#include <lnet/lnet.h>
#include <lnet/lib-lnet.h>
#include <lnet/socklnd.h>
+#include <lnet/lnet-sysctl.h>
#define SOCKNAL_PEER_HASH_SIZE 101 /* # peer lists */
#define SOCKNAL_RESCHED 100 /* # scheduler loops before reschedule */
#define SOCKNAL_ENOMEM_RETRY CFS_TICK /* jiffies between retries */
-#define SOCKNAL_ROUND_ROBIN 0 /* round robin / load balance */
-
#define SOCKNAL_SINGLE_FRAG_TX 0 /* disable multi-fragment sends */
#define SOCKNAL_SINGLE_FRAG_RX 0 /* disable multi-fragment receives */
typedef struct /* per scheduler state */
{
- spinlock_t kss_lock; /* serialise */
+ cfs_spinlock_t kss_lock; /* serialise */
struct list_head kss_rx_conns; /* conn waiting to be read */
struct list_head kss_tx_conns; /* conn waiting to be written */
struct list_head kss_zombie_noop_txs; /* zombie noop tx list */
cfs_waitq_t kss_waitq; /* where scheduler sleeps */
int kss_nconns; /* # connections assigned to this scheduler */
+#if !SOCKNAL_SINGLE_FRAG_RX
+ struct page *kss_rx_scratch_pgs[LNET_MAX_IOV];
+#endif
+#if !SOCKNAL_SINGLE_FRAG_TX || !SOCKNAL_SINGLE_FRAG_RX
+ struct iovec kss_scratch_iov[LNET_MAX_IOV];
+#endif
} ksock_sched_t;
typedef struct
int *ksnd_tx_buffer_size; /* socket tx buffer size */
int *ksnd_rx_buffer_size; /* socket rx buffer size */
int *ksnd_nagle; /* enable NAGLE? */
+ int *ksnd_round_robin; /* round robin for multiple interfaces */
+ int *ksnd_keepalive; /* # secs for sending keepalive NOOP */
int *ksnd_keepalive_idle; /* # idle secs before 1st probe */
int *ksnd_keepalive_count; /* # probes */
int *ksnd_keepalive_intvl; /* time between probes */
int *ksnd_peercredits; /* # concurrent sends to 1 peer */
int *ksnd_enable_csum; /* enable check sum */
int *ksnd_inject_csum_error; /* set non-zero to inject checksum error */
- unsigned int *ksnd_zc_min_frag; /* minimum zero copy frag size */
+ unsigned int *ksnd_zc_min_payload; /* minimum zero copy payload size */
+ int *ksnd_zc_recv; /* enable ZC receive (for Chelsio TOE) */
+ int *ksnd_zc_recv_min_nfrags; /* minimum # of fragments to enable ZC receive */
#ifdef CPU_AFFINITY
int *ksnd_irq_affinity; /* enable IRQ affinity? */
#endif
typedef struct
{
__u64 ksnn_incarnation; /* my epoch */
- spinlock_t ksnn_lock; /* serialise */
+ cfs_spinlock_t ksnn_lock; /* serialise */
int ksnn_npeers; /* # peers */
int ksnn_shutdown; /* shutting down? */
int ksnn_ninterfaces; /* IP interfaces */
int ksnd_init; /* initialisation state */
int ksnd_nnets; /* # networks set up */
- rwlock_t ksnd_global_lock; /* stabilize peer/conn ops */
+ cfs_rwlock_t ksnd_global_lock; /* stabilize peer/conn ops */
struct list_head *ksnd_peers; /* hash table of all my known peers */
int ksnd_peer_hash_size; /* size of ksnd_peers */
int ksnd_nschedulers; /* # schedulers */
ksock_sched_t *ksnd_schedulers; /* their state */
- atomic_t ksnd_nactive_txs; /* #active txs */
+ cfs_atomic_t ksnd_nactive_txs; /* #active txs */
struct list_head ksnd_deathrow_conns; /* conns to close: reaper_lock*/
struct list_head ksnd_zombie_conns; /* conns to free: reaper_lock */
struct list_head ksnd_enomem_conns; /* conns to retry: reaper_lock*/
cfs_waitq_t ksnd_reaper_waitq; /* reaper sleeps here */
cfs_time_t ksnd_reaper_waketime; /* when reaper will wake */
- spinlock_t ksnd_reaper_lock; /* serialise */
+ cfs_spinlock_t ksnd_reaper_lock; /* serialise */
int ksnd_enomem_tx; /* test ENOMEM sender */
int ksnd_stall_tx; /* test sluggish sender */
struct list_head ksnd_connd_routes; /* routes waiting to be connected */
cfs_waitq_t ksnd_connd_waitq; /* connds sleep here */
int ksnd_connd_connecting;/* # connds connecting */
- spinlock_t ksnd_connd_lock; /* serialise */
+ cfs_spinlock_t ksnd_connd_lock; /* serialise */
struct list_head ksnd_idle_noop_txs; /* list head for freed noop tx */
- spinlock_t ksnd_tx_lock; /* serialise, NOT safe in g_lock */
+ cfs_spinlock_t ksnd_tx_lock; /* serialise, NOT safe in g_lock */
ksock_irqinfo_t ksnd_irqinfo[NR_IRQS];/* irq->scheduler lookup */
{
struct list_head tx_list; /* queue on conn for transmission etc */
struct list_head tx_zc_list; /* queue on peer for ZC request */
- atomic_t tx_refcount; /* tx reference count */
+ cfs_atomic_t tx_refcount; /* tx reference count */
int tx_nob; /* # packet bytes */
int tx_resid; /* residual bytes */
int tx_niov; /* # packet iovec frags */
struct iovec *tx_iov; /* packet iovec frags */
int tx_nkiov; /* # packet page frags */
- unsigned int tx_checked_zc; /* Have I checked if I should ZC? */
+ unsigned int tx_zc_capable:1; /* payload is large enough for ZC */
+ unsigned int tx_zc_checked:1; /* Have I checked if I should ZC? */
+ unsigned int tx_nonblk:1; /* it's a non-blocking ACK */
lnet_kiov_t *tx_kiov; /* packet page frags */
struct ksock_conn *tx_conn; /* owning conn */
lnet_msg_t *tx_lnetmsg; /* lnet message for lnet_finalize() */
} tx_frags;
} ksock_tx_t;
-#define KSOCK_NOOP_TX_SIZE offsetof(ksock_tx_t, tx_frags.paged.kiov[0])
+#define KSOCK_NOOP_TX_SIZE ((int)offsetof(ksock_tx_t, tx_frags.paged.kiov[0]))
/* network zero copy callback descriptor embedded in ksock_tx_t */
cfs_socket_t *ksnc_sock; /* actual socket */
void *ksnc_saved_data_ready; /* socket's original data_ready() callback */
void *ksnc_saved_write_space; /* socket's original write_space() callback */
- atomic_t ksnc_conn_refcount; /* conn refcount */
- atomic_t ksnc_sock_refcount; /* sock refcount */
+ cfs_atomic_t ksnc_conn_refcount; /* conn refcount */
+ cfs_atomic_t ksnc_sock_refcount; /* sock refcount */
ksock_sched_t *ksnc_scheduler; /* who schedules this connection */
__u32 ksnc_myipaddr; /* my IP */
__u32 ksnc_ipaddr; /* peer's IP */
/* WRITER */
struct list_head ksnc_tx_list; /* where I enq waiting for output space */
struct list_head ksnc_tx_queue; /* packets waiting to be sent */
- ksock_tx_t *ksnc_tx_mono; /* V2.x only, next mono-packet, mono-packet is :
- * a. lnet packet without piggyback
- * b. noop ZC-ACK packet */
+ ksock_tx_t *ksnc_tx_carrier; /* next TX that can carry a LNet message or ZC-ACK */
cfs_time_t ksnc_tx_deadline; /* when (in jiffies) tx times out */
int ksnc_tx_bufnob; /* send buffer marker */
- atomic_t ksnc_tx_nob; /* # bytes queued */
+ cfs_atomic_t ksnc_tx_nob; /* # bytes queued */
int ksnc_tx_ready; /* write space */
int ksnc_tx_scheduled; /* being progressed */
-
-#if !SOCKNAL_SINGLE_FRAG_RX
- struct iovec ksnc_rx_scratch_iov[LNET_MAX_IOV];
-#endif
-#if !SOCKNAL_SINGLE_FRAG_TX
- struct iovec ksnc_tx_scratch_iov[LNET_MAX_IOV];
-#endif
+ cfs_time_t ksnc_tx_last_post; /* time stamp of the last posted TX */
} ksock_conn_t;
typedef struct ksock_route
struct list_head ksnr_list; /* chain on peer route list */
struct list_head ksnr_connd_list; /* chain on ksnr_connd_routes */
struct ksock_peer *ksnr_peer; /* owning peer */
- atomic_t ksnr_refcount; /* # users */
+ cfs_atomic_t ksnr_refcount; /* # users */
cfs_time_t ksnr_timeout; /* when (in jiffies) reconnection can happen next */
cfs_duration_t ksnr_retry_interval; /* how long between retries */
__u32 ksnr_myipaddr; /* my IP */
int ksnr_conn_count; /* # conns established by this route */
} ksock_route_t;
+#define SOCKNAL_KEEPALIVE_PING 1 /* cookie for keepalive ping */
+
typedef struct ksock_peer
{
struct list_head ksnp_list; /* stash on global peer list */
lnet_process_id_t ksnp_id; /* who's on the other end(s) */
- atomic_t ksnp_refcount; /* # users */
+ cfs_atomic_t ksnp_refcount; /* # users */
int ksnp_sharecount; /* lconf usage counter */
int ksnp_closing; /* being closed */
int ksnp_accepting; /* # passive connections pending */
struct list_head ksnp_conns; /* all active connections */
struct list_head ksnp_routes; /* routes */
struct list_head ksnp_tx_queue; /* waiting packets */
- spinlock_t ksnp_lock; /* serialize, NOT safe in g_lock */
+ cfs_spinlock_t ksnp_lock; /* serialize, NOT safe in g_lock */
struct list_head ksnp_zc_req_list; /* zero copy requests wait for ACK */
cfs_time_t ksnp_last_alive; /* when (in jiffies) I was last alive */
+ cfs_time_t ksnp_send_keepalive; /* time to send keepalive */
lnet_ni_t *ksnp_ni; /* which network */
int ksnp_n_passive_ips; /* # of... */
__u32 ksnp_passive_ips[LNET_MAX_INTERFACES]; /* preferred local interfaces */
extern ksock_nal_data_t ksocknal_data;
extern ksock_tunables_t ksocknal_tunables;
+#define SOCKNAL_MATCH_NO 0 /* TX can't match type of connection */
+#define SOCKNAL_MATCH_YES 1 /* TX matches type of connection */
+#define SOCKNAL_MATCH_MAY 2 /* TX can be sent on the connection, but not preferred */
+
typedef struct ksock_proto
{
- int pro_version; /* version number of protocol */
- int (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *); /* handshake function */
- int (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);/* handshake function */
- void (*pro_pack)(ksock_tx_t *); /* message pack */
- void (*pro_unpack)(ksock_msg_t *); /* message unpack */
+ int pro_version; /* version number of protocol */
+ int (*pro_send_hello)(ksock_conn_t *, ksock_hello_msg_t *); /* handshake function */
+ int (*pro_recv_hello)(ksock_conn_t *, ksock_hello_msg_t *, int);/* handshake function */
+ void (*pro_pack)(ksock_tx_t *); /* message pack */
+ void (*pro_unpack)(ksock_msg_t *); /* message unpack */
+ ksock_tx_t *(*pro_queue_tx_msg)(ksock_conn_t *, ksock_tx_t *); /* queue tx on the connection */
+ int (*pro_queue_tx_zcack)(ksock_conn_t *, ksock_tx_t *, __u64); /* queue ZC ack on the connection */
+ int (*pro_handle_zcreq)(ksock_conn_t *, __u64, int); /* handle ZC request */
+ int (*pro_handle_zcack)(ksock_conn_t *, __u64, __u64); /* handle ZC ACK */
+ int (*pro_match_tx)(ksock_conn_t *, ksock_tx_t *, int); /* msg type matches the connection type:
+ * return value:
+ * return MATCH_NO : no
+ * return MATCH_YES : matching type
+ * return MATCH_MAY : can be backup */
} ksock_proto_t;
extern ksock_proto_t ksocknal_protocol_v1x;
extern ksock_proto_t ksocknal_protocol_v2x;
+extern ksock_proto_t ksocknal_protocol_v3x;
#define KSOCK_PROTO_V1_MAJOR LNET_PROTO_TCP_VERSION_MAJOR
#define KSOCK_PROTO_V1_MINOR LNET_PROTO_TCP_VERSION_MINOR
static inline void
ksocknal_conn_addref (ksock_conn_t *conn)
{
- LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
- atomic_inc(&conn->ksnc_conn_refcount);
+ LASSERT (cfs_atomic_read(&conn->ksnc_conn_refcount) > 0);
+ cfs_atomic_inc(&conn->ksnc_conn_refcount);
}
extern void ksocknal_queue_zombie_conn (ksock_conn_t *conn);
static inline void
ksocknal_conn_decref (ksock_conn_t *conn)
{
- LASSERT (atomic_read(&conn->ksnc_conn_refcount) > 0);
- if (atomic_dec_and_test(&conn->ksnc_conn_refcount))
+ LASSERT (cfs_atomic_read(&conn->ksnc_conn_refcount) > 0);
+ if (cfs_atomic_dec_and_test(&conn->ksnc_conn_refcount))
ksocknal_queue_zombie_conn(conn);
}
{
int rc = -ESHUTDOWN;
- read_lock (&ksocknal_data.ksnd_global_lock);
+ cfs_read_lock (&ksocknal_data.ksnd_global_lock);
if (!conn->ksnc_closing) {
- LASSERT (atomic_read(&conn->ksnc_sock_refcount) > 0);
- atomic_inc(&conn->ksnc_sock_refcount);
+ LASSERT (cfs_atomic_read(&conn->ksnc_sock_refcount) > 0);
+ cfs_atomic_inc(&conn->ksnc_sock_refcount);
rc = 0;
}
- read_unlock (&ksocknal_data.ksnd_global_lock);
+ cfs_read_unlock (&ksocknal_data.ksnd_global_lock);
return (rc);
}
static inline void
ksocknal_connsock_decref (ksock_conn_t *conn)
{
- LASSERT (atomic_read(&conn->ksnc_sock_refcount) > 0);
- if (atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
+ LASSERT (cfs_atomic_read(&conn->ksnc_sock_refcount) > 0);
+ if (cfs_atomic_dec_and_test(&conn->ksnc_sock_refcount)) {
LASSERT (conn->ksnc_closing);
libcfs_sock_release(conn->ksnc_sock);
conn->ksnc_sock = NULL;
static inline void
ksocknal_tx_addref (ksock_tx_t *tx)
{
- LASSERT (atomic_read(&tx->tx_refcount) > 0);
- atomic_inc(&tx->tx_refcount);
+ LASSERT (cfs_atomic_read(&tx->tx_refcount) > 0);
+ cfs_atomic_inc(&tx->tx_refcount);
}
+extern void ksocknal_tx_prep (ksock_conn_t *, ksock_tx_t *tx);
extern void ksocknal_tx_done (lnet_ni_t *ni, ksock_tx_t *tx);
static inline void
ksocknal_tx_decref (ksock_tx_t *tx)
{
- LASSERT (atomic_read(&tx->tx_refcount) > 0);
- if (atomic_dec_and_test(&tx->tx_refcount))
+ LASSERT (cfs_atomic_read(&tx->tx_refcount) > 0);
+ if (cfs_atomic_dec_and_test(&tx->tx_refcount))
ksocknal_tx_done(NULL, tx);
}
static inline void
ksocknal_route_addref (ksock_route_t *route)
{
- LASSERT (atomic_read(&route->ksnr_refcount) > 0);
- atomic_inc(&route->ksnr_refcount);
+ LASSERT (cfs_atomic_read(&route->ksnr_refcount) > 0);
+ cfs_atomic_inc(&route->ksnr_refcount);
}
extern void ksocknal_destroy_route (ksock_route_t *route);
static inline void
ksocknal_route_decref (ksock_route_t *route)
{
- LASSERT (atomic_read (&route->ksnr_refcount) > 0);
- if (atomic_dec_and_test(&route->ksnr_refcount))
+ LASSERT (cfs_atomic_read (&route->ksnr_refcount) > 0);
+ if (cfs_atomic_dec_and_test(&route->ksnr_refcount))
ksocknal_destroy_route (route);
}
static inline void
ksocknal_peer_addref (ksock_peer_t *peer)
{
- LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
- atomic_inc(&peer->ksnp_refcount);
+ LASSERT (cfs_atomic_read (&peer->ksnp_refcount) > 0);
+ cfs_atomic_inc(&peer->ksnp_refcount);
}
extern void ksocknal_destroy_peer (ksock_peer_t *peer);
static inline void
ksocknal_peer_decref (ksock_peer_t *peer)
{
- LASSERT (atomic_read (&peer->ksnp_refcount) > 0);
- if (atomic_dec_and_test(&peer->ksnp_refcount))
+ LASSERT (cfs_atomic_read (&peer->ksnp_refcount) > 0);
+ if (cfs_atomic_dec_and_test(&peer->ksnp_refcount))
ksocknal_destroy_peer (peer);
}
__u32 ipaddr, int why);
extern int ksocknal_close_conn_and_siblings (ksock_conn_t *conn, int why);
extern int ksocknal_close_matching_conns (lnet_process_id_t id, __u32 ipaddr);
-
+extern ksock_conn_t *ksocknal_find_conn_locked(ksock_peer_t *peer,
+ ksock_tx_t *tx, int nonblk);
+
+extern int ksocknal_launch_packet(lnet_ni_t *ni, ksock_tx_t *tx,
+ lnet_process_id_t id);
+extern ksock_tx_t *ksocknal_alloc_tx(int type, int size);
+extern void ksocknal_free_tx (ksock_tx_t *tx);
+extern ksock_tx_t *ksocknal_alloc_tx_noop(__u64 cookie, int nonblk);
+extern void ksocknal_next_tx_carrier(ksock_conn_t *conn);
extern void ksocknal_queue_tx_locked (ksock_tx_t *tx, ksock_conn_t *conn);
extern void ksocknal_txlist_done (lnet_ni_t *ni, struct list_head *txlist, int error);
extern void ksocknal_notify (lnet_ni_t *ni, lnet_nid_t gw_nid, int alive);
extern void ksocknal_read_callback(ksock_conn_t *conn);
extern void ksocknal_write_callback(ksock_conn_t *conn);
-extern int ksocknal_lib_zc_capable(cfs_socket_t *sock);
+extern int ksocknal_lib_zc_capable(ksock_conn_t *conn);
extern void ksocknal_lib_save_callback(cfs_socket_t *sock, ksock_conn_t *conn);
extern void ksocknal_lib_set_callback(cfs_socket_t *sock, ksock_conn_t *conn);
extern void ksocknal_lib_reset_callback(cfs_socket_t *sock, ksock_conn_t *conn);
extern int ksocknal_lib_get_conn_tunables (ksock_conn_t *conn, int *txmem,
int *rxmem, int *nagle);
+extern int ksocknal_tunables_init(void);
+extern void ksocknal_tunables_fini(void);
extern int ksocknal_lib_tunables_init(void);
extern void ksocknal_lib_tunables_fini(void);
extern void ksocknal_lib_csum_tx(ksock_tx_t *tx);
+
+extern int ksocknal_lib_memory_pressure(ksock_conn_t *conn);
+extern __u64 ksocknal_lib_new_incarnation(void);
+extern int ksocknal_lib_bind_thread_to_cpu(int id);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff