* considered full when less than ?_MAXREQSIZE is left in them.
*/
-#define LDLM_THREADS_AUTO_MIN min((int)(smp_num_cpus * smp_num_cpus * 2), 8)
+#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_BL_THREADS LDLM_THREADS_AUTO_MIN
-#define LDLM_NBUFS (64 * smp_num_cpus)
+#define LDLM_NBUFS (64 * num_online_cpus())
#define LDLM_BUFSIZE (8 * 1024)
#define LDLM_MAXREQSIZE (5 * 1024)
#define LDLM_MAXREPSIZE (1024)
/* Absolute limits */
-#define MDS_THREADS_MIN 2UL
-#define MDS_THREADS_MAX 512UL
-/* Dynamic thread limits */
-#define MDS_THREADS_AUTO_MIN max(MDS_THREADS_MIN, min(32UL, \
- smp_num_cpus * num_physpages >> (27 - CFS_PAGE_SHIFT)))
-#define MDS_THREADS_AUTO_MAX min(MDS_THREADS_MAX, MDS_THREADS_AUTO_MIN * 4)
+#define MDS_THREADS_MIN 2
+#define MDS_THREADS_MAX 512
#define MDS_THREADS_MIN_READPAGE 2
-#define MDS_NBUFS (64 * smp_num_cpus)
+#define MDS_NBUFS (64 * num_online_cpus())
#define MDS_BUFSIZE (8 * 1024)
/* Assume file name length = FNAME_MAX = 256 (true for ext3).
* path name length = PATH_MAX = 4096
#define MDS_MAXREQSIZE (5 * 1024)
#define MDS_MAXREPSIZE max(9 * 1024, 280 + LOV_MAX_STRIPE_COUNT * 56)
-#define MGS_THREADS_AUTO_MAX 128UL
-#define MGS_THREADS_AUTO_MIN MDS_THREADS_AUTO_MIN
-#define MGS_NBUFS (64 * smp_num_cpus)
+#define MGS_THREADS_AUTO_MIN 2
+#define MGS_THREADS_AUTO_MAX 32
+#define MGS_NBUFS (64 * num_online_cpus())
#define MGS_BUFSIZE (8 * 1024)
#define MGS_MAXREQSIZE (8 * 1024)
#define MGS_MAXREPSIZE (9 * 1024)
/* Absolute limits */
-#define OSS_THREADS_MIN 2UL
-#define OSS_THREADS_MAX 512UL
-/* Dynamic thread limits */
-#define OSS_THREADS_AUTO_MIN max(OSS_THREADS_MIN, \
- smp_num_cpus * num_physpages >> (27 - CFS_PAGE_SHIFT))
-#define OSS_THREADS_AUTO_MAX min(OSS_THREADS_MAX, OSS_THREADS_AUTO_MIN * 4)
-#define OST_NBUFS (64 * smp_num_cpus)
+#define OSS_THREADS_MIN 2
+#define OSS_THREADS_MAX 512
+#define OST_NBUFS (64 * num_online_cpus())
#define OST_BUFSIZE (8 * 1024)
/* OST_MAXREQSIZE ~= 4768 bytes =
* lustre_msg + obdo + 16 * obd_ioobj + 256 * niobuf_remote
struct ptlrpc_connection {
struct list_head c_link;
+ struct hlist_node c_hash;
lnet_nid_t c_self;
lnet_process_id_t c_peer;
struct obd_uuid c_remote_uuid;
cfs_waitq_t set_waitq;
cfs_waitq_t *set_wakeup_ptr;
struct list_head set_requests;
+ struct list_head set_cblist; /* list of completion callbacks */
set_interpreter_func set_interpret; /* completion callback */
void *set_arg; /* completion context */
/* locked so that any old caller can communicate requests to
struct list_head set_new_requests;
};
+struct ptlrpc_set_cbdata {
+ struct list_head psc_item;
+ set_interpreter_func psc_interpret;
+ void *psc_data;
+};
+
struct ptlrpc_bulk_desc;
/*
struct list_head rs_debug_list;
#endif
/* updates to following flag serialised by srv_request_lock */
- unsigned int rs_difficult:1; /* ACK/commit stuff */
- unsigned int rs_scheduled:1; /* being handled? */
- unsigned int rs_scheduled_ever:1;/* any schedule attempts? */
- unsigned int rs_handled:1; /* been handled yet? */
- unsigned int rs_on_net:1; /* reply_out_callback pending? */
- unsigned int rs_prealloc:1; /* rs from prealloc list */
+ unsigned long rs_difficult:1; /* ACK/commit stuff */
+ unsigned long rs_scheduled:1; /* being handled? */
+ unsigned long rs_scheduled_ever:1;/* any schedule attempts? */
+ unsigned long rs_handled:1; /* been handled yet? */
+ unsigned long rs_on_net:1; /* reply_out_callback pending? */
+ unsigned long rs_prealloc:1; /* rs from prealloc list */
int rs_size;
__u64 rs_transno;
struct ptlrpc_request {
int rq_type; /* one of PTL_RPC_MSG_* */
struct list_head rq_list;
+ struct list_head rq_timed_list; /* server-side early replies */
struct list_head rq_history_list; /* server-side history */
__u64 rq_history_seq; /* history sequence # */
int rq_status;
spinlock_t rq_lock;
- /* client-side flags */
- unsigned int rq_intr:1, rq_replied:1, rq_err:1,
+ /* client-side flags are serialized by rq_lock */
+ unsigned long rq_intr:1, rq_replied:1, rq_err:1,
rq_timedout:1, rq_resend:1, rq_restart:1,
/*
* when ->rq_replay is set, request is kept by the client even
*/
rq_replay:1,
rq_no_resend:1, rq_waiting:1, rq_receiving_reply:1,
- rq_no_delay:1, rq_net_err:1;
+ rq_no_delay:1, rq_net_err:1, rq_early:1, rq_must_unlink:1,
+ /* server-side flags */
+ rq_packed_final:1, /* packed final reply */
+ rq_sent_final:1; /* stop sending early replies */
enum rq_phase rq_phase; /* one of RQ_PHASE_* */
- atomic_t rq_refcount; /* client-side refcount for SENT race */
+ atomic_t rq_refcount; /* client-side refcount for SENT race,
+ server-side refcounf for multiple replies */
struct ptlrpc_thread *rq_svc_thread; /* initial thread servicing req */
int rq_reqlen;
struct lustre_msg *rq_reqmsg;
- int rq_timeout; /* time to wait for reply (seconds) */
int rq_replen;
+ struct lustre_msg *rq_repbuf; /* client only, buf may be bigger than msg */
struct lustre_msg *rq_repmsg;
__u64 rq_transno;
__u64 rq_xid;
int rq_import_generation;
enum lustre_imp_state rq_send_state;
+ int rq_early_count; /* how many early replies (for stats) */
+
/* client+server request */
lnet_handle_md_t rq_req_md_h;
struct ptlrpc_cb_id rq_req_cbid;
void *rq_cb_data;
struct ptlrpc_bulk_desc *rq_bulk; /* client side bulk */
- time_t rq_sent; /* when request sent, seconds */
+ /* client outgoing req */
+ time_t rq_sent; /* when request sent, seconds,
+ * or time when request should
+ * be sent */
+ volatile time_t rq_deadline; /* when request must finish. volatile
+ so that servers' early reply updates to the deadline aren't
+ kept in per-cpu cache */
+ int rq_timeout; /* service time estimate (secs) */
/* Multi-rpc bits */
struct list_head rq_set_chain;
struct ptlrpc_request_set *rq_set;
void *rq_interpret_reply; /* Async completion handler */
union ptlrpc_async_args rq_async_args; /* Async completion context */
- void *rq_ptlrpcd_data;
struct ptlrpc_request_pool *rq_pool; /* Pool if request from
preallocated list */
};
return req->rq_rep_swab_mask & (1 << index);
}
-#define SWAB_PARANOIA 1
-
-#if SWAB_PARANOIA
-/* unpacking: assert idx not unpacked already */
-#define LASSERT_REQSWAB(rq, idx) lustre_set_req_swabbed(rq, idx)
-#define LASSERT_REPSWAB(rq, idx) lustre_set_rep_swabbed(rq, idx)
-
-/* just looking: assert idx already unpacked */
-#define LASSERT_REQSWABBED(rq, idx) LASSERT(lustre_req_swabbed(rq, idx))
-#define LASSERT_REPSWABBED(rq, idx) LASSERT(lustre_rep_swabbed(rq, idx))
-
-#else
-#define LASSERT_REQSWAB(rq, idx)
-#define LASSERT_REPSWAB(rq, idx)
-#define LASSERT_REQSWABBED(rq, idx)
-#define LASSERT_REPSWABBED(rq, idx)
-#endif
-
static inline const char *
ptlrpc_rqphase2str(struct ptlrpc_request *req)
{
/* Spare the preprocessor, spoil the bugs. */
#define FLAG(field, str) (field ? str : "")
-#define DEBUG_REQ_FLAGS(req) \
- ptlrpc_rqphase2str(req), \
- FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
- FLAG(req->rq_err, "E"), \
- FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
- FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
- FLAG(req->rq_no_resend, "N"), \
+#define DEBUG_REQ_FLAGS(req) \
+ ptlrpc_rqphase2str(req), \
+ FLAG(req->rq_intr, "I"), FLAG(req->rq_replied, "R"), \
+ FLAG(req->rq_err, "E"), \
+ FLAG(req->rq_timedout, "X") /* eXpired */, FLAG(req->rq_resend, "S"), \
+ FLAG(req->rq_restart, "T"), FLAG(req->rq_replay, "P"), \
+ FLAG(req->rq_no_resend, "N"), \
FLAG(req->rq_waiting, "W")
#define REQ_FLAGS_FMT "%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, ...);
-
-#define debug_req(cdls, level, req, file, func, line, fmt, a...) \
-do { \
- CHECK_STACK(); \
- \
- 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); \
- } \
+ struct libcfs_debug_msg_data *data, const char *fmt, ...)
+ __attribute__ ((format (printf, 4, 5)));
+
+#define debug_req(cdls, level, req, file, func, line, fmt, a...) \
+do { \
+ CHECK_STACK(); \
+ \
+ 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); \
+ } \
} while(0)
/* for most callers (level is a constant) this is resolved at compile time */
-#define DEBUG_REQ(level, req, fmt, args...) \
-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); \
+#define DEBUG_REQ(level, req, fmt, args...) \
+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); \
} while (0)
struct ptlrpc_bulk_page {
#define BULK_PUT_SOURCE 3
struct ptlrpc_bulk_desc {
- unsigned int bd_success:1; /* completed successfully */
- unsigned int bd_network_rw:1; /* accessible to the network */
- unsigned int bd_type:2; /* {put,get}{source,sink} */
- unsigned int bd_registered:1; /* client side */
+ unsigned long bd_success:1; /* completed successfully */
+ unsigned long bd_network_rw:1; /* accessible to the network */
+ unsigned long bd_type:2; /* {put,get}{source,sink} */
+ unsigned long bd_registered:1; /* client side */
spinlock_t bd_lock; /* serialise with callback */
int bd_import_generation;
struct obd_export *bd_export;
__u64 bd_last_xid;
struct ptlrpc_cb_id bd_cbid; /* network callback info */
- lnet_handle_md_t bd_md_h; /* associated MD */
+ lnet_handle_md_t bd_md_h; /* associated MD */
+ lnet_nid_t bd_sender; /* stash event::sender */
#if defined(__KERNEL__)
lnet_kiov_t bd_iov[0];
int srv_n_difficult_replies; /* # 'difficult' replies */
int srv_n_active_reqs; /* # reqs being served */
cfs_duration_t srv_rqbd_timeout; /* timeout before re-posting reqs, in tick */
- int srv_watchdog_timeout; /* soft watchdog timeout, in ms */
+ int srv_watchdog_factor; /* soft watchdog timeout mutiplier */
unsigned srv_cpu_affinity:1; /* bind threads to CPUs */
+ unsigned srv_at_check:1; /* check early replies */
__u32 srv_req_portal;
__u32 srv_rep_portal;
-
- int srv_n_queued_reqs; /* # reqs waiting to be served */
+
+ /* 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 */
+ cfs_timer_t srv_at_timer; /* early reply timer */
+
+ int srv_n_queued_reqs; /* # reqs in either of the queues below */
+ struct list_head srv_req_in_queue; /* incoming reqs */
struct list_head srv_request_queue; /* reqs waiting for service */
struct list_head srv_request_history; /* request history */
lnet_nid_t self, struct obd_uuid *uuid);
int ptlrpc_put_connection(struct ptlrpc_connection *c);
struct ptlrpc_connection *ptlrpc_connection_addref(struct ptlrpc_connection *);
-void ptlrpc_init_connection(void);
+int ptlrpc_init_connection(void);
void ptlrpc_cleanup_connection(void);
extern lnet_pid_t ptl_get_pid(void);
return (rc);
}
-int ptlrpc_send_reply(struct ptlrpc_request *req, int);
+#define PTLRPC_REPLY_MAYBE_DIFFICULT 0x01
+#define PTLRPC_REPLY_EARLY 0x02
+int ptlrpc_send_reply(struct ptlrpc_request *req, int flags);
int ptlrpc_reply(struct ptlrpc_request *req);
int ptlrpc_error(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);
struct ptlrpc_connection *ptlrpc_uuid_to_connection(struct obd_uuid *uuid);
static inline int
-ptlrpc_client_receiving_reply (struct ptlrpc_request *req)
-{
- int rc;
-
- spin_lock(&req->rq_lock);
- rc = req->rq_receiving_reply;
- spin_unlock(&req->rq_lock);
- return (rc);
-}
-
-static inline int
-ptlrpc_client_replied (struct ptlrpc_request *req)
+ptlrpc_client_recv_or_unlink (struct ptlrpc_request *req)
{
int rc;
spin_lock(&req->rq_lock);
- rc = req->rq_replied;
+ rc = req->rq_receiving_reply || req->rq_must_unlink;
spin_unlock(&req->rq_lock);
return (rc);
}
void ptlrpc_abort_inflight(struct obd_import *imp);
struct ptlrpc_request_set *ptlrpc_prep_set(void);
+int ptlrpc_set_add_cb(struct ptlrpc_request_set *set,
+ set_interpreter_func fn, void *data);
int ptlrpc_set_next_timeout(struct ptlrpc_request_set *);
int ptlrpc_check_set(struct ptlrpc_request_set *set);
int ptlrpc_set_wait(struct ptlrpc_request_set *);
void ptlrpc_add_rqs_to_pool(struct ptlrpc_request_pool *pool, int num_rq);
struct ptlrpc_request_pool *ptlrpc_init_rq_pool(int, int,
void (*populate_pool)(struct ptlrpc_request_pool *, int));
+void ptlrpc_at_set_req_timeout(struct ptlrpc_request *req);
struct ptlrpc_request *ptlrpc_prep_req(struct obd_import *imp, __u32 version,
int opcode, int count, int *lengths,
char **bufs);
struct ptlrpc_service *ptlrpc_init_svc(int nbufs, int bufsize, int max_req_size,
int max_reply_size,
int req_portal, int rep_portal,
- int watchdog_timeout, /* in ms */
+ int watchdog_factor,
svc_handler_t, char *name,
cfs_proc_dir_entry_t *proc_entry,
svcreq_printfn_t,
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 ptlrpc_import_setasync(struct obd_import *imp, int count);
+int ptlrpc_reconnect_import(struct obd_import *imp);
/* ptlrpc/pack_generic.c */
int lustre_msg_swabbed(struct lustre_msg *msg);
int *lens, char **bufs);
int lustre_pack_reply(struct ptlrpc_request *, int count, int *lens,
char **bufs);
+#define LPRFL_EARLY_REPLY 1
+int lustre_pack_reply_flags(struct ptlrpc_request *, int count, int *lens,
+ char **bufs, int flags);
void lustre_shrink_reply(struct ptlrpc_request *req, int segment,
unsigned int newlen, int move_data);
void lustre_free_reply_state(struct ptlrpc_reply_state *rs);
int lustre_msg_size(__u32 magic, int count, int *lengths);
+int lustre_packed_msg_size(struct lustre_msg *msg);
+int lustre_msg_early_size(void);
int lustre_unpack_msg(struct lustre_msg *m, int len);
void *lustre_msg_buf(struct lustre_msg *m, int n, int minlen);
int lustre_msg_buflen(struct lustre_msg *m, int n);
void *swabber);
void *lustre_swab_repbuf(struct ptlrpc_request *req, int n, int minlen,
void *swabber);
+__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 lustre_msg_add_flags(struct lustre_msg *msg, int flags);
void lustre_msg_set_flags(struct lustre_msg *msg, int flags);
__u64 lustre_msg_get_last_xid(struct lustre_msg *msg);
__u64 lustre_msg_get_last_committed(struct lustre_msg *msg);
__u64 lustre_msg_get_transno(struct lustre_msg *msg);
-__u32 lustre_msg_get_status(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_slv(struct lustre_msg *msg, __u64 slv);
+void lustre_msg_set_limit(struct lustre_msg *msg, __u64 limit);
+int lustre_msg_get_status(struct lustre_msg *msg);
__u32 lustre_msg_get_conn_cnt(struct lustre_msg *msg);
+int lustre_msg_is_v1(struct lustre_msg *msg);
__u32 lustre_msg_get_magic(struct lustre_msg *msg);
+__u32 lustre_msg_get_timeout(struct lustre_msg *msg);
+__u32 lustre_msg_get_service_time(struct lustre_msg *msg);
+__u32 lustre_msg_get_cksum(struct lustre_msg *msg);
+__u32 lustre_msg_calc_cksum(struct lustre_msg *msg);
void lustre_msg_set_handle(struct lustre_msg *msg,struct lustre_handle *handle);
void lustre_msg_set_type(struct lustre_msg *msg, __u32 type);
void lustre_msg_set_opc(struct lustre_msg *msg, __u32 opc);
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);
+void lustre_msg_set_timeout(struct lustre_msg *msg, __u32 timeout);
+void lustre_msg_set_service_time(struct lustre_msg *msg, __u32 service_time);
+void lustre_msg_set_cksum(struct lustre_msg *msg, __u32 cksum);
static inline void
ptlrpc_rs_addref(struct ptlrpc_reply_state *rs)
lustre_free_reply_state(rs);
}
+/* Should only be called once per req */
+static inline void ptlrpc_req_drop_rs(struct ptlrpc_request *req)
+{
+ if (req->rq_reply_state == NULL)
+ return; /* shouldn't occur */
+ ptlrpc_rs_decref(req->rq_reply_state);
+ req->rq_reply_state = NULL;
+ req->rq_repmsg = NULL;
+}
+
static inline __u32 lustre_request_magic(struct ptlrpc_request *req)
{
return lustre_msg_get_magic(req->rq_reqmsg);
static inline void
ptlrpc_req_set_repsize(struct ptlrpc_request *req, int count, int *lens)
{
- req->rq_replen = lustre_msg_size(req->rq_reqmsg->lm_magic, count, lens);
+ int size = lustre_msg_size(req->rq_reqmsg->lm_magic, count, lens);
+
+ req->rq_replen = size + lustre_msg_early_size();
if (req->rq_reqmsg->lm_magic == LUSTRE_MSG_MAGIC_V2)
- req->rq_reqmsg->lm_repsize = req->rq_replen;
+ req->rq_reqmsg->lm_repsize = size;
}
/* ldlm/ldlm_lib.c */
int client_obd_setup(struct obd_device *obddev, obd_count len, void *buf);
int client_obd_cleanup(struct obd_device * obddev);
int client_connect_import(struct lustre_handle *conn, struct obd_device *obd,
- struct obd_uuid *cluuid, struct obd_connect_data *);
+ struct obd_uuid *cluuid, struct obd_connect_data *,
+ void *localdata);
int client_disconnect_export(struct obd_export *exp);
int client_import_add_conn(struct obd_import *imp, struct obd_uuid *uuid,
int priority);
void ptlrpcd_decref(void);
/* ptlrpc/lproc_ptlrpc.c */
+const char* ll_opcode2str(__u32 opcode);
#ifdef LPROCFS
void ptlrpc_lprocfs_register_obd(struct obd_device *obd);
void ptlrpc_lprocfs_unregister_obd(struct obd_device *obd);
git://git.whamcloud.com / fs / lustre-release.git / blobdiff