp, tifp)); > >>> + ifp =3D tifp; > >>> goto use_real_interface; > >>> } > >>> if (rs->rs_flow_limit && > >>> @@ -1095,9 +1200,9 @@ use_real_interface: > >>> NET_EPOCH_EXIT(et); > >>> return (NULL); > >>> } > >>> - rte =3D tcp_find_suitable_rate(rs, bytes_per_sec, flags); > >>> + rte =3D tcp_find_suitable_rate(rs, bytes_per_sec, flags, lower_rate= ); > >>> if (rte) { > >>> - err =3D in_pcbattach_txrtlmt(inp, rs->rs_ifp, > >>> + err =3D in_pcbattach_txrtlmt(inp, oifp, > >>> inp->inp_flowtype, > >>> inp->inp_flowid, > >>> rte->rate, > >>> @@ -1107,6 +1212,11 @@ use_real_interface: > >>> if (error) > >>> *error =3D err; > >>> rte =3D NULL; > >>> + } else { > >>> + KASSERT((inp->inp_snd_tag !=3D NULL) , > >>> + ("Setup rate has no snd_tag inp:%p rte:%p rate:%lu rs:%p", > >>> + inp, rte, rte->rate, rs)); > >>> + counter_u64_add(rate_limit_new, 1); > >>> } > >>> } > >>> if (rte) { > >>> @@ -1125,6 +1235,7 @@ tcp_rl_ifnet_link(void *arg __unused, struct if= net > >>> *ifp, int link_state) > >>> { > >>> int error; > >>> struct tcp_rate_set *rs; > >>> + struct epoch_tracker et; > >>>=20 > >>> if (((ifp->if_capenable & IFCAP_TXRTLMT) =3D=3D 0) || > >>> (link_state !=3D LINK_STATE_UP)) { > >>> @@ -1134,53 +1245,56 @@ tcp_rl_ifnet_link(void *arg __unused, struct > >>> ifnet > >>> *ifp, int link_state) > >>> */ > >>> return; > >>> } > >>> + NET_EPOCH_ENTER(et); > >>> mtx_lock(&rs_mtx); > >>> - CK_LIST_FOREACH(rs, &int_rs, next) { > >>> - if ((rs->rs_ifp =3D=3D ifp) && > >>> - (rs->rs_if_dunit =3D=3D ifp->if_dunit)) { > >>> - /* We already have initialized this guy */ > >>> - mtx_unlock(&rs_mtx); > >>> - return; > >>> - } > >>> + rs =3D find_rs_for_ifp(ifp); > >>> + if (rs) { > >>> + /* We already have initialized this guy */ > >>> + mtx_unlock(&rs_mtx); > >>> + NET_EPOCH_EXIT(et); > >>> + return; > >>> } > >>> mtx_unlock(&rs_mtx); > >>> + NET_EPOCH_EXIT(et); > >>> rt_setup_new_rs(ifp, &error); > >>> } > >>>=20 > >>> static void > >>> tcp_rl_ifnet_departure(void *arg __unused, struct ifnet *ifp) > >>> { > >>> - struct tcp_rate_set *rs, *nrs; > >>> + struct tcp_rate_set *rs; > >>> + struct epoch_tracker et; > >>> int i; > >>>=20 > >>> + NET_EPOCH_ENTER(et); > >>> mtx_lock(&rs_mtx); > >>> - CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) { > >>> - if ((rs->rs_ifp =3D=3D ifp) && > >>> - (rs->rs_if_dunit =3D=3D ifp->if_dunit)) { > >>> - CK_LIST_REMOVE(rs, next); > >>> - rs_number_alive--; > >>> - rs->rs_flags |=3D RS_IS_DEAD; > >>> - for (i =3D 0; i < rs->rs_rate_cnt; i++) { > >>> - if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) { > >>> - in_pcbdetach_tag(rs->rs_rlt[i].tag); > >>> - rs->rs_rlt[i].tag =3D NULL; > >>> - } > >>> - rs->rs_rlt[i].flags =3D HDWRPACE_IFPDEPARTED; > >>> + rs =3D find_rs_for_ifp(ifp); > >>> + if (rs) { > >>> + CK_LIST_REMOVE(rs, next); > >>> + rs_number_alive--; > >>> + rs->rs_flags |=3D RS_IS_DEAD; > >>> + for (i =3D 0; i < rs->rs_rate_cnt; i++) { > >>> + if (rs->rs_rlt[i].flags & HDWRPACE_TAGPRESENT) { > >>> + in_pcbdetach_tag(rs->rs_rlt[i].tag); > >>> + rs->rs_rlt[i].tag =3D NULL; > >>> } > >>> - if (rs->rs_flows_using =3D=3D 0) > >>> - rs_defer_destroy(rs); > >>> - break; > >>> + rs->rs_rlt[i].flags =3D HDWRPACE_IFPDEPARTED; > >>> } > >>> + if (rs->rs_flows_using =3D=3D 0) > >>> + rs_defer_destroy(rs); > >>> } > >>> mtx_unlock(&rs_mtx); > >>> + NET_EPOCH_EXIT(et); > >>> } > >>>=20 > >>> static void > >>> tcp_rl_shutdown(void *arg __unused, int howto __unused) > >>> { > >>> struct tcp_rate_set *rs, *nrs; > >>> + struct epoch_tracker et; > >>> int i; > >>>=20 > >>> + NET_EPOCH_ENTER(et); > >>> mtx_lock(&rs_mtx); > >>> CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) { > >>> CK_LIST_REMOVE(rs, next); > >>> @@ -1197,11 +1311,12 @@ tcp_rl_shutdown(void *arg __unused, int howto > >>> __unused) > >>> rs_defer_destroy(rs); > >>> } > >>> mtx_unlock(&rs_mtx); > >>> + NET_EPOCH_EXIT(et); > >>> } > >>>=20 > >>> const struct tcp_hwrate_limit_table * > >>> tcp_set_pacing_rate(struct tcpcb *tp, struct ifnet *ifp, > >>> - uint64_t bytes_per_sec, int flags, int *error) > >>> + uint64_t bytes_per_sec, int flags, int *error, uint64_t *lower_r= ate) > >>> { > >>> const struct tcp_hwrate_limit_table *rte; > >>> #ifdef KERN_TLS > >>> @@ -1233,7 +1348,9 @@ tcp_set_pacing_rate(struct tcpcb *tp, struct if= net > >>> *ifp, > >>> } > >>> } > >>> #endif > >>> - rte =3D rt_setup_rate(tp->t_inpcb, ifp, bytes_per_sec, flags, erro= r); > >>> + rte =3D rt_setup_rate(tp->t_inpcb, ifp, bytes_per_sec, flags, erro= r, > >>> lower_rate); > >>> + if (rte) > >>> + rl_increment_using(rte); > >>> #ifdef KERN_TLS > >>> if (rte !=3D NULL && tls !=3D NULL && tls->snd_tag !=3D NULL) { > >>> /* > >>> @@ -1253,22 +1370,23 @@ tcp_set_pacing_rate(struct tcpcb *tp, struct > >>> ifnet > >>> *ifp, > >>> *error =3D EINVAL; > >>> rte =3D NULL; > >>> } > >>> - tp->t_pacing_rate =3D rte->rate; > >>> - *error =3D 0; > >>> + if (rte !=3D NULL) { > >>> + tp->t_pacing_rate =3D rte->rate; > >>> + *error =3D 0; > >>> + } > >>> return (rte); > >>> } > >>>=20 > >>> const struct tcp_hwrate_limit_table * > >>> tcp_chg_pacing_rate(const struct tcp_hwrate_limit_table *crte, > >>> struct tcpcb *tp, struct ifnet *ifp, > >>> - uint64_t bytes_per_sec, int flags, int *error) > >>> + uint64_t bytes_per_sec, int flags, int *error, uint64_t *lower_r= ate) > >>> { > >>> const struct tcp_hwrate_limit_table *nrte; > >>> const struct tcp_rate_set *rs; > >>> #ifdef KERN_TLS > >>> struct ktls_session *tls =3D NULL; > >>> #endif > >>> - int is_indirect =3D 0; > >>> int err; > >>>=20 > >>> INP_WLOCK_ASSERT(tp->t_inpcb); > >>> @@ -1307,41 +1425,13 @@ tcp_chg_pacing_rate(const struct > >>> tcp_hwrate_limit_table *crte, > >>> if ((rs->rs_flags & RS_IS_DEAD) || > >>> (crte->flags & HDWRPACE_IFPDEPARTED)) { > >>> /* Release the rate, and try anew */ > >>> -re_rate: > >>> + > >>> tcp_rel_pacing_rate(crte, tp); > >>> nrte =3D tcp_set_pacing_rate(tp, ifp, > >>> - bytes_per_sec, flags, error); > >>> *** 386 LINES SKIPPED *** > >>> _______________________________________________ > >>> dev-commits-src-all@freebsd.org mailing list > >>> https://lists.freebsd.org/mailman/listinfo/dev-commits-src-all > >>> To unsubscribe, send any mail to > >>> "dev-commits-src-all-unsubscribe@freebsd.org" > >>>=20 > >>=20 > >>=20 > >> -- > >> Mateusz Guzik <mjguzik gmail.com> > >>=20 > >=20 > >=20 > > --=20 > > Mateusz Guzik <mjguzik gmail.com> >=20 > _______________________________________________ > dev-commits-src-main@freebsd.org mailing list > https://lists.freebsd.org/mailman/listinfo/dev-commits-src-main > To unsubscribe, send any mail to "dev-commits-src-main-unsubscribe@freebs= d.org" --=20 --EB6sWAVXHCNbq2EJ Content-Type: application/pgp-signature; name="signature.asc" -----BEGIN PGP SIGNATURE----- iQGTBAEBCgB9FiEEYbWI0KY5X7yH/Fy4OQX2V8rP09wFAmAWtYdfFIAAAAAALgAo aXNzdWVyLWZwckBub3RhdGlvbnMub3BlbnBncC5maWZ0aGhvcnNlbWFuLm5ldDYx QjU4OEQwQTYzOTVGQkM4N0ZDNUNCODM5MDVGNjU3Q0FDRkQzREMACgkQOQX2V8rP 09yBsggA6Bz/ekU783kfCFO7K2M5IsZxgl9HSzReBAVueYpGWtHhs5X8pnVAaf2s r6ap43+8AjfVK/4VgtuBRYmXh32vL1fiGosm69R3GTsHqzWpkeWxOZ0cda1P+hHO 8N3k/MoCe/GF+ZnDPmprWrcJFY1gcd24Cvi9N9ZTcCIQBWPrfMQLBHRxKG5Nq8fp rIUHZc6zOdjLq9dYxx6zQe2VTzUniwPpUY14a3CG+Vz8JibJbvFRCykWBsxtNnWT NAfbEmAjieEoCXBmWB2lCE9FpVId1BBxt2rJZvBXG4CoA/2XgWpRQrH3HzZZMGK2 RGLwCPyY/7ZYG/2qezSFcR0hKEvfDw== =cVfZ -----END PGP SIGNATURE----- --EB6sWAVXHCNbq2EJ--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?YBa1h6uc1uPaEOCt>