Whamcloud - gitweb
5ead59348796231995c73786c2f23958b72b087b
[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.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, 2017, 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_export PortalRPC export definitions
33  *
34  * @{
35  */
36
37 #ifndef __EXPORT_H
38 #define __EXPORT_H
39
40 /** \defgroup export export
41  *
42  * @{
43  */
44
45 #include <linux/workqueue.h>
46
47 #include <lprocfs_status.h>
48 #include <uapi/linux/lustre/lustre_idl.h>
49 #include <lustre_dlm.h>
50
51 struct mds_client_data;
52 struct mdt_client_data;
53 struct mds_idmap_table;
54 struct mdt_idmap_table;
55
56 /**
57  * Target-specific export data
58  */
59 struct tg_export_data {
60         /** Protects ted_lcd, ted_reply_* and
61          * ted_release_* fields below */
62         struct mutex            ted_lcd_lock;
63         /** Per-client data for each export */
64         struct lsd_client_data  *ted_lcd;
65         /** Offset of record in last_rcvd file */
66         loff_t                  ted_lr_off;
67         /** Client index in last_rcvd file */
68         int                     ted_lr_idx;
69
70         /**
71          * ted_nodemap_lock is used to ensure that the nodemap is not destroyed
72          * between the time that ted_nodemap is checked for NULL, and a
73          * reference is taken. Modifications to ted_nodemap require that the
74          * active_config_lock and the nodemap(s)'s nm_member_list_lock be
75          * taken, as well as ted_nodemap_lock, so the export can be properly
76          * added to or removed from the nodemap's member list. When an export
77          * is added to a nodemap, a reference on that nodemap must be taken.
78          * That reference can be put only after ted_nodemap no longer refers to
79          * it.
80          */
81         spinlock_t              ted_nodemap_lock;
82         struct lu_nodemap       *ted_nodemap;
83         struct list_head        ted_nodemap_member;
84
85         /** last version of nodemap config sent to client */
86         __u64                   ted_nodemap_version;
87
88         /* Every reply data fields below are
89          * protected by ted_lcd_lock */
90         /** List of reply data */
91         struct list_head        ted_reply_list;
92         int                     ted_reply_cnt;
93         /** Reply data with highest transno is retained */
94         struct tg_reply_data    *ted_reply_last;
95         /* Statistics */
96         int                     ted_reply_max; /* high water mark */
97         int                     ted_release_xid;
98         int                     ted_release_tag;
99         /* grants */
100         long                    ted_dirty;    /* in bytes */
101         long                    ted_grant;    /* in bytes */
102         long                    ted_pending;  /* bytes just being written */
103         __u8                    ted_pagebits; /* log2 of client page size */
104 };
105
106 /**
107  * MDT-specific export data
108  */
109 struct mdt_export_data {
110         struct tg_export_data   med_ted;
111         /** List of all files opened by client on this MDT */
112         struct list_head        med_open_head;
113         spinlock_t              med_open_lock; /* med_open_head, mfd_list */
114 };
115
116 struct ec_export_data { /* echo client */
117         struct list_head        eced_locks;
118 };
119
120 /* In-memory access to client data from OST struct */
121 /** Filter (oss-side) specific import data */
122 struct filter_export_data {
123         struct tg_export_data   fed_ted;
124         spinlock_t              fed_lock;       /**< protects fed_mod_list */
125         __u64                   fed_lastid_gen;
126         struct list_head        fed_mod_list; /* files being modified */
127         /* count of SOFT_SYNC RPCs, which will be reset after
128          * ofd_soft_sync_limit number of RPCs, and trigger a sync. */
129         atomic_t                fed_soft_sync_count;
130         int                     fed_mod_count;/* items in fed_writing list */
131         __u32                   fed_group;
132 };
133
134 struct mgs_export_data {
135         struct list_head        med_clients;    /* mgc fs client via this exp */
136         spinlock_t              med_lock;       /* protect med_clients */
137 };
138
139 /**
140  * per-NID statistics structure.
141  * It tracks access patterns to this export on a per-client-NID basis
142  */
143 struct nid_stat {
144         lnet_nid_t               nid;
145         struct hlist_node        nid_hash;
146         struct list_head         nid_list;
147         struct obd_device       *nid_obd;
148         struct proc_dir_entry   *nid_proc;
149         struct lprocfs_stats    *nid_stats;
150         struct lprocfs_stats    *nid_ldlm_stats;
151         atomic_t                 nid_exp_ref_count; /* for obd_nid_stats_hash
152                                                        exp_nid_stats */
153 };
154
155 #define nidstat_getref(nidstat)                                                \
156 do {                                                                           \
157         atomic_inc(&(nidstat)->nid_exp_ref_count);                         \
158 } while(0)
159
160 #define nidstat_putref(nidstat)                                                \
161 do {                                                                           \
162         atomic_dec(&(nidstat)->nid_exp_ref_count);                         \
163         LASSERTF(atomic_read(&(nidstat)->nid_exp_ref_count) >= 0,          \
164                  "stat %p nid_exp_ref_count < 0\n", nidstat);                  \
165 } while(0)
166
167 enum obd_option {
168         OBD_OPT_FORCE =         0x0001,
169         OBD_OPT_FAILOVER =      0x0002,
170         OBD_OPT_ABORT_RECOV =   0x0004,
171 };
172
173 /**
174  * Export structure. Represents target-side of connection in portals.
175  * Also used in Lustre to connect between layers on the same node when
176  * there is no network-connection in-between.
177  * For every connected client there is an export structure on the server
178  * attached to the same obd device.
179  */
180 struct obd_export {
181         /**
182          * Export handle, it's id is provided to client on connect
183          * Subsequent client RPCs contain this handle id to identify
184          * what export they are talking to.
185          */
186         struct portals_handle   exp_handle;
187         atomic_t                exp_refcount;
188         /**
189          * Set of counters below is to track where export references are
190          * kept. The exp_rpc_count is used for reconnect handling also,
191          * the cb_count and locks_count are for debug purposes only for now.
192          * The sum of them should be less than exp_refcount by 3
193          */
194         atomic_t                exp_rpc_count; /* RPC references */
195         atomic_t                exp_cb_count; /* Commit callback references */
196         /** Number of queued replay requests to be processes */
197         atomic_t                exp_replay_count;
198         atomic_t                exp_locks_count; /** Lock references */
199 #if LUSTRE_TRACKS_LOCK_EXP_REFS
200         struct list_head        exp_locks_list;
201         spinlock_t              exp_locks_list_guard;
202 #endif
203         /** UUID of client connected to this export */
204         struct obd_uuid         exp_client_uuid;
205         /** To link all exports on an obd device */
206         struct list_head        exp_obd_chain;
207         /** work_struct for destruction of export */
208         struct work_struct      exp_zombie_work;
209         /* Unlinked export list */
210         struct list_head        exp_stale_list;
211         struct hlist_node       exp_uuid_hash;  /** uuid-export hash*/
212         struct hlist_node       exp_nid_hash;   /** nid-export hash */
213         struct hlist_node       exp_gen_hash;   /** last_rcvd clt gen hash */
214         /**
215          * All exports eligible for ping evictor are linked into a list
216          * through this field in "most time since last request on this export"
217          * order
218          * protected by obd_dev_lock
219          */
220         struct list_head        exp_obd_chain_timed;
221         /** Obd device of this export */
222         struct obd_device      *exp_obd;
223         /**
224          * "reverse" import to send requests (e.g. from ldlm) back to client
225          * exp_lock protect its change
226          */
227         struct obd_import        *exp_imp_reverse;
228         struct nid_stat          *exp_nid_stats;
229         /** Active connetion */
230         struct ptlrpc_connection *exp_connection;
231         /** Connection count value from last successful reconnect rpc */
232         __u32                     exp_conn_cnt;
233         /** Hash list of all ldlm locks granted on this export */
234         struct cfs_hash          *exp_lock_hash;
235         /**
236          * Hash list for Posix lock deadlock detection, added with
237          * ldlm_lock::l_exp_flock_hash.
238          */
239         struct cfs_hash        *exp_flock_hash;
240         struct list_head        exp_outstanding_replies;
241         struct list_head        exp_uncommitted_replies;
242         spinlock_t              exp_uncommitted_replies_lock;
243         /** Last committed transno for this export */
244         __u64                   exp_last_committed;
245         /** When was last request received */
246         time64_t                exp_last_request_time;
247         /** On replay all requests waiting for replay are linked here */
248         struct list_head        exp_req_replay_queue;
249         /**
250          * protects exp_flags, exp_outstanding_replies and the change
251          * of exp_imp_reverse
252          */
253         spinlock_t              exp_lock;
254         /** Compatibility flags for this export are embedded into
255          *  exp_connect_data */
256         struct obd_connect_data exp_connect_data;
257         enum obd_option         exp_flags;
258         unsigned long           exp_failed:1,
259                                 exp_in_recovery:1,
260                                 exp_disconnected:1,
261                                 exp_connecting:1,
262                                 /** VBR: export missed recovery */
263                                 exp_delayed:1,
264                                 /** VBR: failed version checking */
265                                 exp_vbr_failed:1,
266                                 exp_req_replay_needed:1,
267                                 exp_lock_replay_needed:1,
268                                 exp_need_sync:1,
269                                 exp_flvr_changed:1,
270                                 exp_flvr_adapt:1,
271                                 /* if to swap nidtbl entries for 2.2 clients.
272                                  * Only used by the MGS to fix LU-1644. */
273                                 exp_need_mne_swab:1,
274                                 /* The export already got final replay ping
275                                  * request. */
276                                 exp_replay_done:1;
277         /* also protected by exp_lock */
278         enum lustre_sec_part    exp_sp_peer;
279         struct sptlrpc_flavor   exp_flvr;               /* current */
280         struct sptlrpc_flavor   exp_flvr_old[2];        /* about-to-expire */
281         time64_t                exp_flvr_expire[2];     /* seconds */
282
283         /** protects exp_hp_rpcs */
284         spinlock_t              exp_rpc_lock;
285         struct list_head        exp_hp_rpcs;    /* (potential) HP RPCs */
286         struct list_head        exp_reg_rpcs;  /* RPC being handled */
287
288         /** blocking dlm lock list, protected by exp_bl_list_lock */
289         struct list_head        exp_bl_list;
290         spinlock_t              exp_bl_list_lock;
291
292         /** Target specific data */
293         union {
294                 struct tg_export_data     eu_target_data;
295                 struct mdt_export_data    eu_mdt_data;
296                 struct filter_export_data eu_filter_data;
297                 struct ec_export_data     eu_ec_data;
298                 struct mgs_export_data    eu_mgs_data;
299         } u;
300
301         struct adaptive_timeout    exp_bl_lock_at;
302
303         /** highest XID received by export client that has no
304          * unreceived lower-numbered XID
305          */
306         __u64                     exp_last_xid;
307 };
308
309 #define exp_target_data u.eu_target_data
310 #define exp_mdt_data    u.eu_mdt_data
311 #define exp_filter_data u.eu_filter_data
312 #define exp_ec_data     u.eu_ec_data
313
314 static inline __u64 *exp_connect_flags_ptr(struct obd_export *exp)
315 {
316         return &exp->exp_connect_data.ocd_connect_flags;
317 }
318
319 static inline __u64 exp_connect_flags(struct obd_export *exp)
320 {
321         return *exp_connect_flags_ptr(exp);
322 }
323
324 static inline __u64 *exp_connect_flags2_ptr(struct obd_export *exp)
325 {
326         return &exp->exp_connect_data.ocd_connect_flags2;
327 }
328
329 static inline __u64 exp_connect_flags2(struct obd_export *exp)
330 {
331         if (exp_connect_flags(exp) & OBD_CONNECT_FLAGS2)
332                 return *exp_connect_flags2_ptr(exp);
333         return 0;
334 }
335
336 static inline int exp_max_brw_size(struct obd_export *exp)
337 {
338         LASSERT(exp != NULL);
339         if (exp_connect_flags(exp) & OBD_CONNECT_BRW_SIZE)
340                 return exp->exp_connect_data.ocd_brw_size;
341
342         return ONE_MB_BRW_SIZE;
343 }
344
345 static inline int exp_connect_multibulk(struct obd_export *exp)
346 {
347         return exp_max_brw_size(exp) > ONE_MB_BRW_SIZE;
348 }
349
350 static inline int exp_connect_cancelset(struct obd_export *exp)
351 {
352         LASSERT(exp != NULL);
353         return !!(exp_connect_flags(exp) & OBD_CONNECT_CANCELSET);
354 }
355
356 static inline int exp_connect_lru_resize(struct obd_export *exp)
357 {
358         LASSERT(exp != NULL);
359         return !!(exp_connect_flags(exp) & OBD_CONNECT_LRU_RESIZE);
360 }
361
362 static inline int exp_connect_vbr(struct obd_export *exp)
363 {
364         LASSERT(exp != NULL);
365         LASSERT(exp->exp_connection);
366         return !!(exp_connect_flags(exp) & OBD_CONNECT_VBR);
367 }
368
369 static inline int exp_connect_umask(struct obd_export *exp)
370 {
371         return !!(exp_connect_flags(exp) & OBD_CONNECT_UMASK);
372 }
373
374 static inline int imp_connect_lru_resize(struct obd_import *imp)
375 {
376         struct obd_connect_data *ocd;
377
378         LASSERT(imp != NULL);
379         ocd = &imp->imp_connect_data;
380         return !!(ocd->ocd_connect_flags & OBD_CONNECT_LRU_RESIZE);
381 }
382
383 static inline int exp_connect_layout(struct obd_export *exp)
384 {
385         return !!(exp_connect_flags(exp) & OBD_CONNECT_LAYOUTLOCK);
386 }
387
388 static inline bool exp_connect_lvb_type(struct obd_export *exp)
389 {
390         LASSERT(exp != NULL);
391         if (exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE)
392                 return true;
393         else
394                 return false;
395 }
396
397 static inline bool imp_connect_lvb_type(struct obd_import *imp)
398 {
399         struct obd_connect_data *ocd;
400
401         LASSERT(imp != NULL);
402         ocd = &imp->imp_connect_data;
403         if (ocd->ocd_connect_flags & OBD_CONNECT_LVB_TYPE)
404                 return true;
405         else
406                 return false;
407 }
408
409 static inline bool imp_connect_disp_stripe(struct obd_import *imp)
410 {
411         struct obd_connect_data *ocd;
412
413         LASSERT(imp != NULL);
414         ocd = &imp->imp_connect_data;
415         return ocd->ocd_connect_flags & OBD_CONNECT_DISP_STRIPE;
416 }
417
418 static inline bool imp_connect_shortio(struct obd_import *imp)
419 {
420         struct obd_connect_data *ocd = &imp->imp_connect_data;
421
422         return ocd->ocd_connect_flags & OBD_CONNECT_SHORTIO;
423 }
424
425 static inline __u64 exp_connect_ibits(struct obd_export *exp)
426 {
427         struct obd_connect_data *ocd;
428
429         ocd = &exp->exp_connect_data;
430         return ocd->ocd_ibits_known;
431 }
432
433 static inline int exp_connect_large_acl(struct obd_export *exp)
434 {
435         return !!(exp_connect_flags(exp) & OBD_CONNECT_LARGE_ACL);
436 }
437
438 static inline int exp_connect_lockahead_old(struct obd_export *exp)
439 {
440         return !!(exp_connect_flags(exp) & OBD_CONNECT_LOCKAHEAD_OLD);
441 }
442
443 static inline int exp_connect_lockahead(struct obd_export *exp)
444 {
445         return !!(exp_connect_flags2(exp) & OBD_CONNECT2_LOCKAHEAD);
446 }
447
448 static inline int exp_connect_flr(struct obd_export *exp)
449 {
450         return !!(exp_connect_flags2(exp) & OBD_CONNECT2_FLR);
451 }
452
453 static inline int exp_connect_lock_convert(struct obd_export *exp)
454 {
455         return !!(exp_connect_flags2(exp) & OBD_CONNECT2_LOCK_CONVERT);
456 }
457
458 extern struct obd_export *class_conn2export(struct lustre_handle *conn);
459 extern struct obd_device *class_conn2obd(struct lustre_handle *conn);
460
461 #define KKUC_CT_DATA_MAGIC      0x092013cea
462 struct kkuc_ct_data {
463         __u32           kcd_magic;
464         __u32           kcd_archive;
465 };
466
467 /** @} export */
468
469 #endif /* __EXPORT_H */
470 /** @} obd_export */