#ifndef LIB_LNET_CONFIG_API_H
#define LIB_LNET_CONFIG_API_H
+#include <errno.h>
#include <net/if.h>
+#include <stdbool.h>
+#include <sys/socket.h>
+#include <yaml.h>
+#include <netlink/netlink.h>
+#include <netlink/genl/genl.h>
+#include <netlink/genl/ctrl.h>
+
+#include <libcfs/util/ioctl.h>
#include <libcfs/util/string.h>
#include <linux/lnet/lnet-dlc.h>
#include <linux/lnet/nidstr.h>
#define LUSTRE_CFG_RC_MATCH -7
#define LUSTRE_CFG_RC_SKIP -8
#define LUSTRE_CFG_RC_LAST_ELEM -9
+#define LUSTRE_CFG_RC_MARSHAL_FAIL -10
+
+#define CONFIG_CMD "configure"
+#define UNCONFIG_CMD "unconfigure"
+#define ADD_CMD "add"
+#define DEL_CMD "del"
+#define SHOW_CMD "show"
+#define DBG_CMD "dbg"
+#define MANAGE_CMD "manage"
+
+#define MAX_NUM_IPS 128
+
+#define modparam_path "/sys/module/lnet/parameters/"
+#define o2ib_modparam_path "/sys/module/ko2iblnd/parameters/"
+#define gni_nid_path "/proc/cray_xt/"
+
+enum lnetctl_cmd {
+ LNETCTL_CONFIG_CMD = 1,
+ LNETCTL_UNCONFIG_CMD = 2,
+ LNETCTL_ADD_CMD = 3,
+ LNETCTL_DEL_CMD = 4,
+ LNETCTL_SHOW_CMD = 5,
+ LNETCTL_DBG_CMD = 6,
+ LNETCTL_MANAGE_CMD = 7,
+ LNETCTL_LAST_CMD
+};
+
+/*
+ * Max number of nids we'll configure for a single peer via a single DLC
+ * operation
+ */
+#define LNET_MAX_NIDS_PER_PEER 128
struct lnet_dlc_network_descr {
struct list_head network_on_rule;
struct cfs_expr_list *cpt_expr;
};
+/* This UDSP structures need to match the kernel space structures
+ * in order for the marshall and unmarshall functions to be the same.
+ */
+
+/* Net is described as a
+ * 1. net type
+ * 2. num range
+ */
+struct lnet_ud_net_descr {
+ __u32 udn_net_type;
+ struct list_head udn_net_num_range;
+};
+
+/* each NID range is defined as
+ * 1. net descriptor
+ * 2. address range descriptor
+ */
+struct lnet_ud_nid_descr {
+ struct lnet_ud_net_descr ud_net_id;
+ struct list_head ud_addr_range;
+};
+
+/* a UDSP rule can have up to three user defined NID descriptors
+ * - src: defines the local NID range for the rule
+ * - dst: defines the peer NID range for the rule
+ * - rte: defines the router NID range for the rule
+ *
+ * An action union defines the action to take when the rule
+ * is matched
+ */
+struct lnet_udsp {
+ struct list_head udsp_on_list;
+ __u32 udsp_idx;
+ struct lnet_ud_nid_descr udsp_src;
+ struct lnet_ud_nid_descr udsp_dst;
+ struct lnet_ud_nid_descr udsp_rte;
+ enum lnet_udsp_action_type udsp_action_type;
+ union {
+ __u32 udsp_priority;
+ } udsp_action;
+};
+
+/* This union is passed from lnetctl to fill the action union in udsp
+ * structure
+ * TODO: The idea here is if we add extra actions, ex: drop, it can be
+ * added to the union
+ */
+union lnet_udsp_action {
+ int udsp_priority;
+};
+
/* forward declaration of the cYAML structure. */
struct cYAML;
int lustre_lnet_config_ni_healthv(int value, bool all, char *ni_nid,
int seq_no, struct cYAML **err_rc);
+
+/* lustre_lnet_config_ni_conns_per_peer
+ * set the conns_per_peer value of the NI. Valid range is specific to
+ * network type.
+ *
+ * value: conns_per_peer value to set.
+ * all: true to set all local NIs to that value.
+ * ni_nid: NI NID to set its conns_per_peer value. 'all' parameter always
+ * takes precedence
+ * seq_no - sequence number of the request
+ * err_rc - [OUT] struct cYAML tree describing the error. Freed by
+ * caller
+ */
+int lustre_lnet_config_ni_conns_per_peer(int value, bool all, char *ni_nid,
+ int seq_no, struct cYAML **err_rc);
+
/*
* lustre_lnet_config_peer_ni_healthv
* set the health value of the peer NI. -1 resets the value to maximum.
int lustre_lnet_show_retry_count(int seq_no, struct cYAML **show_rc,
struct cYAML **err_rc);
+int lustre_lnet_show_lnd_timeout(int seq_no, struct cYAML **show_rc,
+ struct cYAML **err_rc);
+
int lustre_lnet_show_local_ni_recovq(int seq_no, struct cYAML **show_rc,
struct cYAML **err_rc);
int lustre_lnet_show_peer_ni_recovq(int seq_no, struct cYAML **show_rc,
struct cYAML **err_rc);
+int lustre_lnet_config_response_tracking(int count, int seq_no,
+ struct cYAML **err_rc);
+int lustre_lnet_show_response_tracking(int seq_no, struct cYAML **show_rc,
+ struct cYAML **err_rc);
+int lustre_lnet_config_recovery_limit(int val, int seq_no,
+ struct cYAML **err_rc);
+int lustre_lnet_show_recovery_limit(int seq_no, struct cYAML **show_rc,
+ struct cYAML **err_rc);
+int lustre_lnet_show_max_recovery_ping_interval(int seq_no,
+ struct cYAML **show_rc,
+ struct cYAML **err_rc);
/*
* lustre_lnet_config_max_intf
struct cYAML **err_rc);
/*
+ * lustre_lnet_calc_service_id
+ * Calculate the lustre service id to be used for qos
+ */
+int lustre_lnet_calc_service_id(__u64 *service_id);
+
+/*
+ * lustre_lnet_setup_mrrouting
+ * configure linux routing tables for tcp interfaces
+ *
+ * err_rc - [OUT] struct cYAML tree describing the error. Freed by
+ * caller
+ */
+int lustre_lnet_setup_mrrouting(struct cYAML **err_rc);
+
+/*
+ * lustre_lnet_calc_cpt_of_nid
+ * Return the cpt number of the NID provided
+ */
+int lustre_lnet_calc_cpt_of_nid(char *nidc, long int ncpts);
+
+/*
* lustre_lnet_config_discovery
* Enable or disable peer discovery. Peer discovery is enabled by default.
*
int seq_no, struct cYAML **err_rc);
/*
+ * lustre_lnet_config_max_recovery_ping_interval
+ * Set the maximum recovery ping interval.
+ *
+ * interval - interval value in seconds
+ * seq_no - sequence number of the request
+ * err_rc - [OUT] struct cYAML tree describing the error. Freed by caller
+ */
+int lustre_lnet_config_max_recovery_ping_interval(int interval, int seq_no,
+ struct cYAML **err_rc);
+
+/*
* lustre_lnet_show_routing
* Send down an IOCTL to dump buffers and routing status
* This function is used to dump buffers for all CPU partitions.
struct cYAML **err_rc);
/*
- * lustre_lnet_config_peer_nid
- * Add a peer nid to a peer with primary nid pnid. If no pnid is given
- * then the first nid in the nid list becomes the primary nid for
- * a newly created peer.
- * Otherwise if pnid is provided and it's unique then a new peer is
- * created with pnid as the primary NID and the nids in the nid list as
- * secondary nids.
- * If any of the peers nids provided in with exception to the pnid is
- * not unique the operation fails. Some peer nids might have already
- * been added. It's the role of the caller of this API to remove the
- * added NIDs if they wish.
- *
- * pnid - Primary NID of the peer
- * nid - list of nids to add
- * num_nids - number of nids in the nid array
- * mr - true if this peer is MR capable.
- * ip2nets - true if a list of nid expressions are given to configure
- * multiple peers
- * seq_no - sequence number of the command
+ * lustre_lnet_reset_stats
+ * Resets internal LNET statistics.
+ *
* err_rc - YAML strucutre of the resultant return code.
*/
-int lustre_lnet_config_peer_nid(char *pnid, char **nid, int num_nids,
- bool mr, bool ip2nets, int seq_no,
- struct cYAML **err_rc);
+int lustre_lnet_reset_stats(int seq_no, struct cYAML **err_rc);
/*
- * lustre_lnet_del_peer_nid
- * Delete the nids given in the nid list from the peer with primary NID
- * pnid. If pnid is NULL or it doesn't identify a peer the operation
- * fails and no change happens to the system.
+ * lustre_lnet_modify_peer
+ * Handle a peer config or delete operation.
+ *
+ * Config Operation:
+ * Add a peer NID to a peer with primary NID pnid.
+ * If the provided primary NID is unique, then a new peer is
+ * created with this primary NID, and the NIDs in the NID list are added as
+ * secondary NIDs to this new peer.
+ * If any of the NIDs in the NID list are not unique then the operation
+ * fails. Some peer NIDs might have already been added. It's the responsibility
+ * of the caller of this API to remove the added NIDs if so desired.
+ *
+ * Delete Operation:
+ * Delete the NIDs given in the NID list from the peer with the primary NID
+ * pnid. If pnid is NULL, or it doesn't identify a peer, the operation fails,
+ * and no change happens to the system.
* The operation is aborted on the first NID that fails to be deleted.
*
- * pnid - Primary NID of the peer
- * nid - list of nids to add
- * num_nids - number of nids in the nid array
- * ip2nets - used to specify a range of nids
- * seq_no - sequence number of the command
- * err_rc - YAML strucutre of the resultant return code.
+ * prim_nid - The desired primary NID of a new peer, or the primary NID of
+ * an existing peer.
+ * nids - a comma separated string of nids
+ * is_mr - Specifies whether this peer is MR capable.
+ * cmd - CONFIG or DELETE
+ * seq_no - sequence number of the command
+ * err_rc - YAML structure of the resultant return code
*/
-int lustre_lnet_del_peer_nid(char *pnid, char **nid, int num_nids,
- bool ip2nets, int seq_no, struct cYAML **err_rc);
+int lustre_lnet_modify_peer(char *prim_nid, char *nids, bool is_mr,
+ int cmd, int seq_no, struct cYAML **err_rc);
/*
* lustre_lnet_show_peer
* Ping the nid list, pnids.
*
* pnids - NID list to ping.
+ * src_nidstr - source NID
* timeout - timeout(seconds) for ping.
* seq_no - sequence number of the command.
* show_rc - YAML structure of the resultant show.
* err_rc - YAML strucutre of the resultant return code.
*
*/
-int lustre_lnet_ping_nid(char *pnid, int timeout, int seq_no,
+int lustre_lnet_ping_nid(char *pnid, char *src_nidstr, int timeout, int seq_no,
struct cYAML **show_rc, struct cYAML **err_rc);
/* lustre_lnet_discover_nid
int lustre_yaml_exec(char *f, struct cYAML **show_rc,
struct cYAML **err_rc);
+/**
+ * yaml_emitter_set_output_netlink
+ *
+ * Special handling to integrate LNet handling into libyaml.
+ * This function sets up the ability to take the data stored in @emitter
+ * and formats into a netlink packet to send to the kernel.
+ *
+ * emitter - YAML emitter containing what the user specified
+ * nl - Netlink socket to be used by libyaml
+ * family - Netlink family
+ * version - notify kernel what version user land supports
+ * cmd - Netlink command to execute
+ * flags - Netlink flags
+ */
+int yaml_emitter_set_output_netlink(yaml_emitter_t *emitter, struct nl_sock *nl,
+ char *family, int version, int cmd,
+ int flags);
+
+/**
+ * yaml_parser_set_input_netlink
+ *
+ * Special handling to LNet handling into libyaml.
+ * This function sets up the ability to receive the Netlink data
+ * from the Linux kernel. That data is formated into a YAML document.
+ *
+ * parser - YAML parser that is used to present the data received
+ * from the kernel in Netlink format.
+ * nl - should be the Netlink socket receiving data from
+ * the kernel.
+ * stream - Handle the case of continuous data coming in.
+ */
+int yaml_parser_set_input_netlink(yaml_parser_t *parser, struct nl_sock *nl,
+ bool stream);
+
+/**
+ * yaml_parser_get_reader_proto_version
+ *
+ * This function retrieves the Neltink version the kernel module
+ * is set to.
+ *
+ * parser - YAML parser being used for Netlink communication.
+ */
+int yaml_parser_get_reader_proto_version(yaml_parser_t *parser);
+
+/**
+ * yaml_parser_get_reader_error
+ *
+ * By default libyaml reports a generic write error. We need a way
+ * to report better parser errors so we can track down problems.
+ *
+ * parser - YAML parser that has reported an error.
+ */
+const char *yaml_parser_get_reader_error(yaml_parser_t *parser);
+
+/**
+ * yaml_parser_log_error
+ *
+ * Helper function to report the parser error to @log.
+ *
+ * parser - YAML parser that has reported an error.
+ * log - file descriptor to write the error message out to.
+ * errmsg - Special extra string to append to error message.
+ */
+void yaml_parser_log_error(yaml_parser_t *parser, FILE *log,
+ const char *errmsg);
+
+/**
+ * yaml_emitter_log_error
+ *
+ * Helper function to report the emitter error to @log.
+ *
+ * emitter - YAML emitter that has reported an error.
+ * log - file descriptor to write the error message out to.
+ */
+void yaml_emitter_log_error(yaml_emitter_t *emitter, FILE *log);
+
+
/*
* lustre_lnet_init_nw_descr
* initialize the network descriptor structure for use
struct lnet_dlc_network_descr *nw_descr);
/*
- * lustre_lnet_parse_nids
- * Parse a set of nids into a locally allocated array and return the
- * pointer of the array to the caller. The caller is responsible for
- * freeing the array. If an initial array is provided then copy over
- * the contents of that array into the new array and append to it the
- * new content.
- * The nids can be of the form "nid [,nid, nid, nid]"
- * nids: nids string to be parsed
- * array: initial array of content
- * size: num of elements in the array
- * out_array: [OUT] new allocated array.
- * Returns size of array
- * sets the out_array to NULL on failure.
- */
-int lustre_lnet_parse_nids(char *nids, char **array, int size,
- char ***out_array);
+ * lustre_lnet_parse_nidstr
+ * This is a small wrapper around cfs_parse_nidlist.
+ * nidstr - A string parseable by cfs_parse_nidlist
+ * lnet_nidlist - An array of lnet_nid_t to hold the nids specified
+ * by the nidstring.
+ * max_nids - Size of the lnet_nidlist array, and the maximum number of
+ * nids that can be expressed by the nidstring. If the
+ * nidstring expands to a larger number of nids than max_nids
+ * then an error is returned.
+ * err_str - char pointer where we store an informative error
+ * message when an error is encountered
+ * Returns:
+ * The number (> 0) of lnet_nid_t stored in the supplied array, or
+ * LUSTRE_CFG_RC_BAD_PARAM if:
+ * - nidstr is NULL
+ * - nidstr contains an asterisk. This character is not allowed
+ * because it would cause the size of the expanded nidlist to exceed
+ * the maximum number of nids that is supported by expected callers
+ * of this function.
+ * - cfs_parse_nidlist fails to parse the nidstring
+ * - The nidlist populated by cfs_parse_nidlist is empty
+ * - The nidstring expands to a larger number of nids than max_nids
+ * - The nidstring expands to zero nids
+ * LUSTRE_CFG_RC_OUT_OF_MEM if:
+ * - cfs_expand_nidlist can return ENOMEM. We return out of mem in
+ * this case.
+ */
+int lustre_lnet_parse_nidstr(char *nidstr, lnet_nid_t *lnet_nidlist,
+ int max_nids, char *err_str);
+
+/* lustre_lnet_add_udsp
+ * Add a selection policy.
+ * src - source NID descriptor
+ * dst - destination NID descriptor
+ * rte - router NID descriptor
+ * type - action type
+ * action - union of the action
+ * idx - the index to delete
+ * seq_no - sequence number of the request
+ * err_rc - [OUT] struct cYAML tree describing the error. Freed by
+ * caller
+ */
+int lustre_lnet_add_udsp(char *src, char *dst, char *rte, char *type,
+ union lnet_udsp_action *action, int idx,
+ int seq_no, struct cYAML **err_rc);
+
+/* lustre_lnet_del_udsp
+ * Delete a net selection policy.
+ * idx - the index to delete
+ * seq_no - sequence number of the request
+ * err_rc - [OUT] struct cYAML tree describing the error. Freed by
+ * caller
+ */
+int lustre_lnet_del_udsp(unsigned int idx, int seq_no, struct cYAML **err_rc);
+
+/* lustre_lnet_show_udsp
+ * show selection policy.
+ * idx - the index to show. -1 to show all policies
+ * seq_no - sequence number of the request
+ * err_rc - [IN/OUT] struct cYAML tree containing udsp info
+ * err_rc - [OUT] struct cYAML tree describing the error. Freed by
+ * caller
+ */
+int lustre_lnet_show_udsp(int idx, int seq_no, struct cYAML **show_rc,
+ struct cYAML **err_rc);
+
+/* lustre_lnet_show_peer_debug_info
+ * dump peer debug info
+ * peer_nid - nid of the peer to show
+ * seq_no - sequence number of the request
+ * err_rc - [OUT] struct cYAML tree describing the error. Freed by
+ * caller
+ */
+int lustre_lnet_show_peer_debug_info(char *peer_nid, int seq_no,
+ struct cYAML **err_rc);
#endif /* LIB_LNET_CONFIG_API_H */