Whamcloud - gitweb
LU-3467 mdt: call MDT handlers via unified request handler
[fs/lustre-release.git] / lustre / include / lustre_export.h
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, 2013, 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_export PortalRPC export definitions
37  *
38  * @{
39  */
40
41 #ifndef __EXPORT_H
42 #define __EXPORT_H
43
44 /** \defgroup export export
45  *
46  * @{
47  */
48
49 #include <lprocfs_status.h>
50 #include <lustre/lustre_idl.h>
51 #include <lustre_dlm.h>
52
53 struct mds_client_data;
54 struct mdt_client_data;
55 struct mds_idmap_table;
56 struct mdt_idmap_table;
57
58 /**
59  * Target-specific export data
60  */
61 struct tg_export_data {
62         /** Protects led_lcd below */
63         struct mutex            ted_lcd_lock;
64         /** Per-client data for each export */
65         struct lsd_client_data  *ted_lcd;
66         /** Offset of record in last_rcvd file */
67         loff_t                  ted_lr_off;
68         /** Client index in last_rcvd file */
69         int                     ted_lr_idx;
70 };
71
72 /**
73  * MDT-specific export data
74  */
75 struct mdt_export_data {
76         struct tg_export_data   med_ted;
77         /** List of all files opened by client on this MDT */
78         cfs_list_t              med_open_head;
79         spinlock_t              med_open_lock; /* med_open_head, mfd_list */
80         struct mutex            med_idmap_mutex;
81         struct lustre_idmap_table *med_idmap;
82 };
83
84 struct ec_export_data { /* echo client */
85         cfs_list_t eced_locks;
86 };
87
88 /* In-memory access to client data from OST struct */
89 /** Filter (oss-side) specific import data */
90 struct filter_export_data {
91         struct tg_export_data   fed_ted;
92         spinlock_t              fed_lock;       /**< protects fed_mod_list */
93         long                       fed_dirty;    /* in bytes */
94         long                       fed_grant;    /* in bytes */
95         cfs_list_t                 fed_mod_list; /* files being modified */
96         int                        fed_mod_count;/* items in fed_writing list */
97         long                       fed_pending;  /* bytes just being written */
98         __u32                      fed_group;
99         __u8                       fed_pagesize; /* log2 of client page size */
100 };
101
102 struct mgs_export_data {
103         cfs_list_t              med_clients;    /* mgc fs client via this exp */
104         spinlock_t              med_lock;       /* protect med_clients */
105 };
106
107 /**
108  * per-NID statistics structure.
109  * It tracks access patterns to this export on a per-client-NID basis
110  */
111 struct nid_stat {
112         lnet_nid_t               nid;
113         cfs_hlist_node_t         nid_hash;
114         cfs_list_t               nid_list;
115         struct obd_device       *nid_obd;
116         struct proc_dir_entry   *nid_proc;
117         struct lprocfs_stats    *nid_stats;
118         struct lprocfs_stats    *nid_ldlm_stats;
119         cfs_atomic_t             nid_exp_ref_count; /* for obd_nid_stats_hash
120                                                            exp_nid_stats */
121 };
122
123 #define nidstat_getref(nidstat)                                                \
124 do {                                                                           \
125         cfs_atomic_inc(&(nidstat)->nid_exp_ref_count);                         \
126 } while(0)
127
128 #define nidstat_putref(nidstat)                                                \
129 do {                                                                           \
130         cfs_atomic_dec(&(nidstat)->nid_exp_ref_count);                         \
131         LASSERTF(cfs_atomic_read(&(nidstat)->nid_exp_ref_count) >= 0,          \
132                  "stat %p nid_exp_ref_count < 0\n", nidstat);                  \
133 } while(0)
134
135 enum obd_option {
136         OBD_OPT_FORCE =         0x0001,
137         OBD_OPT_FAILOVER =      0x0002,
138         OBD_OPT_ABORT_RECOV =   0x0004,
139 };
140
141 /**
142  * Export structure. Represents target-side of connection in portals.
143  * Also used in Lustre to connect between layers on the same node when
144  * there is no network-connection in-between.
145  * For every connected client there is an export structure on the server
146  * attached to the same obd device.
147  */
148 struct obd_export {
149         /**
150          * Export handle, it's id is provided to client on connect
151          * Subsequent client RPCs contain this handle id to identify
152          * what export they are talking to.
153          */
154         struct portals_handle     exp_handle;
155         cfs_atomic_t              exp_refcount;
156         /**
157          * Set of counters below is to track where export references are
158          * kept. The exp_rpc_count is used for reconnect handling also,
159          * the cb_count and locks_count are for debug purposes only for now.
160          * The sum of them should be less than exp_refcount by 3
161          */
162         cfs_atomic_t              exp_rpc_count; /* RPC references */
163         cfs_atomic_t              exp_cb_count; /* Commit callback references */
164         /** Number of queued replay requests to be processes */
165         cfs_atomic_t              exp_replay_count;
166         cfs_atomic_t              exp_locks_count; /** Lock references */
167 #if LUSTRE_TRACKS_LOCK_EXP_REFS
168         cfs_list_t                exp_locks_list;
169         spinlock_t                exp_locks_list_guard;
170 #endif
171         /** UUID of client connected to this export */
172         struct obd_uuid           exp_client_uuid;
173         /** To link all exports on an obd device */
174         cfs_list_t                exp_obd_chain;
175         cfs_hlist_node_t          exp_uuid_hash; /** uuid-export hash*/
176         cfs_hlist_node_t          exp_nid_hash; /** nid-export hash */
177         /**
178          * All exports eligible for ping evictor are linked into a list
179          * through this field in "most time since last request on this export"
180          * order
181          * protected by obd_dev_lock
182          */
183         cfs_list_t                exp_obd_chain_timed;
184         /** Obd device of this export */
185         struct obd_device        *exp_obd;
186         /**
187          * "reverse" import to send requests (e.g. from ldlm) back to client
188          * exp_lock protect its change
189          */
190         struct obd_import        *exp_imp_reverse;
191         struct nid_stat          *exp_nid_stats;
192         /** Active connetion */
193         struct ptlrpc_connection *exp_connection;
194         /** Connection count value from last succesful reconnect rpc */
195         __u32                     exp_conn_cnt;
196         /** Hash list of all ldlm locks granted on this export */
197         cfs_hash_t               *exp_lock_hash;
198         /**
199          * Hash list for Posix lock deadlock detection, added with
200          * ldlm_lock::l_exp_flock_hash.
201          */
202         cfs_hash_t               *exp_flock_hash;
203         cfs_list_t                exp_outstanding_replies;
204         cfs_list_t                exp_uncommitted_replies;
205         spinlock_t                exp_uncommitted_replies_lock;
206         /** Last committed transno for this export */
207         __u64                     exp_last_committed;
208         /** When was last request received */
209         cfs_time_t                exp_last_request_time;
210         /** On replay all requests waiting for replay are linked here */
211         cfs_list_t                exp_req_replay_queue;
212         /**
213          * protects exp_flags, exp_outstanding_replies and the change
214          * of exp_imp_reverse
215          */
216         spinlock_t                exp_lock;
217         /** Compatibility flags for this export are embedded into
218          *  exp_connect_data */
219         struct obd_connect_data   exp_connect_data;
220         enum obd_option           exp_flags;
221         unsigned long             exp_failed:1,
222                                   exp_in_recovery:1,
223                                   exp_disconnected:1,
224                                   exp_connecting:1,
225                                   /** VBR: export missed recovery */
226                                   exp_delayed:1,
227                                   /** VBR: failed version checking */
228                                   exp_vbr_failed:1,
229                                   exp_req_replay_needed:1,
230                                   exp_lock_replay_needed:1,
231                                   exp_need_sync:1,
232                                   exp_flvr_changed:1,
233                                   exp_flvr_adapt:1,
234                                   exp_libclient:1, /* liblustre client? */
235                                   /* client timed out and tried to reconnect,
236                                    * but couldn't because of active rpcs */
237                                   exp_abort_active_req:1,
238                                   /* if to swap nidtbl entries for 2.2 clients.
239                                    * Only used by the MGS to fix LU-1644. */
240                                   exp_need_mne_swab:1;
241         /* also protected by exp_lock */
242         enum lustre_sec_part      exp_sp_peer;
243         struct sptlrpc_flavor     exp_flvr;             /* current */
244         struct sptlrpc_flavor     exp_flvr_old[2];      /* about-to-expire */
245         cfs_time_t                exp_flvr_expire[2];   /* seconds */
246
247         /** protects exp_hp_rpcs */
248         spinlock_t                exp_rpc_lock;
249         cfs_list_t                exp_hp_rpcs;  /* (potential) HP RPCs */
250
251         /** blocking dlm lock list, protected by exp_bl_list_lock */
252         cfs_list_t                exp_bl_list;
253         spinlock_t                exp_bl_list_lock;
254
255         /** Target specific data */
256         union {
257                 struct tg_export_data     eu_target_data;
258                 struct mdt_export_data    eu_mdt_data;
259                 struct filter_export_data eu_filter_data;
260                 struct ec_export_data     eu_ec_data;
261                 struct mgs_export_data    eu_mgs_data;
262         } u;
263 };
264
265 #define exp_target_data u.eu_target_data
266 #define exp_mdt_data    u.eu_mdt_data
267 #define exp_filter_data u.eu_filter_data
268 #define exp_ec_data     u.eu_ec_data
269
270 static inline __u64 *exp_connect_flags_ptr(struct obd_export *exp)
271 {
272         return &exp->exp_connect_data.ocd_connect_flags;
273 }
274
275 static inline __u64 exp_connect_flags(struct obd_export *exp)
276 {
277         return *exp_connect_flags_ptr(exp);
278 }
279
280 static inline int exp_max_brw_size(struct obd_export *exp)
281 {
282         LASSERT(exp != NULL);
283         if (exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE)
284                 return exp->exp_connect_data.ocd_brw_size;
285
286         return ONE_MB_BRW_SIZE;
287 }
288
289 static inline int exp_connect_multibulk(struct obd_export *exp)
290 {
291         return exp_max_brw_size(exp) > ONE_MB_BRW_SIZE;
292 }
293
294 static inline int exp_expired(struct obd_export *exp, cfs_duration_t age)
295 {
296         LASSERT(exp->exp_delayed);
297         return cfs_time_before(cfs_time_add(exp->exp_last_request_time, age),
298                                cfs_time_current_sec());
299 }
300
301 static inline int exp_connect_cancelset(struct obd_export *exp)
302 {
303         LASSERT(exp != NULL);
304         return !!(exp_connect_flags(exp) & OBD_CONNECT_CANCELSET);
305 }
306
307 static inline int exp_connect_lru_resize(struct obd_export *exp)
308 {
309         LASSERT(exp != NULL);
310         return !!(exp_connect_flags(exp) & OBD_CONNECT_LRU_RESIZE);
311 }
312
313 static inline int exp_connect_rmtclient(struct obd_export *exp)
314 {
315         LASSERT(exp != NULL);
316         return !!(exp_connect_flags(exp) & OBD_CONNECT_RMT_CLIENT);
317 }
318
319 static inline int client_is_remote(struct obd_export *exp)
320 {
321         struct obd_import *imp = class_exp2cliimp(exp);
322
323         return !!(imp->imp_connect_data.ocd_connect_flags &
324                   OBD_CONNECT_RMT_CLIENT);
325 }
326
327 static inline int exp_connect_vbr(struct obd_export *exp)
328 {
329         LASSERT(exp != NULL);
330         LASSERT(exp->exp_connection);
331         return !!(exp_connect_flags(exp) & OBD_CONNECT_VBR);
332 }
333
334 static inline int exp_connect_som(struct obd_export *exp)
335 {
336         LASSERT(exp != NULL);
337         return !!(exp_connect_flags(exp) & OBD_CONNECT_SOM);
338 }
339
340 static inline int exp_connect_umask(struct obd_export *exp)
341 {
342         return !!(exp_connect_flags(exp) & OBD_CONNECT_UMASK);
343 }
344
345 static inline int imp_connect_lru_resize(struct obd_import *imp)
346 {
347         struct obd_connect_data *ocd;
348
349         LASSERT(imp != NULL);
350         ocd = &imp->imp_connect_data;
351         return !!(ocd->ocd_connect_flags & OBD_CONNECT_LRU_RESIZE);
352 }
353
354 static inline int exp_connect_layout(struct obd_export *exp)
355 {
356         return !!(exp_connect_flags(exp) & OBD_CONNECT_LAYOUTLOCK);
357 }
358
359 static inline bool exp_connect_lvb_type(struct obd_export *exp)
360 {
361         LASSERT(exp != NULL);
362         if (exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE)
363                 return true;
364         else
365                 return false;
366 }
367
368 static inline bool imp_connect_lvb_type(struct obd_import *imp)
369 {
370         struct obd_connect_data *ocd;
371
372         LASSERT(imp != NULL);
373         ocd = &imp->imp_connect_data;
374         if (ocd->ocd_connect_flags & OBD_CONNECT_LVB_TYPE)
375                 return true;
376         else
377                 return false;
378 }
379
380 extern struct obd_export *class_conn2export(struct lustre_handle *conn);
381 extern struct obd_device *class_conn2obd(struct lustre_handle *conn);
382
383 #define KKUC_CT_DATA_MAGIC      0x092013cea
384 struct kkuc_ct_data {
385         __u32           kcd_magic;
386         struct obd_uuid kcd_uuid;
387         __u32           kcd_archive;
388 };
389
390 /** @} export */
391
392 #endif /* __EXPORT_H */
393 /** @} obd_export */