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, 2014, 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
52 #include <libcfs/lucache.h>
53 #include <lustre_net.h>
54 #include <lustre/lustre_idl.h>
55 #include <obd_class.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_req_layout.h>
62 #include <lustre_sec.h>
63 #include <lustre_idmap.h>
64 #include <lustre_eacl.h>
65 #include <lustre_quota.h>
66 #include <lustre_linkea.h>
68 /* check if request's xid is equal to last one or not*/
69 static inline int req_xid_is_last(struct ptlrpc_request *req)
71 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
72 return (req->rq_xid == lcd->lcd_last_xid ||
73 req->rq_xid == lcd->lcd_last_close_xid);
78 /* file data for open files on MDS */
79 struct mdt_file_data {
80 /** portals handle must be first */
81 struct portals_handle mfd_handle;
82 /** open mode provided by client */
84 /** protected by med_open_lock */
85 struct list_head mfd_list;
86 /** xid of the open request */
88 /** old handle in replay case */
89 struct lustre_handle mfd_old_handle;
90 /** point to opened object */
91 struct mdt_object *mfd_object;
94 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
95 #define CDT_NORETRY_ACTION (1ULL << 1)
96 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
97 #define CDT_POLICY_SHIFT_COUNT 2
98 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
101 /* when adding a new policy, do not forget to update
102 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
104 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
106 enum cdt_states { CDT_STOPPED = 0,
112 /* when multiple lock are needed, the lock order is
120 struct ptlrpc_thread cdt_thread; /**< coordinator thread */
121 struct lu_env cdt_env; /**< coordinator lustre
123 struct lu_context cdt_session; /** session for lu_ucred */
124 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
125 __u64 cdt_policy; /**< policy flags */
126 enum cdt_states cdt_state; /**< state */
127 atomic_t cdt_compound_id; /**< compound id
129 __u64 cdt_last_cookie; /**< last cookie
131 struct mutex cdt_llog_lock; /**< protect llog
133 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
134 struct rw_semaphore cdt_request_lock; /**< protect request
136 struct mutex cdt_restore_lock; /**< protect restore
138 cfs_time_t cdt_loop_period; /**< llog scan period */
139 cfs_time_t cdt_grace_delay; /**< request grace
141 cfs_time_t cdt_active_req_timeout; /**< request timeout */
142 __u32 cdt_default_archive_id; /**< archive id used
145 __u64 cdt_max_requests; /**< max count of started
147 atomic_t cdt_request_count; /**< current count of
148 * started requests */
149 struct list_head cdt_requests; /**< list of started
151 struct list_head cdt_agents; /**< list of register
153 struct list_head cdt_restore_hdl; /**< list of restore lock
155 /* Bitmasks indexed by the HSMA_XXX constants. */
156 __u64 cdt_user_request_mask;
157 __u64 cdt_group_request_mask;
158 __u64 cdt_other_request_mask;
161 /* mdt state flag bits */
162 #define MDT_FL_CFGLOG 0
163 #define MDT_FL_SYNCED 1
167 struct lu_device mdt_lu_dev;
168 struct seq_server_site mdt_seq_site;
169 /* DLM name-space for meta-data locks maintained by this server */
170 struct ldlm_namespace *mdt_namespace;
171 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
172 struct ptlrpc_client *mdt_ldlm_client;
173 /* underlying device */
174 struct obd_export *mdt_child_exp;
175 struct md_device *mdt_child;
176 struct dt_device *mdt_bottom;
177 struct obd_export *mdt_bottom_exp;
179 struct lu_target mdt_lut;
181 * Options bit-fields.
184 unsigned int mo_user_xattr:1,
189 /* mdt state flags */
190 unsigned long mdt_state;
191 /* lock to protect IOepoch */
192 spinlock_t mdt_ioepoch_lock;
195 /* transaction callbacks */
196 struct dt_txn_callback mdt_txn_cb;
198 /* these values should be updated from lov if necessary.
199 * or should be placed somewhere else. */
204 struct upcall_cache *mdt_identity_cache;
206 /* capability keys */
207 unsigned long mdt_capa_timeout;
209 struct dt_object *mdt_ck_obj;
210 unsigned long mdt_ck_timeout;
211 unsigned long mdt_ck_expiry;
212 struct timer_list mdt_ck_timer;
213 struct ptlrpc_thread mdt_ck_thread;
214 struct lustre_capa_key mdt_capa_keys[2];
215 unsigned int mdt_capa_conf:1,
217 /* Enable remote dir on non-MDT0 */
218 mdt_enable_remote_dir:1;
220 gid_t mdt_enable_remote_dir_gid;
221 /* statfs optimization: we cache a bit */
222 struct obd_statfs mdt_osfs;
224 spinlock_t mdt_osfs_lock;
227 struct root_squash_info mdt_squash;
229 struct rename_stats mdt_rename_stats;
230 struct lu_fid mdt_md_root_fid;
232 /* connection to quota master */
233 struct obd_export *mdt_qmt_exp;
234 /* quota master device associated with this MDT */
235 struct lu_device *mdt_qmt_dev;
237 struct coordinator mdt_coordinator;
240 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
241 #define MDT_COS_DEFAULT (0)
244 struct lu_object_header mot_header;
245 struct lu_object mot_obj;
248 int mot_ioepoch_count;
250 /* Lock to protect object's IO epoch. */
251 struct mutex mot_ioepoch_mutex;
252 /* Lock to protect create_data */
253 struct mutex mot_lov_mutex;
254 /* Lock to protect lease open.
255 * Lease open acquires write lock; normal open acquires read lock */
256 struct rw_semaphore mot_open_sem;
257 atomic_t mot_lease_count;
258 atomic_t mot_open_count;
261 enum mdt_object_flags {
262 /** SOM attributes are changed. */
263 MOF_SOM_CHANGE = (1 << 0),
265 * The SOM recovery state for mdt object.
266 * This state is an in-memory equivalent of an absent SOM EA, used
267 * instead of invalidating SOM EA while IOEpoch is still opened when
268 * a client eviction occurs or a client fails to obtain SOM attributes.
269 * It indicates that the last IOEpoch holder will need to obtain SOM
270 * attributes under [0;EOF] extent lock to flush all the client's
271 * cached of evicted from MDS clients (but not necessary evicted from
272 * OST) before taking ost attributes.
274 MOF_SOM_RECOV = (1 << 1),
275 /** File has been just created. */
276 MOF_SOM_CREATED = (1 << 2),
277 /** lov object has been created. */
278 MOF_LOV_CREATED = (1 << 3),
281 struct mdt_lock_handle {
282 /* Lock type, reg for cross-ref use or pdo lock. */
286 struct lustre_handle mlh_reg_lh;
287 ldlm_mode_t mlh_reg_mode;
290 struct lustre_handle mlh_pdo_lh;
291 ldlm_mode_t mlh_pdo_mode;
292 unsigned int mlh_pdo_hash;
294 /* Remote regular lock */
295 struct lustre_handle mlh_rreg_lh;
296 ldlm_mode_t mlh_rreg_mode;
300 MDT_LH_PARENT, /* parent lockh */
301 MDT_LH_CHILD, /* child lockh */
302 MDT_LH_OLD, /* old lockh for rename */
303 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
304 MDT_LH_NEW, /* new lockh for rename */
305 MDT_LH_RMT, /* used for return lh to caller */
306 MDT_LH_LOCAL, /* local lock never return to client */
315 /* Special magical errno for communicaiton between mdt_reint_open()
316 * and mdt_intent_reint() which means return the lock to the client
317 * for subsequent cross ref open. Previously we used plain -EREMOTE
318 * but other functions called in that path might return it too and
319 * confuse us. This is not returned to the client. See LU-5370. */
320 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
322 struct mdt_reint_record {
323 mdt_reint_t rr_opcode;
324 const struct lustre_handle *rr_handle;
325 const struct lu_fid *rr_fid1;
326 const struct lu_fid *rr_fid2;
327 struct lu_name rr_name;
328 struct lu_name rr_tgt_name;
334 enum mdt_reint_flag {
335 MRF_OPEN_TRUNC = 1 << 0,
339 * Common data shared by mdt-level handlers. This is allocated per-thread to
340 * reduce stack consumption.
342 struct mdt_thread_info {
345 * The following members will be filled explicitly
346 * with specific data in mdt_thread_info_init().
348 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
349 * request handling may migrate from one server thread to another.
351 struct req_capsule *mti_pill;
353 /* although we have export in req, there are cases when it is not
354 * available, e.g. closing files upon export destroy */
355 struct obd_export *mti_exp;
357 * A couple of lock handles.
359 struct mdt_lock_handle mti_lh[MDT_LH_NR];
361 struct mdt_device *mti_mdt;
362 const struct lu_env *mti_env;
364 /* transaction number of current request */
370 * The following members will be filled expilictly
371 * with zero in mdt_thread_info_init(). These members may be used
378 struct md_attr mti_attr;
380 * Body for "habeo corpus" operations.
382 const struct mdt_body *mti_body;
384 * Host object. This is released at the end of mdt_handler().
386 struct mdt_object *mti_object;
388 * Lock request for "habeo clavis" operations.
390 const struct ldlm_request *mti_dlm_req;
392 __u32 mti_has_trans:1, /* has txn already? */
395 /* opdata for mdt_reint_open(), has the same as
396 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
397 * value onto disk for recovery when mdt_trans_stop_cb() is called.
403 * The following members will be filled explicitly
404 * with zero in mdt_reint_unpack(), because they are only used
405 * by reint requests (including mdt_reint_open()).
409 * reint record. contains information for reint operations.
411 struct mdt_reint_record mti_rr;
413 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
415 * Operation specification (currently create and lookup)
417 struct md_op_spec mti_spec;
421 * The following members will _NOT_ be initialized at all.
422 * DO NOT expect them to contain any valid value.
423 * They should be initialized explicitly by the user themselves.
426 /* XXX: If something is in a union, make sure they do not conflict */
428 struct lu_fid mti_tmp_fid1;
429 struct lu_fid mti_tmp_fid2;
430 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
431 * mdt_rename_lock() */
432 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
435 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
436 char ns_name[48]; /* for mdt_init0() */
437 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
438 struct obd_statfs osfs; /* for mdt_statfs() */
440 /* for mdt_readpage() */
441 struct lu_rdpg mti_rdpg;
442 /* for mdt_sendpage() */
443 struct l_wait_info mti_wait_info;
447 struct md_som_data data;
451 /* IO epoch related stuff. */
452 struct mdt_ioepoch *mti_ioepoch;
453 __u64 mti_replayepoch;
456 struct lu_buf mti_buf;
457 struct lu_buf mti_big_buf;
458 struct lustre_capa_key mti_capa_key;
460 /* Ops object filename */
461 struct lu_name mti_name;
462 /* per-thread values, can be re-used */
465 /* big_lmm buffer was used and must be used in reply */
466 int mti_big_lmm_used;
467 /* should be enough to fit lustre_mdt_attrs */
468 char mti_xattr_buf[128];
469 struct ldlm_enqueue_info mti_einfo;
472 extern struct lu_context_key mdt_thread_key;
474 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
476 struct mdt_thread_info *mti;
478 lu_env_refill((void *)env);
479 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
484 struct cdt_req_progress {
485 struct mutex crp_lock; /**< protect tree */
486 struct interval_node *crp_root; /**< tree to track extent
488 struct interval_node **crp_node; /**< buffer for tree nodes
489 * vector of fixed size
491 int crp_cnt; /**< # of used nodes */
492 int crp_max; /**< # of allocated nodes */
495 struct cdt_agent_req {
496 struct list_head car_request_list; /**< to chain all the req. */
497 atomic_t car_refcount; /**< reference counter */
498 __u64 car_compound_id; /**< compound id */
499 __u64 car_flags; /**< request original flags */
500 struct obd_uuid car_uuid; /**< agent doing the req. */
501 __u32 car_archive_id; /**< archive id */
502 int car_canceled; /**< request was canceled */
503 cfs_time_t car_req_start; /**< start time */
504 cfs_time_t car_req_update; /**< last update time */
505 struct hsm_action_item *car_hai; /**< req. to the agent */
506 struct cdt_req_progress car_progress; /**< track data mvt
509 extern struct kmem_cache *mdt_hsm_car_kmem;
512 struct list_head ha_list; /**< to chain the agents */
513 struct obd_uuid ha_uuid; /**< agent uuid */
514 __u32 *ha_archive_id; /**< archive id */
515 int ha_archive_cnt; /**< number of archive entries
516 * 0 means any archive */
517 atomic_t ha_requests; /**< current request count */
518 atomic_t ha_success; /**< number of successful
520 atomic_t ha_failure; /**< number of failed actions */
523 struct cdt_restore_handle {
524 struct list_head crh_list; /**< to chain the handle */
525 struct lu_fid crh_fid; /**< fid of the object */
526 struct ldlm_extent crh_extent; /**< extent of the restore */
527 struct mdt_lock_handle crh_lh; /**< lock handle */
529 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
531 static inline const struct md_device_operations *
532 mdt_child_ops(struct mdt_device * m)
534 LASSERT(m->mdt_child);
535 return m->mdt_child->md_ops;
538 static inline struct md_object *mdt_object_child(struct mdt_object *o)
541 return lu2md(lu_object_next(&o->mot_obj));
544 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
546 return info->mti_pill ? info->mti_pill->rc_req : NULL;
549 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
551 LASSERT(info->mti_exp);
552 return exp_connect_flags(info->mti_exp);
555 static inline void mdt_object_get(const struct lu_env *env,
556 struct mdt_object *o)
559 lu_object_get(&o->mot_obj);
563 static inline void mdt_object_put(const struct lu_env *env,
564 struct mdt_object *o)
567 lu_object_put(env, &o->mot_obj);
571 static inline int mdt_object_exists(const struct mdt_object *o)
573 return lu_object_exists(&o->mot_obj);
576 static inline int mdt_object_remote(const struct mdt_object *o)
578 return lu_object_remote(&o->mot_obj);
581 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
583 return lu_object_fid(&o->mot_obj);
586 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
588 return mdt->mdt_lu_dev.ld_site;
591 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
593 return &mdt->mdt_seq_site;
596 static inline void mdt_export_evict(struct obd_export *exp)
598 class_fail_export(exp);
599 class_export_put(exp);
602 /* Here we use LVB_TYPE to check dne client, because it is
603 * also landed on 2.4. */
604 static inline bool mdt_is_dne_client(struct obd_export *exp)
606 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
609 static inline bool mdt_is_striped_client(struct obd_export *exp)
611 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
614 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
615 void mdt_set_disposition(struct mdt_thread_info *info,
616 struct ldlm_reply *rep, int flag);
617 void mdt_clear_disposition(struct mdt_thread_info *info,
618 struct ldlm_reply *rep, int flag);
620 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
621 ldlm_mode_t lock_mode,
622 const struct lu_name *lname);
624 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
627 int mdt_lock_setup(struct mdt_thread_info *info,
628 struct mdt_object *o,
629 struct mdt_lock_handle *lh);
631 int mdt_check_resent_lock(struct mdt_thread_info *info,
632 struct mdt_object *mo,
633 struct mdt_lock_handle *lhc);
635 int mdt_object_lock(struct mdt_thread_info *,
637 struct mdt_lock_handle *,
640 int mdt_object_lock_try(struct mdt_thread_info *,
642 struct mdt_lock_handle *,
645 void mdt_object_unlock(struct mdt_thread_info *,
647 struct mdt_lock_handle *,
650 struct mdt_object *mdt_object_new(const struct lu_env *,
652 const struct lu_fid *);
653 struct mdt_object *mdt_object_find(const struct lu_env *,
655 const struct lu_fid *);
656 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
657 const struct lu_fid *,
658 struct mdt_lock_handle *,
660 void mdt_object_unlock_put(struct mdt_thread_info *,
662 struct mdt_lock_handle *,
665 void mdt_client_compatibility(struct mdt_thread_info *info);
667 int mdt_remote_object_lock(struct mdt_thread_info *mti,
668 struct mdt_object *o, const struct lu_fid *fid,
669 struct lustre_handle *lh,
670 ldlm_mode_t mode, __u64 ibits);
672 enum mdt_name_flags {
676 int mdt_name_unpack(struct req_capsule *pill,
677 const struct req_msg_field *field,
679 enum mdt_name_flags flags);
680 int mdt_close_unpack(struct mdt_thread_info *info);
681 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
682 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
683 #ifdef CONFIG_FS_POSIX_ACL
684 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
685 struct mdt_object *o, struct lu_nodemap *nodemap);
687 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
688 const struct lu_attr *attr, const struct lu_fid *fid);
690 int mdt_getxattr(struct mdt_thread_info *info);
691 int mdt_reint_setxattr(struct mdt_thread_info *info,
692 struct mdt_lock_handle *lh);
694 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
695 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
697 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
698 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
699 struct mdt_lock_handle *lhc);
701 extern void target_recovery_fini(struct obd_device *obd);
702 extern void target_recovery_init(struct lu_target *lut,
703 svc_handler_t handler);
704 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
705 struct obd_device *, struct lustre_sb_info *lsi);
706 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
708 int mdt_export_stats_init(struct obd_device *obd,
709 struct obd_export *exp,
712 int mdt_lock_new_child(struct mdt_thread_info *info,
713 struct mdt_object *o,
714 struct mdt_lock_handle *child_lockh);
716 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
719 int mdt_reint_open(struct mdt_thread_info *info,
720 struct mdt_lock_handle *lhc);
722 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
723 const struct lustre_handle *handle,
727 MDT_IOEPOCH_CLOSED = 0,
728 MDT_IOEPOCH_OPENED = 1,
729 MDT_IOEPOCH_GETATTR = 2,
737 int mdt_get_info(struct tgt_session_info *tsi);
738 int mdt_attr_get_complex(struct mdt_thread_info *info,
739 struct mdt_object *o, struct md_attr *ma);
740 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
742 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
743 struct md_attr *ma, const char *name);
744 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
746 int mdt_object_is_som_enabled(struct mdt_object *mo);
747 int mdt_write_get(struct mdt_object *o);
748 void mdt_write_put(struct mdt_object *o);
749 int mdt_write_read(struct mdt_object *o);
750 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
751 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
752 void mdt_mfd_free(struct mdt_file_data *mfd);
753 int mdt_close(struct tgt_session_info *tsi);
754 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
756 int mdt_done_writing(struct tgt_session_info *tsi);
757 int mdt_fix_reply(struct mdt_thread_info *info);
758 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
759 const struct md_attr *);
760 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
762 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
763 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
764 const void *area, ssize_t len);
766 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
767 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
769 int mdt_check_ucred(struct mdt_thread_info *);
770 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
771 int mdt_init_ucred_reint(struct mdt_thread_info *);
772 void mdt_exit_ucred(struct mdt_thread_info *);
773 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
774 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
775 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
777 void mdt_thread_info_init(struct ptlrpc_request *req,
778 struct mdt_thread_info *mti);
779 void mdt_thread_info_fini(struct mdt_thread_info *mti);
780 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
782 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
783 const struct md_hsm *mh);
785 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
786 void *data, int flag);
787 int mdt_links_read(struct mdt_thread_info *info,
788 struct mdt_object *mdt_obj,
789 struct linkea_data *ldata);
791 int mdt_init_idmap(struct tgt_session_info *tsi);
792 void mdt_cleanup_idmap(struct mdt_export_data *);
793 int mdt_handle_idmap(struct tgt_session_info *tsi);
794 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
795 struct ptlrpc_user_desc *);
796 void mdt_body_reverse_idmap(struct mdt_thread_info *,
798 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
799 struct mdt_remote_perm *);
800 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
802 static inline struct mdt_device *mdt_dev(struct lu_device *d)
804 return container_of0(d, struct mdt_device, mdt_lu_dev);
807 static inline struct mdt_object *mdt_obj(struct lu_object *o)
809 return container_of0(o, struct mdt_object, mot_obj);
812 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
814 struct lu_object *lo;
815 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
817 lo = lu_object_locate(mo->mot_obj.lo_header,
818 mdt->mdt_bottom->dd_lu_dev.ld_type);
823 /* mdt/mdt_identity.c */
824 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
826 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
828 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
830 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
832 void mdt_flush_identity(struct upcall_cache *, int);
834 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
836 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
839 int mdt_hsm_state_get(struct tgt_session_info *tsi);
840 int mdt_hsm_state_set(struct tgt_session_info *tsi);
841 int mdt_hsm_action(struct tgt_session_info *tsi);
842 int mdt_hsm_progress(struct tgt_session_info *tsi);
843 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
844 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
845 int mdt_hsm_request(struct tgt_session_info *tsi);
847 /* mdt/mdt_hsm_cdt_actions.c */
848 extern const struct file_operations mdt_hsm_actions_fops;
849 void dump_llog_agent_req_rec(const char *prefix,
850 const struct llog_agent_req_rec *larr);
851 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
852 llog_cb_t cb, void *data);
853 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
854 __u64 compound_id, __u32 archive_id,
855 __u64 flags, struct hsm_action_item *hai);
856 int mdt_agent_record_update(const struct lu_env *env,
857 struct mdt_device *mdt, __u64 *cookies,
858 int cookies_count, enum agent_req_status status);
859 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
860 struct llog_handle *llh,
861 struct llog_agent_req_rec *larr);
863 /* mdt/mdt_hsm_cdt_agent.c */
864 extern const struct file_operations mdt_hsm_agent_fops;
865 int mdt_hsm_agent_register(struct mdt_thread_info *info,
866 const struct obd_uuid *uuid,
867 int nr_archives, __u32 *archive_num);
868 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
869 const struct obd_uuid *uuid,
871 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
872 const struct obd_uuid *uuid);
873 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
874 int succ_rq, int fail_rq, int new_rq,
875 const struct obd_uuid *uuid);
876 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
877 struct obd_uuid *uuid);
878 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
880 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
881 struct hsm_progress_kernel *pgs);
882 /* mdt/mdt_hsm_cdt_client.c */
883 int mdt_hsm_add_actions(struct mdt_thread_info *info,
884 struct hsm_action_list *hal, __u64 *compound_id);
885 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
886 struct hsm_action_list *hal);
887 int mdt_hsm_get_running(struct mdt_thread_info *mti,
888 struct hsm_action_list *hal);
889 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
890 const struct lu_fid *fid);
891 /* mdt/mdt_hsm_cdt_requests.c */
892 extern const struct file_operations mdt_hsm_active_requests_fops;
893 void dump_requests(char *prefix, struct coordinator *cdt);
894 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
895 __u64 flags, struct obd_uuid *uuid,
896 struct hsm_action_item *hai);
897 void mdt_cdt_free_request(struct cdt_agent_req *car);
898 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
899 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
901 const struct lu_fid *fid);
902 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
903 void mdt_cdt_get_request(struct cdt_agent_req *car);
904 void mdt_cdt_put_request(struct cdt_agent_req *car);
905 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
906 const struct hsm_progress_kernel *pgs);
907 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
908 /* mdt/mdt_coordinator.c */
909 void mdt_hsm_dump_hal(int level, const char *prefix,
910 struct hsm_action_list *hal);
911 /* coordinator management */
912 int mdt_hsm_cdt_init(struct mdt_device *mdt);
913 int mdt_hsm_cdt_start(struct mdt_device *mdt);
914 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
915 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
916 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
918 /* coordinator control /proc interface */
919 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file, const char *buffer,
920 size_t count, loff_t *off);
921 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
922 int hsm_cdt_procfs_init(struct mdt_device *mdt);
923 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
924 struct lprocfs_seq_vars *hsm_cdt_get_proc_vars(void);
926 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
927 const struct lu_fid *fid,
929 /* actions/request helpers */
930 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
931 struct hsm_action_list *hal, struct obd_uuid *uuid);
932 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
933 const int hal_an, const __u64 rq_flags,
934 const struct md_hsm *hsm);
935 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
936 struct hsm_progress_kernel *pgs,
937 const int update_record);
939 extern struct lu_context_key mdt_thread_key;
941 /* debug issues helper starts here*/
942 static inline int mdt_fail_write(const struct lu_env *env,
943 struct dt_device *dd, int id)
945 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
946 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
948 return dd->dd_ops->dt_ro(env, dd);
949 /* We set FAIL_ONCE because we never "un-fail" a device */
955 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
957 return &req->rq_export->exp_mdt_data;
960 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
962 return mdt_dev(exp->exp_obd->obd_lu_dev);
965 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
966 struct mdt_lock_handle *lhc);
967 static inline int mdt_check_resent(struct mdt_thread_info *info,
968 mdt_reconstruct_t reconstruct,
969 struct mdt_lock_handle *lhc)
971 struct ptlrpc_request *req = mdt_info_req(info);
974 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
975 if (req_xid_is_last(req)) {
976 reconstruct(info, lhc);
979 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
980 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
985 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
986 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
988 static inline int is_identity_get_disabled(struct upcall_cache *cache)
990 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
993 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
995 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
996 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
997 struct lustre_handle *lh,
999 ldlm_policy_data_t *policy,
1000 const struct ldlm_res_id *res_id,
1001 __u64 flags, const __u64 *client_cookie)
1005 LASSERT(ns != NULL);
1006 LASSERT(lh != NULL);
1008 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1009 mode, &flags, mdt_blocking_ast,
1010 ldlm_completion_ast, NULL, NULL, 0,
1011 LVB_T_NONE, client_cookie, lh);
1012 return rc == ELDLM_OK ? 0 : -EIO;
1015 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1018 ldlm_lock_decref(lh, mode);
1021 extern mdl_mode_t mdt_mdl_lock_modes[];
1022 extern ldlm_mode_t mdt_dlm_lock_modes[];
1024 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1026 LASSERT(IS_PO2(mode));
1027 return mdt_mdl_lock_modes[mode];
1030 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1032 LASSERT(IS_PO2(mode));
1033 return mdt_dlm_lock_modes[mode];
1037 extern struct ldlm_valblock_ops mdt_lvbo;
1039 void mdt_enable_cos(struct mdt_device *, int);
1040 int mdt_cos_is_enabled(struct mdt_device *);
1058 LPROC_MDT_SAMEDIR_RENAME,
1059 LPROC_MDT_CROSSDIR_RENAME,
1062 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1063 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1064 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1065 void mdt_procfs_fini(struct mdt_device *mdt);
1067 /* lustre/mdt_mdt_lproc.c */
1068 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1070 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1071 struct mdt_device *mdt,
1072 struct ptlrpc_request *req,
1073 struct mdt_object *src, struct mdt_object *tgt);
1076 int mdt_ck_thread_start(struct mdt_device *mdt);
1077 void mdt_ck_thread_stop(struct mdt_device *mdt);
1078 void mdt_ck_timer_callback(unsigned long castmeharder);
1079 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1080 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1081 const struct lu_fid *fid, struct lustre_capa *capa);
1082 void mdt_dump_capainfo(struct mdt_thread_info *info);
1084 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1086 return mdt->mdt_lu_dev.ld_obd;
1089 extern const struct lu_device_operations mdt_lu_ops;
1091 static inline char *mdt_obd_name(struct mdt_device *mdt)
1093 return mdt->mdt_lu_dev.ld_obd->obd_name;
1096 int mds_mod_init(void);
1097 void mds_mod_exit(void);
1099 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1101 struct obd_export *exp = req->rq_export;
1104 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1105 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1110 #endif /* _MDT_INTERNAL_H */