* considered full when less than ?_MAXREQSIZE is left in them.
*/
-#define LDLM_THREADS_AUTO_MIN \
- min((int)(num_online_cpus() * num_online_cpus() * 2), 8)
-#define LDLM_THREADS_AUTO_MAX (LDLM_THREADS_AUTO_MIN * 16)
+#define LDLM_THREADS_AUTO_MIN (2)
+#define LDLM_THREADS_AUTO_MAX min(num_online_cpus()*num_online_cpus()*32, 128)
#define LDLM_BL_THREADS LDLM_THREADS_AUTO_MIN
#define LDLM_NBUFS (64 * num_online_cpus())
#define LDLM_BUFSIZE (8 * 1024)
* 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, 280 + LOV_MAX_STRIPE_COUNT * 56)
+#define MDS_MAXREPSIZE max(9 * 1024, 362 + LOV_MAX_STRIPE_COUNT * 56)
/* FLD_MAXREQSIZE == lustre_msg + __u32 padding + ptlrpc_body + opc + md_fld */
#define FLD_MAXREQSIZE (160)
#define MGS_MAXREPSIZE (9 * 1024)
/* Absolute limits */
-#define OSS_THREADS_MIN 2
+#define OSS_THREADS_MIN 3 /* difficult replies, HPQ, others */
#define OSS_THREADS_MAX 512
#define OST_NBUFS (64 * num_online_cpus())
#define OST_BUFSIZE (8 * 1024)
#if RS_DEBUG
struct list_head rs_debug_list;
#endif
- /* updates to following flag serialised by srv_request_lock */
+ /* A spinlock to protect the reply state flags */
+ spinlock_t rs_lock;
+ /* Reply state flags */
unsigned long rs_difficult:1; /* ACK/commit stuff */
unsigned long rs_no_ack:1; /* no ACK, even for
difficult requests */
unsigned long rs_prealloc:1; /* rs from prealloc list */
int rs_size;
+ __u32 rs_opc;
__u64 rs_transno;
__u64 rs_xid;
struct obd_export *rs_export;
struct lu_context;
struct lu_env;
+struct ldlm_lock;
+
+struct ptlrpc_hpreq_ops {
+ /**
+ * Check if the lock handle of the given lock is the same as
+ * taken from the request.
+ */
+ int (*hpreq_lock_match)(struct ptlrpc_request *, struct ldlm_lock *);
+ /**
+ * Check if the request is a high priority one.
+ */
+ int (*hpreq_check)(struct ptlrpc_request *);
+};
+
/**
* Represents remote procedure call.
*/
struct list_head rq_list;
struct list_head rq_timed_list; /* server-side early replies */
struct list_head rq_history_list; /* server-side history */
+ struct list_head rq_exp_list; /* server-side per-export list */
+ struct ptlrpc_hpreq_ops *rq_ops; /* server-side hp handlers */
__u64 rq_history_seq; /* history sequence # */
+ /* the index of service's srv_at_array into which request is linked */
+ time_t rq_at_index;
int rq_status;
spinlock_t rq_lock;
/* client-side flags are serialized by rq_lock */
* after server commits corresponding transaction. This is
* used for operations that require sequence of multiple
* requests to be replayed. The only example currently is file
- * open/close/dw/setattr. When last request in such a sequence
- * is committed, ->rq_replay is cleared on all requests in the
+ * open/close. When last request in such a sequence is
+ * committed, ->rq_replay is cleared on all requests in the
* sequence.
*/
rq_replay:1,
- /* this is the last request in the sequence. */
- rq_sequence:1,
rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
rq_no_delay:1, rq_net_err:1, rq_wait_ctx:1,
rq_early:1, rq_must_unlink:1,
+ rq_fake:1, /* this fake req */
/* server-side flags */
rq_packed_final:1, /* packed final reply */
- rq_sent_final:1; /* stop sending early replies */
+ rq_sent_final:1, /* stop sending early replies */
+ rq_hp:1, /* high priority RPC */
+ rq_at_linked:1; /* link into service's srv_at_array */
enum rq_phase rq_phase; /* one of RQ_PHASE_* */
enum rq_phase rq_next_phase; /* one of RQ_PHASE_* to be used next */
__u64 rq_transno;
__u64 rq_xid;
struct list_head rq_replay_list;
- struct list_head rq_mod_list;
struct ptlrpc_cli_ctx *rq_cli_ctx; /* client's half ctx */
struct ptlrpc_svc_ctx *rq_svc_ctx; /* server's half ctx */
so that servers' early reply updates to the deadline aren't
kept in per-cpu cache */
time_t rq_reply_deadline; /* when req reply unlink must finish. */
+ time_t rq_bulk_deadline; /* when req bulk unlink must finish. */
int rq_timeout; /* service time estimate (secs) */
/* Multi-rpc bits */
struct req_capsule rq_pill;
};
-static inline void ptlrpc_close_replay_seq(struct ptlrpc_request *req)
+static inline int ptlrpc_req_interpret(const struct lu_env *env,
+ struct ptlrpc_request *req, int rc)
{
- spin_lock(&req->rq_lock);
- req->rq_replay = 0;
- req->rq_sequence = 1;
- spin_unlock(&req->rq_lock);
+ if (req->rq_interpret_reply != NULL) {
+ req->rq_status = req->rq_interpret_reply(env, req,
+ &req->rq_async_args,
+ rc);
+ return req->rq_status;
+ }
+ return rc;
}
static inline void lustre_set_req_swabbed(struct ptlrpc_request *req, int index)
FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
FLAG(req->rq_no_resend, "N"), \
FLAG(req->rq_waiting, "W"), \
- FLAG(req->rq_wait_ctx, "C")
+ FLAG(req->rq_wait_ctx, "C"), FLAG(req->rq_hp, "H")
-#define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s"
+#define REQ_FLAGS_FMT "%s:%s%s%s%s%s%s%s%s%s%s%s"
void _debug_req(struct ptlrpc_request *req, __u32 mask,
struct libcfs_debug_msg_data *data, const char *fmt, ...)
lnet_handle_md_t bd_md_h; /* associated MD */
lnet_nid_t bd_sender; /* stash event::sender */
- cfs_page_t **bd_enc_pages;
#if defined(__KERNEL__)
+ /*
+ * encrypt iov, size is either 0 or bd_iov_count.
+ */
+ lnet_kiov_t *bd_enc_iov;
+
lnet_kiov_t bd_iov[0];
#else
lnet_md_iovec_t bd_iov[0];
* put watchdog in the structure per thread b=14840
*/
struct lc_watchdog *t_watchdog;
+ /**
+ * the svc this thread belonged to b=18582
+ */
+ struct ptlrpc_service *t_svc;
cfs_waitq_t t_ctl_waitq;
struct lu_env *t_env;
};
typedef int (*svc_handler_t)(struct ptlrpc_request *req);
typedef void (*svcreq_printfn_t)(void *, struct ptlrpc_request *);
+typedef int (*svc_hpreq_handler_t)(struct ptlrpc_request *);
+
+#define PTLRPC_SVC_HP_RATIO 10
struct ptlrpc_service {
struct list_head srv_list; /* chain thru all services */
int srv_threads_max; /* thread upper limit */
int srv_threads_started; /* index of last started thread */
int srv_threads_running; /* # running threads */
- int srv_n_difficult_replies; /* # 'difficult' replies */
+ atomic_t srv_n_difficult_replies; /* # 'difficult' replies */
int srv_n_active_reqs; /* # reqs being served */
+ int srv_n_hpreq; /* # HPreqs being served */
cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
int srv_watchdog_factor; /* soft watchdog timeout mutiplier */
unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
/* AT stuff */
struct adaptive_timeout srv_at_estimate;/* estimated rpc service time */
spinlock_t srv_at_lock;
- struct list_head srv_at_list; /* reqs waiting for replies */
+ struct ptlrpc_at_array srv_at_array; /* reqs waiting for replies */
cfs_timer_t srv_at_timer; /* early reply timer */
int srv_n_queued_reqs; /* # reqs in either of the queues below */
+ int srv_hpreq_count; /* # hp requests handled */
+ int srv_hpreq_ratio; /* # hp per lp reqs to handle */
struct list_head srv_req_in_queue; /* incoming reqs */
struct list_head srv_request_queue; /* reqs waiting for service */
+ struct list_head srv_request_hpq; /* high priority queue */
struct list_head srv_request_history; /* request history */
__u64 srv_request_seq; /* next request sequence # */
atomic_t srv_outstanding_replies;
struct list_head srv_active_replies; /* all the active replies */
+#ifndef __KERNEL__
struct list_head srv_reply_queue; /* replies waiting for service */
-
+#endif
cfs_waitq_t srv_waitq; /* all threads sleep on this. This
* wait-queue is signalled when new
* incoming request arrives and when
struct list_head srv_threads; /* service thread list */
svc_handler_t srv_handler;
+ svc_hpreq_handler_t srv_hpreq_handler; /* hp request handler */
char *srv_name; /* only statically allocated strings here; we don't clean them */
char *srv_thread_name; /* only statically allocated strings here; we don't clean them */
struct ptlrpcd_ctl {
/**
- * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_STOP_FORCE)
+ * Ptlrpc thread control flags (LIOD_START, LIOD_STOP, LIOD_FORCE)
*/
unsigned long pc_flags;
/**
*/
LIOD_STOP = 1 << 1,
/**
- * Ptlrpc thread stop force flag. This will cause also
- * aborting any inflight rpcs handled by thread.
+ * Ptlrpc thread force flag (only stop force so far).
+ * This will cause aborting any inflight rpcs handled
+ * by thread if LIOD_STOP is specified.
*/
- LIOD_STOP_FORCE = 1 << 2,
+ LIOD_FORCE = 1 << 2,
/**
* This is a recovery ptlrpc thread.
*/
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);
-void ptlrpc_unregister_bulk (struct ptlrpc_request *req);
+int ptlrpc_unregister_bulk(struct ptlrpc_request *req, int async);
-static inline int ptlrpc_bulk_active (struct ptlrpc_bulk_desc *desc)
+static inline int ptlrpc_server_bulk_active(struct ptlrpc_bulk_desc *desc)
{
- int rc;
+ int rc;
+
+ LASSERT(desc != NULL);
spin_lock(&desc->bd_lock);
rc = desc->bd_network_rw;
spin_unlock(&desc->bd_lock);
- return (rc);
+ return rc;
+}
+
+static inline int ptlrpc_client_bulk_active(struct ptlrpc_request *req)
+{
+ struct ptlrpc_bulk_desc *desc = req->rq_bulk;
+ int rc;
+
+ LASSERT(req != NULL);
+
+ if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_BULK_UNLINK) &&
+ req->rq_bulk_deadline > cfs_time_current_sec())
+ return 1;
+
+ if (!desc)
+ return 0;
+
+ spin_lock(&desc->bd_lock);
+ rc = desc->bd_network_rw;
+ spin_unlock(&desc->bd_lock);
+ return rc;
}
#define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
int ptlrpc_unregister_reply(struct ptlrpc_request *req, int async);
void ptlrpc_restart_req(struct ptlrpc_request *req);
void ptlrpc_abort_inflight(struct obd_import *imp);
+void ptlrpc_cleanup_imp(struct obd_import *imp);
void ptlrpc_abort_set(struct ptlrpc_request_set *set);
struct ptlrpc_request_set *ptlrpc_prep_set(void);
int ptlrpc_request_bufs_pack(struct ptlrpc_request *request,
__u32 version, int opcode, char **bufs,
struct ptlrpc_cli_ctx *ctx);
+struct ptlrpc_request *ptlrpc_prep_fakereq(struct obd_import *imp,
+ unsigned int timeout,
+ ptlrpc_interpterer_t interpreter);
+void ptlrpc_fakereq_finished(struct ptlrpc_request *req);
+
struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
int opcode, int count, __u32 *lengths,
char **bufs);
/* ptlrpc/service.c */
void ptlrpc_save_lock (struct ptlrpc_request *req,
struct lustre_handle *lock, int mode, int no_ack);
-void ptlrpc_commit_replies (struct obd_device *obd);
+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);
struct ptlrpc_service *ptlrpc_init_svc_conf(struct ptlrpc_service_conf *c,
svc_handler_t h, char *name,
cfs_proc_dir_entry_t *proc_entry,
svcreq_printfn_t,
int min_threads, int max_threads,
- char *threadname, __u32 ctx_tags);
+ char *threadname, __u32 ctx_tags,
+ svc_hpreq_handler_t);
void ptlrpc_stop_all_threads(struct ptlrpc_service *svc);
int ptlrpc_start_threads(struct obd_device *dev, struct ptlrpc_service *svc);
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);
+#ifdef __KERNEL__
+int ptlrpc_hr_init(void);
+void ptlrpc_hr_fini(void);
+#else
+# define ptlrpc_hr_init() (0)
+# define ptlrpc_hr_fini() do {} while(0)
+#endif
struct ptlrpc_svc_data {
char *name;
__u32 lustre_msg_get_opc(struct lustre_msg *msg);
__u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
__u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
+__u64 *lustre_msg_get_versions(struct lustre_msg *msg);
__u64 lustre_msg_get_transno(struct lustre_msg *msg);
__u64 lustre_msg_get_slv(struct lustre_msg *msg);
__u32 lustre_msg_get_limit(struct lustre_msg *msg);
void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
void lustre_msg_set_last_xid(struct lustre_msg *msg, __u64 last_xid);
void lustre_msg_set_last_committed(struct lustre_msg *msg,__u64 last_committed);
+void lustre_msg_set_versions(struct lustre_msg *msg, __u64 *versions);
void lustre_msg_set_transno(struct lustre_msg *msg, __u64 transno);
void lustre_msg_set_status(struct lustre_msg *msg, __u32 status);
void lustre_msg_set_conn_cnt(struct lustre_msg *msg, __u32 conn_cnt);
{
if (req->rq_phase == new_phase)
return;
-
+
if (new_phase == RQ_PHASE_UNREGISTERING) {
req->rq_next_phase = req->rq_phase;
if (req->rq_import)
atomic_inc(&req->rq_import->imp_unregistering);
}
-
+
if (req->rq_phase == RQ_PHASE_UNREGISTERING) {
if (req->rq_import)
atomic_dec(&req->rq_import->imp_unregistering);
}
- DEBUG_REQ(D_RPCTRACE, req, "move req \"%s\" -> \"%s\"",
+ DEBUG_REQ(D_RPCTRACE, req, "move req \"%s\" -> \"%s\"",
ptlrpc_rqphase2str(req), ptlrpc_phase2str(new_phase));
req->rq_phase = new_phase;
static inline int
ptlrpc_client_early(struct ptlrpc_request *req)
{
- if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_UNLINK) &&
+ if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
req->rq_reply_deadline > cfs_time_current_sec())
return 0;
return req->rq_early;
static inline int
ptlrpc_client_replied(struct ptlrpc_request *req)
{
- if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_UNLINK) &&
+ if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
req->rq_reply_deadline > cfs_time_current_sec())
return 0;
return req->rq_replied;
static inline int
ptlrpc_client_recv(struct ptlrpc_request *req)
{
- if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_UNLINK) &&
+ if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
req->rq_reply_deadline > cfs_time_current_sec())
return 1;
return req->rq_receiving_reply;
int rc;
spin_lock(&req->rq_lock);
- if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_UNLINK) &&
+ if (OBD_FAIL_CHECK(OBD_FAIL_PTLRPC_LONG_REPL_UNLINK) &&
req->rq_reply_deadline > cfs_time_current_sec()) {
spin_unlock(&req->rq_lock);
return 1;
int client_obd_setup(struct obd_device *obddev, struct lustre_cfg *lcfg);
int client_obd_cleanup(struct obd_device *obddev);
int client_connect_import(const struct lu_env *env,
- struct lustre_handle *conn, struct obd_device *obd,
+ struct obd_export **exp, struct obd_device *obd,
struct obd_uuid *cluuid, struct obd_connect_data *,
void *localdata);
int client_disconnect_export(struct obd_export *exp);
int priority);
int client_import_del_conn(struct obd_import *imp, struct obd_uuid *uuid);
int import_set_conn_priority(struct obd_import *imp, struct obd_uuid *uuid);
+void client_destroy_import(struct obd_import *imp);
/* ptlrpc/pinger.c */
+enum timeout_event {
+ TIMEOUT_GRANT = 1
+};
+struct timeout_item;
+typedef int (*timeout_cb_t)(struct timeout_item *, void *);
int ptlrpc_pinger_add_import(struct obd_import *imp);
int ptlrpc_pinger_del_import(struct obd_import *imp);
+int ptlrpc_add_timeout_client(int time, enum timeout_event event,
+ timeout_cb_t cb, void *data,
+ struct list_head *obd_list);
+int ptlrpc_del_timeout_client(struct list_head *obd_list,
+ enum timeout_event event);
+struct ptlrpc_request * ptlrpc_prep_ping(struct obd_import *imp);
+int ptlrpc_obd_ping(struct obd_device *obd);
cfs_time_t ptlrpc_suspend_wakeup_time(void);
#ifdef __KERNEL__
void ping_evictor_start(void);
int ptlrpcd_start(const char *name, struct ptlrpcd_ctl *pc);
void ptlrpcd_stop(struct ptlrpcd_ctl *pc, int force);
void ptlrpcd_wake(struct ptlrpc_request *req);
-void ptlrpcd_add_req(struct ptlrpc_request *req, enum ptlrpcd_scope scope);
+int ptlrpcd_add_req(struct ptlrpc_request *req, enum ptlrpcd_scope scope);
+void ptlrpcd_add_rqset(struct ptlrpc_request_set *set);
int ptlrpcd_addref(void);
void ptlrpcd_decref(void);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff