lustre_import.h revision 727543d639efe4abb4be8ce56a1065c32dbac918
1/*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
15 *
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
19 *
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26/*
27 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2011, 2012, Intel Corporation.
31 */
32/*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36/** \defgroup obd_import PtlRPC import definitions
37 * Imports are client-side representation of remote obd target.
38 *
39 * @{
40 */
41
42#ifndef __IMPORT_H
43#define __IMPORT_H
44
45/** \defgroup export export
46 *
47 * @{
48 */
49
50#include "lustre_handles.h"
51#include "lustre/lustre_idl.h"
52
53
54/**
55 * Adaptive Timeout stuff
56 *
57 * @{
58 */
59#define D_ADAPTTO D_OTHER
60#define AT_BINS 4		  /* "bin" means "N seconds of history" */
61#define AT_FLG_NOHIST 0x1	  /* use last reported value only */
62
63struct adaptive_timeout {
64	time_t		at_binstart;	 /* bin start time */
65	unsigned int	at_hist[AT_BINS];    /* timeout history bins */
66	unsigned int	at_flags;
67	unsigned int	at_current;	  /* current timeout value */
68	unsigned int	at_worst_ever;       /* worst-ever timeout value */
69	time_t		at_worst_time;       /* worst-ever timeout timestamp */
70	spinlock_t	at_lock;
71};
72
73struct ptlrpc_at_array {
74	struct list_head       *paa_reqs_array; /** array to hold requests */
75	__u32	     paa_size;       /** the size of array */
76	__u32	     paa_count;      /** the total count of reqs */
77	time_t	    paa_deadline;   /** the earliest deadline of reqs */
78	__u32	    *paa_reqs_count; /** the count of reqs in each entry */
79};
80
81#define IMP_AT_MAX_PORTALS 8
82struct imp_at {
83	int		     iat_portal[IMP_AT_MAX_PORTALS];
84	struct adaptive_timeout iat_net_latency;
85	struct adaptive_timeout iat_service_estimate[IMP_AT_MAX_PORTALS];
86};
87
88
89/** @} */
90
91/** Possible import states */
92enum lustre_imp_state {
93	LUSTRE_IMP_CLOSED     = 1,
94	LUSTRE_IMP_NEW	= 2,
95	LUSTRE_IMP_DISCON     = 3,
96	LUSTRE_IMP_CONNECTING = 4,
97	LUSTRE_IMP_REPLAY     = 5,
98	LUSTRE_IMP_REPLAY_LOCKS = 6,
99	LUSTRE_IMP_REPLAY_WAIT  = 7,
100	LUSTRE_IMP_RECOVER    = 8,
101	LUSTRE_IMP_FULL       = 9,
102	LUSTRE_IMP_EVICTED    = 10,
103};
104
105/** Returns test string representation of numeric import state \a state */
106static inline char *ptlrpc_import_state_name(enum lustre_imp_state state)
107{
108	static char *import_state_names[] = {
109		"<UNKNOWN>", "CLOSED",  "NEW", "DISCONN",
110		"CONNECTING", "REPLAY", "REPLAY_LOCKS", "REPLAY_WAIT",
111		"RECOVER", "FULL", "EVICTED",
112	};
113
114	LASSERT (state <= LUSTRE_IMP_EVICTED);
115	return import_state_names[state];
116}
117
118/**
119 * List of import event types
120 */
121enum obd_import_event {
122	IMP_EVENT_DISCON     = 0x808001,
123	IMP_EVENT_INACTIVE   = 0x808002,
124	IMP_EVENT_INVALIDATE = 0x808003,
125	IMP_EVENT_ACTIVE     = 0x808004,
126	IMP_EVENT_OCD	= 0x808005,
127	IMP_EVENT_DEACTIVATE = 0x808006,
128	IMP_EVENT_ACTIVATE   = 0x808007,
129};
130
131/**
132 * Definition of import connection structure
133 */
134struct obd_import_conn {
135	/** Item for linking connections together */
136	struct list_head		oic_item;
137	/** Pointer to actual PortalRPC connection */
138	struct ptlrpc_connection *oic_conn;
139	/** uuid of remote side */
140	struct obd_uuid	   oic_uuid;
141	/**
142	 * Time (64 bit jiffies) of last connection attempt on this connection
143	 */
144	__u64		     oic_last_attempt;
145};
146
147/* state history */
148#define IMP_STATE_HIST_LEN 16
149struct import_state_hist {
150	enum lustre_imp_state ish_state;
151	time_t		ish_time;
152};
153
154/**
155 * Definition of PortalRPC import structure.
156 * Imports are representing client-side view to remote target.
157 */
158struct obd_import {
159	/** Local handle (== id) for this import. */
160	struct portals_handle     imp_handle;
161	/** Reference counter */
162	atomic_t	      imp_refcount;
163	struct lustre_handle      imp_dlm_handle; /* client's ldlm export */
164	/** Currently active connection */
165	struct ptlrpc_connection *imp_connection;
166	/** PortalRPC client structure for this import */
167	struct ptlrpc_client     *imp_client;
168	/** List element for linking into pinger chain */
169	struct list_head		imp_pinger_chain;
170	/** List element for linking into chain for destruction */
171	struct list_head		imp_zombie_chain;
172
173	/**
174	 * Lists of requests that are retained for replay, waiting for a reply,
175	 * or waiting for recovery to complete, respectively.
176	 * @{
177	 */
178	struct list_head		imp_replay_list;
179	struct list_head		imp_sending_list;
180	struct list_head		imp_delayed_list;
181	/** @} */
182
183	/**
184	 * List of requests that are retained for committed open replay. Once
185	 * open is committed, open replay request will be moved from the
186	 * imp_replay_list into the imp_committed_list.
187	 * The imp_replay_cursor is for accelerating searching during replay.
188	 * @{
189	 */
190	struct list_head		imp_committed_list;
191	struct list_head	       *imp_replay_cursor;
192	/** @} */
193
194	/** obd device for this import */
195	struct obd_device	*imp_obd;
196
197	/**
198	 * some seciruty-related fields
199	 * @{
200	 */
201	struct ptlrpc_sec	*imp_sec;
202	struct mutex		  imp_sec_mutex;
203	unsigned long		imp_sec_expire;
204	/** @} */
205
206	/** Wait queue for those who need to wait for recovery completion */
207	wait_queue_head_t	       imp_recovery_waitq;
208
209	/** Number of requests currently in-flight */
210	atomic_t	      imp_inflight;
211	/** Number of requests currently unregistering */
212	atomic_t	      imp_unregistering;
213	/** Number of replay requests inflight */
214	atomic_t	      imp_replay_inflight;
215	/** Number of currently happening import invalidations */
216	atomic_t	      imp_inval_count;
217	/** Numbner of request timeouts */
218	atomic_t	      imp_timeouts;
219	/** Current import state */
220	enum lustre_imp_state     imp_state;
221	/** History of import states */
222	struct import_state_hist  imp_state_hist[IMP_STATE_HIST_LEN];
223	int		       imp_state_hist_idx;
224	/** Current import generation. Incremented on every reconnect */
225	int		       imp_generation;
226	/** Incremented every time we send reconnection request */
227	__u32		     imp_conn_cnt;
228       /**
229	* \see ptlrpc_free_committed remembers imp_generation value here
230	* after a check to save on unnecessary replay list iterations
231	*/
232	int		       imp_last_generation_checked;
233	/** Last transno we replayed */
234	__u64		     imp_last_replay_transno;
235	/** Last transno committed on remote side */
236	__u64		     imp_peer_committed_transno;
237	/**
238	 * \see ptlrpc_free_committed remembers last_transno since its last
239	 * check here and if last_transno did not change since last run of
240	 * ptlrpc_free_committed and import generation is the same, we can
241	 * skip looking for requests to remove from replay list as optimisation
242	 */
243	__u64		     imp_last_transno_checked;
244	/**
245	 * Remote export handle. This is how remote side knows what export
246	 * we are talking to. Filled from response to connect request
247	 */
248	struct lustre_handle      imp_remote_handle;
249	/** When to perform next ping. time in jiffies. */
250	unsigned long		imp_next_ping;
251	/** When we last successfully connected. time in 64bit jiffies */
252	__u64		     imp_last_success_conn;
253
254	/** List of all possible connection for import. */
255	struct list_head		imp_conn_list;
256	/**
257	 * Current connection. \a imp_connection is imp_conn_current->oic_conn
258	 */
259	struct obd_import_conn   *imp_conn_current;
260
261	/** Protects flags, level, generation, conn_cnt, *_list */
262	spinlock_t		  imp_lock;
263
264	/* flags */
265	unsigned long	     imp_no_timeout:1, /* timeouts are disabled */
266				  imp_invalid:1,    /* evicted */
267				  /* administratively disabled */
268				  imp_deactive:1,
269				  /* try to recover the import */
270				  imp_replayable:1,
271				  /* don't run recovery (timeout instead) */
272				  imp_dlm_fake:1,
273				  /* use 1/2 timeout on MDS' OSCs */
274				  imp_server_timeout:1,
275				  /* VBR: imp in delayed recovery */
276				  imp_delayed_recovery:1,
277				  /* VBR: if gap was found then no lock replays
278				   */
279				  imp_no_lock_replay:1,
280				  /* recovery by versions was failed */
281				  imp_vbr_failed:1,
282				  /* force an immediate ping */
283				  imp_force_verify:1,
284				  /* force a scheduled ping */
285				  imp_force_next_verify:1,
286				  /* pingable */
287				  imp_pingable:1,
288				  /* resend for replay */
289				  imp_resend_replay:1,
290				  /* disable normal recovery, for test only. */
291				  imp_no_pinger_recover:1,
292				  /* need IR MNE swab */
293				  imp_need_mne_swab:1,
294				  /* import must be reconnected instead of
295				   * chose new connection */
296				  imp_force_reconnect:1,
297				  /* import has tried to connect with server */
298				  imp_connect_tried:1;
299	__u32		     imp_connect_op;
300	struct obd_connect_data   imp_connect_data;
301	__u64		     imp_connect_flags_orig;
302	int		       imp_connect_error;
303
304	__u32		     imp_msg_magic;
305	__u32		     imp_msghdr_flags;       /* adjusted based on server capability */
306
307	struct ptlrpc_request_pool *imp_rq_pool;	  /* emergency request pool */
308
309	struct imp_at	     imp_at;		 /* adaptive timeout data */
310	time_t		    imp_last_reply_time;    /* for health check */
311};
312
313typedef void (*obd_import_callback)(struct obd_import *imp, void *closure,
314				    int event, void *event_arg, void *cb_data);
315
316/**
317 * Structure for import observer.
318 * It is possible to register "observer" on an import and every time
319 * something happens to an import (like connect/evict/disconnect)
320 * obderver will get its callback called with event type
321 */
322struct obd_import_observer {
323	struct list_head	   oio_chain;
324	obd_import_callback  oio_cb;
325	void		*oio_cb_data;
326};
327
328void class_observe_import(struct obd_import *imp, obd_import_callback cb,
329			  void *cb_data);
330void class_unobserve_import(struct obd_import *imp, obd_import_callback cb,
331			    void *cb_data);
332void class_notify_import_observers(struct obd_import *imp, int event,
333				   void *event_arg);
334
335/* import.c */
336static inline unsigned int at_est2timeout(unsigned int val)
337{
338	/* add an arbitrary minimum: 125% +5 sec */
339	return (val + (val >> 2) + 5);
340}
341
342static inline unsigned int at_timeout2est(unsigned int val)
343{
344	/* restore estimate value from timeout: e=4/5(t-5) */
345	LASSERT(val);
346	return (max((val << 2) / 5, 5U) - 4);
347}
348
349static inline void at_reset(struct adaptive_timeout *at, int val)
350{
351	spin_lock(&at->at_lock);
352	at->at_current = val;
353	at->at_worst_ever = val;
354	at->at_worst_time = get_seconds();
355	spin_unlock(&at->at_lock);
356}
357static inline void at_init(struct adaptive_timeout *at, int val, int flags)
358{
359	memset(at, 0, sizeof(*at));
360	spin_lock_init(&at->at_lock);
361	at->at_flags = flags;
362	at_reset(at, val);
363}
364extern unsigned int at_min;
365static inline int at_get(struct adaptive_timeout *at)
366{
367	return (at->at_current > at_min) ? at->at_current : at_min;
368}
369int at_measured(struct adaptive_timeout *at, unsigned int val);
370int import_at_get_index(struct obd_import *imp, int portal);
371extern unsigned int at_max;
372#define AT_OFF (at_max == 0)
373
374/* genops.c */
375struct obd_export;
376extern struct obd_import *class_exp2cliimp(struct obd_export *);
377extern struct obd_import *class_conn2cliimp(struct lustre_handle *);
378
379/** @} import */
380
381#endif /* __IMPORT_H */
382
383/** @} obd_import */
384