/*      $NetBSD: pktqueue.c,v 1.22 2023/05/28 08:09:34 andvar Exp $     */

/*-
* Copyright (c) 2014 The NetBSD Foundation, Inc.
* All rights reserved.
*
* This code is derived from software contributed to The NetBSD Foundation
* by Mindaugas Rasiukevicius.
*
* 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 THE NETBSD FOUNDATION, INC. 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 FOUNDATION 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.
*/

/*
* The packet queue (pktqueue) interface is a lockless IP input queue
* which also abstracts and handles network ISR scheduling.  It provides
* a mechanism to enable receiver-side packet steering (RPS).
*/

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: pktqueue.c,v 1.22 2023/05/28 08:09:34 andvar Exp $");

#ifdef _KERNEL_OPT
#include "opt_net_mpsafe.h"
#endif

#include <sys/param.h>
#include <sys/types.h>

#include <sys/atomic.h>
#include <sys/cpu.h>
#include <sys/pcq.h>
#include <sys/intr.h>
#include <sys/mbuf.h>
#include <sys/proc.h>
#include <sys/percpu.h>
#include <sys/xcall.h>
#include <sys/once.h>
#include <sys/queue.h>
#include <sys/rwlock.h>

#include <net/pktqueue.h>
#include <net/rss_config.h>

#include <netinet/in.h>
#include <netinet/ip.h>
#include <netinet/ip6.h>

struct pktqueue {
       /*
        * The lock used for a barrier mechanism.  The barrier counter,
        * as well as the drop counter, are managed atomically though.
        * Ensure this group is in a separate cache line.
        */
       union {
               struct {
                       kmutex_t        pq_lock;
                       volatile u_int  pq_barrier;
               };
               uint8_t  _pad[COHERENCY_UNIT];
       };

       /* The size of the queue, counters and the interrupt handler. */
       u_int           pq_maxlen;
       percpu_t *      pq_counters;
       void *          pq_sih;

       /* The per-CPU queues. */
       struct percpu * pq_pcq; /* struct pcq * */

       /* The linkage on the list of all pktqueues. */
       LIST_ENTRY(pktqueue) pq_list;
};

/* The counters of the packet queue. */
#define PQCNT_ENQUEUE   0
#define PQCNT_DEQUEUE   1
#define PQCNT_DROP      2
#define PQCNT_NCOUNTERS 3

typedef struct {
       uint64_t        count[PQCNT_NCOUNTERS];
} pktq_counters_t;

/* Special marker value used by pktq_barrier() mechanism. */
#define PKTQ_MARKER     ((void *)(~0ULL))

/*
* This is a list of all pktqueues.  This list is used by
* pktq_ifdetach() to issue a barrier on every pktqueue.
*
* The r/w lock is acquired for writing in pktq_create() and
* pktq_destroy(), and for reading in pktq_ifdetach().
*
* This list is not performance critical, and will seldom be
* accessed.
*/
static LIST_HEAD(, pktqueue) pktqueue_list      __read_mostly;
static krwlock_t pktqueue_list_lock             __read_mostly;
static once_t pktqueue_list_init_once           __read_mostly;

static int
pktqueue_list_init(void)
{
       LIST_INIT(&pktqueue_list);
       rw_init(&pktqueue_list_lock);
       return 0;
}

static void
pktq_init_cpu(void *vqp, void *vpq, struct cpu_info *ci)
{
       struct pcq **qp = vqp;
       struct pktqueue *pq = vpq;

       *qp = pcq_create(pq->pq_maxlen, KM_SLEEP);
}

static void
pktq_fini_cpu(void *vqp, void *vpq, struct cpu_info *ci)
{
       struct pcq **qp = vqp, *q = *qp;

       KASSERT(pcq_peek(q) == NULL);
       pcq_destroy(q);
       *qp = NULL;             /* paranoia */
}

static struct pcq *
pktq_pcq(struct pktqueue *pq, struct cpu_info *ci)
{
       struct pcq **qp, *q;

       /*
        * As long as preemption is disabled, the xcall to swap percpu
        * buffers can't complete, so it is safe to read the pointer.
        */
       KASSERT(kpreempt_disabled());

       qp = percpu_getptr_remote(pq->pq_pcq, ci);
       q = *qp;

       return q;
}

pktqueue_t *
pktq_create(size_t maxlen, void (*intrh)(void *), void *sc)
{
       const u_int sflags = SOFTINT_NET | SOFTINT_MPSAFE | SOFTINT_RCPU;
       pktqueue_t *pq;
       percpu_t *pc;
       void *sih;

       RUN_ONCE(&pktqueue_list_init_once, pktqueue_list_init);

       pc = percpu_alloc(sizeof(pktq_counters_t));
       if ((sih = softint_establish(sflags, intrh, sc)) == NULL) {
               percpu_free(pc, sizeof(pktq_counters_t));
               return NULL;
       }

       pq = kmem_zalloc(sizeof(*pq), KM_SLEEP);
       mutex_init(&pq->pq_lock, MUTEX_DEFAULT, IPL_NONE);
       pq->pq_maxlen = maxlen;
       pq->pq_counters = pc;
       pq->pq_sih = sih;
       pq->pq_pcq = percpu_create(sizeof(struct pcq *),
           pktq_init_cpu, pktq_fini_cpu, pq);

       rw_enter(&pktqueue_list_lock, RW_WRITER);
       LIST_INSERT_HEAD(&pktqueue_list, pq, pq_list);
       rw_exit(&pktqueue_list_lock);

       return pq;
}

void
pktq_destroy(pktqueue_t *pq)
{

       KASSERT(pktqueue_list_init_once.o_status == ONCE_DONE);

       rw_enter(&pktqueue_list_lock, RW_WRITER);
       LIST_REMOVE(pq, pq_list);
       rw_exit(&pktqueue_list_lock);

       percpu_free(pq->pq_pcq, sizeof(struct pcq *));
       percpu_free(pq->pq_counters, sizeof(pktq_counters_t));
       softint_disestablish(pq->pq_sih);
       mutex_destroy(&pq->pq_lock);
       kmem_free(pq, sizeof(*pq));
}

/*
* - pktq_inc_counter: increment the counter given an ID.
* - pktq_collect_counts: handler to sum up the counts from each CPU.
* - pktq_getcount: return the effective count given an ID.
*/

static inline void
pktq_inc_count(pktqueue_t *pq, u_int i)
{
       percpu_t *pc = pq->pq_counters;
       pktq_counters_t *c;

       c = percpu_getref(pc);
       c->count[i]++;
       percpu_putref(pc);
}

static void
pktq_collect_counts(void *mem, void *arg, struct cpu_info *ci)
{
       const pktq_counters_t *c = mem;
       pktq_counters_t *sum = arg;

       int s = splnet();

       for (u_int i = 0; i < PQCNT_NCOUNTERS; i++) {
               sum->count[i] += c->count[i];
       }

       splx(s);
}

static uint64_t
pktq_get_count(pktqueue_t *pq, pktq_count_t c)
{
       pktq_counters_t sum;

       if (c != PKTQ_MAXLEN) {
               memset(&sum, 0, sizeof(sum));
               percpu_foreach_xcall(pq->pq_counters,
                   XC_HIGHPRI_IPL(IPL_SOFTNET), pktq_collect_counts, &sum);
       }
       switch (c) {
       case PKTQ_NITEMS:
               return sum.count[PQCNT_ENQUEUE] - sum.count[PQCNT_DEQUEUE];
       case PKTQ_DROPS:
               return sum.count[PQCNT_DROP];
       case PKTQ_MAXLEN:
               return pq->pq_maxlen;
       }
       return 0;
}

uint32_t
pktq_rps_hash(const pktq_rps_hash_func_t *funcp, const struct mbuf *m)
{
       pktq_rps_hash_func_t func = atomic_load_relaxed(funcp);

       KASSERT(func != NULL);

       return (*func)(m);
}

static uint32_t
pktq_rps_hash_zero(const struct mbuf *m __unused)
{

       return 0;
}

static uint32_t
pktq_rps_hash_curcpu(const struct mbuf *m __unused)
{

       return cpu_index(curcpu());
}

static uint32_t
pktq_rps_hash_toeplitz(const struct mbuf *m)
{
       struct ip *ip;
       /*
        * Disable UDP port - IP fragments aren't currently being handled
        * and so we end up with a mix of 2-tuple and 4-tuple
        * traffic.
        */
       const u_int flag = RSS_TOEPLITZ_USE_TCP_PORT;

       /* glance IP version */
       if ((m->m_flags & M_PKTHDR) == 0)
               return 0;

       ip = mtod(m, struct ip *);
       if (ip->ip_v == IPVERSION) {
               if (__predict_false(m->m_len < sizeof(struct ip)))
                       return 0;
               return rss_toeplitz_hash_from_mbuf_ipv4(m, flag);
       } else if (ip->ip_v == 6) {
               if (__predict_false(m->m_len < sizeof(struct ip6_hdr)))
                       return 0;
               return rss_toeplitz_hash_from_mbuf_ipv6(m, flag);
       }

       return 0;
}

/*
* toeplitz without curcpu.
* Generally, this has better performance than toeplitz.
*/
static uint32_t
pktq_rps_hash_toeplitz_othercpus(const struct mbuf *m)
{
       uint32_t hash;

       if (ncpu == 1)
               return 0;

       hash = pktq_rps_hash_toeplitz(m);
       hash %= ncpu - 1;
       if (hash >= cpu_index(curcpu()))
               return hash + 1;
       else
               return hash;
}

static struct pktq_rps_hash_table {
       const char* prh_type;
       pktq_rps_hash_func_t prh_func;
} const pktq_rps_hash_tab[] = {
       { "zero", pktq_rps_hash_zero },
       { "curcpu", pktq_rps_hash_curcpu },
       { "toeplitz", pktq_rps_hash_toeplitz },
       { "toeplitz-othercpus", pktq_rps_hash_toeplitz_othercpus },
};
const pktq_rps_hash_func_t pktq_rps_hash_default =
#ifdef NET_MPSAFE
       pktq_rps_hash_curcpu;
#else
       pktq_rps_hash_zero;
#endif

static const char *
pktq_get_rps_hash_type(pktq_rps_hash_func_t func)
{

       for (int i = 0; i < __arraycount(pktq_rps_hash_tab); i++) {
               if (func == pktq_rps_hash_tab[i].prh_func) {
                       return pktq_rps_hash_tab[i].prh_type;
               }
       }

       return NULL;
}

static int
pktq_set_rps_hash_type(pktq_rps_hash_func_t *func, const char *type)
{

       if (strcmp(type, pktq_get_rps_hash_type(*func)) == 0)
               return 0;

       for (int i = 0; i < __arraycount(pktq_rps_hash_tab); i++) {
               if (strcmp(type, pktq_rps_hash_tab[i].prh_type) == 0) {
                       atomic_store_relaxed(func, pktq_rps_hash_tab[i].prh_func);
                       return 0;
               }
       }

       return ENOENT;
}

int
sysctl_pktq_rps_hash_handler(SYSCTLFN_ARGS)
{
       struct sysctlnode node;
       pktq_rps_hash_func_t *func;
       int error;
       char type[PKTQ_RPS_HASH_NAME_LEN];

       node = *rnode;
       func = node.sysctl_data;

       strlcpy(type, pktq_get_rps_hash_type(*func), PKTQ_RPS_HASH_NAME_LEN);

       node.sysctl_data = &type;
       node.sysctl_size = sizeof(type);
       error = sysctl_lookup(SYSCTLFN_CALL(&node));
       if (error || newp == NULL)
               return error;

       error = pktq_set_rps_hash_type(func, type);

       return error;
}

/*
* pktq_enqueue: inject the packet into the end of the queue.
*
* => Must be called from the interrupt or with the preemption disabled.
* => Consumes the packet and returns true on success.
* => Returns false on failure; caller is responsible to free the packet.
*/
bool
pktq_enqueue(pktqueue_t *pq, struct mbuf *m, const u_int hash __unused)
{
#if defined(_RUMPKERNEL) || defined(_RUMP_NATIVE_ABI)
       struct cpu_info *ci = curcpu();
#else
       struct cpu_info *ci = cpu_lookup(hash % ncpu);
#endif

       KASSERT(kpreempt_disabled());

       if (__predict_false(!pcq_put(pktq_pcq(pq, ci), m))) {
               pktq_inc_count(pq, PQCNT_DROP);
               return false;
       }
       softint_schedule_cpu(pq->pq_sih, ci);
       pktq_inc_count(pq, PQCNT_ENQUEUE);
       return true;
}

/*
* pktq_dequeue: take a packet from the queue.
*
* => Must be called with preemption disabled.
* => Must ensure there are not concurrent dequeue calls.
*/
struct mbuf *
pktq_dequeue(pktqueue_t *pq)
{
       struct cpu_info *ci = curcpu();
       struct mbuf *m;

       KASSERT(kpreempt_disabled());

       m = pcq_get(pktq_pcq(pq, ci));
       if (__predict_false(m == PKTQ_MARKER)) {
               /* Note the marker entry. */
               atomic_inc_uint(&pq->pq_barrier);

               /* Get the next queue entry. */
               m = pcq_get(pktq_pcq(pq, ci));

               /*
                * There can only be one barrier operation pending
                * on a pktqueue at any given time, so we can assert
                * that the next item is not a marker.
                */
               KASSERT(m != PKTQ_MARKER);
       }
       if (__predict_true(m != NULL)) {
               pktq_inc_count(pq, PQCNT_DEQUEUE);
       }
       return m;
}

/*
* pktq_barrier: waits for a grace period when all packets enqueued at
* the moment of calling this routine will be processed.  This is used
* to ensure that e.g. packets referencing some interface were drained.
*/
void
pktq_barrier(pktqueue_t *pq)
{
       CPU_INFO_ITERATOR cii;
       struct cpu_info *ci;
       u_int pending = 0;

       mutex_enter(&pq->pq_lock);
       KASSERT(pq->pq_barrier == 0);

       for (CPU_INFO_FOREACH(cii, ci)) {
               struct pcq *q;

               kpreempt_disable();
               q = pktq_pcq(pq, ci);
               kpreempt_enable();

               /* If the queue is empty - nothing to do. */
               if (pcq_peek(q) == NULL) {
                       continue;
               }
               /* Otherwise, put the marker and entry. */
               while (!pcq_put(q, PKTQ_MARKER)) {
                       kpause("pktqsync", false, 1, NULL);
               }
               kpreempt_disable();
               softint_schedule_cpu(pq->pq_sih, ci);
               kpreempt_enable();
               pending++;
       }

       /* Wait for each queue to process the markers. */
       while (pq->pq_barrier != pending) {
               kpause("pktqsync", false, 1, NULL);
       }
       pq->pq_barrier = 0;
       mutex_exit(&pq->pq_lock);
}

/*
* pktq_ifdetach: issue a barrier on all pktqueues when a network
* interface is detached.
*/
void
pktq_ifdetach(void)
{
       pktqueue_t *pq;

       /* Just in case no pktqueues have been created yet... */
       RUN_ONCE(&pktqueue_list_init_once, pktqueue_list_init);

       rw_enter(&pktqueue_list_lock, RW_READER);
       LIST_FOREACH(pq, &pktqueue_list, pq_list) {
               pktq_barrier(pq);
       }
       rw_exit(&pktqueue_list_lock);
}

/*
* pktq_flush: free mbufs in all queues.
*
* => The caller must ensure there are no concurrent writers or flush calls.
*/
void
pktq_flush(pktqueue_t *pq)
{
       CPU_INFO_ITERATOR cii;
       struct cpu_info *ci;
       struct mbuf *m, *m0 = NULL;

       ASSERT_SLEEPABLE();

       /*
        * Run a dummy softint at IPL_SOFTNET on all CPUs to ensure that any
        * already running handler for this pktqueue is no longer running.
        */
       xc_barrier(XC_HIGHPRI_IPL(IPL_SOFTNET));

       /*
        * Acquire the barrier lock.  While the caller ensures that
        * no explicit pktq_barrier() calls will be issued, this holds
        * off any implicit pktq_barrier() calls that would happen
        * as the result of pktq_ifdetach().
        */
       mutex_enter(&pq->pq_lock);

       for (CPU_INFO_FOREACH(cii, ci)) {
               struct pcq *q;

               kpreempt_disable();
               q = pktq_pcq(pq, ci);
               kpreempt_enable();

               /*
                * Pull the packets off the pcq and chain them into
                * a list to be freed later.
                */
               while ((m = pcq_get(q)) != NULL) {
                       pktq_inc_count(pq, PQCNT_DEQUEUE);
                       m->m_nextpkt = m0;
                       m0 = m;
               }
       }

       mutex_exit(&pq->pq_lock);

       /* Free the packets now that the critical section is over. */
       while ((m = m0) != NULL) {
               m0 = m->m_nextpkt;
               m_freem(m);
       }
}

static void
pktq_set_maxlen_cpu(void *vpq, void *vqs)
{
       struct pktqueue *pq = vpq;
       struct pcq **qp, *q, **qs = vqs;
       unsigned i = cpu_index(curcpu());
       int s;

       s = splnet();
       qp = percpu_getref(pq->pq_pcq);
       q = *qp;
       *qp = qs[i];
       qs[i] = q;
       percpu_putref(pq->pq_pcq);
       splx(s);
}

/*
* pktq_set_maxlen: create per-CPU queues using a new size and replace
* the existing queues without losing any packets.
*
* XXX ncpu must remain stable throughout.
*/
int
pktq_set_maxlen(pktqueue_t *pq, size_t maxlen)
{
       const u_int slotbytes = ncpu * sizeof(pcq_t *);
       pcq_t **qs;

       if (!maxlen || maxlen > PCQ_MAXLEN)
               return EINVAL;
       if (pq->pq_maxlen == maxlen)
               return 0;

       /* First, allocate the new queues. */
       qs = kmem_zalloc(slotbytes, KM_SLEEP);
       for (u_int i = 0; i < ncpu; i++) {
               qs[i] = pcq_create(maxlen, KM_SLEEP);
       }

       /*
        * Issue an xcall to replace the queue pointers on each CPU.
        * This implies all the necessary memory barriers.
        */
       mutex_enter(&pq->pq_lock);
       xc_wait(xc_broadcast(XC_HIGHPRI, pktq_set_maxlen_cpu, pq, qs));
       pq->pq_maxlen = maxlen;
       mutex_exit(&pq->pq_lock);

       /*
        * At this point, the new packets are flowing into the new
        * queues.  However, the old queues may have some packets
        * present which are no longer being processed.  We are going
        * to re-enqueue them.  This may change the order of packet
        * arrival, but it is not considered an issue.
        *
        * There may be in-flight interrupts calling pktq_dequeue()
        * which reference the old queues.  Issue a barrier to ensure
        * that we are going to be the only pcq_get() callers on the
        * old queues.
        */
       pktq_barrier(pq);

       for (u_int i = 0; i < ncpu; i++) {
               struct pcq *q;
               struct mbuf *m;

               kpreempt_disable();
               q = pktq_pcq(pq, cpu_lookup(i));
               kpreempt_enable();

               while ((m = pcq_get(qs[i])) != NULL) {
                       while (!pcq_put(q, m)) {
                               kpause("pktqrenq", false, 1, NULL);
                       }
               }
               pcq_destroy(qs[i]);
       }

       /* Well, that was fun. */
       kmem_free(qs, slotbytes);
       return 0;
}

static int
sysctl_pktq_maxlen(SYSCTLFN_ARGS)
{
       struct sysctlnode node = *rnode;
       pktqueue_t * const pq = node.sysctl_data;
       u_int nmaxlen = pktq_get_count(pq, PKTQ_MAXLEN);
       int error;

       node.sysctl_data = &nmaxlen;
       error = sysctl_lookup(SYSCTLFN_CALL(&node));
       if (error || newp == NULL)
               return error;
       return pktq_set_maxlen(pq, nmaxlen);
}

static int
sysctl_pktq_count(SYSCTLFN_ARGS, u_int count_id)
{
       struct sysctlnode node = *rnode;
       pktqueue_t * const pq = node.sysctl_data;
       uint64_t count = pktq_get_count(pq, count_id);

       node.sysctl_data = &count;
       return sysctl_lookup(SYSCTLFN_CALL(&node));
}

static int
sysctl_pktq_nitems(SYSCTLFN_ARGS)
{
       return sysctl_pktq_count(SYSCTLFN_CALL(rnode), PKTQ_NITEMS);
}

static int
sysctl_pktq_drops(SYSCTLFN_ARGS)
{
       return sysctl_pktq_count(SYSCTLFN_CALL(rnode), PKTQ_DROPS);
}

/*
* pktqueue_sysctl_setup: set up the sysctl nodes for a pktqueue
* using standardized names at the specified parent node and
* node ID (or CTL_CREATE).
*/
void
pktq_sysctl_setup(pktqueue_t * const pq, struct sysctllog ** const clog,
                 const struct sysctlnode * const parent_node, const int qid)
{
       const struct sysctlnode *rnode = parent_node, *cnode;

       KASSERT(pq != NULL);
       KASSERT(parent_node != NULL);
       KASSERT(qid == CTL_CREATE || qid >= 0);

       /* Create the "ifq" node below the parent node. */
       sysctl_createv(clog, 0, &rnode, &cnode,
                      CTLFLAG_PERMANENT,
                      CTLTYPE_NODE, "ifq",
                      SYSCTL_DESCR("Protocol input queue controls"),
                      NULL, 0, NULL, 0,
                      qid, CTL_EOL);

       /* Now create the standard child nodes below "ifq". */
       rnode = cnode;

       sysctl_createv(clog, 0, &rnode, &cnode,
                      CTLFLAG_PERMANENT,
                      CTLTYPE_QUAD, "len",
                      SYSCTL_DESCR("Current input queue length"),
                      sysctl_pktq_nitems, 0, (void *)pq, 0,
                      IFQCTL_LEN, CTL_EOL);
       sysctl_createv(clog, 0, &rnode, &cnode,
                      CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
                      CTLTYPE_INT, "maxlen",
                      SYSCTL_DESCR("Maximum allowed input queue length"),
                      sysctl_pktq_maxlen, 0, (void *)pq, 0,
                      IFQCTL_MAXLEN, CTL_EOL);
       sysctl_createv(clog, 0, &rnode, &cnode,
                      CTLFLAG_PERMANENT,
                      CTLTYPE_QUAD, "drops",
                      SYSCTL_DESCR("Packets dropped due to full input queue"),
                      sysctl_pktq_drops, 0, (void *)pq, 0,
                      IFQCTL_DROPS, CTL_EOL);
}

You are viewing proxied material from ftp.icm.edu.pl. The copyright of proxied material belongs to its original authors. Any comments or complaints in relation to proxied material should be directed to the original authors of the content concerned. Please see the disclaimer for more details.