*/
/*
* This file is part of Lustre, http://www.lustre.org/
- * Lustre is a trademark of Sun Microsystems, Inc.
*/
#ifndef __UAPI_LNET_TYPES_H__
#define __UAPI_LNET_TYPES_H__
+#include <linux/types.h>
+#include <linux/string.h>
+#include <asm/byteorder.h>
+#ifndef __KERNEL__
+#include <stdbool.h>
+#endif
+
/** \addtogroup lnet
* @{ */
-#include <linux/types.h>
+#include <linux/lnet/lnet-idl.h>
+
/** \addtogroup lnet_addr
* @{ */
*/
#define LNET_RESERVED_PORTAL 0
-/**
- * Address of an end-point in an LNet network.
- *
- * A node can have multiple end-points and hence multiple addresses.
- * An LNet network can be a simple network (e.g. tcp0) or a network of
- * LNet networks connected by LNet routers. Therefore an end-point address
- * has two parts: network ID, and address within a network.
- *
- * \see LNET_NIDNET, LNET_NIDADDR, and LNET_MKNID.
- */
-typedef __u64 lnet_nid_t;
-
-/**
- * ID of a process in a node. Shortened as PID to distinguish from
- * lnet_process_id, the global process ID.
- */
-typedef __u32 lnet_pid_t;
-
/** wildcard NID that matches any end-point address */
-#define LNET_NID_ANY ((lnet_nid_t) -1)
+#define LNET_NID_ANY (~(lnet_nid_t) 0)
/** wildcard PID that matches any lnet_pid_t */
-#define LNET_PID_ANY ((lnet_pid_t) -1)
+#define LNET_PID_ANY (~(lnet_pid_t) 0)
+
+static inline int LNET_NID_IS_ANY(const struct lnet_nid *nid)
+{
+ /* A NULL pointer can be used to mean "ANY" */
+ return !nid || nid->nid_type == 0xFF;
+}
+
+#define LNET_ANY_NID ((struct lnet_nid) \
+ {0xFF, 0xFF, ~0, {~0, ~0, ~0, ~0} })
#define LNET_PID_RESERVED 0xf0000000 /* reserved bits in PID */
#define LNET_PID_USERFLAG 0x80000000 /* set in userspace peers */
return (nid >> 32) & 0xffffffff;
}
-static inline lnet_nid_t LNET_MKNID(__u32 net, __u32 addr)
-{
- return (((__u64)net) << 32) | addr;
-}
-
static inline __u32 LNET_NETNUM(__u32 net)
{
return net & 0xffff;
static inline __u32 LNET_NETTYP(__u32 net)
{
- return (net >> 16) & 0xffff;
+ return (net >> 16) & 0xff;
}
static inline __u32 LNET_MKNET(__u32 type, __u32 num)
return (type << 16) | num;
}
-#define WIRE_ATTR __attribute__((packed))
+static inline lnet_nid_t LNET_MKNID(__u32 net, __u32 addr)
+{
+ return (((__u64)net) << 32) | addr;
+}
-/* Packed version of struct lnet_process_id to transfer via network */
-struct lnet_process_id_packed {
- lnet_nid_t nid;
- lnet_pid_t pid; /* node id / process id */
-} WIRE_ATTR;
-
-/* The wire handle's interface cookie only matches one network interface in
- * one epoch (i.e. new cookie when the interface restarts or the node
- * reboots). The object cookie only matches one object on that interface
- * during that object's lifetime (i.e. no cookie re-use). */
-struct lnet_handle_wire {
- __u64 wh_interface_cookie;
- __u64 wh_object_cookie;
-} WIRE_ATTR;
-
-enum lnet_msg_type {
- LNET_MSG_ACK = 0,
- LNET_MSG_PUT,
- LNET_MSG_GET,
- LNET_MSG_REPLY,
- LNET_MSG_HELLO,
-};
+/** The lolnd NID (i.e. myself) */
+#define LNET_NID_LO_0 LNET_MKNID(LNET_MKNET(LOLND, 0), 0)
-/* The variant fields of the portals message header are aligned on an 8
- * byte boundary in the message header. Note that all types used in these
- * wire structs MUST be fixed size and the smaller types are placed at the
- * end. */
-struct lnet_ack {
- struct lnet_handle_wire dst_wmd;
- __u64 match_bits;
- __u32 mlength;
-} WIRE_ATTR;
+#define LNET_NET_ANY LNET_NIDNET(LNET_NID_ANY)
-struct lnet_put {
- struct lnet_handle_wire ack_wmd;
- __u64 match_bits;
- __u64 hdr_data;
- __u32 ptl_index;
- __u32 offset;
-} WIRE_ATTR;
+static inline bool nid_is_nid4(const struct lnet_nid *nid)
+{
+ return NID_ADDR_BYTES(nid) == 4;
+}
-struct lnet_get {
- struct lnet_handle_wire return_wmd;
- __u64 match_bits;
- __u32 ptl_index;
- __u32 src_offset;
- __u32 sink_length;
-} WIRE_ATTR;
-
-struct lnet_reply {
- struct lnet_handle_wire dst_wmd;
-} WIRE_ATTR;
-
-struct lnet_hello {
- __u64 incarnation;
- __u32 type;
-} WIRE_ATTR;
-
-struct lnet_hdr {
- lnet_nid_t dest_nid;
- lnet_nid_t src_nid;
- lnet_pid_t dest_pid;
- lnet_pid_t src_pid;
- __u32 type; /* enum lnet_msg_type */
- __u32 payload_length; /* payload data to follow */
- /*<------__u64 aligned------->*/
- union {
- struct lnet_ack ack;
- struct lnet_put put;
- struct lnet_get get;
- struct lnet_reply reply;
- struct lnet_hello hello;
- } msg;
-} WIRE_ATTR;
-
-/* A HELLO message contains a magic number and protocol version
- * code in the header's dest_nid, the peer's NID in the src_nid, and
- * LNET_MSG_HELLO in the type field. All other common fields are zero
- * (including payload_size; i.e. no payload).
- * This is for use by byte-stream LNDs (e.g. TCP/IP) to check the peer is
- * running the same protocol and to find out its NID. These LNDs should
- * exchange HELLO messages when a connection is first established. Individual
- * LNDs can put whatever else they fancy in lnet_hdr::msg.
- */
-struct lnet_magicversion {
- __u32 magic; /* LNET_PROTO_TCP_MAGIC */
- __u16 version_major; /* increment on incompatible change */
- __u16 version_minor; /* increment on compatible change */
-} WIRE_ATTR;
-
-/* PROTO MAGIC for LNDs */
-#define LNET_PROTO_IB_MAGIC 0x0be91b91
-#define LNET_PROTO_GNI_MAGIC 0xb00fbabe /* ask Kim */
-#define LNET_PROTO_TCP_MAGIC 0xeebc0ded
-#define LNET_PROTO_ACCEPTOR_MAGIC 0xacce7100
-#define LNET_PROTO_PING_MAGIC 0x70696E67 /* 'ping' */
-
-/* Placeholder for a future "unified" protocol across all LNDs */
-/* Current LNDs that receive a request with this magic will respond
- * with a "stub" reply using their current protocol */
-#define LNET_PROTO_MAGIC 0x45726963 /* ! */
-
-#define LNET_PROTO_TCP_VERSION_MAJOR 1
-#define LNET_PROTO_TCP_VERSION_MINOR 0
-
-/* Acceptor connection request */
-struct lnet_acceptor_connreq {
- __u32 acr_magic; /* PTL_ACCEPTOR_PROTO_MAGIC */
- __u32 acr_version; /* protocol version */
- __u64 acr_nid; /* target NID */
-} WIRE_ATTR;
-
-#define LNET_PROTO_ACCEPTOR_VERSION 1
-
-struct lnet_counters_common {
- __u32 lcc_msgs_alloc;
- __u32 lcc_msgs_max;
- __u32 lcc_errors;
- __u32 lcc_send_count;
- __u32 lcc_recv_count;
- __u32 lcc_route_count;
- __u32 lcc_drop_count;
- __u64 lcc_send_length;
- __u64 lcc_recv_length;
- __u64 lcc_route_length;
- __u64 lcc_drop_length;
-} WIRE_ATTR;
+/* check for address set */
+static inline bool nid_addr_is_set(const struct lnet_nid *nid)
+{
+ __u8 *addr = (__u8 *)(&nid->nid_addr[0]);
+ int i;
+
+ for (i = 0; i < NID_ADDR_BYTES(nid); i++)
+ if (addr[i])
+ return true;
+
+ return false;
+}
+
+/* LOLND may not be defined yet, so we cannot use an inline */
+#define nid_is_lo0(__nid) \
+ ((__nid)->nid_type == LOLND && \
+ nid_is_nid4(__nid) && \
+ (__nid)->nid_num == 0 && \
+ (__nid)->nid_addr[0] == 0)
+
+static inline __u32 LNET_NID_NET(const struct lnet_nid *nid)
+{
+ if (LNET_NID_IS_ANY(nid))
+ return LNET_NET_ANY;
+ else
+ return LNET_MKNET(nid->nid_type, __be16_to_cpu(nid->nid_num));
+}
+
+static inline void lnet_nid4_to_nid(lnet_nid_t nid4, struct lnet_nid *nid)
+{
+ if (nid4 == LNET_NID_ANY) {
+ *nid = LNET_ANY_NID;
+ return;
+ }
+
+ nid->nid_size = 0;
+ nid->nid_type = LNET_NETTYP(LNET_NIDNET(nid4));
+ nid->nid_num = __cpu_to_be16(LNET_NETNUM(LNET_NIDNET(nid4)));
+ nid->nid_addr[0] = __cpu_to_be32(LNET_NIDADDR(nid4));
+ nid->nid_addr[1] = nid->nid_addr[2] = nid->nid_addr[3] = 0;
+}
+
+static inline lnet_nid_t lnet_nid_to_nid4(const struct lnet_nid *nid)
+{
+ if (LNET_NID_IS_ANY(nid))
+ return LNET_NID_ANY;
+
+ return LNET_MKNID(LNET_NID_NET(nid), __be32_to_cpu(nid->nid_addr[0]));
+}
+
+static inline int nid_same(const struct lnet_nid *n1,
+ const struct lnet_nid *n2)
+{
+ return n1->nid_size == n2->nid_size &&
+ n1->nid_type == n2->nid_type &&
+ n1->nid_num == n2->nid_num &&
+ n1->nid_addr[0] == n2->nid_addr[0] &&
+ n1->nid_addr[1] == n2->nid_addr[1] &&
+ n1->nid_addr[2] == n2->nid_addr[2] &&
+ n1->nid_addr[3] == n2->nid_addr[3];
+}
+
+/* This can be used when we need to hash a nid */
+static inline unsigned long nidhash(const struct lnet_nid *nid)
+{
+ int i;
+ unsigned long hash = 0;
+
+ hash ^= LNET_NID_NET(nid);
+ for (i = 0; i < 4; i++)
+ hash ^= nid->nid_addr[i];
+ return hash;
+}
struct lnet_counters_health {
__u32 lch_rst_alloc;
struct lnet_counters_health lct_health;
};
-#define LNET_NI_STATUS_UP 0x15aac0de
-#define LNET_NI_STATUS_DOWN 0xdeadface
-#define LNET_NI_STATUS_INVALID 0x00000000
-
-struct lnet_ni_status {
- lnet_nid_t ns_nid;
- __u32 ns_status;
- __u32 ns_unused;
-} WIRE_ATTR;
-
-/*
- * NB: value of these features equal to LNET_PROTO_PING_VERSION_x
- * of old LNet, so there shouldn't be any compatibility issue
- */
-#define LNET_PING_FEAT_INVAL (0) /* no feature */
-#define LNET_PING_FEAT_BASE (1 << 0) /* just a ping */
-#define LNET_PING_FEAT_NI_STATUS (1 << 1) /* return NI status */
-#define LNET_PING_FEAT_RTE_DISABLED (1 << 2) /* Routing enabled */
-#define LNET_PING_FEAT_MULTI_RAIL (1 << 3) /* Multi-Rail aware */
-#define LNET_PING_FEAT_DISCOVERY (1 << 4) /* Supports Discovery */
-
-/*
- * All ping feature bits fit to hit the wire.
- * In lnet_assert_wire_constants() this is compared against its open-coded
- * value, and in lnet_ping_target_update() it is used to verify that no
- * unknown bits have been set.
- * New feature bits can be added, just be aware that this does change the
- * over-the-wire protocol.
- */
-#define LNET_PING_FEAT_BITS (LNET_PING_FEAT_BASE | \
- LNET_PING_FEAT_NI_STATUS | \
- LNET_PING_FEAT_RTE_DISABLED | \
- LNET_PING_FEAT_MULTI_RAIL | \
- LNET_PING_FEAT_DISCOVERY)
-
-struct lnet_ping_info {
- __u32 pi_magic;
- __u32 pi_features;
- lnet_pid_t pi_pid;
- __u32 pi_nnis;
- struct lnet_ni_status pi_ni[0];
-} WIRE_ATTR;
-
-#define LNET_PING_INFO_SIZE(NNIDS) \
- offsetof(struct lnet_ping_info, pi_ni[NNIDS])
-#define LNET_PING_INFO_LONI(PINFO) ((PINFO)->pi_ni[0].ns_nid)
-#define LNET_PING_INFO_SEQNO(PINFO) ((PINFO)->pi_ni[0].ns_status)
-
/*
* This is a hard-coded limit on the number of interfaces supported by
* the interface bonding implemented by the ksocknal LND. It must be
/**
* Objects maintained by the LNet are accessed through handles. Handle types
* have names of the form lnet_handle_xx, where xx is one of the two letter
- * object type codes ('eq' for event queue, 'md' for memory descriptor, and
+ * object type codes ('md' for memory descriptor, and
* 'me' for match entry). Each type of object is given a unique handle type
* to enhance type checking.
*/
-#define LNET_WIRE_HANDLE_COOKIE_NONE (-1)
+#define LNET_WIRE_HANDLE_COOKIE_NONE (~0ULL)
struct lnet_handle_md {
__u64 cookie;
}
/**
- * Check whether eq handle \a h is invalid.
+ * Check whether handler \a h is invalid.
*
* \return 1 if handle is invalid, 0 if valid.
*/
/** process id */
lnet_pid_t pid;
};
+
+/**
+ * Global process ID - with large addresses
+ */
+struct lnet_processid {
+ /** node id */
+ struct lnet_nid nid;
+ /** process id */
+ lnet_pid_t pid;
+};
+
+static inline void
+lnet_pid4_to_pid(struct lnet_process_id pid4, struct lnet_processid *pid)
+{
+ pid->pid = pid4.pid;
+ lnet_nid4_to_nid(pid4.nid, &pid->nid);
+}
+
+static inline struct lnet_process_id
+lnet_pid_to_pid4(struct lnet_processid *pid)
+{
+ struct lnet_process_id ret;
+
+ ret.pid = pid->pid;
+ ret.nid = lnet_nid_to_nid4(&pid->nid);
+ return ret;
+}
+
/** @} lnet_addr */
/** \addtogroup lnet_me
/** \addtogroup lnet_md
* @{ */
+struct lnet_hdr_nid16 {
+ char _bytes[sizeof(struct lnet_hdr)];
+} __attribute__((packed));
+
+/**
+ * Event queue handler function type.
+ *
+ * The EQ handler runs for each event that is deposited into the EQ. The
+ * handler is supplied with a pointer to the event that triggered the
+ * handler invocation.
+ *
+ * The handler must not block, must be reentrant, and must not call any LNet
+ * API functions. It should return as quickly as possible.
+ */
+struct lnet_event;
+typedef void (*lnet_handler_t)(struct lnet_event *event);
+
/**
* Defines the visible parts of a memory descriptor. Values of this type
* are used to initialize memory descriptors.
* the number of entries in the array. The length can't be bigger
* than LNET_MAX_IOV. The struct bio_vec is used to describe page-based
* fragments that are not necessarily mapped in virtal memory.
- * - LNET_MD_IOVEC bit set: The start field points to the starting
- * address of an array of struct kvec and the length field specifies
- * the number of entries in the array. The length can't be bigger
- * than LNET_MAX_IOV. The struct kvec is used to describe fragments
- * that have virtual addresses.
* - Otherwise: The memory region is contiguous. The start field
* specifies the starting address for the memory region and the
* length field specifies its length.
* The data sent in the REPLY serves as an implicit acknowledgment.
* - LNET_MD_KIOV: The start and length fields specify an array of
* struct bio_vec.
- * - LNET_MD_IOVEC: The start and length fields specify an array of
- * struct iovec.
* - LNET_MD_MAX_SIZE: The max_size field is valid.
* - LNET_MD_BULK_HANDLE: The bulk_handle field is valid.
+ * - LNET_MD_TRACK_RESPONSE: Enable response tracking on this MD
+ * regardless of the value of the lnet_response_tracking param.
+ * - LNET_MD_NO_TRACK_RESPONSE: Disable response tracking on this MD
+ * regardless of the value of the lnet_response_tracking param.
+ * - LNET_MD_GNILND: Disable warning about exceeding LNET_MAX_IOV.
*
* Note:
- * - LNET_MD_KIOV or LNET_MD_IOVEC allows for a scatter/gather
- * capability for memory descriptors. They can't be both set.
+ * - LNET_MD_KIOV allows for a scatter/gather capability for memory
+ * descriptors.
* - When LNET_MD_MAX_SIZE is set, the total length of the memory
* region (i.e. sum of all fragment lengths) must not be less than
* \a max_size.
*/
void *user_ptr;
/**
- * A handle for the event queue used to log the operations performed on
- * the memory region. If this argument is a NULL handle operations
+ * The event handler used to log the operations performed on
+ * the memory region. If this argument is NULL operations
* performed on this memory descriptor are not logged.
*/
- struct lnet_eq *eq_handle;
+ lnet_handler_t handler;
/**
* The bulk MD handle which was registered to describe the buffers
* either to be used to transfer data to the peer or receive data
/** See struct lnet_md::options. */
#define LNET_MD_ACK_DISABLE (1 << 5)
/** See struct lnet_md::options. */
-#define LNET_MD_IOVEC (1 << 6)
+/* deprecated #define LNET_MD_IOVEC (1 << 6) */
/** See struct lnet_md::options. */
#define LNET_MD_MAX_SIZE (1 << 7)
/** See struct lnet_md::options. */
#define LNET_MD_KIOV (1 << 8)
/** See struct lnet_md::options. */
#define LNET_MD_BULK_HANDLE (1 << 9)
-
-/* For compatibility with Cray Portals */
-#define LNET_MD_PHYS 0
+/** See struct lnet_md::options. */
+#define LNET_MD_TRACK_RESPONSE (1 << 10)
+/** See struct lnet_md::options. */
+#define LNET_MD_NO_TRACK_RESPONSE (1 << 11)
+/** See struct lnet_md::options. */
+#define LNET_MD_GNILND (1 << 12)
+/** Special page mapping handling */
+#define LNET_MD_GPU_ADDR (1 << 13)
/** Infinite threshold on MD operations. See struct lnet_md::threshold */
#define LNET_MD_THRESH_INF (-1)
*/
struct lnet_event {
/** The identifier (nid, pid) of the target. */
- struct lnet_process_id target;
+ struct lnet_processid target;
/** The identifier (nid, pid) of the initiator. */
- struct lnet_process_id initiator;
+ struct lnet_processid initiator;
/** The source NID on the initiator. */
- struct lnet_process_id source;
+ struct lnet_processid source;
/**
* The NID of the immediate sender. If the request has been forwarded
* by routers, this is the NID of the last hop; otherwise it's the
* same as the source.
*/
- lnet_nid_t sender;
+ struct lnet_nid sender;
/** Indicates the type of the event. */
enum lnet_event_kind type;
/** The portal table index specified in the request */
- unsigned int pt_index;
+ unsigned int pt_index;
/** A copy of the match bits specified in the request. */
- __u64 match_bits;
+ __u64 match_bits;
/** The length (in bytes) specified in the request. */
- unsigned int rlength;
+ unsigned int rlength;
/**
* The length (in bytes) of the data that was manipulated by the
* operation. For truncated operations, the manipulated length will be
* see struct lnet_md). For all other operations, the manipulated length
* will be the length of the requested operation, i.e. rlength.
*/
- unsigned int mlength;
+ unsigned int mlength;
/**
* The handle to the MD associated with the event. The handle may be
* invalid if the MD has been unlinked.
*/
struct lnet_handle_md md_handle;
/**
- * A snapshot of the state of the MD immediately after the event has
- * been processed. In particular, the threshold field in md will
- * reflect the value of the threshold after the operation occurred.
+ * A snapshot of relevant state of the MD immediately after the event
+ * has been processed.
*/
- struct lnet_md md;
+ void *md_start;
+ void *md_user_ptr;
+ unsigned int md_options;
/**
* 64 bits of out-of-band user data. Only valid for LNET_EVENT_PUT.
* \see LNetPut
*/
- __u64 hdr_data;
+ __u64 hdr_data;
/**
* The message type, to ensure a handler for LNET_EVENT_SEND can
* distinguish between LNET_MSG_GET and LNET_MSG_PUT.
*/
- __u32 msg_type;
+ __u32 msg_type;
/**
* Indicates the completion status of the operation. It's 0 for
* successful operations, otherwise it's an error code.
*/
- int status;
+ int status;
/**
* Indicates whether the MD has been unlinked. Note that:
* - An event with unlinked set is the last event on the MD.
* - This field is also set for an explicit LNET_EVENT_UNLINK event.
* \see LNetMDUnlink
*/
- int unlinked;
+ int unlinked;
/**
* The displacement (in bytes) into the memory region that the
* operation used. The offset can be determined by the operation for
* a remote managed MD or by the local MD.
* \see struct lnet_md::options
*/
- unsigned int offset;
+ unsigned int offset;
/**
* The sequence number for this event. Sequence numbers are unique
* to each event.
*/
- volatile unsigned long sequence;
+ volatile unsigned long sequence;
};
-/**
- * Event queue handler function type.
- *
- * The EQ handler runs for each event that is deposited into the EQ. The
- * handler is supplied with a pointer to the event that triggered the
- * handler invocation.
- *
- * The handler must not block, must be reentrant, and must not call any LNet
- * API functions. It should return as quickly as possible.
- */
-typedef void (*lnet_eq_handler_t)(struct lnet_event *event);
-#define LNET_EQ_HANDLER_NONE NULL
-/** @} lnet_eq */
-
/** \addtogroup lnet_data
* @{ */
/** Request that no acknowledgment should be generated. */
LNET_NOACK_REQ
};
+
+/**
+ * UDSP action types. There are two available actions:
+ * 1. PRIORITY - set priority of matching LNet constructs
+ * 2. PREFERRED LIST - set preferred list of matching LNet constructs
+ */
+enum lnet_udsp_action_type {
+ EN_LNET_UDSP_ACTION_NONE = 0,
+ /** assign a priority to matching constructs */
+ EN_LNET_UDSP_ACTION_PRIORITY = 1,
+ /** assign a preferred list of NIDs to matching constructs */
+ EN_LNET_UDSP_ACTION_PREFERRED_LIST = 2,
+};
+
/** @} lnet_data */
/** @} lnet */
git://git.whamcloud.com / fs / lustre-release.git / blobdiff