/*
* Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
* Use is subject to license terms.
+ *
+ * Copyright (c) 2011, 2012, Whamcloud, Inc.
*/
/*
* This file is part of Lustre, http://www.lustre.org/
#define MDT_MAX_THREADS 512UL
#endif
#define MDS_NBUFS (64 * cfs_num_online_cpus())
-#define MDS_BUFSIZE (8 * 1024)
/**
* Assume file name length = FNAME_MAX = 256 (true for ext3).
* path name length = PATH_MAX = 4096
- * LOV MD size max = EA_MAX = 4000
+ * LOV MD size max = EA_MAX = 48000 (2000 stripes)
* symlink: FNAME_MAX + PATH_MAX <- largest
* link: FNAME_MAX + PATH_MAX (mds_rec_link < mds_rec_create)
* rename: FNAME_MAX + FNAME_MAX
* open: FNAME_MAX + EA_MAX
*
* MDS_MAXREQSIZE ~= 4736 bytes =
- * lustre_msg + ldlm_request + mds_body + mds_rec_create + FNAME_MAX + PATH_MAX
+ * lustre_msg + ldlm_request + mdt_body + mds_rec_create + FNAME_MAX + PATH_MAX
* MDS_MAXREPSIZE ~= 8300 bytes = lustre_msg + llog_header
* or, for mds_close() and mds_reint_unlink() on a many-OST filesystem:
- * = 9210 bytes = lustre_msg + mds_body + 160 * (easize + cookiesize)
+ * = 9210 bytes = lustre_msg + mdt_body + 160 * (easize + cookiesize)
*
* Realistic size is about 512 bytes (20 character name + 128 char symlink),
* except in the open case where there are a large number of OSTs in a LOV.
*/
-#define MDS_MAXREQSIZE (5 * 1024)
-#define MDS_MAXREPSIZE max(9 * 1024, 362 + LOV_MAX_STRIPE_COUNT * 56)
+#define MDS_MAXREPSIZE max(10 * 1024, 362 + LOV_MAX_STRIPE_COUNT * 56)
+#define MDS_MAXREQSIZE MDS_MAXREPSIZE
+
+/** MDS_BUFSIZE = max_reqsize + max sptlrpc payload size */
+#define MDS_BUFSIZE (MDS_MAXREQSIZE + 1024)
-/** FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + md_fld */
+/** FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc */
#define FLD_MAXREQSIZE (160)
-/** FLD_MAXREPSIZE == lustre_msg + ptlrpc_body + md_fld */
+/** FLD_MAXREPSIZE == lustre_msg + ptlrpc_body */
#define FLD_MAXREPSIZE (152)
/**
* returned.
*/
struct ptlrpc_request_set {
+ cfs_atomic_t set_refcount;
+ /** number of in queue requests */
+ cfs_atomic_t set_new_count;
/** number of uncompleted requests */
cfs_atomic_t set_remaining;
/** wait queue to wait on for request events */
* Check if the request is a high priority one.
*/
int (*hpreq_check)(struct ptlrpc_request *);
+ /**
+ * Called after the request has been handled.
+ */
+ void (*hpreq_fini)(struct ptlrpc_request *);
};
/**
rq_reply_truncate:1,
rq_committed:1,
/* whether the "rq_set" is a valid one */
- rq_invalid_rqset:1;
+ rq_invalid_rqset:1,
+ rq_generation_set:1;
enum rq_phase rq_phase; /* one of RQ_PHASE_* */
enum rq_phase rq_next_phase; /* one of RQ_PHASE_* to be used next */
#define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s%s"
-void _debug_req(struct ptlrpc_request *req, __u32 mask,
+void _debug_req(struct ptlrpc_request *req,
struct libcfs_debug_msg_data *data, const char *fmt, ...)
- __attribute__ ((format (printf, 4, 5)));
+ __attribute__ ((format (printf, 3, 4)));
/**
* Helper that decides if we need to print request accordig to current debug
* level settings
*/
-#define debug_req(cdls, level, req, file, func, line, fmt, a...) \
+#define debug_req(msgdata, mask, cdls, req, fmt, a...) \
do { \
- CFS_CHECK_STACK(); \
+ CFS_CHECK_STACK(msgdata, mask, cdls); \
\
- if (((level) & D_CANTMASK) != 0 || \
- ((libcfs_debug & (level)) != 0 && \
- (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
- static struct libcfs_debug_msg_data _req_dbg_data = \
- DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, file, func, line); \
- _debug_req((req), (level), &_req_dbg_data, fmt, ##a); \
- } \
+ if (((mask) & D_CANTMASK) != 0 || \
+ ((libcfs_debug & (mask)) != 0 && \
+ (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
+ _debug_req((req), msgdata, fmt, ##a); \
} while(0)
/**
do { \
if ((level) & (D_ERROR | D_WARNING)) { \
static cfs_debug_limit_state_t cdls; \
- debug_req(&cdls, level, req, __FILE__, __func__, __LINE__, \
- "@@@ "fmt" ", ## args); \
- } else \
- debug_req(NULL, level, req, __FILE__, __func__, __LINE__, \
- "@@@ "fmt" ", ## args); \
+ LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, &cdls); \
+ debug_req(&msgdata, level, &cdls, req, "@@@ "fmt" ", ## args);\
+ } else { \
+ LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, level, NULL); \
+ debug_req(&msgdata, level, NULL, req, "@@@ "fmt" ", ## args); \
+ } \
} while (0)
/** @} */
struct lu_env *t_env;
};
+static inline int thread_is_stopped(struct ptlrpc_thread *thread)
+{
+ return !!(thread->t_flags & SVC_STOPPED);
+}
+
+static inline int thread_is_stopping(struct ptlrpc_thread *thread)
+{
+ return !!(thread->t_flags & SVC_STOPPING);
+}
+
+static inline int thread_is_starting(struct ptlrpc_thread *thread)
+{
+ return !!(thread->t_flags & SVC_STARTING);
+}
+
+static inline int thread_is_running(struct ptlrpc_thread *thread)
+{
+ return !!(thread->t_flags & SVC_RUNNING);
+}
+
+static inline int thread_is_event(struct ptlrpc_thread *thread)
+{
+ return !!(thread->t_flags & SVC_EVENT);
+}
+
+static inline int thread_is_signal(struct ptlrpc_thread *thread)
+{
+ return !!(thread->t_flags & SVC_SIGNAL);
+}
+
+static inline void thread_clear_flags(struct ptlrpc_thread *thread, __u32 flags)
+{
+ thread->t_flags &= ~flags;
+}
+
+static inline void thread_set_flags(struct ptlrpc_thread *thread, __u32 flags)
+{
+ thread->t_flags = flags;
+}
+
+static inline void thread_add_flags(struct ptlrpc_thread *thread, __u32 flags)
+{
+ thread->t_flags |= flags;
+}
+
+static inline int thread_test_and_clear_flags(struct ptlrpc_thread *thread,
+ __u32 flags)
+{
+ if (thread->t_flags & flags) {
+ thread->t_flags &= ~flags;
+ return 1;
+ }
+ return 0;
+}
+
/**
* Request buffer descriptor structure.
* This is a structure that contains one posted request buffer for service.
* Environment for request interpreters to run in.
*/
struct lu_env pc_env;
+ /**
+ * Index of ptlrpcd thread in the array.
+ */
+ int pc_index;
+ /**
+ * Number of the ptlrpcd's partners.
+ */
+ int pc_npartners;
+ /**
+ * Pointer to the array of partners' ptlrpcd_ctl structure.
+ */
+ struct ptlrpcd_ctl **pc_partners;
+ /**
+ * Record the partner index to be processed next.
+ */
+ int pc_cursor;
#ifndef __KERNEL__
/**
* Async rpcs flag to make sure that ptlrpcd_check() is called only
/**
* This is a recovery ptlrpc thread.
*/
- LIOD_RECOVERY = 1 << 3
+ LIOD_RECOVERY = 1 << 3,
+ /**
+ * The ptlrpcd is bound to some CPU core.
+ */
+ LIOD_BIND = 1 << 4,
};
/* ptlrpc/events.c */
* underlying buffer
* @{
*/
-extern void request_out_callback (lnet_event_t *ev);
+extern void request_out_callback(lnet_event_t *ev);
extern void reply_in_callback(lnet_event_t *ev);
-extern void client_bulk_callback (lnet_event_t *ev);
+extern void client_bulk_callback(lnet_event_t *ev);
extern void request_in_callback(lnet_event_t *ev);
extern void reply_out_callback(lnet_event_t *ev);
-extern void server_bulk_callback (lnet_event_t *ev);
+#ifdef HAVE_SERVER_SUPPORT
+extern void server_bulk_callback(lnet_event_t *ev);
+#endif
/** @} */
/* ptlrpc/connection.c */
* Actual interfacing with LNet to put/get/register/unregister stuff
* @{
*/
+#ifdef HAVE_SERVER_SUPPORT
+struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
+ int npages, int type, int portal);
int ptlrpc_start_bulk_transfer(struct ptlrpc_bulk_desc *desc);
void ptlrpc_abort_bulk(struct ptlrpc_bulk_desc *desc);
-int ptlrpc_register_bulk(struct ptlrpc_request *req);
-int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
{
cfs_spin_unlock(&desc->bd_lock);
return rc;
}
+#endif
+
+int ptlrpc_register_bulk(struct ptlrpc_request *req);
+int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
{
void ptlrpc_resend_req(struct ptlrpc_request *request);
int ptlrpc_at_get_net_latency(struct ptlrpc_request *req);
int ptl_send_rpc(struct ptlrpc_request *request, int noreply);
-int ptlrpc_register_rqbd (struct ptlrpc_request_buffer_desc *rqbd);
+int ptlrpc_register_rqbd(struct ptlrpc_request_buffer_desc *rqbd);
/** @} */
/* ptlrpc/client.c */
void ptlrpc_mark_interrupted(struct ptlrpc_request *req);
void ptlrpc_set_destroy(struct ptlrpc_request_set *);
void ptlrpc_set_add_req(struct ptlrpc_request_set *, struct ptlrpc_request *);
-int ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
- struct ptlrpc_request *req);
+void ptlrpc_set_add_new_req(struct ptlrpcd_ctl *pc,
+ struct ptlrpc_request *req);
void ptlrpc_free_rq_pool(struct ptlrpc_request_pool *pool);
void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
void ptlrpc_req_finished(struct ptlrpc_request *request);
void ptlrpc_req_finished_with_imp_lock(struct ptlrpc_request *request);
struct ptlrpc_request *ptlrpc_request_addref(struct ptlrpc_request *req);
-struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp (struct ptlrpc_request *req,
- int npages, int type, int portal);
-struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_exp(struct ptlrpc_request *req,
+struct ptlrpc_bulk_desc *ptlrpc_prep_bulk_imp(struct ptlrpc_request *req,
int npages, int type, int portal);
void ptlrpc_free_bulk(struct ptlrpc_bulk_desc *bulk);
void ptlrpc_prep_bulk_page(struct ptlrpc_bulk_desc *desc,
__u64 ptlrpc_sample_next_xid(void);
__u64 ptlrpc_req_xid(struct ptlrpc_request *request);
+/* Set of routines to run a function in ptlrpcd context */
+void *ptlrpcd_alloc_work(struct obd_import *imp,
+ int (*cb)(const struct lu_env *, void *), void *data);
+void ptlrpcd_destroy_work(void *handler);
+int ptlrpcd_queue_work(void *handler);
+
/** @} */
struct ptlrpc_service_conf {
*
* @{
*/
-void ptlrpc_save_lock (struct ptlrpc_request *req,
- struct lustre_handle *lock, int mode, int no_ack);
+void ptlrpc_save_lock(struct ptlrpc_request *req,
+ struct lustre_handle *lock, int mode, int no_ack);
void ptlrpc_commit_replies(struct obd_export *exp);
-void ptlrpc_dispatch_difficult_reply (struct ptlrpc_reply_state *rs);
-void ptlrpc_schedule_difficult_reply (struct ptlrpc_reply_state *rs);
+void ptlrpc_dispatch_difficult_reply(struct ptlrpc_reply_state *rs);
+void ptlrpc_schedule_difficult_reply(struct ptlrpc_reply_state *rs);
struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
svc_handler_t h, char *name,
struct proc_dir_entry *proc_entry,
int ptlrpc_start_threads(struct ptlrpc_service *svc);
int ptlrpc_start_thread(struct ptlrpc_service *svc);
int ptlrpc_unregister_service(struct ptlrpc_service *service);
-int liblustre_check_services (void *arg);
+int liblustre_check_services(void *arg);
void ptlrpc_daemonize(char *name);
int ptlrpc_service_health_check(struct ptlrpc_service *);
void ptlrpc_hpreq_reorder(struct ptlrpc_request *req);
int ptlrpc_init_import(struct obd_import *imp);
int ptlrpc_disconnect_import(struct obd_import *imp, int noclose);
int ptlrpc_import_recovery_state_machine(struct obd_import *imp);
+void deuuidify(char *uuid, const char *prefix, char **uuid_start,
+ int *uuid_len);
/* ptlrpc/pack_generic.c */
int ptlrpc_reconnect_import(struct obd_import *imp);
int lustre_msg_buflen(struct lustre_msg *m, int n);
void lustre_msg_set_buflen(struct lustre_msg *m, int n, int len);
int lustre_msg_bufcount(struct lustre_msg *m);
-char *lustre_msg_string (struct lustre_msg *m, int n, int max_len);
+char *lustre_msg_string(struct lustre_msg *m, int n, int max_len);
__u32 lustre_msghdr_get_flags(struct lustre_msg *msg);
void lustre_msghdr_set_flags(struct lustre_msg *msg, __u32 flags);
__u32 lustre_msg_get_flags(struct lustre_msg *msg);
void client_destroy_import(struct obd_import *imp);
/** @} */
+#ifdef HAVE_SERVER_SUPPORT
int server_disconnect_export(struct obd_export *exp);
+#endif
/* ptlrpc/pinger.c */
/**
int ptlrpc_check_and_wait_suspend(struct ptlrpc_request *req);
/** @} */
-/* ptlrpc/ptlrpcd.c */
-
-/**
- * Ptlrpcd scope is a set of two threads: ptlrpcd-foo and ptlrpcd-foo-rcv,
- * these threads are used to asynchronously send requests queued with
- * ptlrpcd_add_req(req, PCSOPE_FOO), and to handle completion call-backs for
- * such requests. Multiple scopes are needed to avoid dead-locks.
- */
-enum ptlrpcd_scope {
- /** Scope of bulk read-write rpcs. */
- PSCOPE_BRW,
- /** Everything else. */
- PSCOPE_OTHER,
- PSCOPE_NR
-};
+/* ptlrpc daemon bind policy */
+typedef enum {
+ /* all ptlrpcd threads are free mode */
+ PDB_POLICY_NONE = 1,
+ /* all ptlrpcd threads are bound mode */
+ PDB_POLICY_FULL = 2,
+ /* <free1 bound1> <free2 bound2> ... <freeN boundN> */
+ PDB_POLICY_PAIR = 3,
+ /* <free1 bound1> <bound1 free2> ... <freeN boundN> <boundN free1>,
+ * means each ptlrpcd[X] has two partners: thread[X-1] and thread[X+1].
+ * If kernel supports NUMA, pthrpcd threads are binded and
+ * grouped by NUMA node */
+ PDB_POLICY_NEIGHBOR = 4,
+} pdb_policy_t;
+
+/* ptlrpc daemon load policy
+ * It is caller's duty to specify how to push the async RPC into some ptlrpcd
+ * queue, but it is not enforced, affected by "ptlrpcd_bind_policy". If it is
+ * "PDB_POLICY_FULL", then the RPC will be processed by the selected ptlrpcd,
+ * Otherwise, the RPC may be processed by the selected ptlrpcd or its partner,
+ * depends on which is scheduled firstly, to accelerate the RPC processing. */
+typedef enum {
+ /* on the same CPU core as the caller */
+ PDL_POLICY_SAME = 1,
+ /* within the same CPU partition, but not the same core as the caller */
+ PDL_POLICY_LOCAL = 2,
+ /* round-robin on all CPU cores, but not the same core as the caller */
+ PDL_POLICY_ROUND = 3,
+ /* the specified CPU core is preferred, but not enforced */
+ PDL_POLICY_PREFERRED = 4,
+} pdl_policy_t;
-int ptlrpcd_start(const char *name, struct ptlrpcd_ctl *pc);
+/* ptlrpc/ptlrpcd.c */
void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
void ptlrpcd_wake(struct ptlrpc_request *req);
-int ptlrpcd_add_req(struct ptlrpc_request *req, enum ptlrpcd_scope scope);
+void ptlrpcd_add_req(struct ptlrpc_request *req, pdl_policy_t policy, int idx);
void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
int ptlrpcd_addref(void);
void ptlrpcd_decref(void);