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;
58 /* all of sub updates are linked here */
59 struct list_head dtrq_sub_list;
60 spinlock_t dtrq_sub_list_lock;
63 /* Each one represents a sub replay item under a distribute
64 * transaction. A distribute transaction will be operated in
65 * two or more MDTs, and updates on each MDT will be represented
66 * by this structure */
67 struct distribute_txn_replay_req_sub {
68 __u32 dtrqs_mdt_index;
69 struct llog_cookie dtrqs_llog_cookie;
70 struct list_head dtrqs_list;
73 struct target_distribute_txn_data;
74 typedef int (*distribute_txn_replay_handler_t)(struct lu_env *env,
75 struct target_distribute_txn_data *tdtd,
76 struct distribute_txn_replay_req *dtrq);
77 struct target_distribute_txn_data {
78 /* Distribution ID is used to identify updates log on different
79 * MDTs for one operation */
80 spinlock_t tdtd_batchid_lock;
82 struct lu_target *tdtd_lut;
83 struct dt_object *tdtd_batchid_obj;
84 struct dt_device *tdtd_dt;
86 /* Committed batchid for distribute transaction */
87 __u64 tdtd_committed_batchid;
89 /* List for distribute transaction */
90 struct list_head tdtd_list;
92 /* Threads to manage distribute transaction */
93 wait_queue_head_t tdtd_commit_thread_waitq;
94 atomic_t tdtd_refcount;
97 distribute_txn_replay_handler_t tdtd_replay_handler;
98 struct list_head tdtd_replay_list;
99 spinlock_t tdtd_replay_list_lock;
100 /* last replay update transno */
101 __u64 tdtd_last_update_transno;
102 __u32 tdtd_replay_ready:1;
107 struct obd_device *lut_obd;
108 struct dt_device *lut_bottom;
110 struct target_distribute_txn_data *lut_tdtd;
111 struct ptlrpc_thread lut_tdtd_commit_thread;
113 /* supported opcodes and handlers for this target */
114 struct tgt_opc_slice *lut_slice;
115 __u32 lut_reply_fail_id;
116 __u32 lut_request_fail_id;
119 rwlock_t lut_sptlrpc_lock;
120 struct sptlrpc_rule_set lut_sptlrpc_rset;
121 spinlock_t lut_flags_lock;
123 unsigned int lut_syncjournal:1,
124 lut_sync_lock_cancel:2,
126 lut_no_reconstruct:1;
127 /** last_rcvd file */
128 struct dt_object *lut_last_rcvd;
129 /* transaction callbacks */
130 struct dt_txn_callback lut_txn_cb;
131 /** server data in last_rcvd file */
132 struct lr_server_data lut_lsd;
133 /** Server last transaction number */
134 __u64 lut_last_transno;
135 /** Lock protecting last transaction number */
136 spinlock_t lut_translock;
137 /** Lock protecting client bitmap */
138 spinlock_t lut_client_bitmap_lock;
139 /** Bitmap of known clients */
140 unsigned long *lut_client_bitmap;
143 extern struct lu_context_key tgt_session_key;
145 struct tgt_session_info {
147 * The following members will be filled explicitly
148 * with specific data in tgt_ses_init().
150 struct req_capsule *tsi_pill;
153 * Lock request for "habeo clavis" operations.
155 struct ldlm_request *tsi_dlm_req;
157 /* although we have export in req, there are cases when it is not
158 * available, e.g. closing files upon export destroy */
159 struct obd_export *tsi_exp;
160 const struct lu_env *tsi_env;
161 struct lu_target *tsi_tgt;
163 const struct mdt_body *tsi_mdt_body;
164 struct ost_body *tsi_ost_body;
165 struct lu_object *tsi_corpus;
167 struct lu_fid tsi_fid;
168 struct ldlm_res_id tsi_resid;
170 /* object affected by VBR, for last_rcvd_update */
171 struct dt_object *tsi_vbr_obj;
172 /* opdata for mdt_reint_open(), has the same value as
173 * ldlm_reply:lock_policy_res1. The tgt_update_last_rcvd() stores
174 * this value onto disk for recovery when tgt_txn_stop_cb() is called.
179 * Additional fail id that can be set by handler.
181 int tsi_reply_fail_id;
182 bool tsi_preprocessed;
187 static inline struct tgt_session_info *tgt_ses_info(const struct lu_env *env)
189 struct tgt_session_info *tsi;
191 LASSERT(env->le_ses != NULL);
192 tsi = lu_context_key_get(env->le_ses, &tgt_session_key);
197 static inline void tgt_vbr_obj_set(const struct lu_env *env,
198 struct dt_object *obj)
200 struct tgt_session_info *tsi;
202 if (env->le_ses != NULL) {
203 tsi = tgt_ses_info(env);
204 tsi->tsi_vbr_obj = obj;
208 static inline void tgt_opdata_set(const struct lu_env *env, __u64 flags)
210 struct tgt_session_info *tsi;
212 if (env->le_ses != NULL) {
213 tsi = tgt_ses_info(env);
214 tsi->tsi_opdata |= flags;
218 static inline void tgt_opdata_clear(const struct lu_env *env, __u64 flags)
220 struct tgt_session_info *tsi;
222 if (env->le_ses != NULL) {
223 tsi = tgt_ses_info(env);
224 tsi->tsi_opdata &= ~flags;
229 * Generic unified target support.
231 enum tgt_handler_flags {
233 * struct *_body is passed in the incoming message, and object
234 * identified by this fid exists on disk.
236 * "habeo corpus" == "I have a body"
238 HABEO_CORPUS = (1 << 0),
240 * struct ldlm_request is passed in the incoming message.
242 * "habeo clavis" == "I have a key"
244 HABEO_CLAVIS = (1 << 1),
246 * this request has fixed reply format, so that reply message can be
247 * packed by generic code.
249 * "habeo refero" == "I have a reply"
251 HABEO_REFERO = (1 << 2),
253 * this request will modify something, so check whether the file system
254 * is readonly or not, then return -EROFS to client asap if necessary.
256 * "mutabor" == "I shall modify"
262 /* The name of this handler. */
264 /* Fail id, check at the beginning */
268 /* Flags in enum tgt_handler_flags */
270 /* Request version for this opcode */
272 /* Handler function */
273 int (*th_act)(struct tgt_session_info *tsi);
274 /* Handler function for high priority requests */
275 void (*th_hp)(struct tgt_session_info *tsi);
276 /* Request format for this request */
277 const struct req_format *th_fmt;
280 struct tgt_opc_slice {
281 __u32 tos_opc_start; /* First op code */
282 __u32 tos_opc_end; /* Last op code */
283 struct tgt_handler *tos_hs; /* Registered handler */
286 static inline struct ptlrpc_request *tgt_ses_req(struct tgt_session_info *tsi)
288 return tsi->tsi_pill ? tsi->tsi_pill->rc_req : NULL;
291 static inline __u64 tgt_conn_flags(struct tgt_session_info *tsi)
293 LASSERT(tsi->tsi_exp);
294 return exp_connect_flags(tsi->tsi_exp);
297 static inline int req_is_replay(struct ptlrpc_request *req)
299 LASSERT(req->rq_reqmsg);
300 return !!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
303 /* target/tgt_handler.c */
304 int tgt_request_handle(struct ptlrpc_request *req);
305 char *tgt_name(struct lu_target *tgt);
306 void tgt_counter_incr(struct obd_export *exp, int opcode);
307 int tgt_connect_check_sptlrpc(struct ptlrpc_request *req,
308 struct obd_export *exp);
309 int tgt_adapt_sptlrpc_conf(struct lu_target *tgt, int initial);
310 int tgt_connect(struct tgt_session_info *tsi);
311 int tgt_disconnect(struct tgt_session_info *uti);
312 int tgt_obd_ping(struct tgt_session_info *tsi);
313 int tgt_enqueue(struct tgt_session_info *tsi);
314 int tgt_convert(struct tgt_session_info *tsi);
315 int tgt_bl_callback(struct tgt_session_info *tsi);
316 int tgt_cp_callback(struct tgt_session_info *tsi);
317 int tgt_llog_open(struct tgt_session_info *tsi);
318 int tgt_llog_close(struct tgt_session_info *tsi);
319 int tgt_llog_destroy(struct tgt_session_info *tsi);
320 int tgt_llog_read_header(struct tgt_session_info *tsi);
321 int tgt_llog_next_block(struct tgt_session_info *tsi);
322 int tgt_llog_prev_block(struct tgt_session_info *tsi);
323 int tgt_sec_ctx_init(struct tgt_session_info *tsi);
324 int tgt_sec_ctx_init_cont(struct tgt_session_info *tsi);
325 int tgt_sec_ctx_fini(struct tgt_session_info *tsi);
326 int tgt_sendpage(struct tgt_session_info *tsi, struct lu_rdpg *rdpg, int nob);
327 int tgt_validate_obdo(struct tgt_session_info *tsi, struct obdo *oa);
328 int tgt_sync(const struct lu_env *env, struct lu_target *tgt,
329 struct dt_object *obj, __u64 start, __u64 end);
331 int tgt_io_thread_init(struct ptlrpc_thread *thread);
332 void tgt_io_thread_done(struct ptlrpc_thread *thread);
334 int tgt_extent_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
335 __u64 start, __u64 end, struct lustre_handle *lh,
336 int mode, __u64 *flags);
337 void tgt_extent_unlock(struct lustre_handle *lh, ldlm_mode_t mode);
338 int tgt_brw_lock(struct ldlm_namespace *ns, struct ldlm_res_id *res_id,
339 struct obd_ioobj *obj, struct niobuf_remote *nb,
340 struct lustre_handle *lh, int mode);
341 void tgt_brw_unlock(struct obd_ioobj *obj, struct niobuf_remote *niob,
342 struct lustre_handle *lh, int mode);
343 int tgt_brw_read(struct tgt_session_info *tsi);
344 int tgt_brw_write(struct tgt_session_info *tsi);
345 int tgt_hpreq_handler(struct ptlrpc_request *req);
346 void tgt_register_lfsck_in_notify(int (*notify)(const struct lu_env *,
348 struct lfsck_request *,
350 void tgt_register_lfsck_query(int (*query)(const struct lu_env *,
352 struct lfsck_request *));
354 extern struct tgt_handler tgt_lfsck_handlers[];
355 extern struct tgt_handler tgt_obd_handlers[];
356 extern struct tgt_handler tgt_dlm_handlers[];
357 extern struct tgt_handler tgt_llog_handlers[];
358 extern struct tgt_handler tgt_out_handlers[];
359 extern struct tgt_handler fld_handlers[];
360 extern struct tgt_handler seq_handlers[];
362 typedef void (*tgt_cb_t)(struct lu_target *lut, __u64 transno,
363 void *data, int err);
364 struct tgt_commit_cb {
365 tgt_cb_t tgt_cb_func;
369 int tgt_hpreq_handler(struct ptlrpc_request *req);
371 /* target/tgt_main.c */
372 void tgt_boot_epoch_update(struct lu_target *lut);
373 int tgt_last_commit_cb_add(struct thandle *th, struct lu_target *lut,
374 struct obd_export *exp, __u64 transno);
375 int tgt_new_client_cb_add(struct thandle *th, struct obd_export *exp);
376 int tgt_init(const struct lu_env *env, struct lu_target *lut,
377 struct obd_device *obd, struct dt_device *dt,
378 struct tgt_opc_slice *slice,
379 int request_fail_id, int reply_fail_id);
380 void tgt_fini(const struct lu_env *env, struct lu_target *lut);
381 int tgt_client_alloc(struct obd_export *exp);
382 void tgt_client_free(struct obd_export *exp);
383 int tgt_client_del(const struct lu_env *env, struct obd_export *exp);
384 int tgt_client_add(const struct lu_env *env, struct obd_export *exp, int);
385 int tgt_client_new(const struct lu_env *env, struct obd_export *exp);
386 int tgt_client_data_read(const struct lu_env *env, struct lu_target *tg,
387 struct lsd_client_data *lcd, loff_t *off, int index);
388 int tgt_client_data_write(const struct lu_env *env, struct lu_target *tg,
389 struct lsd_client_data *lcd, loff_t *off, struct thandle *th);
390 int tgt_server_data_read(const struct lu_env *env, struct lu_target *tg);
391 int tgt_server_data_write(const struct lu_env *env, struct lu_target *tg,
393 int tgt_server_data_update(const struct lu_env *env, struct lu_target *tg,
395 int tgt_truncate_last_rcvd(const struct lu_env *env, struct lu_target *tg,
398 /* target/update_trans.c */
399 int distribute_txn_init(const struct lu_env *env,
400 struct lu_target *lut,
401 struct target_distribute_txn_data *tdtd,
403 void distribute_txn_fini(const struct lu_env *env,
404 struct target_distribute_txn_data *tdtd);
406 /* target/update_recovery.c */
407 int insert_update_records_to_replay_list(struct target_distribute_txn_data *,
408 struct llog_update_record *,
409 struct llog_cookie *, __u32);
410 void dtrq_list_dump(struct target_distribute_txn_data *tdtd,
412 void dtrq_list_destroy(struct target_distribute_txn_data *tdtd);
413 int distribute_txn_replay_handle(struct lu_env *env,
414 struct target_distribute_txn_data *tdtd,
415 struct distribute_txn_replay_req *dtrq);
416 __u64 distribute_txn_get_next_transno(struct target_distribute_txn_data *tdtd);
417 struct distribute_txn_replay_req *
418 distribute_txn_get_next_req(struct target_distribute_txn_data *tdtd);
419 void dtrq_destroy(struct distribute_txn_replay_req *dtrq);
420 struct distribute_txn_replay_req_sub *
421 dtrq_sub_lookup(struct distribute_txn_replay_req *dtrq, __u32 mdt_index);
427 static inline int err_serious(int rc)
430 return -(-rc | ESERIOUS);
433 static inline int clear_serious(int rc)
436 rc = -(-rc & ~ESERIOUS);
440 static inline int is_serious(int rc)
442 return (rc < 0 && -rc & ESERIOUS);
446 * Do not return server-side uid/gid to remote client
448 static inline void tgt_drop_id(struct obd_export *exp, struct obdo *oa)
450 if (unlikely(exp_connect_rmtclient(exp))) {
453 oa->o_valid &= ~(OBD_MD_FLUID | OBD_MD_FLGID);
458 * Unified target generic handers macros and generic functions.
460 #define TGT_RPC_HANDLER_HP(base, flags, opc, fn, hp, fmt, version) \
463 .th_fail_id = OBD_FAIL_ ## opc ## _NET, \
468 .th_version = version, \
471 #define TGT_RPC_HANDLER(base, flags, opc, fn, fmt, version) \
472 TGT_RPC_HANDLER_HP(base, flags, opc, fn, NULL, fmt, version)
474 /* MDT Request with a format known in advance */
475 #define TGT_MDT_HDL(flags, name, fn) \
476 TGT_RPC_HANDLER(MDS_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
478 /* Request with a format we do not yet know */
479 #define TGT_MDT_HDL_VAR(flags, name, fn) \
480 TGT_RPC_HANDLER(MDS_FIRST_OPC, flags, name, fn, NULL, \
483 /* OST Request with a format known in advance */
484 #define TGT_OST_HDL(flags, name, fn) \
485 TGT_RPC_HANDLER(OST_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
487 #define TGT_OST_HDL_HP(flags, name, fn, hp) \
488 TGT_RPC_HANDLER_HP(OST_FIRST_OPC, flags, name, fn, hp, \
489 &RQF_ ## name, LUSTRE_OST_VERSION)
491 /* MGS request with a format known in advance */
492 #define TGT_MGS_HDL(flags, name, fn) \
493 TGT_RPC_HANDLER(MGS_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
495 #define TGT_MGS_HDL_VAR(flags, name, fn) \
496 TGT_RPC_HANDLER(MGS_FIRST_OPC, flags, name, fn, NULL, \
500 * OBD handler macros and generic functions.
502 #define TGT_OBD_HDL(flags, name, fn) \
503 TGT_RPC_HANDLER(OBD_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
505 #define TGT_OBD_HDL_VAR(flags, name, fn) \
506 TGT_RPC_HANDLER(OBD_FIRST_OPC, flags, name, fn, NULL, \
510 * DLM handler macros and generic functions.
512 #define TGT_DLM_HDL_VAR(flags, name, fn) \
513 TGT_RPC_HANDLER(LDLM_FIRST_OPC, flags, name, fn, NULL, \
515 #define TGT_DLM_HDL(flags, name, fn) \
516 TGT_RPC_HANDLER(LDLM_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
520 * LLOG handler macros and generic functions.
522 #define TGT_LLOG_HDL_VAR(flags, name, fn) \
523 TGT_RPC_HANDLER(LLOG_FIRST_OPC, flags, name, fn, NULL, \
525 #define TGT_LLOG_HDL(flags, name, fn) \
526 TGT_RPC_HANDLER(LLOG_FIRST_OPC, flags, name, fn, &RQF_ ## name, \
530 * Sec context handler macros and generic functions.
532 #define TGT_SEC_HDL_VAR(flags, name, fn) \
533 TGT_RPC_HANDLER(SEC_FIRST_OPC, flags, name, fn, NULL, \
536 #define TGT_QUOTA_HDL(flags, name, fn) \
537 TGT_RPC_HANDLER(QUOTA_DQACQ, flags, name, fn, &RQF_ ## name, \
540 /* Sequence service handlers */
541 #define TGT_SEQ_HDL(flags, name, fn) \
542 TGT_RPC_HANDLER(SEQ_QUERY, flags, name, fn, &RQF_ ## name, \
545 /* FID Location Database handlers */
546 #define TGT_FLD_HDL_VAR(flags, name, fn) \
547 TGT_RPC_HANDLER(FLD_QUERY, flags, name, fn, NULL, \
551 #define TGT_LFSCK_HDL(flags, name, fn) \
552 TGT_RPC_HANDLER(LFSCK_FIRST_OPC, flags, name, fn, \
553 &RQF_ ## name, LUSTRE_OBD_VERSION)
555 /* Request with a format known in advance */
556 #define TGT_UPDATE_HDL(flags, name, fn) \
557 TGT_RPC_HANDLER(OUT_UPDATE, flags, name, fn, &RQF_ ## name, \
560 #endif /* __LUSTRE_LU_TARGET_H */