/* * GPL HEADER START * * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License version 2 only, * as published by the Free Software Foundation. * * This program is distributed in the hope that it will be useful, but * WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License version 2 for more details (a copy is included * in the LICENSE file that accompanied this code). * * You should have received a copy of the GNU General Public License * version 2 along with this program; If not, see * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf * * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, * CA 95054 USA or visit www.sun.com if you need additional information or * have any questions. * * GPL HEADER END */ /* * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. * Use is subject to license terms. * * Copyright (c) 2012, 2014, Intel Corporation. */ /* * This file is part of Lustre, http://www.lustre.org/ * Lustre is a trademark of Sun Microsystems, Inc. */ #ifndef __LNET_TYPES_H__ #define __LNET_TYPES_H__ /** \addtogroup lnet * @{ */ #include /** \addtogroup lnet_addr * @{ */ #define LNET_VERSION "0.7.0" /** Portal reserved for LNet's own use. * \see lustre/include/lustre/lustre_idl.h for Lustre portal assignments. */ #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_t, 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) /** wildcard PID that matches any lnet_pid_t */ #define LNET_PID_ANY ((lnet_pid_t) -1) #define LNET_PID_RESERVED 0xf0000000 /* reserved bits in PID */ #define LNET_PID_USERFLAG 0x80000000 /* set in userspace peers */ #define LNET_PID_LUSTRE 12345 #define LNET_TIME_FOREVER (-1) /* how an LNET NID encodes net:address */ /** extract the address part of an lnet_nid_t */ static inline __u32 LNET_NIDADDR(lnet_nid_t nid) { return nid & 0xffffffff; } static inline __u32 LNET_NIDNET(lnet_nid_t nid) { 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; } static inline __u32 LNET_MKNET(__u32 type, __u32 num) { return (type << 16) | num; } #define WIRE_ATTR __attribute__((packed)) /* Packed version of lnet_process_id_t to transfer via network */ typedef struct { lnet_nid_t nid; lnet_pid_t pid; /* node id / process id */ } WIRE_ATTR lnet_process_id_packed_t; /* 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). */ typedef struct { __u64 wh_interface_cookie; __u64 wh_object_cookie; } WIRE_ATTR lnet_handle_wire_t; typedef enum { LNET_MSG_ACK = 0, LNET_MSG_PUT, LNET_MSG_GET, LNET_MSG_REPLY, LNET_MSG_HELLO, } lnet_msg_type_t; /* 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. */ typedef struct lnet_ack { lnet_handle_wire_t dst_wmd; __u64 match_bits; __u32 mlength; } WIRE_ATTR lnet_ack_t; typedef struct lnet_put { lnet_handle_wire_t ack_wmd; __u64 match_bits; __u64 hdr_data; __u32 ptl_index; __u32 offset; } WIRE_ATTR lnet_put_t; typedef struct lnet_get { lnet_handle_wire_t return_wmd; __u64 match_bits; __u32 ptl_index; __u32 src_offset; __u32 sink_length; } WIRE_ATTR lnet_get_t; typedef struct lnet_reply { lnet_handle_wire_t dst_wmd; } WIRE_ATTR lnet_reply_t; typedef struct lnet_hello { __u64 incarnation; __u32 type; } WIRE_ATTR lnet_hello_t; typedef struct { lnet_nid_t dest_nid; lnet_nid_t src_nid; lnet_pid_t dest_pid; lnet_pid_t src_pid; __u32 type; /* lnet_msg_type_t */ __u32 payload_length; /* payload data to follow */ /*<------__u64 aligned------->*/ union { lnet_ack_t ack; lnet_put_t put; lnet_get_t get; lnet_reply_t reply; lnet_hello_t hello; } msg; } WIRE_ATTR lnet_hdr_t; /* 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_t::msg. */ typedef struct { __u32 magic; /* LNET_PROTO_TCP_MAGIC */ __u16 version_major; /* increment on incompatible change */ __u16 version_minor; /* increment on compatible change */ } WIRE_ATTR lnet_magicversion_t; /* 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 */ typedef struct { __u32 acr_magic; /* PTL_ACCEPTOR_PROTO_MAGIC */ __u32 acr_version; /* protocol version */ __u64 acr_nid; /* target NID */ } WIRE_ATTR lnet_acceptor_connreq_t; #define LNET_PROTO_ACCEPTOR_VERSION 1 typedef struct lnet_counters { __u32 msgs_alloc; __u32 msgs_max; __u32 errors; __u32 send_count; __u32 recv_count; __u32 route_count; __u32 drop_count; __u64 send_length; __u64 recv_length; __u64 route_length; __u64 drop_length; } WIRE_ATTR lnet_counters_t; #define LNET_NI_STATUS_UP 0x15aac0de #define LNET_NI_STATUS_DOWN 0xdeadface #define LNET_NI_STATUS_INVALID 0x00000000 #define LNET_MAX_INTERFACES 16 /** * Objects maintained by the LNet are accessed through handles. Handle types * have names of the form lnet_handle_xx_t, where xx is one of the two letter * object type codes ('eq' for event queue, 'md' for memory descriptor, and * 'me' for match entry). * Each type of object is given a unique handle type to enhance type checking. * The type lnet_handle_any_t can be used when a generic handle is needed. * Every handle value can be converted into a value of type lnet_handle_any_t * without loss of information. */ typedef struct { __u64 cookie; } lnet_handle_any_t; typedef lnet_handle_any_t lnet_handle_eq_t; typedef lnet_handle_any_t lnet_handle_md_t; typedef lnet_handle_any_t lnet_handle_me_t; #define LNET_WIRE_HANDLE_COOKIE_NONE (-1) /** * Invalidate handle \a h. */ static inline void LNetInvalidateHandle(lnet_handle_any_t *h) { h->cookie = LNET_WIRE_HANDLE_COOKIE_NONE; } /** * Compare handles \a h1 and \a h2. * * \return 1 if handles are equal, 0 if otherwise. */ static inline int LNetHandleIsEqual (lnet_handle_any_t h1, lnet_handle_any_t h2) { return (h1.cookie == h2.cookie); } /** * Check whether handle \a h is invalid. * * \return 1 if handle is invalid, 0 if valid. */ static inline int LNetHandleIsInvalid(lnet_handle_any_t h) { return (LNET_WIRE_HANDLE_COOKIE_NONE == h.cookie); } /** * Global process ID. */ typedef struct { /** node id */ lnet_nid_t nid; /** process id */ lnet_pid_t pid; } lnet_process_id_t; /** @} lnet_addr */ /** \addtogroup lnet_me * @{ */ /** * Specifies whether the match entry or memory descriptor should be unlinked * automatically (LNET_UNLINK) or not (LNET_RETAIN). */ typedef enum { LNET_RETAIN = 0, LNET_UNLINK } lnet_unlink_t; /** * Values of the type lnet_ins_pos_t are used to control where a new match * entry is inserted. The value LNET_INS_BEFORE is used to insert the new * entry before the current entry or before the head of the list. The value * LNET_INS_AFTER is used to insert the new entry after the current entry * or after the last item in the list. */ typedef enum { /** insert ME before current position or head of the list */ LNET_INS_BEFORE, /** insert ME after current position or tail of the list */ LNET_INS_AFTER, /** attach ME at tail of local CPU partition ME list */ LNET_INS_LOCAL } lnet_ins_pos_t; /** @} lnet_me */ /** \addtogroup lnet_md * @{ */ /** * Defines the visible parts of a memory descriptor. Values of this type * are used to initialize memory descriptors. */ typedef struct { /** * Specify the memory region associated with the memory descriptor. * If the options field has: * - LNET_MD_KIOV bit set: The start field points to the starting * address of an array of lnet_kiov_t and the length field specifies * the number of entries in the array. The length can't be bigger * than LNET_MAX_IOV. The lnet_kiov_t 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. * * When the memory region is fragmented, all fragments but the first * one must start on page boundary, and all but the last must end on * page boundary. */ void *start; unsigned int length; /** * Specifies the maximum number of operations that can be performed * on the memory descriptor. An operation is any action that could * possibly generate an event. In the usual case, the threshold value * is decremented for each operation on the MD. When the threshold * drops to zero, the MD becomes inactive and does not respond to * operations. A threshold value of LNET_MD_THRESH_INF indicates that * there is no bound on the number of operations that may be applied * to a MD. */ int threshold; /** * Specifies the largest incoming request that the memory descriptor * should respond to. When the unused portion of a MD (length - * local offset) falls below this value, the MD becomes inactive and * does not respond to further operations. This value is only used * if the LNET_MD_MAX_SIZE option is set. */ int max_size; /** * Specifies the behavior of the memory descriptor. A bitwise OR * of the following values can be used: * - LNET_MD_OP_PUT: The LNet PUT operation is allowed on this MD. * - LNET_MD_OP_GET: The LNet GET operation is allowed on this MD. * - LNET_MD_MANAGE_REMOTE: The offset used in accessing the memory * region is provided by the incoming request. By default, the * offset is maintained locally. When maintained locally, the * offset is incremented by the length of the request so that * the next operation (PUT or GET) will access the next part of * the memory region. Note that only one offset variable exists * per memory descriptor. If both PUT and GET operations are * performed on a memory descriptor, the offset is updated each time. * - LNET_MD_TRUNCATE: The length provided in the incoming request can * be reduced to match the memory available in the region (determined * by subtracting the offset from the length of the memory region). * By default, if the length in the incoming operation is greater * than the amount of memory available, the operation is rejected. * - LNET_MD_ACK_DISABLE: An acknowledgment should not be sent for * incoming PUT operations, even if requested. By default, * acknowledgments are sent for PUT operations that request an * acknowledgment. Acknowledgments are never sent for GET operations. * The data sent in the REPLY serves as an implicit acknowledgment. * - LNET_MD_KIOV: The start and length fields specify an array of * lnet_kiov_t. * - LNET_MD_IOVEC: The start and length fields specify an array of * struct iovec. * - LNET_MD_MAX_SIZE: The max_size field is valid. * * Note: * - LNET_MD_KIOV or LNET_MD_IOVEC allows for a scatter/gather * capability for memory descriptors. They can't be both set. * - 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. */ unsigned int options; /** * A user-specified value that is associated with the memory * descriptor. The value does not need to be a pointer, but must fit * in the space used by a pointer. This value is recorded in events * associated with operations on this MD. */ 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 (i.e. nullified * by LNetInvalidateHandle()), operations performed on this memory * descriptor are not logged. */ lnet_handle_eq_t eq_handle; } lnet_md_t; /* Max Transfer Unit (minimum supported everywhere). * CAVEAT EMPTOR, with multinet (i.e. routers forwarding between networks) * these limits are system wide and not interface-local. */ #define LNET_MTU_BITS 20 #define LNET_MTU (1 << LNET_MTU_BITS) /** limit on the number of fragments in discontiguous MDs */ #define LNET_MAX_IOV 256 /** * Options for the MD structure. See lnet_md_t::options. */ #define LNET_MD_OP_PUT (1 << 0) /** See lnet_md_t::options. */ #define LNET_MD_OP_GET (1 << 1) /** See lnet_md_t::options. */ #define LNET_MD_MANAGE_REMOTE (1 << 2) /* unused (1 << 3) */ /** See lnet_md_t::options. */ #define LNET_MD_TRUNCATE (1 << 4) /** See lnet_md_t::options. */ #define LNET_MD_ACK_DISABLE (1 << 5) /** See lnet_md_t::options. */ #define LNET_MD_IOVEC (1 << 6) /** See lnet_md_t::options. */ #define LNET_MD_MAX_SIZE (1 << 7) /** See lnet_md_t::options. */ #define LNET_MD_KIOV (1 << 8) /* For compatibility with Cray Portals */ #define LNET_MD_PHYS 0 /** Infinite threshold on MD operations. See lnet_md_t::threshold */ #define LNET_MD_THRESH_INF (-1) /* NB lustre portals uses struct iovec internally! */ typedef struct iovec lnet_md_iovec_t; /** * A page-based fragment of a MD. */ typedef struct { /** Pointer to the page where the fragment resides */ struct page *kiov_page; /** Length in bytes of the fragment */ unsigned int kiov_len; /** * Starting offset of the fragment within the page. Note that the * end of the fragment must not pass the end of the page; i.e., * kiov_len + kiov_offset <= PAGE_CACHE_SIZE. */ unsigned int kiov_offset; } lnet_kiov_t; /** @} lnet_md */ /** \addtogroup lnet_eq * @{ */ /** * Six types of events can be logged in an event queue. */ typedef enum { /** An incoming GET operation has completed on the MD. */ LNET_EVENT_GET = 1, /** * An incoming PUT operation has completed on the MD. The * underlying layers will not alter the memory (on behalf of this * operation) once this event has been logged. */ LNET_EVENT_PUT, /** * A REPLY operation has completed. This event is logged after the * data (if any) from the REPLY has been written into the MD. */ LNET_EVENT_REPLY, /** An acknowledgment has been received. */ LNET_EVENT_ACK, /** * An outgoing send (PUT or GET) operation has completed. This event * is logged after the entire buffer has been sent and it is safe for * the caller to reuse the buffer. * * Note: * - The LNET_EVENT_SEND doesn't guarantee message delivery. It can * happen even when the message has not yet been put out on wire. * - It's unsafe to assume that in an outgoing GET operation * the LNET_EVENT_SEND event would happen before the * LNET_EVENT_REPLY event. The same holds for LNET_EVENT_SEND and * LNET_EVENT_ACK events in an outgoing PUT operation. */ LNET_EVENT_SEND, /** * A MD has been unlinked. Note that LNetMDUnlink() does not * necessarily trigger an LNET_EVENT_UNLINK event. * \see LNetMDUnlink */ LNET_EVENT_UNLINK, } lnet_event_kind_t; #define LNET_SEQ_BASETYPE long typedef unsigned LNET_SEQ_BASETYPE lnet_seq_t; #define LNET_SEQ_GT(a,b) (((signed LNET_SEQ_BASETYPE)((a) - (b))) > 0) /** * Information about an event on a MD. */ typedef struct { /** The identifier (nid, pid) of the target. */ lnet_process_id_t target; /** The identifier (nid, pid) of the initiator. */ lnet_process_id_t initiator; /** * 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 initiator. */ lnet_nid_t sender; /** Indicates the type of the event. */ lnet_event_kind_t type; /** The portal table index specified in the request */ unsigned int pt_index; /** A copy of the match bits specified in the request. */ __u64 match_bits; /** The length (in bytes) specified in the request. */ unsigned int rlength; /** * The length (in bytes) of the data that was manipulated by the * operation. For truncated operations, the manipulated length will be * the number of bytes specified by the MD (possibly with an offset, * see lnet_md_t). For all other operations, the manipulated length * will be the length of the requested operation, i.e. rlength. */ unsigned int mlength; /** * The handle to the MD associated with the event. The handle may be * invalid if the MD has been unlinked. */ lnet_handle_md_t 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. */ lnet_md_t md; /** * 64 bits of out-of-band user data. Only valid for LNET_EVENT_PUT. * \see LNetPut */ __u64 hdr_data; /** * Indicates the completion status of the operation. It's 0 for * successful operations, otherwise it's an error code. */ 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; /** * 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 lnet_md_t::options */ unsigned int offset; /** * The sequence number for this event. Sequence numbers are unique * to each event. */ volatile lnet_seq_t sequence; } lnet_event_t; /** * 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)(lnet_event_t *event); #define LNET_EQ_HANDLER_NONE NULL /** @} lnet_eq */ /** \addtogroup lnet_data * @{ */ /** * Specify whether an acknowledgment should be sent by target when the PUT * operation completes (i.e., when the data has been written to a MD of the * target process). * * \see lnet_md_t::options for the discussion on LNET_MD_ACK_DISABLE by which * acknowledgments can be disabled for a MD. */ typedef enum { /** Request an acknowledgment */ LNET_ACK_REQ, /** Request that no acknowledgment should be generated. */ LNET_NOACK_REQ } lnet_ack_req_t; /** @} lnet_data */ /** @} lnet */ #endif