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) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_internal.h
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
48 #ifndef _MDT_INTERNAL_H
49 #define _MDT_INTERNAL_H
51 #if defined(__KERNEL__)
53 #include <obd_class.h>
54 #include <lustre_net.h>
55 #include <lustre/lustre_idl.h>
56 #include <lustre_disk.h>
57 #include <lu_target.h>
58 #include <md_object.h>
59 #include <lustre_fid.h>
60 #include <lustre_fld.h>
61 #include <lustre_mdt.h>
62 #include <lustre_req_layout.h>
63 #include <lustre_sec.h>
65 #include <lustre_idmap.h>
66 #include <lustre_eacl.h>
67 #include <lustre_quota.h>
69 /* check if request's xid is equal to last one or not*/
70 static inline int req_xid_is_last(struct ptlrpc_request *req)
72 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
73 return (req->rq_xid == lcd->lcd_last_xid ||
74 req->rq_xid == lcd->lcd_last_close_xid);
78 /* file data for open files on MDS */
79 struct mdt_file_data {
80 struct portals_handle mfd_handle; /* must be first */
81 int mfd_mode; /* open mode provided by client */
82 cfs_list_t mfd_list; /* protected by med_open_lock */
83 __u64 mfd_xid; /* xid of the open request */
84 struct lustre_handle mfd_old_handle; /* old handle in replay case */
85 struct mdt_object *mfd_object; /* point to opened object */
88 /* mdt state flag bits */
89 #define MDT_FL_CFGLOG 0
90 #define MDT_FL_SYNCED 1
94 struct lu_device mdt_lu_dev;
95 struct seq_server_site mdt_seq_site;
96 /* DLM name-space for meta-data locks maintained by this server */
97 struct ldlm_namespace *mdt_namespace;
98 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
99 struct ptlrpc_client *mdt_ldlm_client;
100 /* underlying device */
101 struct obd_export *mdt_child_exp;
102 struct md_device *mdt_child;
103 struct dt_device *mdt_bottom;
104 struct obd_export *mdt_bottom_exp;
106 struct lu_target mdt_lut;
108 * Options bit-fields.
111 unsigned int mo_user_xattr:1,
118 /* mdt state flags */
119 unsigned long mdt_state;
120 /* lock to protect IOepoch */
121 spinlock_t mdt_ioepoch_lock;
124 /* transaction callbacks */
125 struct dt_txn_callback mdt_txn_cb;
127 /* these values should be updated from lov if necessary.
128 * or should be placed somewhere else. */
130 int mdt_max_cookiesize;
132 struct upcall_cache *mdt_identity_cache;
135 rwlock_t mdt_sptlrpc_lock;
136 struct sptlrpc_rule_set mdt_sptlrpc_rset;
138 /* capability keys */
139 unsigned long mdt_capa_timeout;
141 struct dt_object *mdt_ck_obj;
142 unsigned long mdt_ck_timeout;
143 unsigned long mdt_ck_expiry;
144 cfs_timer_t mdt_ck_timer;
145 struct ptlrpc_thread mdt_ck_thread;
146 struct lustre_capa_key mdt_capa_keys[2];
147 unsigned int mdt_capa_conf:1,
149 /* Enable remote dir on non-MDT0 */
150 mdt_enable_remote_dir:1;
152 gid_t mdt_enable_remote_dir_gid;
153 /* statfs optimization: we cache a bit */
154 struct obd_statfs mdt_osfs;
156 spinlock_t mdt_osfs_lock;
159 uid_t mdt_squash_uid;
160 gid_t mdt_squash_gid;
161 cfs_list_t mdt_nosquash_nids;
162 char *mdt_nosquash_str;
163 int mdt_nosquash_strlen;
164 struct rw_semaphore mdt_squash_sem;
167 struct rename_stats mdt_rename_stats;
168 struct lu_fid mdt_md_root_fid;
170 /* connection to quota master */
171 struct obd_export *mdt_qmt_exp;
172 /* quota master device associated with this MDT */
173 struct lu_device *mdt_qmt_dev;
176 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
177 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
178 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
179 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
180 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
181 #define MDT_COS_DEFAULT (0)
184 struct lu_object_header mot_header;
185 struct lu_object mot_obj;
188 int mot_ioepoch_count;
190 /* Lock to protect object's IO epoch. */
191 struct mutex mot_ioepoch_mutex;
192 /* Lock to protect create_data */
193 struct mutex mot_lov_mutex;
196 enum mdt_object_flags {
197 /** SOM attributes are changed. */
198 MOF_SOM_CHANGE = (1 << 0),
200 * The SOM recovery state for mdt object.
201 * This state is an in-memory equivalent of an absent SOM EA, used
202 * instead of invalidating SOM EA while IOEpoch is still opened when
203 * a client eviction occurs or a client fails to obtain SOM attributes.
204 * It indicates that the last IOEpoch holder will need to obtain SOM
205 * attributes under [0;EOF] extent lock to flush all the client's
206 * cached of evicted from MDS clients (but not necessary evicted from
207 * OST) before taking ost attributes.
209 MOF_SOM_RECOV = (1 << 1),
210 /** File has been just created. */
211 MOF_SOM_CREATED = (1 << 2),
212 /** lov object has been created. */
213 MOF_LOV_CREATED = (1 << 3),
216 struct mdt_lock_handle {
217 /* Lock type, reg for cross-ref use or pdo lock. */
221 struct lustre_handle mlh_reg_lh;
222 ldlm_mode_t mlh_reg_mode;
225 struct lustre_handle mlh_pdo_lh;
226 ldlm_mode_t mlh_pdo_mode;
227 unsigned int mlh_pdo_hash;
229 /* Remote regular lock */
230 struct lustre_handle mlh_rreg_lh;
231 ldlm_mode_t mlh_rreg_mode;
235 MDT_LH_PARENT, /* parent lockh */
236 MDT_LH_CHILD, /* child lockh */
237 MDT_LH_OLD, /* old lockh for rename */
238 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
239 MDT_LH_NEW, /* new lockh for rename */
240 MDT_LH_RMT, /* used for return lh to caller */
249 struct mdt_reint_record {
250 mdt_reint_t rr_opcode;
251 const struct lustre_handle *rr_handle;
252 const struct lu_fid *rr_fid1;
253 const struct lu_fid *rr_fid2;
258 const void *rr_eadata;
261 const struct llog_cookie *rr_logcookies;
265 enum mdt_reint_flag {
266 MRF_OPEN_TRUNC = 1 << 0,
269 struct mdt_thread_info;
271 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
275 tx_exec_func_t exec_fn;
276 tx_exec_func_t undo_fn;
277 struct dt_object *object;
279 struct update_reply *reply;
284 const struct dt_rec *rec;
285 const struct dt_key *key;
300 struct dt_allocation_hint hint;
301 struct dt_object_format dof;
309 struct ost_body *body;
314 #define TX_MAX_OPS 10
315 struct thandle_exec_args {
316 struct thandle *ta_handle;
317 struct dt_device *ta_dev;
319 struct tx_arg ta_args[TX_MAX_OPS];
320 int ta_argno; /* used args */
324 * Common data shared by mdt-level handlers. This is allocated per-thread to
325 * reduce stack consumption.
327 struct mdt_thread_info {
330 * The following members will be filled explicitly
331 * with specific data in mdt_thread_info_init().
333 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
334 * request handling may migrate from one server thread to another.
336 struct req_capsule *mti_pill;
338 /* although we have export in req, there are cases when it is not
339 * available, e.g. closing files upon export destroy */
340 struct obd_export *mti_exp;
342 * A couple of lock handles.
344 struct mdt_lock_handle mti_lh[MDT_LH_NR];
346 struct mdt_device *mti_mdt;
347 const struct lu_env *mti_env;
350 * Additional fail id that can be set by handler. Passed to
351 * target_send_reply().
355 /* transaction number of current request */
361 * The following members will be filled expilictly
362 * with zero in mdt_thread_info_init(). These members may be used
369 struct md_attr mti_attr;
371 * Body for "habeo corpus" operations.
373 const struct mdt_body *mti_body;
375 * Host object. This is released at the end of mdt_handler().
377 struct mdt_object *mti_object;
379 * Lock request for "habeo clavis" operations.
381 const struct ldlm_request *mti_dlm_req;
383 __u32 mti_has_trans:1, /* has txn already? */
386 /* opdata for mdt_reint_open(), has the same as
387 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
388 * value onto disk for recovery when mdt_trans_stop_cb() is called.
394 * The following members will be filled explicitly
395 * with zero in mdt_reint_unpack(), because they are only used
396 * by reint requests (including mdt_reint_open()).
400 * reint record. contains information for reint operations.
402 struct mdt_reint_record mti_rr;
404 /** md objects included in operation */
405 struct mdt_object *mti_mos;
406 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
408 * Operation specification (currently create and lookup)
410 struct md_op_spec mti_spec;
414 * The following members will _NOT_ be initialized at all.
415 * DO NOT expect them to contain any valid value.
416 * They should be initialized explicitly by the user themselves.
419 /* XXX: If something is in a union, make sure they do not conflict */
421 struct lu_fid mti_tmp_fid1;
422 struct lu_fid mti_tmp_fid2;
423 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
424 * mdt_rename_lock() */
425 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
428 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
429 char ns_name[48]; /* for mdt_init0() */
430 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
431 struct obd_statfs osfs; /* for mdt_statfs() */
433 /* for mdt_readpage() */
434 struct lu_rdpg mti_rdpg;
435 /* for mdt_sendpage() */
436 struct l_wait_info mti_wait_info;
440 struct md_som_data data;
443 struct dt_object_format mti_update_dof;
444 struct update_reply *mti_update_reply;
445 struct update *mti_update;
446 int mti_update_reply_index;
447 struct obdo mti_obdo;
448 struct dt_object *mti_dt_object;
452 /* IO epoch related stuff. */
453 struct mdt_ioepoch *mti_ioepoch;
454 __u64 mti_replayepoch;
457 struct lu_buf mti_buf;
458 struct lu_buf mti_big_buf;
459 struct lustre_capa_key mti_capa_key;
461 /* Ops object filename */
462 struct lu_name mti_name;
463 /* per-thread values, can be re-used */
466 /* big_lmm buffer was used and must be used in reply */
467 int mti_big_lmm_used;
468 /* should be enough to fit lustre_mdt_attrs */
469 char mti_xattr_buf[128];
470 struct thandle_exec_args mti_handle;
471 struct ldlm_enqueue_info mti_einfo;
474 /* ptlrpc request handler for MDT. All handlers are
475 * grouped into several slices - struct mdt_opc_slice,
476 * and stored in an array - mdt_handlers[].
479 /* The name of this handler. */
481 /* Fail id for this handler, checked at the beginning of this handler*/
483 /* Operation code for this handler */
485 /* flags are listed in enum mdt_handler_flags below. */
487 /* The actual handler function to execute. */
488 int (*mh_act)(struct mdt_thread_info *info);
489 /* Request format for this request. */
490 const struct req_format *mh_fmt;
493 enum mdt_handler_flags {
495 * struct mdt_body is passed in the incoming message, and object
496 * identified by this fid exists on disk.
498 * "habeo corpus" == "I have a body"
500 HABEO_CORPUS = (1 << 0),
502 * struct ldlm_request is passed in the incoming message.
504 * "habeo clavis" == "I have a key"
506 HABEO_CLAVIS = (1 << 1),
508 * this request has fixed reply format, so that reply message can be
509 * packed by generic code.
511 * "habeo refero" == "I have a reply"
513 HABEO_REFERO = (1 << 2),
515 * this request will modify something, so check whether the filesystem
516 * is readonly or not, then return -EROFS to client asap if necessary.
518 * "mutabor" == "I shall modify"
523 struct mdt_opc_slice {
526 struct mdt_handler *mos_hs;
529 static inline const struct md_device_operations *
530 mdt_child_ops(struct mdt_device * m)
532 LASSERT(m->mdt_child);
533 return m->mdt_child->md_ops;
536 static inline struct md_object *mdt_object_child(struct mdt_object *o)
539 return lu2md(lu_object_next(&o->mot_obj));
542 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
544 return info->mti_pill ? info->mti_pill->rc_req : NULL;
547 static inline int req_is_replay(struct ptlrpc_request *req)
549 LASSERT(req->rq_reqmsg);
550 return !!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
553 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
555 LASSERT(info->mti_exp);
556 return exp_connect_flags(info->mti_exp);
559 static inline void mdt_object_get(const struct lu_env *env,
560 struct mdt_object *o)
563 lu_object_get(&o->mot_obj);
567 static inline void mdt_object_put(const struct lu_env *env,
568 struct mdt_object *o)
571 lu_object_put(env, &o->mot_obj);
575 static inline int mdt_object_exists(const struct mdt_object *o)
577 return lu_object_exists(&o->mot_obj);
580 static inline int mdt_object_remote(const struct mdt_object *o)
582 return lu_object_remote(&o->mot_obj);
585 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
587 return lu_object_fid(&o->mot_obj);
590 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
592 return mdt->mdt_lu_dev.ld_site;
595 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
597 return &mdt->mdt_seq_site;
600 static inline void mdt_export_evict(struct obd_export *exp)
602 class_fail_export(exp);
603 class_export_put(exp);
606 /* Here we use LVB_TYPE to check dne client, because it is
607 * also landed on 2.4. */
608 static inline int mdt_is_dne_client(struct obd_export *exp)
610 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
613 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
614 void mdt_set_disposition(struct mdt_thread_info *info,
615 struct ldlm_reply *rep, int flag);
616 void mdt_clear_disposition(struct mdt_thread_info *info,
617 struct ldlm_reply *rep, int flag);
619 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
620 ldlm_mode_t lm, const char *name,
623 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
626 int mdt_lock_setup(struct mdt_thread_info *info,
627 struct mdt_object *o,
628 struct mdt_lock_handle *lh);
630 int mdt_object_lock(struct mdt_thread_info *,
632 struct mdt_lock_handle *,
635 int mdt_object_lock_try(struct mdt_thread_info *,
637 struct mdt_lock_handle *,
640 void mdt_object_unlock(struct mdt_thread_info *,
642 struct mdt_lock_handle *,
645 struct mdt_object *mdt_object_new(const struct lu_env *,
647 const struct lu_fid *);
648 struct mdt_object *mdt_object_find(const struct lu_env *,
650 const struct lu_fid *);
651 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
652 const struct lu_fid *,
653 struct mdt_lock_handle *,
655 void mdt_object_unlock_put(struct mdt_thread_info *,
657 struct mdt_lock_handle *,
660 void mdt_client_compatibility(struct mdt_thread_info *info);
662 int mdt_remote_object_lock(struct mdt_thread_info *mti,
663 struct mdt_object *o, struct lustre_handle *lh,
664 ldlm_mode_t mode, __u64 ibits);
665 int mdt_close_unpack(struct mdt_thread_info *info);
666 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
667 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
668 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
669 const struct lu_attr *attr, const struct lu_fid *fid);
671 int mdt_getxattr(struct mdt_thread_info *info);
672 int mdt_reint_setxattr(struct mdt_thread_info *info,
673 struct mdt_lock_handle *lh);
675 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
676 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
678 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
679 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
680 struct mdt_lock_handle *lhc);
682 extern void target_recovery_fini(struct obd_device *obd);
683 extern void target_recovery_init(struct lu_target *lut,
684 svc_handler_t handler);
685 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
686 struct obd_device *, struct lustre_sb_info *lsi);
687 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
689 int mdt_export_stats_init(struct obd_device *obd,
690 struct obd_export *exp,
693 int mdt_pin(struct mdt_thread_info* info);
695 int mdt_lock_new_child(struct mdt_thread_info *info,
696 struct mdt_object *o,
697 struct mdt_lock_handle *child_lockh);
699 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
702 int mdt_reint_open(struct mdt_thread_info *info,
703 struct mdt_lock_handle *lhc);
705 struct mdt_file_data *mdt_handle2mfd(struct mdt_thread_info *,
706 const struct lustre_handle *);
709 MDT_IOEPOCH_CLOSED = 0,
710 MDT_IOEPOCH_OPENED = 1,
711 MDT_IOEPOCH_GETATTR = 2,
719 int mdt_attr_get_complex(struct mdt_thread_info *info,
720 struct mdt_object *o, struct md_attr *ma);
721 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
723 int mdt_object_is_som_enabled(struct mdt_object *mo);
724 int mdt_write_get(struct mdt_object *o);
725 void mdt_write_put(struct mdt_object *o);
726 int mdt_write_read(struct mdt_object *o);
727 struct mdt_file_data *mdt_mfd_new(void);
728 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
729 void mdt_mfd_free(struct mdt_file_data *mfd);
730 int mdt_close(struct mdt_thread_info *info);
731 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
732 struct md_attr *ma, int flags);
733 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
735 int mdt_done_writing(struct mdt_thread_info *info);
736 int mdt_fix_reply(struct mdt_thread_info *info);
737 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
738 const struct md_attr *);
739 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
741 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
742 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
743 const void *area, ssize_t len);
745 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
747 int mdt_check_ucred(struct mdt_thread_info *);
748 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
749 int mdt_init_ucred_reint(struct mdt_thread_info *);
750 void mdt_exit_ucred(struct mdt_thread_info *);
751 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
752 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
753 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
755 int mdt_handle_common(struct ptlrpc_request *req,
756 struct mdt_opc_slice *supported);
757 int mdt_connect(struct mdt_thread_info *info);
758 int mdt_disconnect(struct mdt_thread_info *info);
759 int mdt_set_info(struct mdt_thread_info *info);
760 int mdt_get_info(struct mdt_thread_info *info);
761 int mdt_getstatus(struct mdt_thread_info *info);
762 int mdt_getattr(struct mdt_thread_info *info);
763 int mdt_getattr_name(struct mdt_thread_info *info);
764 int mdt_statfs(struct mdt_thread_info *info);
765 int mdt_reint(struct mdt_thread_info *info);
766 int mdt_sync(struct mdt_thread_info *info);
767 int mdt_is_subdir(struct mdt_thread_info *info);
768 int mdt_obd_ping(struct mdt_thread_info *info);
769 int mdt_obd_log_cancel(struct mdt_thread_info *info);
770 int mdt_obd_qc_callback(struct mdt_thread_info *info);
771 int mdt_enqueue(struct mdt_thread_info *info);
772 int mdt_convert(struct mdt_thread_info *info);
773 int mdt_bl_callback(struct mdt_thread_info *info);
774 int mdt_cp_callback(struct mdt_thread_info *info);
775 int mdt_llog_create(struct mdt_thread_info *info);
776 int mdt_llog_destroy(struct mdt_thread_info *info);
777 int mdt_llog_read_header(struct mdt_thread_info *info);
778 int mdt_llog_next_block(struct mdt_thread_info *info);
779 int mdt_llog_prev_block(struct mdt_thread_info *info);
780 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
781 int mdt_readpage(struct mdt_thread_info *info);
782 int mdt_obd_idx_read(struct mdt_thread_info *info);
784 extern struct mdt_opc_slice mdt_regular_handlers[];
785 extern struct mdt_opc_slice mdt_seq_handlers[];
786 extern struct mdt_opc_slice mdt_fld_handlers[];
788 int mdt_quotacheck(struct mdt_thread_info *info);
789 int mdt_quotactl(struct mdt_thread_info *info);
790 int mdt_quota_dqacq(struct mdt_thread_info *info);
791 int mdt_swap_layouts(struct mdt_thread_info *info);
793 extern struct lprocfs_vars lprocfs_mds_module_vars[];
794 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
796 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
799 struct mdt_handler *mdt_handler_find(__u32 opc,
800 struct mdt_opc_slice *supported);
802 int mdt_init_sec_level(struct mdt_thread_info *);
803 int mdt_init_idmap(struct mdt_thread_info *);
804 void mdt_cleanup_idmap(struct mdt_export_data *);
805 int mdt_handle_idmap(struct mdt_thread_info *);
806 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
807 struct ptlrpc_user_desc *);
808 void mdt_body_reverse_idmap(struct mdt_thread_info *,
810 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
811 struct mdt_remote_perm *);
812 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
814 static inline struct mdt_device *mdt_dev(struct lu_device *d)
816 return container_of0(d, struct mdt_device, mdt_lu_dev);
819 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
821 struct lu_object *lo;
822 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
824 lo = lu_object_locate(mo->mot_obj.lo_header,
825 mdt->mdt_bottom->dd_lu_dev.ld_type);
830 /* mdt/mdt_identity.c */
831 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
833 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
835 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
837 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
839 void mdt_flush_identity(struct upcall_cache *, int);
841 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
843 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
846 int mdt_hsm_state_get(struct mdt_thread_info *info);
847 int mdt_hsm_state_set(struct mdt_thread_info *info);
848 int mdt_hsm_action(struct mdt_thread_info *info);
849 int mdt_hsm_progress(struct mdt_thread_info *info);
850 int mdt_hsm_ct_register(struct mdt_thread_info *info);
851 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
852 int mdt_hsm_request(struct mdt_thread_info *info);
854 extern struct lu_context_key mdt_thread_key;
855 /* debug issues helper starts here*/
856 static inline int mdt_fail_write(const struct lu_env *env,
857 struct dt_device *dd, int id)
859 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
860 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
862 return dd->dd_ops->dt_ro(env, dd);
863 /* We set FAIL_ONCE because we never "un-fail" a device */
869 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
871 return &req->rq_export->exp_mdt_data;
874 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
875 struct mdt_lock_handle *lhc);
876 static inline int mdt_check_resent(struct mdt_thread_info *info,
877 mdt_reconstruct_t reconstruct,
878 struct mdt_lock_handle *lhc)
880 struct ptlrpc_request *req = mdt_info_req(info);
883 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
884 if (req_xid_is_last(req)) {
885 reconstruct(info, lhc);
888 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
889 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
894 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
895 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
897 static inline int is_identity_get_disabled(struct upcall_cache *cache)
899 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
902 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
904 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
905 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
906 struct lustre_handle *lh,
908 ldlm_policy_data_t *policy,
909 const struct ldlm_res_id *res_id,
910 __u64 flags, const __u64 *client_cookie)
917 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
918 mode, &flags, mdt_blocking_ast,
919 ldlm_completion_ast, NULL, NULL, 0,
920 LVB_T_NONE, client_cookie, lh);
921 return rc == ELDLM_OK ? 0 : -EIO;
924 static inline void mdt_fid_unlock(struct lustre_handle *lh,
927 ldlm_lock_decref(lh, mode);
930 extern mdl_mode_t mdt_mdl_lock_modes[];
931 extern ldlm_mode_t mdt_dlm_lock_modes[];
933 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
935 LASSERT(IS_PO2(mode));
936 return mdt_mdl_lock_modes[mode];
939 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
941 LASSERT(IS_PO2(mode));
942 return mdt_dlm_lock_modes[mode];
946 extern struct ldlm_valblock_ops mdt_lvbo;
948 static inline struct lu_name *mdt_name(const struct lu_env *env,
949 char *name, int namelen)
951 struct lu_name *lname;
952 struct mdt_thread_info *mti;
954 LASSERT(namelen > 0);
955 /* trailing '\0' in buffer */
956 LASSERT(name[namelen] == '\0');
958 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
959 lname = &mti->mti_name;
960 lname->ln_name = name;
961 lname->ln_namelen = namelen;
965 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
966 struct lu_name *slname)
971 tlname->ln_name = slname->ln_name;
972 tlname->ln_namelen = slname->ln_namelen;
976 void mdt_enable_cos(struct mdt_device *, int);
977 int mdt_cos_is_enabled(struct mdt_device *);
978 int mdt_hsm_copytool_send(struct obd_export *exp);
996 LPROC_MDT_SAMEDIR_RENAME,
997 LPROC_MDT_CROSSDIR_RENAME,
1000 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1001 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1002 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1003 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1004 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1005 void mdt_procfs_fini(struct mdt_device *mdt);
1006 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1007 struct mdt_device *mdt,
1008 struct ptlrpc_request *req,
1009 struct mdt_object *src, struct mdt_object *tgt);
1012 int mdt_ck_thread_start(struct mdt_device *mdt);
1013 void mdt_ck_thread_stop(struct mdt_device *mdt);
1014 void mdt_ck_timer_callback(unsigned long castmeharder);
1015 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1017 static inline void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1018 const struct lu_fid *fid,
1019 struct lustre_capa *capa)
1021 struct md_capainfo *ci;
1023 LASSERT(offset >= 0 && offset < MD_CAPAINFO_MAX);
1024 if (!info->mti_mdt->mdt_opts.mo_mds_capa ||
1025 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
1028 ci = md_capainfo(info->mti_env);
1030 ci->mc_fid[offset] = *fid;
1031 ci->mc_capa[offset] = capa;
1034 static inline void mdt_dump_capainfo(struct mdt_thread_info *info)
1036 struct md_capainfo *ci = md_capainfo(info->mti_env);
1041 for (i = 0; i < MD_CAPAINFO_MAX; i++) {
1042 if (!ci->mc_capa[i]) {
1043 CERROR("no capa for index %d "DFID"\n",
1044 i, PFID(&ci->mc_fid[i]));
1047 if (ci->mc_capa[i] == BYPASS_CAPA) {
1048 CERROR("bypass for index %d "DFID"\n",
1049 i, PFID(&ci->mc_fid[i]));
1052 DEBUG_CAPA(D_ERROR, ci->mc_capa[i], "index %d", i);
1056 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1058 return mdt->mdt_lu_dev.ld_obd;
1061 extern const struct lu_device_operations mdt_lu_ops;
1063 static inline int lu_device_is_mdt(struct lu_device *d)
1065 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1068 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1070 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1071 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1072 return container_of0(d, struct mdt_device, mdt_lu_dev);
1075 static inline char *mdt_obd_name(struct mdt_device *mdt)
1077 return mdt->mdt_lu_dev.ld_obd->obd_name;
1080 int mds_mod_init(void);
1081 void mds_mod_exit(void);
1083 /* Update handlers */
1084 int out_handle(struct mdt_thread_info *info);
1086 #define out_tx_create(info, obj, attr, fid, dof, th, reply, idx) \
1087 __out_tx_create(info, obj, attr, fid, dof, th, reply, idx, \
1090 #define out_tx_attr_set(info, obj, attr, th, reply, idx) \
1091 __out_tx_attr_set(info, obj, attr, th, reply, idx, \
1094 #define out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx) \
1095 __out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx, \
1098 #define out_tx_ref_add(info, obj, th, reply, idx) \
1099 __out_tx_ref_add(info, obj, th, reply, idx, __FILE__, __LINE__)
1101 #define out_tx_ref_del(info, obj, th, reply, idx) \
1102 __out_tx_ref_del(info, obj, th, reply, idx, __FILE__, __LINE__)
1104 #define out_tx_index_insert(info, obj, th, name, fid, reply, idx) \
1105 __out_tx_index_insert(info, obj, th, name, fid, reply, idx, \
1108 #define out_tx_index_delete(info, obj, th, name, reply, idx) \
1109 __out_tx_index_delete(info, obj, th, name, reply, idx, \
1112 #define out_tx_destroy(info, obj, th, reply, idx) \
1113 __out_tx_destroy(info, obj, th, reply, idx, __FILE__, __LINE__)
1115 #endif /* __KERNEL__ */