• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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  * Adaptive Timeout stuff
55  *
56  * @{
57  */
58 #define D_ADAPTTO D_OTHER
59 #define AT_BINS 4		  /* "bin" means "N seconds of history" */
60 #define AT_FLG_NOHIST 0x1	  /* use last reported value only */
61 
62 struct adaptive_timeout {
63 	time64_t	at_binstart;	 /* bin start time */
64 	unsigned int	at_hist[AT_BINS];    /* timeout history bins */
65 	unsigned int	at_flags;
66 	unsigned int	at_current;	  /* current timeout value */
67 	unsigned int	at_worst_ever;       /* worst-ever timeout value */
68 	time64_t	at_worst_time;       /* worst-ever timeout timestamp */
69 	spinlock_t	at_lock;
70 };
71 
72 struct ptlrpc_at_array {
73 	struct list_head       *paa_reqs_array; /** array to hold requests */
74 	__u32	     paa_size;       /** the size of array */
75 	__u32	     paa_count;      /** the total count of reqs */
76 	time64_t     paa_deadline;   /** the earliest deadline of reqs */
77 	__u32	    *paa_reqs_count; /** the count of reqs in each entry */
78 };
79 
80 #define IMP_AT_MAX_PORTALS 8
81 struct imp_at {
82 	int		     iat_portal[IMP_AT_MAX_PORTALS];
83 	struct adaptive_timeout iat_net_latency;
84 	struct adaptive_timeout iat_service_estimate[IMP_AT_MAX_PORTALS];
85 };
86 
87 /** @} */
88 
89 /** Possible import states */
90 enum lustre_imp_state {
91 	LUSTRE_IMP_CLOSED     = 1,
92 	LUSTRE_IMP_NEW	= 2,
93 	LUSTRE_IMP_DISCON     = 3,
94 	LUSTRE_IMP_CONNECTING = 4,
95 	LUSTRE_IMP_REPLAY     = 5,
96 	LUSTRE_IMP_REPLAY_LOCKS = 6,
97 	LUSTRE_IMP_REPLAY_WAIT  = 7,
98 	LUSTRE_IMP_RECOVER    = 8,
99 	LUSTRE_IMP_FULL       = 9,
100 	LUSTRE_IMP_EVICTED    = 10,
101 };
102 
103 /** Returns test string representation of numeric import state \a state */
ptlrpc_import_state_name(enum lustre_imp_state state)104 static inline char *ptlrpc_import_state_name(enum lustre_imp_state state)
105 {
106 	static char *import_state_names[] = {
107 		"<UNKNOWN>", "CLOSED",  "NEW", "DISCONN",
108 		"CONNECTING", "REPLAY", "REPLAY_LOCKS", "REPLAY_WAIT",
109 		"RECOVER", "FULL", "EVICTED",
110 	};
111 
112 	LASSERT (state <= LUSTRE_IMP_EVICTED);
113 	return import_state_names[state];
114 }
115 
116 /**
117  * List of import event types
118  */
119 enum obd_import_event {
120 	IMP_EVENT_DISCON     = 0x808001,
121 	IMP_EVENT_INACTIVE   = 0x808002,
122 	IMP_EVENT_INVALIDATE = 0x808003,
123 	IMP_EVENT_ACTIVE     = 0x808004,
124 	IMP_EVENT_OCD	= 0x808005,
125 	IMP_EVENT_DEACTIVATE = 0x808006,
126 	IMP_EVENT_ACTIVATE   = 0x808007,
127 };
128 
129 /**
130  * Definition of import connection structure
131  */
132 struct obd_import_conn {
133 	/** Item for linking connections together */
134 	struct list_head		oic_item;
135 	/** Pointer to actual PortalRPC connection */
136 	struct ptlrpc_connection *oic_conn;
137 	/** uuid of remote side */
138 	struct obd_uuid	   oic_uuid;
139 	/**
140 	 * Time (64 bit jiffies) of last connection attempt on this connection
141 	 */
142 	__u64		     oic_last_attempt;
143 };
144 
145 /* state history */
146 #define IMP_STATE_HIST_LEN 16
147 struct import_state_hist {
148 	enum lustre_imp_state ish_state;
149 	time64_t	ish_time;
150 };
151 
152 /**
153  * Definition of PortalRPC import structure.
154  * Imports are representing client-side view to remote target.
155  */
156 struct obd_import {
157 	/** Local handle (== id) for this import. */
158 	struct portals_handle     imp_handle;
159 	/** Reference counter */
160 	atomic_t	      imp_refcount;
161 	struct lustre_handle      imp_dlm_handle; /* client's ldlm export */
162 	/** Currently active connection */
163 	struct ptlrpc_connection *imp_connection;
164 	/** PortalRPC client structure for this import */
165 	struct ptlrpc_client     *imp_client;
166 	/** List element for linking into pinger chain */
167 	struct list_head		imp_pinger_chain;
168 	/** List element for linking into chain for destruction */
169 	struct list_head		imp_zombie_chain;
170 
171 	/**
172 	 * Lists of requests that are retained for replay, waiting for a reply,
173 	 * or waiting for recovery to complete, respectively.
174 	 * @{
175 	 */
176 	struct list_head		imp_replay_list;
177 	struct list_head		imp_sending_list;
178 	struct list_head		imp_delayed_list;
179 	/** @} */
180 
181 	/**
182 	 * List of requests that are retained for committed open replay. Once
183 	 * open is committed, open replay request will be moved from the
184 	 * imp_replay_list into the imp_committed_list.
185 	 * The imp_replay_cursor is for accelerating searching during replay.
186 	 * @{
187 	 */
188 	struct list_head		imp_committed_list;
189 	struct list_head	       *imp_replay_cursor;
190 	/** @} */
191 
192 	/** obd device for this import */
193 	struct obd_device	*imp_obd;
194 
195 	/**
196 	 * some seciruty-related fields
197 	 * @{
198 	 */
199 	struct ptlrpc_sec	*imp_sec;
200 	struct mutex		  imp_sec_mutex;
201 	time64_t		imp_sec_expire;
202 	/** @} */
203 
204 	/** Wait queue for those who need to wait for recovery completion */
205 	wait_queue_head_t	       imp_recovery_waitq;
206 
207 	/** Number of requests currently in-flight */
208 	atomic_t	      imp_inflight;
209 	/** Number of requests currently unregistering */
210 	atomic_t	      imp_unregistering;
211 	/** Number of replay requests inflight */
212 	atomic_t	      imp_replay_inflight;
213 	/** Number of currently happening import invalidations */
214 	atomic_t	      imp_inval_count;
215 	/** Numbner of request timeouts */
216 	atomic_t	      imp_timeouts;
217 	/** Current import state */
218 	enum lustre_imp_state     imp_state;
219 	/** Last replay state */
220 	enum lustre_imp_state	  imp_replay_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 imp_at	     imp_at;		 /* adaptive timeout data */
308 	time64_t	     imp_last_reply_time;    /* for health check */
309 };
310 
311 typedef void (*obd_import_callback)(struct obd_import *imp, void *closure,
312 				    int event, void *event_arg, void *cb_data);
313 
314 /**
315  * Structure for import observer.
316  * It is possible to register "observer" on an import and every time
317  * something happens to an import (like connect/evict/disconnect)
318  * obderver will get its callback called with event type
319  */
320 struct obd_import_observer {
321 	struct list_head	   oio_chain;
322 	obd_import_callback  oio_cb;
323 	void		*oio_cb_data;
324 };
325 
326 void class_observe_import(struct obd_import *imp, obd_import_callback cb,
327 			  void *cb_data);
328 void class_unobserve_import(struct obd_import *imp, obd_import_callback cb,
329 			    void *cb_data);
330 void class_notify_import_observers(struct obd_import *imp, int event,
331 				   void *event_arg);
332 
333 /* import.c */
at_est2timeout(unsigned int val)334 static inline unsigned int at_est2timeout(unsigned int val)
335 {
336 	/* add an arbitrary minimum: 125% +5 sec */
337 	return (val + (val >> 2) + 5);
338 }
339 
at_timeout2est(unsigned int val)340 static inline unsigned int at_timeout2est(unsigned int val)
341 {
342 	/* restore estimate value from timeout: e=4/5(t-5) */
343 	LASSERT(val);
344 	return (max((val << 2) / 5, 5U) - 4);
345 }
346 
at_reset(struct adaptive_timeout * at,int val)347 static inline void at_reset(struct adaptive_timeout *at, int val)
348 {
349 	spin_lock(&at->at_lock);
350 	at->at_current = val;
351 	at->at_worst_ever = val;
352 	at->at_worst_time = ktime_get_real_seconds();
353 	spin_unlock(&at->at_lock);
354 }
355 
at_init(struct adaptive_timeout * at,int val,int flags)356 static inline void at_init(struct adaptive_timeout *at, int val, int flags)
357 {
358 	memset(at, 0, sizeof(*at));
359 	spin_lock_init(&at->at_lock);
360 	at->at_flags = flags;
361 	at_reset(at, val);
362 }
363 
364 extern unsigned int at_min;
at_get(struct adaptive_timeout * at)365 static inline int at_get(struct adaptive_timeout *at)
366 {
367 	return (at->at_current > at_min) ? at->at_current : at_min;
368 }
369 
370 int at_measured(struct adaptive_timeout *at, unsigned int val);
371 int import_at_get_index(struct obd_import *imp, int portal);
372 extern unsigned int at_max;
373 #define AT_OFF (at_max == 0)
374 
375 /* genops.c */
376 struct obd_export;
377 struct obd_import *class_exp2cliimp(struct obd_export *);
378 
379 /** @} import */
380 
381 #endif /* __IMPORT_H */
382 
383 /** @} obd_import */
384