#include "socknal.h"
+nal_t ksocknal_api;
+ksock_nal_data_t ksocknal_data;
ptl_handle_ni_t ksocknal_ni;
-static nal_t ksocknal_api;
-#if (LINUX_VERSION_CODE >= KERNEL_VERSION(2,5,0))
-ksock_nal_data_t ksocknal_data;
-#else
-static ksock_nal_data_t ksocknal_data;
-#endif
+ksock_tunables_t ksocknal_tunables;
kpr_nal_interface_t ksocknal_router_interface = {
kprni_nalid: SOCKNAL,
kprni_notify: ksocknal_notify,
};
+#ifdef CONFIG_SYSCTL
#define SOCKNAL_SYSCTL 200
#define SOCKNAL_SYSCTL_TIMEOUT 1
static ctl_table ksocknal_ctl_table[] = {
{SOCKNAL_SYSCTL_TIMEOUT, "timeout",
- &ksocknal_data.ksnd_io_timeout, sizeof (int),
+ &ksocknal_tunables.ksnd_io_timeout, sizeof (int),
0644, NULL, &proc_dointvec},
{SOCKNAL_SYSCTL_EAGER_ACK, "eager_ack",
- &ksocknal_data.ksnd_eager_ack, sizeof (int),
+ &ksocknal_tunables.ksnd_eager_ack, sizeof (int),
0644, NULL, &proc_dointvec},
#if SOCKNAL_ZC
- {SOCKNAL_SYSCTL_EAGER_ACK, "zero_copy",
- &ksocknal_data.ksnd_zc_min_frag, sizeof (int),
+ {SOCKNAL_SYSCTL_ZERO_COPY, "zero_copy",
+ &ksocknal_tunables.ksnd_zc_min_frag, sizeof (int),
0644, NULL, &proc_dointvec},
#endif
{SOCKNAL_SYSCTL_TYPED, "typed",
- &ksocknal_data.ksnd_typed_conns, sizeof (int),
+ &ksocknal_tunables.ksnd_typed_conns, sizeof (int),
0644, NULL, &proc_dointvec},
{SOCKNAL_SYSCTL_MIN_BULK, "min_bulk",
- &ksocknal_data.ksnd_min_bulk, sizeof (int),
+ &ksocknal_tunables.ksnd_min_bulk, sizeof (int),
0644, NULL, &proc_dointvec},
{ 0 }
};
{SOCKNAL_SYSCTL, "socknal", NULL, 0, 0555, ksocknal_ctl_table},
{ 0 }
};
+#endif
int
ksocknal_api_forward(nal_t *nal, int id, void *args, size_t args_len,
return PTL_OK;
}
-int
-ksocknal_api_shutdown(nal_t *nal, int ni)
-{
- CDEBUG (D_NET, "closing all connections\n");
-
- ksocknal_del_route (PTL_NID_ANY, 0, 0, 0);
- ksocknal_close_matching_conns (PTL_NID_ANY, 0);
- return PTL_OK;
-}
-
-void
-ksocknal_api_yield(nal_t *nal)
-{
- our_cond_resched();
- return;
-}
-
void
ksocknal_api_lock(nal_t *nal, unsigned long *flags)
{
nal_cb->cb_sti(nal_cb,flags);
}
-nal_t *
-ksocknal_init(int interface, ptl_pt_index_t ptl_size,
- ptl_ac_index_t ac_size, ptl_pid_t requested_pid)
+int
+ksocknal_api_yield(nal_t *nal, unsigned long *flags, int milliseconds)
{
- CDEBUG(D_NET, "calling lib_init with nid "LPX64"\n", (ptl_nid_t)0);
- lib_init(&ksocknal_lib, (ptl_nid_t)0, 0, 10, ptl_size, ac_size);
- return (&ksocknal_api);
+ /* NB called holding statelock */
+ wait_queue_t wait;
+ unsigned long now = jiffies;
+
+ CDEBUG (D_NET, "yield\n");
+
+ if (milliseconds == 0) {
+ our_cond_resched();
+ return 0;
+ }
+
+ init_waitqueue_entry(&wait, current);
+ set_current_state (TASK_INTERRUPTIBLE);
+ add_wait_queue (&ksocknal_data.ksnd_yield_waitq, &wait);
+
+ ksocknal_api_unlock(nal, flags);
+
+ if (milliseconds < 0)
+ schedule ();
+ else
+ schedule_timeout((milliseconds * HZ) / 1000);
+
+ ksocknal_api_lock(nal, flags);
+
+ remove_wait_queue (&ksocknal_data.ksnd_yield_waitq, &wait);
+
+ if (milliseconds > 0) {
+ milliseconds -= ((jiffies - now) * 1000) / HZ;
+ if (milliseconds < 0)
+ milliseconds = 0;
+ }
+
+ return (milliseconds);
}
-/*
- * EXTRA functions follow
- */
-
int
ksocknal_set_mynid(ptl_nid_t nid)
{
/* FIXME: Find a better method of setting IRQ affinity...
*/
- call_usermodehelper (argv[0], argv, envp);
+ USERMODEHELPER(argv[0], argv, envp);
#endif
}
ksock_route_t *
ksocknal_create_route (__u32 ipaddr, int port, int buffer_size,
- int nonagel, int irq_affinity, int eager)
+ int irq_affinity, int eager)
{
ksock_route_t *route;
route->ksnr_port = port;
route->ksnr_buffer_size = buffer_size;
route->ksnr_irq_affinity = irq_affinity;
- route->ksnr_nonagel = nonagel;
route->ksnr_eager = eager;
route->ksnr_connecting = 0;
route->ksnr_connected = 0;
INIT_LIST_HEAD (&peer->ksnp_routes);
INIT_LIST_HEAD (&peer->ksnp_tx_queue);
- /* Can't unload while peers exist; ensures all I/O has terminated
- * before unload attempts */
- PORTAL_MODULE_USE;
atomic_inc (&ksocknal_data.ksnd_npeers);
return (peer);
}
* that _all_ state to do with this peer has been cleaned up when
* its refcount drops to zero. */
atomic_dec (&ksocknal_data.ksnd_npeers);
- PORTAL_MODULE_UNUSE;
}
void
int
ksocknal_add_route (ptl_nid_t nid, __u32 ipaddr, int port, int bufnob,
- int nonagle, int bind_irq, int share, int eager)
+ int bind_irq, int share, int eager)
{
unsigned long flags;
ksock_peer_t *peer;
return (-ENOMEM);
route = ksocknal_create_route (ipaddr, port, bufnob,
- nonagle, bind_irq, eager);
+ bind_irq, eager);
if (route == NULL) {
ksocknal_put_peer (peer);
return (-ENOMEM);
ksocknal_get_peer_addr (conn);
+ CWARN("New conn nid:"LPX64" ip:%08x/%d incarnation:"LPX64"\n",
+ nid, conn->ksnc_ipaddr, conn->ksnc_port, incarnation);
+
irq = ksocknal_conn_irq (conn);
write_lock_irqsave (&ksocknal_data.ksnd_global_lock, flags);
peer->ksnp_last_alive = jiffies;
peer->ksnp_error = 0;
+ /* Set the deadline for the outgoing HELLO to drain */
+ conn->ksnc_tx_deadline = jiffies +
+ ksocknal_tunables.ksnd_io_timeout * HZ;
+
list_add (&conn->ksnc_list, &peer->ksnp_conns);
atomic_inc (&conn->ksnc_refcount);
write_unlock_irqrestore (&ksocknal_data.ksnd_global_lock, flags);
if (rc != 0)
- CERROR ("Closed %d stale conns to "LPX64"\n", rc, nid);
+ CERROR ("Closed %d stale conns to nid "LPX64" ip %d.%d.%d.%d\n",
+ rc, conn->ksnc_peer->ksnp_nid,
+ HIPQUAD(conn->ksnc_ipaddr));
if (bind_irq) /* irq binding required */
ksocknal_bind_irq (irq);
ksocknal_data_ready (sock->sk, 0);
ksocknal_write_space (sock->sk);
- CDEBUG(D_IOCTL, "conn [%p] registered for nid "LPX64"\n",
- conn, conn->ksnc_peer->ksnp_nid);
+ CDEBUG(D_IOCTL, "conn [%p] registered for nid "LPX64" ip %d.%d.%d.%d\n",
+ conn, conn->ksnc_peer->ksnp_nid, HIPQUAD(conn->ksnc_ipaddr));
ksocknal_put_conn (conn);
return (0);
ksocknal_close_conn_locked (ksock_conn_t *conn, int error)
{
/* This just does the immmediate housekeeping, and queues the
- * connection for the reaper to terminate.
+ * connection for the reaper to terminate.
* Caller holds ksnd_global_lock exclusively in irq context */
ksock_peer_t *peer = conn->ksnc_peer;
ksock_route_t *route;
/* complete current receive if any */
switch (conn->ksnc_rx_state) {
case SOCKNAL_RX_BODY:
-#if 0
- lib_finalize (&ksocknal_lib, NULL, conn->ksnc_cookie);
-#else
- CERROR ("Refusing to complete a partial receive from "
- LPX64", ip %08x\n", conn->ksnc_peer->ksnp_nid,
- conn->ksnc_ipaddr);
- CERROR ("This may hang communications and "
- "prevent modules from unloading\n");
-#endif
+ CERROR("Completing partial receive from "LPX64
+ ", ip %d.%d.%d.%d:%d, with error\n",
+ conn->ksnc_peer->ksnp_nid,
+ HIPQUAD(conn->ksnc_ipaddr), conn->ksnc_port);
+ lib_finalize (&ksocknal_lib, NULL, conn->ksnc_cookie, PTL_FAIL);
break;
case SOCKNAL_RX_BODY_FWD:
ksocknal_fmb_callback (conn->ksnc_cookie, -ECONNABORTED);
if (conn->ksnc_incarnation == incarnation)
continue;
+
+ CWARN("Closing stale conn nid:"LPX64" ip:%08x/%d "
+ "incarnation:"LPX64"("LPX64")\n",
+ peer->ksnp_nid, conn->ksnc_ipaddr, conn->ksnc_port,
+ conn->ksnc_incarnation, incarnation);
count++;
ksocknal_close_conn_locked (conn, -ESTALE);
pcfg->pcfg_count = route->ksnr_conn_count;
pcfg->pcfg_size = route->ksnr_buffer_size;
pcfg->pcfg_wait = route->ksnr_sharecount;
- pcfg->pcfg_flags = (route->ksnr_nonagel ? 1 : 0) |
- (route->ksnr_irq_affinity ? 2 : 0) |
+ pcfg->pcfg_flags = (route->ksnr_irq_affinity ? 2 : 0) |
(route->ksnr_eager ? 4 : 0);
ksocknal_put_route (route);
}
case NAL_CMD_ADD_AUTOCONN: {
rc = ksocknal_add_route (pcfg->pcfg_nid, pcfg->pcfg_id,
pcfg->pcfg_misc, pcfg->pcfg_size,
- (pcfg->pcfg_flags & 0x01) != 0,
(pcfg->pcfg_flags & 0x02) != 0,
(pcfg->pcfg_flags & 0x04) != 0,
(pcfg->pcfg_flags & 0x08) != 0);
}
void
-ksocknal_free_buffers (void)
+ksocknal_free_fmbs (ksock_fmb_pool_t *p)
{
- if (ksocknal_data.ksnd_fmbs != NULL) {
- ksock_fmb_t *fmb = (ksock_fmb_t *)ksocknal_data.ksnd_fmbs;
- int i;
- int j;
-
- for (i = 0;
- i < (SOCKNAL_SMALL_FWD_NMSGS + SOCKNAL_LARGE_FWD_NMSGS);
- i++, fmb++)
- for (j = 0; j < fmb->fmb_npages; j++)
- if (fmb->fmb_pages[j] != NULL)
- __free_page (fmb->fmb_pages[j]);
-
- PORTAL_FREE (ksocknal_data.ksnd_fmbs,
- sizeof (ksock_fmb_t) * (SOCKNAL_SMALL_FWD_NMSGS +
- SOCKNAL_LARGE_FWD_NMSGS));
+ int npages = p->fmp_buff_pages;
+ ksock_fmb_t *fmb;
+ int i;
+
+ LASSERT (list_empty(&p->fmp_blocked_conns));
+ LASSERT (p->fmp_nactive_fmbs == 0);
+
+ while (!list_empty(&p->fmp_idle_fmbs)) {
+
+ fmb = list_entry(p->fmp_idle_fmbs.next,
+ ksock_fmb_t, fmb_list);
+
+ for (i = 0; i < npages; i++)
+ if (fmb->fmb_kiov[i].kiov_page != NULL)
+ __free_page(fmb->fmb_kiov[i].kiov_page);
+
+ list_del(&fmb->fmb_list);
+ PORTAL_FREE(fmb, offsetof(ksock_fmb_t, fmb_kiov[npages]));
}
+}
+
+void
+ksocknal_free_buffers (void)
+{
+ ksocknal_free_fmbs(&ksocknal_data.ksnd_small_fmp);
+ ksocknal_free_fmbs(&ksocknal_data.ksnd_large_fmp);
- LASSERT (ksocknal_data.ksnd_active_ltxs == 0);
- if (ksocknal_data.ksnd_ltxs != NULL)
- PORTAL_FREE (ksocknal_data.ksnd_ltxs,
- sizeof (ksock_ltx_t) * (SOCKNAL_NLTXS +
- SOCKNAL_NNBLK_LTXS));
+ LASSERT (atomic_read(&ksocknal_data.ksnd_nactive_ltxs) == 0);
if (ksocknal_data.ksnd_schedulers != NULL)
PORTAL_FREE (ksocknal_data.ksnd_schedulers,
}
void
-ksocknal_module_fini (void)
+ksocknal_api_shutdown (nal_t *nal)
{
int i;
+ if (nal->nal_refct != 0) {
+ /* This module got the first ref */
+ PORTAL_MODULE_UNUSE;
+ return;
+ }
+
CDEBUG(D_MALLOC, "before NAL cleanup: kmem %d\n",
atomic_read (&portal_kmemory));
+ LASSERT(nal == &ksocknal_api);
+
switch (ksocknal_data.ksnd_init) {
default:
LASSERT (0);
case SOCKNAL_INIT_ALL:
-#if CONFIG_SYSCTL
- if (ksocknal_data.ksnd_sysctl != NULL)
- unregister_sysctl_table (ksocknal_data.ksnd_sysctl);
-#endif
- kportal_nal_unregister(SOCKNAL);
- PORTAL_SYMBOL_UNREGISTER (ksocknal_ni);
+ libcfs_nal_cmd_unregister(SOCKNAL);
+
+ ksocknal_data.ksnd_init = SOCKNAL_INIT_LIB;
/* fall through */
- case SOCKNAL_INIT_PTL:
- PtlNIFini(ksocknal_ni);
+ case SOCKNAL_INIT_LIB:
+ /* No more calls to ksocknal_cmd() to create new
+ * autoroutes/connections since we're being unloaded. */
+
+ /* Delete all autoroute entries */
+ ksocknal_del_route(PTL_NID_ANY, 0, 0, 0);
+
+ /* Delete all connections */
+ ksocknal_close_matching_conns (PTL_NID_ANY, 0);
+
+ /* Wait for all peer state to clean up */
+ i = 2;
+ while (atomic_read (&ksocknal_data.ksnd_npeers) != 0) {
+ i++;
+ CDEBUG(((i & (-i)) == i) ? D_WARNING : D_NET, /* power of 2? */
+ "waiting for %d peers to disconnect\n",
+ atomic_read (&ksocknal_data.ksnd_npeers));
+ set_current_state (TASK_UNINTERRUPTIBLE);
+ schedule_timeout (HZ);
+ }
+
+ /* Tell lib we've stopped calling into her. */
lib_fini(&ksocknal_lib);
+
+ ksocknal_data.ksnd_init = SOCKNAL_INIT_DATA;
/* fall through */
case SOCKNAL_INIT_DATA:
for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++) {
LASSERT (list_empty (&ksocknal_data.ksnd_peers[i]));
}
+ LASSERT (list_empty (&ksocknal_data.ksnd_enomem_conns));
LASSERT (list_empty (&ksocknal_data.ksnd_zombie_conns));
LASSERT (list_empty (&ksocknal_data.ksnd_autoconnectd_routes));
LASSERT (list_empty (&ksocknal_data.ksnd_small_fmp.fmp_blocked_conns));
kpr_deregister (&ksocknal_data.ksnd_router);
ksocknal_free_buffers();
+
+ ksocknal_data.ksnd_init = SOCKNAL_INIT_NOTHING;
/* fall through */
case SOCKNAL_INIT_NOTHING:
}
-void __init
+void
ksocknal_init_incarnation (void)
{
struct timeval tv;
(((__u64)tv.tv_sec) * 1000000) + tv.tv_usec;
}
-int __init
-ksocknal_module_init (void)
+int
+ksocknal_api_startup (nal_t *nal, ptl_pid_t requested_pid,
+ ptl_ni_limits_t *requested_limits,
+ ptl_ni_limits_t *actual_limits)
{
- int pkmem = atomic_read(&portal_kmemory);
- int rc;
- int i;
- int j;
+ ptl_process_id_t process_id;
+ int pkmem = atomic_read(&portal_kmemory);
+ int rc;
+ int i;
+ int j;
- /* packet descriptor must fit in a router descriptor's scratchpad */
- LASSERT(sizeof (ksock_tx_t) <= sizeof (kprfd_scratch_t));
- /* the following must be sizeof(int) for proc_dointvec() */
- LASSERT(sizeof (ksocknal_data.ksnd_io_timeout) == sizeof (int));
- LASSERT(sizeof (ksocknal_data.ksnd_eager_ack) == sizeof (int));
- /* check ksnr_connected/connecting field large enough */
- LASSERT(SOCKNAL_CONN_NTYPES <= 4);
-
- LASSERT (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
+ LASSERT (nal == &ksocknal_api);
- ksocknal_api.forward = ksocknal_api_forward;
- ksocknal_api.shutdown = ksocknal_api_shutdown;
- ksocknal_api.yield = ksocknal_api_yield;
- ksocknal_api.validate = NULL; /* our api validate is a NOOP */
- ksocknal_api.lock = ksocknal_api_lock;
- ksocknal_api.unlock = ksocknal_api_unlock;
- ksocknal_api.nal_data = &ksocknal_data;
+ if (nal->nal_refct != 0) {
+ if (actual_limits != NULL)
+ *actual_limits = ksocknal_lib.ni.actual_limits;
+ /* This module got the first ref */
+ PORTAL_MODULE_USE;
+ return (PTL_OK);
+ }
- ksocknal_lib.nal_data = &ksocknal_data;
+ LASSERT (ksocknal_data.ksnd_init == SOCKNAL_INIT_NOTHING);
memset (&ksocknal_data, 0, sizeof (ksocknal_data)); /* zero pointers */
- ksocknal_data.ksnd_io_timeout = SOCKNAL_IO_TIMEOUT;
- ksocknal_data.ksnd_eager_ack = SOCKNAL_EAGER_ACK;
- ksocknal_data.ksnd_typed_conns = SOCKNAL_TYPED_CONNS;
- ksocknal_data.ksnd_min_bulk = SOCKNAL_MIN_BULK;
-#if SOCKNAL_ZC
- ksocknal_data.ksnd_zc_min_frag = SOCKNAL_ZC_MIN_FRAG;
-#endif
ksocknal_init_incarnation();
ksocknal_data.ksnd_peer_hash_size = SOCKNAL_PEER_HASH_SIZE;
PORTAL_ALLOC (ksocknal_data.ksnd_peers,
sizeof (struct list_head) * ksocknal_data.ksnd_peer_hash_size);
if (ksocknal_data.ksnd_peers == NULL)
- RETURN (-ENOMEM);
+ return (-ENOMEM);
for (i = 0; i < ksocknal_data.ksnd_peer_hash_size; i++)
INIT_LIST_HEAD(&ksocknal_data.ksnd_peers[i]);
ksocknal_data.ksnd_nal_cb = &ksocknal_lib;
spin_lock_init (&ksocknal_data.ksnd_nal_cb_lock);
-
+ init_waitqueue_head(&ksocknal_data.ksnd_yield_waitq);
+
spin_lock_init(&ksocknal_data.ksnd_small_fmp.fmp_lock);
INIT_LIST_HEAD(&ksocknal_data.ksnd_small_fmp.fmp_idle_fmbs);
INIT_LIST_HEAD(&ksocknal_data.ksnd_small_fmp.fmp_blocked_conns);
+ ksocknal_data.ksnd_small_fmp.fmp_buff_pages = SOCKNAL_SMALL_FWD_PAGES;
spin_lock_init(&ksocknal_data.ksnd_large_fmp.fmp_lock);
INIT_LIST_HEAD(&ksocknal_data.ksnd_large_fmp.fmp_idle_fmbs);
INIT_LIST_HEAD(&ksocknal_data.ksnd_large_fmp.fmp_blocked_conns);
-
- spin_lock_init(&ksocknal_data.ksnd_idle_ltx_lock);
- INIT_LIST_HEAD(&ksocknal_data.ksnd_idle_nblk_ltx_list);
- INIT_LIST_HEAD(&ksocknal_data.ksnd_idle_ltx_list);
- init_waitqueue_head(&ksocknal_data.ksnd_idle_ltx_waitq);
+ ksocknal_data.ksnd_large_fmp.fmp_buff_pages = SOCKNAL_LARGE_FWD_PAGES;
spin_lock_init (&ksocknal_data.ksnd_reaper_lock);
+ INIT_LIST_HEAD (&ksocknal_data.ksnd_enomem_conns);
INIT_LIST_HEAD (&ksocknal_data.ksnd_zombie_conns);
INIT_LIST_HEAD (&ksocknal_data.ksnd_deathrow_conns);
init_waitqueue_head(&ksocknal_data.ksnd_reaper_waitq);
PORTAL_ALLOC(ksocknal_data.ksnd_schedulers,
sizeof(ksock_sched_t) * SOCKNAL_N_SCHED);
if (ksocknal_data.ksnd_schedulers == NULL) {
- ksocknal_module_fini ();
- RETURN(-ENOMEM);
+ ksocknal_api_shutdown (&ksocknal_api);
+ return (-ENOMEM);
}
for (i = 0; i < SOCKNAL_N_SCHED; i++) {
init_waitqueue_head (&kss->kss_waitq);
}
- CDEBUG (D_MALLOC, "ltx "LPSZ", total "LPSZ"\n", sizeof (ksock_ltx_t),
- sizeof (ksock_ltx_t) * (SOCKNAL_NLTXS + SOCKNAL_NNBLK_LTXS));
-
- PORTAL_ALLOC(ksocknal_data.ksnd_ltxs,
- sizeof(ksock_ltx_t) * (SOCKNAL_NLTXS +SOCKNAL_NNBLK_LTXS));
- if (ksocknal_data.ksnd_ltxs == NULL) {
- ksocknal_module_fini ();
- return (-ENOMEM);
- }
-
- /* Deterministic bugs please */
- memset (ksocknal_data.ksnd_ltxs, 0xeb,
- sizeof (ksock_ltx_t) * (SOCKNAL_NLTXS + SOCKNAL_NNBLK_LTXS));
-
- for (i = 0; i < SOCKNAL_NLTXS + SOCKNAL_NNBLK_LTXS; i++) {
- ksock_ltx_t *ltx = &((ksock_ltx_t *)ksocknal_data.ksnd_ltxs)[i];
-
- ltx->ltx_tx.tx_hdr = <x->ltx_hdr;
- ltx->ltx_idle = i < SOCKNAL_NLTXS ?
- &ksocknal_data.ksnd_idle_ltx_list :
- &ksocknal_data.ksnd_idle_nblk_ltx_list;
- list_add (<x->ltx_tx.tx_list, ltx->ltx_idle);
- }
-
- rc = PtlNIInit(ksocknal_init, 32, 4, 0, &ksocknal_ni);
- if (rc != 0) {
- CERROR("ksocknal: PtlNIInit failed: error %d\n", rc);
- ksocknal_module_fini ();
- RETURN (rc);
+ /* NB we have to wait to be told our true NID... */
+ process_id.pid = 0;
+ process_id.nid = 0;
+
+ rc = lib_init(&ksocknal_lib, process_id,
+ requested_limits, actual_limits);
+ if (rc != PTL_OK) {
+ CERROR("lib_init failed: error %d\n", rc);
+ ksocknal_api_shutdown (&ksocknal_api);
+ return (rc);
}
- PtlNIDebug(ksocknal_ni, ~0);
- ksocknal_data.ksnd_init = SOCKNAL_INIT_PTL; // flag PtlNIInit() called
+ ksocknal_data.ksnd_init = SOCKNAL_INIT_LIB; // flag lib_init() called
for (i = 0; i < SOCKNAL_N_SCHED; i++) {
rc = ksocknal_thread_start (ksocknal_scheduler,
if (rc != 0) {
CERROR("Can't spawn socknal scheduler[%d]: %d\n",
i, rc);
- ksocknal_module_fini ();
- RETURN (rc);
+ ksocknal_api_shutdown (&ksocknal_api);
+ return (rc);
}
}
rc = ksocknal_thread_start (ksocknal_autoconnectd, (void *)((long)i));
if (rc != 0) {
CERROR("Can't spawn socknal autoconnectd: %d\n", rc);
- ksocknal_module_fini ();
- RETURN (rc);
+ ksocknal_api_shutdown (&ksocknal_api);
+ return (rc);
}
}
rc = ksocknal_thread_start (ksocknal_reaper, NULL);
if (rc != 0) {
CERROR ("Can't spawn socknal reaper: %d\n", rc);
- ksocknal_module_fini ();
- RETURN (rc);
+ ksocknal_api_shutdown (&ksocknal_api);
+ return (rc);
}
rc = kpr_register(&ksocknal_data.ksnd_router,
CDEBUG(D_NET, "Can't initialise routing interface "
"(rc = %d): not routing\n", rc);
} else {
- /* Only allocate forwarding buffers if I'm on a gateway */
-
- PORTAL_ALLOC(ksocknal_data.ksnd_fmbs,
- sizeof(ksock_fmb_t) * (SOCKNAL_SMALL_FWD_NMSGS +
- SOCKNAL_LARGE_FWD_NMSGS));
- if (ksocknal_data.ksnd_fmbs == NULL) {
- ksocknal_module_fini ();
- RETURN(-ENOMEM);
- }
-
- /* NULL out buffer pointers etc */
- memset(ksocknal_data.ksnd_fmbs, 0,
- sizeof(ksock_fmb_t) * (SOCKNAL_SMALL_FWD_NMSGS +
- SOCKNAL_LARGE_FWD_NMSGS));
+ /* Only allocate forwarding buffers if there's a router */
for (i = 0; i < (SOCKNAL_SMALL_FWD_NMSGS +
SOCKNAL_LARGE_FWD_NMSGS); i++) {
- ksock_fmb_t *fmb =
- &((ksock_fmb_t *)ksocknal_data.ksnd_fmbs)[i];
-
- if (i < SOCKNAL_SMALL_FWD_NMSGS) {
- fmb->fmb_npages = SOCKNAL_SMALL_FWD_PAGES;
- fmb->fmb_pool = &ksocknal_data.ksnd_small_fmp;
- } else {
- fmb->fmb_npages = SOCKNAL_LARGE_FWD_PAGES;
- fmb->fmb_pool = &ksocknal_data.ksnd_large_fmp;
+ ksock_fmb_t *fmb;
+ ksock_fmb_pool_t *pool;
+
+
+ if (i < SOCKNAL_SMALL_FWD_NMSGS)
+ pool = &ksocknal_data.ksnd_small_fmp;
+ else
+ pool = &ksocknal_data.ksnd_large_fmp;
+
+ PORTAL_ALLOC(fmb, offsetof(ksock_fmb_t,
+ fmb_kiov[pool->fmp_buff_pages]));
+ if (fmb == NULL) {
+ ksocknal_api_shutdown(&ksocknal_api);
+ return (-ENOMEM);
}
- LASSERT (fmb->fmb_npages > 0);
- for (j = 0; j < fmb->fmb_npages; j++) {
- fmb->fmb_pages[j] = alloc_page(GFP_KERNEL);
+ fmb->fmb_pool = pool;
+
+ for (j = 0; j < pool->fmp_buff_pages; j++) {
+ fmb->fmb_kiov[j].kiov_page = alloc_page(GFP_KERNEL);
- if (fmb->fmb_pages[j] == NULL) {
- ksocknal_module_fini ();
+ if (fmb->fmb_kiov[j].kiov_page == NULL) {
+ ksocknal_api_shutdown (&ksocknal_api);
return (-ENOMEM);
}
- LASSERT(page_address (fmb->fmb_pages[j]) !=
- NULL);
+ LASSERT(page_address(fmb->fmb_kiov[j].kiov_page) != NULL);
}
- list_add(&fmb->fmb_list, &fmb->fmb_pool->fmp_idle_fmbs);
+ list_add(&fmb->fmb_list, &pool->fmp_idle_fmbs);
}
}
- rc = kportal_nal_register(SOCKNAL, &ksocknal_cmd, NULL);
+ rc = libcfs_nal_cmd_register(SOCKNAL, &ksocknal_cmd, NULL);
if (rc != 0) {
CERROR ("Can't initialise command interface (rc = %d)\n", rc);
- ksocknal_module_fini ();
+ ksocknal_api_shutdown (&ksocknal_api);
return (rc);
}
- PORTAL_SYMBOL_REGISTER(ksocknal_ni);
-
-#ifdef CONFIG_SYSCTL
- /* Press on regardless even if registering sysctl doesn't work */
- ksocknal_data.ksnd_sysctl = register_sysctl_table (ksocknal_top_ctl_table, 0);
-#endif
/* flag everything initialised */
ksocknal_data.ksnd_init = SOCKNAL_INIT_ALL;
printk(KERN_INFO "Lustre: Routing socket NAL loaded "
- "(Routing %s, initial mem %d)\n",
+ "(Routing %s, initial mem %d, incarnation "LPX64")\n",
kpr_routing (&ksocknal_data.ksnd_router) ?
- "enabled" : "disabled", pkmem);
+ "enabled" : "disabled", pkmem, ksocknal_data.ksnd_incarnation);
+
+ return (0);
+}
+void __exit
+ksocknal_module_fini (void)
+{
+#ifdef CONFIG_SYSCTL
+ if (ksocknal_tunables.ksnd_sysctl != NULL)
+ unregister_sysctl_table (ksocknal_tunables.ksnd_sysctl);
+#endif
+ PtlNIFini(ksocknal_ni);
+
+ ptl_unregister_nal(SOCKNAL);
+}
+
+int __init
+ksocknal_module_init (void)
+{
+ int rc;
+
+ /* packet descriptor must fit in a router descriptor's scratchpad */
+ LASSERT(sizeof (ksock_tx_t) <= sizeof (kprfd_scratch_t));
+ /* the following must be sizeof(int) for proc_dointvec() */
+ LASSERT(sizeof (ksocknal_tunables.ksnd_io_timeout) == sizeof (int));
+ LASSERT(sizeof (ksocknal_tunables.ksnd_eager_ack) == sizeof (int));
+ LASSERT(sizeof (ksocknal_tunables.ksnd_typed_conns) == sizeof (int));
+ LASSERT(sizeof (ksocknal_tunables.ksnd_min_bulk) == sizeof (int));
+#if SOCKNAL_ZC
+ LASSERT(sizeof (ksocknal_tunables.ksnd_zc_min_frag) == sizeof (int));
+#endif
+ /* check ksnr_connected/connecting field large enough */
+ LASSERT(SOCKNAL_CONN_NTYPES <= 4);
+
+ ksocknal_api.startup = ksocknal_api_startup;
+ ksocknal_api.forward = ksocknal_api_forward;
+ ksocknal_api.shutdown = ksocknal_api_shutdown;
+ ksocknal_api.lock = ksocknal_api_lock;
+ ksocknal_api.unlock = ksocknal_api_unlock;
+ ksocknal_api.nal_data = &ksocknal_data;
+
+ ksocknal_lib.nal_data = &ksocknal_data;
+
+ /* Initialise dynamic tunables to defaults once only */
+ ksocknal_tunables.ksnd_io_timeout = SOCKNAL_IO_TIMEOUT;
+ ksocknal_tunables.ksnd_eager_ack = SOCKNAL_EAGER_ACK;
+ ksocknal_tunables.ksnd_typed_conns = SOCKNAL_TYPED_CONNS;
+ ksocknal_tunables.ksnd_min_bulk = SOCKNAL_MIN_BULK;
+#if SOCKNAL_ZC
+ ksocknal_tunables.ksnd_zc_min_frag = SOCKNAL_ZC_MIN_FRAG;
+#endif
+
+ rc = ptl_register_nal(SOCKNAL, &ksocknal_api);
+ if (rc != PTL_OK) {
+ CERROR("Can't register SOCKNAL: %d\n", rc);
+ return (-ENOMEM); /* or something... */
+ }
+
+ /* Pure gateways want the NAL started up at module load time... */
+ rc = PtlNIInit(SOCKNAL, 0, NULL, NULL, &ksocknal_ni);
+ if (rc != PTL_OK && rc != PTL_IFACE_DUP) {
+ ptl_unregister_nal(SOCKNAL);
+ return (-ENODEV);
+ }
+
+#ifdef CONFIG_SYSCTL
+ /* Press on regardless even if registering sysctl doesn't work */
+ ksocknal_tunables.ksnd_sysctl =
+ register_sysctl_table (ksocknal_top_ctl_table, 0);
+#endif
return (0);
}
module_init(ksocknal_module_init);
module_exit(ksocknal_module_fini);
-EXPORT_SYMBOL (ksocknal_ni);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff