Skip site navigation (1)Skip section navigation (2)
Date:      Sun, 31 Jan 2021 12:09:58 +0100
From:      Mateusz Guzik <mjguzik@gmail.com>
To:        Randall Stewart <rrs@freebsd.org>
Cc:        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:  <CAGudoHH=Eu7zRLWezj9mOtXH827A4k4rFsVhRn45aK589mDWvQ@mail.gmail.com>
In-Reply-To: <CAGudoHFh9HMQNUOKpfrj3tBnO2%2BhUyhnnLFVF5c1NnPD=31OFA@mail.gmail.com>
References:  <202101281653.10SGrJWf002301@gitrepo.freebsd.org> <CAGudoHFh9HMQNUOKpfrj3tBnO2%2BhUyhnnLFVF5c1NnPD=31OFA@mail.gmail.com>

next in thread | previous in thread | raw e-mail | index | archive | help
ping? I'm going to revert this on Tuesday to unbreak the build

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
>
>
> On 1/28/21, Randall Stewart <rrs@freebsd.org> wrote:
>> The branch main has been updated by rrs:
>>
>> URL:
>> https://cgit.FreeBSD.org/src/commit/?id=1a714ff204193b9eb810426048e03f5d76e9730e
>>
>> 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
>>
>>     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.
>>
>>         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(-)
>>
>> 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 = lagg_snd_tag_modify;
>>  	ifp->if_snd_tag_query = lagg_snd_tag_query;
>>  	ifp->if_snd_tag_free = lagg_snd_tag_free;
>> +	ifp->if_next_snd_tag = lagg_next_snd_tag;
>>  	ifp->if_ratelimit_query = lagg_ratelimit_query;
>>  #endif
>>  	ifp->if_capenable = ifp->if_capabilities = IFCAP_HWSTATS;
>> @@ -1834,6 +1836,15 @@ lagg_snd_tag_alloc(struct ifnet *ifp,
>>  	return (0);
>>  }
>>
>> +static struct m_snd_tag *
>> +lagg_next_snd_tag(struct m_snd_tag *mst)
>> +{
>> +	struct lagg_snd_tag *lst;
>> +
>> +	lst = 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;
>>
>> 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 *name,
>> size_t len, caddr_t params)
>>  	ifp->if_snd_tag_modify = vlan_snd_tag_modify;
>>  	ifp->if_snd_tag_query = vlan_snd_tag_query;
>>  	ifp->if_snd_tag_free = vlan_snd_tag_free;
>> +	ifp->if_next_snd_tag = vlan_next_snd_tag;
>> +	ifp->if_ratelimit_query = vlan_ratelimit_query;
>>  #endif
>>  	ifp->if_flags = VLAN_IFFLAGS;
>>  	ether_ifattach(ifp, eaddr);
>> @@ -2073,6 +2078,15 @@ vlan_snd_tag_alloc(struct ifnet *ifp,
>>  	return (0);
>>  }
>>
>> +static struct m_snd_tag *
>> +vlan_next_snd_tag(struct m_snd_tag *mst)
>> +{
>> +	struct vlan_snd_tag *vst;
>> +
>> +	vst = 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 = NULL;
>> +	q->flags = RT_IS_INDIRECT;
>> +	q->max_flows = 0;
>> +	q->number_of_rates = 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");
>>
>>  #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 */
>>
>>  #endif /* INET */
>> @@ -3591,6 +3598,8 @@ in_pcboutput_eagain(struct inpcb *inp)
>>  static void
>>  rl_init(void *st)
>>  {
>> +	rate_limit_new = counter_u64_alloc(M_WAITOK);
>> +	rate_limit_chg = counter_u64_alloc(M_WAITOK);
>>  	rate_limit_active = counter_u64_alloc(M_WAITOK);
>>  	rate_limit_alloc_fail = counter_u64_alloc(M_WAITOK);
>>  	rate_limit_set_ok = 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[] = {
>>  	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 */
>>  };
>>
>>  #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 = 0;
>> +static uint32_t wait_time_floor = 8000;	/* 8 ms */
>> +static uint32_t rs_hw_floor_mss = 16;
>> +static uint32_t num_of_waits_allowed = 1; /* How many time blocks are we
>> willing to wait */
>>
>>  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 don'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 willing
>> 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?");
>> +
>>
>>  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)
>>  }
>>
>>  #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) == 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) == RT_IS_FIXED_TABLE) {
>>  		/* Mellanox C4 likely */
>> @@ -671,6 +678,8 @@ bail:
>>  		 */
>>  		rs->rs_rlt[i].ptbl = rs;
>>  		rs->rs_rlt[i].tag = NULL;
>> +		rs->rs_rlt[i].using = 0;
>> +		rs->rs_rlt[i].rs_num_enobufs = 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);
>>  }
>>
>> +/*
>> + * 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 = NULL, *rte = NULL;
>> -	uint64_t mbits_per_sec, ind_calc;
>> +	uint64_t mbits_per_sec, ind_calc, previous_rate = 0;
>>  	int i;
>>
>>  	mbits_per_sec = (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 = 0;
>>  			for(i = rs->rs_lowest_valid; i < 3; i++) {
>>  				if (bytes_per_sec <= rs->rs_rlt[i].rate) {
>>  					rte = &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 = &rs->rs_rlt[i];
>>  				}
>> +				previous_rate = 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 = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)];
>>  			else
>>  				arte = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)];
>> +			previous_rate = 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 = ALL_HARDWARE_RATES-1;
>>  		}
>>  		if ((ind_calc >= rs->rs_lowest_valid) &&
>> -		    (ind_calc <= rs->rs_highest_valid))
>> -		rte = &rs->rs_rlt[ind_calc];
>> +		    (ind_calc <= rs->rs_highest_valid)) {
>> +			rte = &rs->rs_rlt[ind_calc];
>> +			if (ind_calc >= 1)
>> +				previous_rate = 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 <= 2)){
>> @@ -840,10 +861,16 @@ tcp_int_find_suitable_rate(const struct
>> tcp_rate_set
>> *rs,
>>  			for (i=2; i>=rs->rs_lowest_valid; i--) {
>>  				if (bytes_per_sec < rs->rs_rlt[i].rate) {
>>  					rte = &rs->rs_rlt[i];
>> +					if (i >= 1) {
>> +						previous_rate = rs->rs_rlt[(i-1)].rate;
>> +					}
>>  					break;
>>  				} else if ((flags & RS_PACING_GEQ) &&
>>  					   (bytes_per_sec == rs->rs_rlt[i].rate)) {
>>  					rte = &rs->rs_rlt[i];
>> +					if (i >= 1) {
>> +						previous_rate = rs->rs_rlt[(i-1)].rate;
>> +					}
>>  					break;
>>  				} else {
>>  					arte = &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 = &rs->rs_rlt[(ALL_HARDWARE_RATES-1)];
>>  			}
>> +			previous_rate = 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 = (ALL_HARDWARE_RATES-1);
>>  					}
>>  					rte = &rs->rs_rlt[ind_calc];
>> +					if (ind_calc >= 1)
>> +						previous_rate = 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 = ALL_HARDWARE_RATES-1;
>>  			}
>> -			if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED)
>> +			if (rs->rs_rlt[ind_calc].flags & HDWRPACE_INITED) {
>>  				rte = &rs->rs_rlt[ind_calc];
>> +				if (ind_calc >= 1)
>> +					previous_rate = rs->rs_rlt[(ind_calc-1)].rate;
>> +			}
>>  		}
>>  	}
>>  done:
>> @@ -893,11 +926,17 @@ done:
>>  		/* We can use the substitute */
>>  		rte = arte;
>>  	}
>> +	if (lower_rate)
>> +		*lower_rate = previous_rate;
>>  	return (rte);
>>  }
>>
>> +/*
>> + * 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, uint64_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 = NULL;
>> +	uint64_t previous_rate = 0;
>>
>>  	if ((rs->rs_flags & RS_INT_TBL) &&
>>  	    (rs->rs_rate_cnt >= 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 == rs->rs_rlt[i].rate)) {
>>  				rte = &rs->rs_rlt[i];
>>  				matched = 1;
>> +				if (lower_rate != NULL)
>> +					*lower_rate = previous_rate;
>>  				break;
>>  			} else if ((flags & RS_PACING_LT) &&
>>  			    (bytes_per_sec <= rs->rs_rlt[i].rate)) {
>>  				rte = &rs->rs_rlt[i];
>>  				matched = 1;
>> +				if (lower_rate != NULL)
>> +					*lower_rate = previous_rate;
>>  				break;
>>  			}
>> +			previous_rate = 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 != NULL)
>> +					*lower_rate = rs->rs_rlt[i].rate;
>>  				break;
>>  			}
>>  		}
>> @@ -995,10 +1042,10 @@ static struct ifnet *
>>  rt_find_real_interface(struct ifnet *ifp, struct inpcb *inp, int *error)
>>  {
>>  	struct ifnet *tifp;
>> -	struct m_snd_tag *tag;
>> +	struct m_snd_tag *tag, *ntag;
>>  	union if_snd_tag_alloc_params params = {
>>  		.rate_limit.hdr.type = IF_SND_TAG_TYPE_RATE_LIMIT,
>> -		.rate_limit.hdr.flowid = 1,
>> +		.rate_limit.hdr.flowid = inp->inp_flowid,
>>  		.rate_limit.hdr.numa_domain = inp->inp_numa_domain,
>>  		.rate_limit.max_rate = COMMON_RATE,
>>  		.rate_limit.flags = M_NOWAIT,
>> @@ -1017,38 +1064,92 @@ rt_find_real_interface(struct ifnet *ifp, struct
>> inpcb *inp, int *error)
>>  			*error = err;
>>  		return (NULL);
>>  	}
>> -	tifp = tag->ifp;
>> +	ntag = tag;
>> +	while(ntag->ifp->if_next_snd_tag != NULL) {
>> +		ntag = ntag->ifp->if_next_snd_tag(ntag);
>> +	}
>> +	tifp = ntag->ifp;
>>  	m_snd_tag_rele(tag);
>>  	return (tifp);
>>  }
>>
>> +static void
>> +rl_increment_using(const struct tcp_hwrate_limit_table *rte)
>> +{
>> +	struct tcp_hwrate_limit_table *decon_rte;
>> +
>> +	decon_rte = __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 = __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 = __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 == ifp) &&
>> +		    (rs->rs_if_dunit == 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 == 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 = ifp;
>>  	int err;
>>
>>  	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 == ifp) &&
>> -		    (rs->rs_if_dunit == ifp->if_dunit)) {
>> -			/* Ok we found it */
>> -			break;
>> -		}
>> -	}
>> +	rs = find_rs_for_ifp(ifp);
>>  	if ((rs == 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 = ENODEV;
>>  		NET_EPOCH_EXIT(et);
>>  		return (NULL);
>>  	}
>>
>>  	if ((rs == NULL) || (rs->rs_disable != 0)) {
>> -		if (rs->rs_disable && error)
>> +		if (error)
>>  			*error = ENOSPC;
>>  		NET_EPOCH_EXIT(et);
>>  		return (NULL);
>> @@ -1086,6 +1187,10 @@ use_real_interface:
>>  			NET_EPOCH_EXIT(et);
>>  			return (NULL);
>>  		}
>> +		KASSERT((tifp != ifp),
>> +			("Lookup failure ifp:%p inp:%p rt_find_real_interface() returns the
>> same
>> interface tifp:%p?\n",
>> +			 ifp, inp, tifp));
>> +		ifp = tifp;
>>  		goto use_real_interface;
>>  	}
>>  	if (rs->rs_flow_limit &&
>> @@ -1095,9 +1200,9 @@ use_real_interface:
>>  		NET_EPOCH_EXIT(et);
>>  		return (NULL);
>>  	}
>> -	rte = tcp_find_suitable_rate(rs, bytes_per_sec, flags);
>> +	rte = tcp_find_suitable_rate(rs, bytes_per_sec, flags, lower_rate);
>>  	if (rte) {
>> -		err = in_pcbattach_txrtlmt(inp, rs->rs_ifp,
>> +		err = in_pcbattach_txrtlmt(inp, oifp,
>>  		    inp->inp_flowtype,
>>  		    inp->inp_flowid,
>>  		    rte->rate,
>> @@ -1107,6 +1212,11 @@ use_real_interface:
>>  			if (error)
>>  				*error = err;
>>  			rte = NULL;
>> +		} else {
>> +			KASSERT((inp->inp_snd_tag != 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 ifnet
>> *ifp, int link_state)
>>  {
>>  	int error;
>>  	struct tcp_rate_set *rs;
>> +	struct epoch_tracker et;
>>
>>  	if (((ifp->if_capenable & IFCAP_TXRTLMT) == 0) ||
>>  	    (link_state != 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 == ifp) &&
>> -		    (rs->rs_if_dunit == ifp->if_dunit)) {
>> -			/* We already have initialized this guy */
>> -			mtx_unlock(&rs_mtx);
>> -			return;
>> -		}
>> +	rs = 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);
>>  }
>>
>>  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;
>>
>> +	NET_EPOCH_ENTER(et);
>>  	mtx_lock(&rs_mtx);
>> -	CK_LIST_FOREACH_SAFE(rs, &int_rs, next, nrs) {
>> -		if ((rs->rs_ifp == ifp) &&
>> -		    (rs->rs_if_dunit == ifp->if_dunit)) {
>> -			CK_LIST_REMOVE(rs, next);
>> -			rs_number_alive--;
>> -			rs->rs_flags |= RS_IS_DEAD;
>> -			for (i = 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 = NULL;
>> -				}
>> -				rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED;
>> +	rs = find_rs_for_ifp(ifp);
>> +	if (rs) {
>> +		CK_LIST_REMOVE(rs, next);
>> +		rs_number_alive--;
>> +		rs->rs_flags |= RS_IS_DEAD;
>> +		for (i = 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 = NULL;
>>  			}
>> -			if (rs->rs_flows_using == 0)
>> -				rs_defer_destroy(rs);
>> -			break;
>> +			rs->rs_rlt[i].flags = HDWRPACE_IFPDEPARTED;
>>  		}
>> +		if (rs->rs_flows_using == 0)
>> +			rs_defer_destroy(rs);
>>  	}
>>  	mtx_unlock(&rs_mtx);
>> +	NET_EPOCH_EXIT(et);
>>  }
>>
>>  static void
>>  tcp_rl_shutdown(void *arg __unused, int howto __unused)
>>  {
>>  	struct tcp_rate_set *rs, *nrs;
>> +	struct epoch_tracker et;
>>  	int i;
>>
>> +	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);
>>  }
>>
>>  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_rate)
>>  {
>>  	const struct tcp_hwrate_limit_table *rte;
>>  #ifdef KERN_TLS
>> @@ -1233,7 +1348,9 @@ tcp_set_pacing_rate(struct tcpcb *tp, struct ifnet
>> *ifp,
>>  			}
>>  		}
>>  #endif
>> -		rte = rt_setup_rate(tp->t_inpcb, ifp, bytes_per_sec, flags, error);
>> +		rte = rt_setup_rate(tp->t_inpcb, ifp, bytes_per_sec, flags, error,
>> lower_rate);
>> +		if (rte)
>> +			rl_increment_using(rte);
>>  #ifdef KERN_TLS
>>  		if (rte != NULL && tls != NULL && tls->snd_tag != NULL) {
>>  			/*
>> @@ -1253,22 +1370,23 @@ tcp_set_pacing_rate(struct tcpcb *tp, struct
>> ifnet
>> *ifp,
>>  			*error = EINVAL;
>>  		rte = NULL;
>>  	}
>> -	tp->t_pacing_rate = rte->rate;
>> -	*error = 0;
>> +	if (rte != NULL) {
>> +		tp->t_pacing_rate = rte->rate;
>> +		*error = 0;
>> +	}
>>  	return (rte);
>>  }
>>
>>  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_rate)
>>  {
>>  	const struct tcp_hwrate_limit_table *nrte;
>>  	const struct tcp_rate_set *rs;
>>  #ifdef KERN_TLS
>>  	struct ktls_session *tls = NULL;
>>  #endif
>> -	int is_indirect = 0;
>>  	int err;
>>
>>  	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 = 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"
>>
>
>
> --
> Mateusz Guzik <mjguzik gmail.com>
>


-- 
Mateusz Guzik <mjguzik gmail.com>



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAGudoHH=Eu7zRLWezj9mOtXH827A4k4rFsVhRn45aK589mDWvQ>