4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
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.
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).
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
23 * Copyright (c) 2002, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 /** \defgroup obd_import PtlRPC import definitions
33 * Imports are client-side representation of remote obd target.
41 /** \defgroup export export
45 #include <linux/atomic.h>
46 #include <linux/list.h>
47 #include <linux/mutex.h>
48 #include <linux/refcount.h>
49 #include <linux/spinlock.h>
50 #include <linux/time.h>
51 #include <linux/types.h>
52 #include <linux/workqueue.h>
53 #include <libcfs/libcfs.h>
54 #include <uapi/linux/lustre/lustre_idl.h>
57 * Adaptive Timeout stuff
61 #define D_ADAPTTO D_OTHER
62 #define AT_BINS 4 /* "bin" means "N seconds of history" */
63 #define AT_FLG_NOHIST 0x1 /* use last reported value only */
65 struct adaptive_timeout {
66 time64_t at_binstart; /* bin start time */
67 unsigned int at_hist[AT_BINS]; /* timeout history bins */
68 unsigned int at_flags;
69 timeout_t at_current_timeout; /* current timeout value */
70 timeout_t at_worst_timeout_ever; /* worst-ever timeout delta
73 time64_t at_worst_timestamp; /* worst-ever timeout
79 enum lustre_at_flags {
84 struct ptlrpc_at_array {
85 struct list_head *paa_reqs_array; /** array to hold requests */
86 __u32 paa_size; /** the size of array */
87 __u32 paa_count; /** the total count of reqs */
88 time64_t paa_deadline; /** the earliest deadline of reqs */
89 __u32 *paa_reqs_count; /** the count of reqs in each entry */
92 #define IMP_AT_MAX_PORTALS 8
94 int iat_portal[IMP_AT_MAX_PORTALS];
95 struct adaptive_timeout iat_net_latency;
96 struct adaptive_timeout iat_service_estimate[IMP_AT_MAX_PORTALS];
102 /** Possible import states */
103 enum lustre_imp_state {
104 LUSTRE_IMP_CLOSED = 1,
106 LUSTRE_IMP_DISCON = 3,
107 LUSTRE_IMP_CONNECTING = 4,
108 LUSTRE_IMP_REPLAY = 5,
109 LUSTRE_IMP_REPLAY_LOCKS = 6,
110 LUSTRE_IMP_REPLAY_WAIT = 7,
111 LUSTRE_IMP_RECOVER = 8,
113 LUSTRE_IMP_EVICTED = 10,
114 LUSTRE_IMP_IDLE = 11,
118 /** Returns test string representation of numeric import state \a state */
119 static inline const char *ptlrpc_import_state_name(enum lustre_imp_state state)
121 static const char * const import_state_names[] = {
122 "<UNKNOWN>", "CLOSED", "NEW", "DISCONN",
123 "CONNECTING", "REPLAY", "REPLAY_LOCKS", "REPLAY_WAIT",
124 "RECOVER", "FULL", "EVICTED", "IDLE",
127 LASSERT(state < LUSTRE_IMP_LAST);
128 return import_state_names[state];
132 * List of import event types
134 enum obd_import_event {
135 IMP_EVENT_DISCON = 0x808001,
136 IMP_EVENT_INACTIVE = 0x808002,
137 IMP_EVENT_INVALIDATE = 0x808003,
138 IMP_EVENT_ACTIVE = 0x808004,
139 IMP_EVENT_OCD = 0x808005,
140 IMP_EVENT_DEACTIVATE = 0x808006,
141 IMP_EVENT_ACTIVATE = 0x808007,
145 * Definition of import connection structure
147 struct obd_import_conn {
148 /** Item for linking connections together */
149 struct list_head oic_item;
150 /** Pointer to actual PortalRPC connection */
151 struct ptlrpc_connection *oic_conn;
152 /** uuid of remote side */
153 struct obd_uuid oic_uuid;
155 * Time (64 bit seconds) of last connection attempt on this connection
157 time64_t oic_last_attempt;
161 #define IMP_STATE_HIST_LEN 16
162 struct import_state_hist {
163 enum lustre_imp_state ish_state;
168 * Defintion of PortalRPC import structure.
169 * Imports are representing client-side view to remote target.
172 /** Reference counter */
173 refcount_t imp_refcount;
174 struct lustre_handle imp_dlm_handle; /* client's ldlm export */
175 /** Currently active connection */
176 struct ptlrpc_connection *imp_connection;
177 /** PortalRPC client structure for this import */
178 struct ptlrpc_client *imp_client;
179 /** List element for linking into pinger chain */
180 struct list_head imp_pinger_chain;
181 /** work struct for destruction of import */
182 struct work_struct imp_zombie_work;
185 * Lists of requests that are retained for replay, waiting for a reply,
186 * or waiting for recovery to complete, respectively.
189 struct list_head imp_replay_list;
190 struct list_head imp_sending_list;
191 struct list_head imp_delayed_list;
195 * List of requests that are retained for committed open replay. Once
196 * open is committed, open replay request will be moved from the
197 * imp_replay_list into the imp_committed_list.
198 * The imp_replay_cursor is for accelerating searching during replay.
201 struct list_head imp_committed_list;
202 struct list_head *imp_replay_cursor;
205 /** List of not replied requests */
206 struct list_head imp_unreplied_list;
207 /** Known maximal replied XID */
208 __u64 imp_known_replied_xid;
210 /** obd device for this import */
211 struct obd_device *imp_obd;
214 * some seciruty-related fields
217 struct ptlrpc_sec *imp_sec;
218 rwlock_t imp_sec_lock;
219 time64_t imp_sec_expire;
220 pid_t imp_sec_refpid;
223 /** Wait queue for those who need to wait for recovery completion */
224 wait_queue_head_t imp_recovery_waitq;
226 /** Number of requests allocated */
228 /** Number of requests currently in-flight */
229 atomic_t imp_inflight;
230 /** Number of requests currently unregistering */
231 atomic_t imp_unregistering;
232 /** Number of replay requests inflight */
233 atomic_t imp_replay_inflight;
234 /** Number of currently happening import invalidations */
235 atomic_t imp_inval_count;
236 /** Numbner of request timeouts */
237 atomic_t imp_timeouts;
238 /** Current import state */
239 enum lustre_imp_state imp_state;
240 /** Last replay state */
241 enum lustre_imp_state imp_replay_state;
242 /** History of import states */
243 struct import_state_hist imp_state_hist[IMP_STATE_HIST_LEN];
244 int imp_state_hist_idx;
245 /** Current import generation. Incremented on every reconnect */
247 /** Idle connection initiated at this generation */
248 int imp_initiated_at;
249 /** Incremented every time we send reconnection request */
252 * \see ptlrpc_free_committed remembers imp_generation value here
253 * after a check to save on unnecessary replay list iterations
255 int imp_last_generation_checked;
256 /** Last tranno we replayed */
257 __u64 imp_last_replay_transno;
258 /** Last transno committed on remote side */
259 __u64 imp_peer_committed_transno;
261 * \see ptlrpc_free_committed remembers last_transno since its last
262 * check here and if last_transno did not change since last run of
263 * ptlrpc_free_committed and import generation is the same, we can
264 * skip looking for requests to remove from replay list as optimisation
266 __u64 imp_last_transno_checked;
268 * Remote export handle. This is how remote side knows what export
269 * we are talking to. Filled from response to connect request
271 struct lustre_handle imp_remote_handle;
272 /** When to perform next ping. time in jiffies. */
273 time64_t imp_next_ping;
274 /** When we last successfully connected. time in 64bit jiffies */
275 time64_t imp_last_success_conn;
277 /** List of all possible connection for import. */
278 struct list_head imp_conn_list;
280 * Current connection. \a imp_connection is imp_conn_current->oic_conn
282 struct obd_import_conn *imp_conn_current;
284 /** Protects flags, level, generation, conn_cnt, *_list */
288 unsigned long imp_invalid:1, /* evicted */
289 /* administratively disabled */
291 /* try to recover the import */
293 /* don't run recovery (timeout instead) */
295 /* use 1/2 timeout on MDS' OSCs */
296 imp_server_timeout:1,
297 /* VBR: imp in delayed recovery */
298 imp_delayed_recovery:1,
299 /* recovery by versions was failed */
301 /* force an immidiate ping */
303 /* force a scheduled ping */
304 imp_force_next_verify:1,
307 /* resend for replay */
309 /* disable normal recovery, for test only. */
310 imp_no_pinger_recover:1,
311 /* import must be reconnected instead of
312 * chouse new connection */
313 imp_force_reconnect:1,
314 /* import has tried to connect with server */
316 /* connected but not FULL yet */
318 /* grant shrink disabled */
319 imp_grant_shrink_disabled:1,
320 /* to supress LCONSOLE() at conn.restore */
323 u32 imp_idle_timeout;
325 struct obd_connect_data imp_connect_data;
326 __u64 imp_connect_flags_orig;
327 __u64 imp_connect_flags2_orig;
328 int imp_connect_error;
330 enum lustre_msg_magic imp_msg_magic;
331 /* adjusted based on server capability */
332 enum lustre_msghdr imp_msghdr_flags;
334 /* adaptive timeout data */
335 struct imp_at imp_at;
336 time64_t imp_last_reply_time; /* for health check */
339 /* import.c : adaptive timeout handling.
341 * Lustre tracks how long RPCs take to complete. This information is reported
342 * back to clients who utilize the information to estimate the time needed
343 * for future requests and set appropriate RPC timeouts. Minimum and maximum
344 * service times can be configured via the at_min and at_max kernel module
345 * parameters, respectively.
347 * Since this information is transmitted between nodes the timeouts are in
348 * seconds not jiffies which can vary from node to node. To avoid confusion
349 * the timeout is handled in timeout_t (s32) instead of time64_t or
352 static inline timeout_t at_est2timeout(timeout_t timeout)
354 /* add an arbitrary minimum: 125% +5 sec */
355 return timeout + (timeout >> 2) + 5;
358 static inline timeout_t at_timeout2est(timeout_t timeout)
360 /* restore estimate value from timeout: e=4/5(t-5) */
361 LASSERT(timeout > 0);
362 return max((timeout << 2) / 5, 5) - 4;
365 static inline void at_reset_nolock(struct adaptive_timeout *at,
368 at->at_current_timeout = timeout;
369 at->at_worst_timeout_ever = timeout;
370 at->at_worst_timestamp = ktime_get_real_seconds();
373 static inline void at_reset(struct adaptive_timeout *at, timeout_t timeout)
375 spin_lock(&at->at_lock);
376 at_reset_nolock(at, timeout);
377 spin_unlock(&at->at_lock);
380 static inline void at_init(struct adaptive_timeout *at, timeout_t timeout,
383 memset(at, 0, sizeof(*at));
384 spin_lock_init(&at->at_lock);
385 at->at_flags = flags;
386 at_reset(at, timeout);
389 static inline void at_reinit(struct adaptive_timeout *at, timeout_t timeout,
392 spin_lock(&at->at_lock);
394 memset(at->at_hist, 0, sizeof(at->at_hist));
395 at->at_flags = flags;
396 at_reset_nolock(at, timeout);
397 spin_unlock(&at->at_lock);
400 extern unsigned int at_min;
401 extern unsigned int at_max;
402 #define AT_OFF (at_max == 0)
404 static inline timeout_t at_get(struct adaptive_timeout *at)
406 return (at->at_current_timeout > at_min) ?
407 at->at_current_timeout : at_min;
410 timeout_t at_measured(struct adaptive_timeout *at, timeout_t timeout);
411 int import_at_get_index(struct obd_import *imp, int portal);
415 extern struct obd_import *class_exp2cliimp(struct obd_export *);
419 #endif /* __IMPORT_H */