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.sun.com/software/products/lustre/docs/GPLv2.pdf
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
27 * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef _LUSTRE_LU_TARGET_H
38 #define _LUSTRE_LU_TARGET_H
40 #include <dt_object.h>
41 #include <lustre_export.h>
42 #include <lustre_update.h>
43 #include <lustre_disk.h>
44 #include <lustre_lfsck.h>
46 /* Each one represents a distribute transaction replay
47 * operation, and updates on each MDTs are linked to
49 struct distribute_txn_replay_req {
51 struct llog_update_record *dtrq_lur;
54 /* linked to the distribute transaction replay
55 * list (tdtd_replay_list) */
56 struct list_head dtrq_list;
57 __u64 dtrq_master_transno;
61 /* all of sub updates are linked here */
62 struct list_head dtrq_sub_list;
63 spinlock_t dtrq_sub_list_lock;
65 /* If the local update has been executed during replay */
66 __u32 dtrq_local_update_executed:1;
69 /* Each one represents a sub replay item under a distribute
70 * transaction. A distribute transaction will be operated in
71 * two or more MDTs, and updates on each MDT will be represented
72 * by this structure */
73 struct distribute_txn_replay_req_sub {
74 __u32 dtrqs_mdt_index;
76 /* All of cookies for the update will be linked here */
77 spinlock_t dtrqs_cookie_list_lock;
78 struct list_head dtrqs_cookie_list;
79 struct list_head dtrqs_list;
82 struct target_distribute_txn_data;
83 typedef int (*distribute_txn_replay_handler_t)(struct lu_env *env,
84 struct target_distribute_txn_data *tdtd,
85 struct distribute_txn_replay_req *dtrq);
86 struct target_distribute_txn_data {
87 /* Distribution ID is used to identify updates log on different
88 * MDTs for one operation */
89 spinlock_t tdtd_batchid_lock;
91 struct lu_target *tdtd_lut;
92 struct dt_object *tdtd_batchid_obj;
93 struct dt_device *tdtd_dt;
95 /* Committed batchid for distribute transaction */
96 __u64 tdtd_committed_batchid;
98 /* List for distribute transaction */
99 struct list_head tdtd_list;
101 /* Threads to manage distribute transaction */
102 wait_queue_head_t tdtd_commit_thread_waitq;
103 atomic_t tdtd_refcount;
105 /* recovery update */
106 distribute_txn_replay_handler_t tdtd_replay_handler;
107 struct list_head tdtd_replay_list;
108 struct list_head tdtd_replay_finish_list;
109 spinlock_t tdtd_replay_list_lock;
110 /* last replay update transno */
111 __u32 tdtd_replay_ready:1;
115 struct obd_device *lut_obd;
116 struct dt_device *lut_bottom;
118 struct target_distribute_txn_data *lut_tdtd;
119 struct ptlrpc_thread lut_tdtd_commit_thread;
121 /* supported opcodes and handlers for this target */
122 struct tgt_opc_slice *lut_slice;
123 __u32 lut_reply_fail_id;
124 __u32 lut_request_fail_id;
127 rwlock_t lut_sptlrpc_lock;
128 struct sptlrpc_rule_set lut_sptlrpc_rset;
129 spinlock_t lut_flags_lock;
131 unsigned int lut_syncjournal:1,
132 lut_sync_lock_cancel:2,
134 lut_no_reconstruct:1;
135 /** last_rcvd file */
136 struct dt_object *lut_last_rcvd;
137 /* transaction callbacks */
138 struct dt_txn_callback lut_txn_cb;
139 /** server data in last_rcvd file */
140 struct lr_server_data lut_lsd;
141 /** Server last transaction number */
142 __u64 lut_last_transno;
143 /** Lock protecting last transaction number */
144 spinlock_t lut_translock;
145 /** Lock protecting client bitmap */
146 spinlock_t lut_client_bitmap_lock;
147 /** Bitmap of known clients */
148 unsigned long *lut_client_bitmap;
149 /* Number of clients supporting multiple modify RPCs
150 * recorded in the bitmap */
151 atomic_t lut_num_clients;
152 /* Client generation to identify client slot reuse */
153 atomic_t lut_client_generation;
154 /** reply_data file */
155 struct dt_object *lut_reply_data;
156 /** Bitmap of used slots in the reply data file */
157 unsigned long **lut_reply_bitmap;
160 /* number of slots in reply bitmap */
161 #define LUT_REPLY_SLOTS_PER_CHUNK (1<<20)
162 #define LUT_REPLY_SLOTS_MAX_CHUNKS 16
167 struct tg_reply_data {
168 /** chain of reply data anchored in tg_export_data */
169 struct list_head trd_list;
170 /** copy of on-disk reply data */
171 struct lsd_reply_data trd_reply;
172 /** versions for Version Based Recovery */
173 __u64 trd_pre_versions[4];
174 /** slot index in reply_data file */
176 /** tag the client used */
180 extern struct lu_context_key tgt_session_key;
182 struct tgt_session_info {
184 * The following members will be filled explicitly
185 * with specific data in tgt_ses_init().
187 struct req_capsule *tsi_pill;
190 * Lock request for "habeo clavis" operations.
192 struct ldlm_request *tsi_dlm_req;
194 /* although we have export in req, there are cases when it is not
195 * available, e.g. closing files upon export destroy */
196 struct obd_export *tsi_exp;
197 const struct lu_env *tsi_env;
198 struct lu_target *tsi_tgt;
200 const struct mdt_body *tsi_mdt_body;
201 struct ost_body *tsi_ost_body;
202 struct lu_object *tsi_corpus;
204 struct lu_fid tsi_fid;
205 struct ldlm_res_id tsi_resid;
207 /* object affected by VBR, for last_rcvd_update */
208 struct dt_object *tsi_vbr_obj;
209 /* opdata for mdt_reint_open(), has the same value as
210 * ldlm_reply:lock_policy_res1. The tgt_update_last_rcvd() stores
211 * this value onto disk for recovery when tgt_txn_stop_cb() is called.
216 * Additional fail id that can be set by handler.
218 int tsi_reply_fail_id;
219 bool tsi_preprocessed;
226 __u32 tsi_client_gen;
229 static inline struct tgt_session_info *tgt_ses_info(const struct lu_env *env)
231 struct tgt_session_info *tsi;
233 LASSERT(env->le_ses != NULL);
234 tsi = lu_context_key_get(env->le_ses, &tgt_session_key);
239 static inline void tgt_vbr_obj_set(const struct lu_env *env,
240 struct dt_object *obj)
242 struct tgt_session_info *tsi;
244 if (env->le_ses != NULL) {
245 tsi = tgt_ses_info(env);
246 tsi->tsi_vbr_obj = obj;
250 static inline void tgt_opdata_set(const struct lu_env *env, __u64 flags)
252 struct tgt_session_info *tsi;
254 if (env->le_ses != NULL) {
255 tsi = tgt_ses_info(env);
256 tsi->tsi_opdata |= flags;
260 static inline void tgt_opdata_clear(const struct lu_env *env, __u64 flags)
262 struct tgt_session_info *tsi;
264 if (env->le_ses != NULL) {
265 tsi = tgt_ses_info(env);
266 tsi->tsi_opdata &= ~flags;
271 * Generic unified target support.
273 enum tgt_handler_flags {
275 * struct *_body is passed in the incoming message, and object
276 * identified by this fid exists on disk.
278 * "habeo corpus" == "I have a body"
280 HABEO_CORPUS = (1 << 0),
282 * struct ldlm_request is passed in the incoming message.
284 * "habeo clavis" == "I have a key"
286 HABEO_CLAVIS = (1 << 1),
288 * this request has fixed reply format, so that reply message can be
289 * packed by generic code.
291 * "habeo refero" == "I have a reply"
293 HABEO_REFERO = (1 << 2),
295 * this request will modify something, so check whether the file system
296 * is readonly or not, then return -EROFS to client asap if necessary.
298 * "mutabor" == "I shall modify"
304 /* The name of this handler. */
306 /* Fail id, check at the beginning */
310 /* Flags in enum tgt_handler_flags */
312 /* Request version for this opcode */
314 /* Handler function */
315 int (*th_act)(struct tgt_session_info *tsi);
316 /* Handler function for high priority requests */
317 void (*th_hp)(struct tgt_session_info *tsi);
318 /* Request format for this request */
319 const struct req_format *th_fmt;
322 struct tgt_opc_slice {
323 __u32 tos_opc_start; /* First op code */
324 __u32 tos_opc_end; /* Last op code */
325 struct tgt_handler *tos_hs; /* Registered handler */
328 static inline struct ptlrpc_request *tgt_ses_req(struct tgt_session_info *tsi)
330 return tsi->tsi_pill ? tsi->tsi_pill->rc_req : NULL;
333 static inline __u64 tgt_conn_flags(struct tgt_session_info *tsi)
335 LASSERT(tsi->tsi_exp);
336 return exp_connect_flags(tsi->tsi_exp);
339 static inline int req_is_replay(struct ptlrpc_request *req)
341 LASSERT(req->rq_reqmsg);
342 return !!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
345 static inline bool tgt_is_multimodrpcs_client(struct obd_export *exp)
347 return exp_connect_flags(exp) & OBD_CONNECT_MULTIMODRPCS;
351 /* target/tgt_handler.c */
352 int tgt_request_handle(struct ptlrpc_request *req);
353 char *tgt_name(struct lu_target *tgt);
354 void tgt_counter_incr(struct obd_export *exp, int opcode);
355 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req,
356 struct obd_export *exp);
357 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial);
358 int tgt_connect(struct tgt_session_info *tsi);
359 int tgt_disconnect(struct tgt_session_info *uti);
360 int tgt_obd_ping(struct tgt_session_info *tsi);
361 int tgt_enqueue(struct tgt_session_info *tsi);
362 int tgt_convert(struct tgt_session_info *tsi);
363 int tgt_bl_callback(struct tgt_session_info *tsi);
364 int tgt_cp_callback(struct tgt_session_info *tsi);
365 int tgt_llog_open(struct tgt_session_info *tsi);
366 int tgt_llog_close(struct tgt_session_info *tsi);
367 int tgt_llog_destroy(struct tgt_session_info *tsi);
368 int tgt_llog_read_header(struct tgt_session_info *tsi);
369 int tgt_llog_next_block(struct tgt_session_info *tsi);
370 int tgt_llog_prev_block(struct tgt_session_info *tsi);
371 int tgt_sec_ctx_init(struct tgt_session_info *tsi);
372 int tgt_sec_ctx_init_cont(struct tgt_session_info *tsi);
373 int tgt_sec_ctx_fini(struct tgt_session_info *tsi);
374 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob);
375 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa);
376 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
377 struct dt_object *obj, __u64 start, __u64 end);
379 int tgt_io_thread_init(struct ptlrpc_thread *thread);
380 void tgt_io_thread_done(struct ptlrpc_thread *thread);
382 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
383 __u64 start, __u64 end, struct lustre_handle *lh,
384 int mode, __u64 *flags);
385 void tgt_extent_unlock(struct lustre_handle *lh, enum ldlm_mode mode);
386 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
387 struct obd_ioobj *obj, struct niobuf_remote *nb,
388 struct lustre_handle *lh, enum ldlm_mode mode);
389 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
390 struct lustre_handle *lh, enum ldlm_mode mode);
391 int tgt_brw_read(struct tgt_session_info *tsi);
392 int tgt_brw_write(struct tgt_session_info *tsi);
393 int tgt_hpreq_handler(struct ptlrpc_request *req);
394 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
396 struct lfsck_request *,
398 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
400 struct lfsck_request *));
401 bool req_can_reconstruct(struct ptlrpc_request *req, struct tg_reply_data *trd);
403 extern struct tgt_handler tgt_lfsck_handlers[];
404 extern struct tgt_handler tgt_obd_handlers[];
405 extern struct tgt_handler tgt_dlm_handlers[];
406 extern struct tgt_handler tgt_llog_handlers[];
407 extern struct tgt_handler tgt_out_handlers[];
408 extern struct tgt_handler fld_handlers[];
409 extern struct tgt_handler seq_handlers[];
411 typedef void (*tgt_cb_t)(struct lu_target *lut, __u64 transno,
412 void *data, int err);
413 struct tgt_commit_cb {
414 tgt_cb_t tgt_cb_func;
418 int tgt_hpreq_handler(struct ptlrpc_request *req);
420 /* target/tgt_main.c */
421 void tgt_boot_epoch_update(struct lu_target *lut);
422 int tgt_last_commit_cb_add(struct thandle *th, struct lu_target *lut,
423 struct obd_export *exp, __u64 transno);
424 int tgt_new_client_cb_add(struct thandle *th, struct obd_export *exp);
425 int tgt_init(const struct lu_env *env, struct lu_target *lut,
426 struct obd_device *obd, struct dt_device *dt,
427 struct tgt_opc_slice *slice,
428 int request_fail_id, int reply_fail_id);
429 void tgt_fini(const struct lu_env *env, struct lu_target *lut);
430 int tgt_client_alloc(struct obd_export *exp);
431 void tgt_client_free(struct obd_export *exp);
432 int tgt_client_del(const struct lu_env *env, struct obd_export *exp);
433 int tgt_client_add(const struct lu_env *env, struct obd_export *exp, int);
434 int tgt_client_new(const struct lu_env *env, struct obd_export *exp);
435 int tgt_client_data_read(const struct lu_env *env, struct lu_target *tg,
436 struct lsd_client_data *lcd, loff_t *off, int index);
437 int tgt_client_data_write(const struct lu_env *env, struct lu_target *tg,
438 struct lsd_client_data *lcd, loff_t *off, struct thandle *th);
439 int tgt_server_data_read(const struct lu_env *env, struct lu_target *tg);
440 int tgt_server_data_write(const struct lu_env *env, struct lu_target *tg,
442 int tgt_server_data_update(const struct lu_env *env, struct lu_target *tg,
444 int tgt_truncate_last_rcvd(const struct lu_env *env, struct lu_target *tg,
446 int tgt_reply_data_init(const struct lu_env *env, struct lu_target *tgt);
447 bool tgt_lookup_reply(struct ptlrpc_request *req, struct tg_reply_data *trd);
448 int tgt_add_reply_data(const struct lu_env *env, struct lu_target *tgt,
449 struct tg_export_data *ted, struct tg_reply_data *trd,
450 struct thandle *th, bool update_lrd_file);
451 struct tg_reply_data *tgt_lookup_reply_by_xid(struct tg_export_data *ted,
454 /* target/update_trans.c */
455 int distribute_txn_init(const struct lu_env *env,
456 struct lu_target *lut,
457 struct target_distribute_txn_data *tdtd,
459 void distribute_txn_fini(const struct lu_env *env,
460 struct target_distribute_txn_data *tdtd);
462 /* target/update_recovery.c */
463 int insert_update_records_to_replay_list(struct target_distribute_txn_data *,
464 struct llog_update_record *,
465 struct llog_cookie *, __u32);
466 void dtrq_list_dump(struct target_distribute_txn_data *tdtd,
468 void dtrq_list_destroy(struct target_distribute_txn_data *tdtd);
469 int distribute_txn_replay_handle(struct lu_env *env,
470 struct target_distribute_txn_data *tdtd,
471 struct distribute_txn_replay_req *dtrq);
472 __u64 distribute_txn_get_next_transno(struct target_distribute_txn_data *tdtd);
473 struct distribute_txn_replay_req *
474 distribute_txn_get_next_req(struct target_distribute_txn_data *tdtd);
475 void dtrq_destroy(struct distribute_txn_replay_req *dtrq);
476 struct distribute_txn_replay_req_sub *
477 dtrq_sub_lookup(struct distribute_txn_replay_req *dtrq, __u32 mdt_index);
478 struct distribute_txn_replay_req *
479 distribute_txn_lookup_finish_list(struct target_distribute_txn_data *tdtd,
481 bool is_req_replayed_by_update(struct ptlrpc_request *req);
486 static inline int err_serious(int rc)
489 return -(-rc | ESERIOUS);
492 static inline int clear_serious(int rc)
495 rc = -(-rc & ~ESERIOUS);
499 static inline int is_serious(int rc)
501 return (rc < 0 && -rc & ESERIOUS);
505 * Do not return server-side uid/gid to remote client
507 static inline void tgt_drop_id(struct obd_export *exp, struct obdo *oa)
509 if (unlikely(exp_connect_rmtclient(exp))) {
512 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
517 * Unified target generic handers macros and generic functions.
519 #define TGT_RPC_HANDLER_HP(base, flags, opc, fn, hp, fmt, version) \
522 .th_fail_id = OBD_FAIL_ ## opc ## _NET, \
527 .th_version = version, \
530 #define TGT_RPC_HANDLER(base, flags, opc, fn, fmt, version) \
531 TGT_RPC_HANDLER_HP(base, flags, opc, fn, NULL, fmt, version)
533 /* MDT Request with a format known in advance */
534 #define TGT_MDT_HDL(flags, name, fn) \
535 TGT_RPC_HANDLER(MDS_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
537 /* Request with a format we do not yet know */
538 #define TGT_MDT_HDL_VAR(flags, name, fn) \
539 TGT_RPC_HANDLER(MDS_FIRST_OPC, flags, name, fn, NULL, \
542 /* OST Request with a format known in advance */
543 #define TGT_OST_HDL(flags, name, fn) \
544 TGT_RPC_HANDLER(OST_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
546 #define TGT_OST_HDL_HP(flags, name, fn, hp) \
547 TGT_RPC_HANDLER_HP(OST_FIRST_OPC, flags, name, fn, hp, \
548 &RQF_ ## name, LUSTRE_OST_VERSION)
550 /* MGS request with a format known in advance */
551 #define TGT_MGS_HDL(flags, name, fn) \
552 TGT_RPC_HANDLER(MGS_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
554 #define TGT_MGS_HDL_VAR(flags, name, fn) \
555 TGT_RPC_HANDLER(MGS_FIRST_OPC, flags, name, fn, NULL, \
559 * OBD handler macros and generic functions.
561 #define TGT_OBD_HDL(flags, name, fn) \
562 TGT_RPC_HANDLER(OBD_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
564 #define TGT_OBD_HDL_VAR(flags, name, fn) \
565 TGT_RPC_HANDLER(OBD_FIRST_OPC, flags, name, fn, NULL, \
569 * DLM handler macros and generic functions.
571 #define TGT_DLM_HDL_VAR(flags, name, fn) \
572 TGT_RPC_HANDLER(LDLM_FIRST_OPC, flags, name, fn, NULL, \
574 #define TGT_DLM_HDL(flags, name, fn) \
575 TGT_RPC_HANDLER(LDLM_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
579 * LLOG handler macros and generic functions.
581 #define TGT_LLOG_HDL_VAR(flags, name, fn) \
582 TGT_RPC_HANDLER(LLOG_FIRST_OPC, flags, name, fn, NULL, \
584 #define TGT_LLOG_HDL(flags, name, fn) \
585 TGT_RPC_HANDLER(LLOG_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
589 * Sec context handler macros and generic functions.
591 #define TGT_SEC_HDL_VAR(flags, name, fn) \
592 TGT_RPC_HANDLER(SEC_FIRST_OPC, flags, name, fn, NULL, \
595 #define TGT_QUOTA_HDL(flags, name, fn) \
596 TGT_RPC_HANDLER(QUOTA_DQACQ, flags, name, fn, &RQF_ ## name, \
599 /* Sequence service handlers */
600 #define TGT_SEQ_HDL(flags, name, fn) \
601 TGT_RPC_HANDLER(SEQ_QUERY, flags, name, fn, &RQF_ ## name, \
604 /* FID Location Database handlers */
605 #define TGT_FLD_HDL_VAR(flags, name, fn) \
606 TGT_RPC_HANDLER(FLD_QUERY, flags, name, fn, NULL, \
610 #define TGT_LFSCK_HDL(flags, name, fn) \
611 TGT_RPC_HANDLER(LFSCK_FIRST_OPC, flags, name, fn, \
612 &RQF_ ## name, LUSTRE_OBD_VERSION)
614 /* Request with a format known in advance */
615 #define TGT_UPDATE_HDL(flags, name, fn) \
616 TGT_RPC_HANDLER(OUT_UPDATE, flags, name, fn, &RQF_ ## name, \
619 #endif /* __LUSTRE_LU_TARGET_H */