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.gnu.org/licenses/gpl-2.0.html
19 *
20 * GPL HEADER END
21 */
22 /*
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
25 *
26 * Copyright (c) 2011, 2012, Intel Corporation.
27 */
28 /*
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
31 */
32 /** \defgroup obd_import PtlRPC import definitions
33 * Imports are client-side representation of remote obd target.
34 *
35 * @{
36 */
37
38 #ifndef __IMPORT_H
39 #define __IMPORT_H
40
41 /** \defgroup export export
42 *
43 * @{
44 */
45
46 #include <lustre_handles.h>
47 #include <uapi/linux/lustre/lustre_idl.h>
48
49 /**
50 * Adaptive Timeout stuff
51 *
52 * @{
53 */
54 #define D_ADAPTTO D_OTHER
55 #define AT_BINS 4 /* "bin" means "N seconds of history" */
56 #define AT_FLG_NOHIST 0x1 /* use last reported value only */
57
58 struct adaptive_timeout {
59 time64_t at_binstart; /* bin start time */
60 unsigned int at_hist[AT_BINS]; /* timeout history bins */
61 unsigned int at_flags;
62 unsigned int at_current; /* current timeout value */
63 unsigned int at_worst_ever; /* worst-ever timeout value */
64 time64_t at_worst_time; /* worst-ever timeout timestamp */
65 spinlock_t at_lock;
66 };
67
68 struct ptlrpc_at_array {
69 struct list_head *paa_reqs_array; /** array to hold requests */
70 __u32 paa_size; /** the size of array */
71 __u32 paa_count; /** the total count of reqs */
72 time64_t paa_deadline; /** the earliest deadline of reqs */
73 __u32 *paa_reqs_count; /** the count of reqs in each entry */
74 };
75
76 #define IMP_AT_MAX_PORTALS 8
77 struct imp_at {
78 int iat_portal[IMP_AT_MAX_PORTALS];
79 struct adaptive_timeout iat_net_latency;
80 struct adaptive_timeout iat_service_estimate[IMP_AT_MAX_PORTALS];
81 };
82
83 /** @} */
84
85 /** Possible import states */
86 enum lustre_imp_state {
87 LUSTRE_IMP_CLOSED = 1,
88 LUSTRE_IMP_NEW = 2,
89 LUSTRE_IMP_DISCON = 3,
90 LUSTRE_IMP_CONNECTING = 4,
91 LUSTRE_IMP_REPLAY = 5,
92 LUSTRE_IMP_REPLAY_LOCKS = 6,
93 LUSTRE_IMP_REPLAY_WAIT = 7,
94 LUSTRE_IMP_RECOVER = 8,
95 LUSTRE_IMP_FULL = 9,
96 LUSTRE_IMP_EVICTED = 10,
97 };
98
99 /** Returns test string representation of numeric import state \a state */
ptlrpc_import_state_name(enum lustre_imp_state state)100 static inline char *ptlrpc_import_state_name(enum lustre_imp_state state)
101 {
102 static char *import_state_names[] = {
103 "<UNKNOWN>", "CLOSED", "NEW", "DISCONN",
104 "CONNECTING", "REPLAY", "REPLAY_LOCKS", "REPLAY_WAIT",
105 "RECOVER", "FULL", "EVICTED",
106 };
107
108 LASSERT(state <= LUSTRE_IMP_EVICTED);
109 return import_state_names[state];
110 }
111
112 /**
113 * List of import event types
114 */
115 enum obd_import_event {
116 IMP_EVENT_DISCON = 0x808001,
117 IMP_EVENT_INACTIVE = 0x808002,
118 IMP_EVENT_INVALIDATE = 0x808003,
119 IMP_EVENT_ACTIVE = 0x808004,
120 IMP_EVENT_OCD = 0x808005,
121 IMP_EVENT_DEACTIVATE = 0x808006,
122 IMP_EVENT_ACTIVATE = 0x808007,
123 };
124
125 /**
126 * Definition of import connection structure
127 */
128 struct obd_import_conn {
129 /** Item for linking connections together */
130 struct list_head oic_item;
131 /** Pointer to actual PortalRPC connection */
132 struct ptlrpc_connection *oic_conn;
133 /** uuid of remote side */
134 struct obd_uuid oic_uuid;
135 /**
136 * Time (64 bit jiffies) of last connection attempt on this connection
137 */
138 __u64 oic_last_attempt;
139 };
140
141 /* state history */
142 #define IMP_STATE_HIST_LEN 16
143 struct import_state_hist {
144 enum lustre_imp_state ish_state;
145 time64_t ish_time;
146 };
147
148 /**
149 * Definition of PortalRPC import structure.
150 * Imports are representing client-side view to remote target.
151 */
152 struct obd_import {
153 /** Local handle (== id) for this import. */
154 struct portals_handle imp_handle;
155 /** Reference counter */
156 atomic_t imp_refcount;
157 struct lustre_handle imp_dlm_handle; /* client's ldlm export */
158 /** Currently active connection */
159 struct ptlrpc_connection *imp_connection;
160 /** PortalRPC client structure for this import */
161 struct ptlrpc_client *imp_client;
162 /** List element for linking into pinger chain */
163 struct list_head imp_pinger_chain;
164 /** List element for linking into chain for destruction */
165 struct list_head imp_zombie_chain;
166
167 /**
168 * Lists of requests that are retained for replay, waiting for a reply,
169 * or waiting for recovery to complete, respectively.
170 * @{
171 */
172 struct list_head imp_replay_list;
173 struct list_head imp_sending_list;
174 struct list_head imp_delayed_list;
175 /** @} */
176
177 /**
178 * List of requests that are retained for committed open replay. Once
179 * open is committed, open replay request will be moved from the
180 * imp_replay_list into the imp_committed_list.
181 * The imp_replay_cursor is for accelerating searching during replay.
182 * @{
183 */
184 struct list_head imp_committed_list;
185 struct list_head *imp_replay_cursor;
186 /** @} */
187
188 /** List of not replied requests */
189 struct list_head imp_unreplied_list;
190 /** Known maximal replied XID */
191 __u64 imp_known_replied_xid;
192
193 /** obd device for this import */
194 struct obd_device *imp_obd;
195
196 /**
197 * some seciruty-related fields
198 * @{
199 */
200 struct ptlrpc_sec *imp_sec;
201 struct mutex imp_sec_mutex;
202 time64_t imp_sec_expire;
203 /** @} */
204
205 /** Wait queue for those who need to wait for recovery completion */
206 wait_queue_head_t imp_recovery_waitq;
207
208 /** Number of requests currently in-flight */
209 atomic_t imp_inflight;
210 /** Number of requests currently unregistering */
211 atomic_t imp_unregistering;
212 /** Number of replay requests inflight */
213 atomic_t imp_replay_inflight;
214 /** Number of currently happening import invalidations */
215 atomic_t imp_inval_count;
216 /** Numbner of request timeouts */
217 atomic_t imp_timeouts;
218 /** Current import state */
219 enum lustre_imp_state imp_state;
220 /** Last replay state */
221 enum lustre_imp_state imp_replay_state;
222 /** History of import states */
223 struct import_state_hist imp_state_hist[IMP_STATE_HIST_LEN];
224 int imp_state_hist_idx;
225 /** Current import generation. Incremented on every reconnect */
226 int imp_generation;
227 /** Incremented every time we send reconnection request */
228 __u32 imp_conn_cnt;
229 /**
230 * \see ptlrpc_free_committed remembers imp_generation value here
231 * after a check to save on unnecessary replay list iterations
232 */
233 int imp_last_generation_checked;
234 /** Last transno we replayed */
235 __u64 imp_last_replay_transno;
236 /** Last transno committed on remote side */
237 __u64 imp_peer_committed_transno;
238 /**
239 * \see ptlrpc_free_committed remembers last_transno since its last
240 * check here and if last_transno did not change since last run of
241 * ptlrpc_free_committed and import generation is the same, we can
242 * skip looking for requests to remove from replay list as optimisation
243 */
244 __u64 imp_last_transno_checked;
245 /**
246 * Remote export handle. This is how remote side knows what export
247 * we are talking to. Filled from response to connect request
248 */
249 struct lustre_handle imp_remote_handle;
250 /** When to perform next ping. time in jiffies. */
251 unsigned long imp_next_ping;
252 /** When we last successfully connected. time in 64bit jiffies */
253 __u64 imp_last_success_conn;
254
255 /** List of all possible connection for import. */
256 struct list_head imp_conn_list;
257 /**
258 * Current connection. \a imp_connection is imp_conn_current->oic_conn
259 */
260 struct obd_import_conn *imp_conn_current;
261
262 /** Protects flags, level, generation, conn_cnt, *_list */
263 spinlock_t imp_lock;
264
265 /* flags */
266 unsigned long imp_no_timeout:1, /* timeouts are disabled */
267 imp_invalid:1, /* evicted */
268 /* administratively disabled */
269 imp_deactive:1,
270 /* try to recover the import */
271 imp_replayable:1,
272 /* don't run recovery (timeout instead) */
273 imp_dlm_fake:1,
274 /* use 1/2 timeout on MDS' OSCs */
275 imp_server_timeout:1,
276 /* VBR: imp in delayed recovery */
277 imp_delayed_recovery:1,
278 /* VBR: if gap was found then no lock replays
279 */
280 imp_no_lock_replay:1,
281 /* recovery by versions was failed */
282 imp_vbr_failed:1,
283 /* force an immediate ping */
284 imp_force_verify:1,
285 /* force a scheduled ping */
286 imp_force_next_verify:1,
287 /* pingable */
288 imp_pingable:1,
289 /* resend for replay */
290 imp_resend_replay:1,
291 /* disable normal recovery, for test only. */
292 imp_no_pinger_recover:1,
293 #if OBD_OCD_VERSION(3, 0, 53, 0) > LUSTRE_VERSION_CODE
294 /* need IR MNE swab */
295 imp_need_mne_swab:1,
296 #endif
297 /* import must be reconnected instead of
298 * chosing new connection
299 */
300 imp_force_reconnect:1,
301 /* import has tried to connect with server */
302 imp_connect_tried:1,
303 /* connected but not FULL yet */
304 imp_connected:1;
305 __u32 imp_connect_op;
306 struct obd_connect_data imp_connect_data;
307 __u64 imp_connect_flags_orig;
308 int imp_connect_error;
309
310 __u32 imp_msg_magic;
311 __u32 imp_msghdr_flags; /* adjusted based on server capability */
312
313 struct imp_at imp_at; /* adaptive timeout data */
314 time64_t imp_last_reply_time; /* for health check */
315 };
316
317 /* import.c */
at_est2timeout(unsigned int val)318 static inline unsigned int at_est2timeout(unsigned int val)
319 {
320 /* add an arbitrary minimum: 125% +5 sec */
321 return (val + (val >> 2) + 5);
322 }
323
at_timeout2est(unsigned int val)324 static inline unsigned int at_timeout2est(unsigned int val)
325 {
326 /* restore estimate value from timeout: e=4/5(t-5) */
327 LASSERT(val);
328 return (max((val << 2) / 5, 5U) - 4);
329 }
330
at_reset(struct adaptive_timeout * at,int val)331 static inline void at_reset(struct adaptive_timeout *at, int val)
332 {
333 spin_lock(&at->at_lock);
334 at->at_current = val;
335 at->at_worst_ever = val;
336 at->at_worst_time = ktime_get_real_seconds();
337 spin_unlock(&at->at_lock);
338 }
339
at_init(struct adaptive_timeout * at,int val,int flags)340 static inline void at_init(struct adaptive_timeout *at, int val, int flags)
341 {
342 memset(at, 0, sizeof(*at));
343 spin_lock_init(&at->at_lock);
344 at->at_flags = flags;
345 at_reset(at, val);
346 }
347
348 extern unsigned int at_min;
at_get(struct adaptive_timeout * at)349 static inline int at_get(struct adaptive_timeout *at)
350 {
351 return (at->at_current > at_min) ? at->at_current : at_min;
352 }
353
354 int at_measured(struct adaptive_timeout *at, unsigned int val);
355 int import_at_get_index(struct obd_import *imp, int portal);
356 extern unsigned int at_max;
357 #define AT_OFF (at_max == 0)
358
359 /* genops.c */
360 struct obd_export;
361 struct obd_import *class_exp2cliimp(struct obd_export *);
362
363 /** @} import */
364
365 #endif /* __IMPORT_H */
366
367 /** @} obd_import */
368