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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdt/mdt_internal.h
34 * Lustre Metadata Target (mdt) request handler
36 * Author: Peter Braam <braam@clusterfs.com>
37 * Author: Andreas Dilger <adilger@clusterfs.com>
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Mike Shaver <shaver@clusterfs.com>
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Author: Huang Hua <huanghua@clusterfs.com>
44 #ifndef _MDT_INTERNAL_H
45 #define _MDT_INTERNAL_H
47 #include <libcfs/libcfs.h>
48 #include <libcfs/libcfs_hash.h>
49 #include <upcall_cache.h>
50 #include <lustre_net.h>
51 #include <lustre/lustre_idl.h>
52 #include <obd_class.h>
53 #include <lustre_disk.h>
54 #include <lu_target.h>
55 #include <md_object.h>
56 #include <lustre_fid.h>
57 #include <lustre_fld.h>
58 #include <lustre_req_layout.h>
59 #include <lustre_sec.h>
60 #include <lustre_idmap.h>
61 #include <lustre_eacl.h>
62 #include <lustre_quota.h>
63 #include <lustre_linkea.h>
67 /* file data for open files on MDS */
68 struct mdt_file_data {
69 /** portals handle must be first */
70 struct portals_handle mfd_handle;
71 /** open mode provided by client */
73 /** protected by med_open_lock */
74 struct list_head mfd_list;
75 /** xid of the open request */
77 /** old handle in replay case */
78 struct lustre_handle mfd_old_handle;
79 /** point to opened object */
80 struct mdt_object *mfd_object;
83 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
84 #define CDT_NORETRY_ACTION (1ULL << 1)
85 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
86 #define CDT_POLICY_SHIFT_COUNT 2
87 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
90 /* when adding a new policy, do not forget to update
91 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
93 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
95 enum cdt_states { CDT_STOPPED = 0,
101 /* when multiple lock are needed, the lock order is
109 wait_queue_head_t cdt_waitq; /**< cdt wait queue */
110 unsigned int cdt_flags; /**< cdt event flags */
111 struct lu_env cdt_env; /**< coordinator lustre
113 struct lu_context cdt_session; /** session for lu_ucred */
114 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
115 __u64 cdt_policy; /**< policy flags */
116 enum cdt_states cdt_state; /**< state */
117 atomic_t cdt_compound_id; /**< compound id
119 __u64 cdt_last_cookie; /**< last cookie
121 struct mutex cdt_llog_lock; /**< protect llog
123 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
124 struct rw_semaphore cdt_request_lock; /**< protect request
126 struct mutex cdt_restore_lock; /**< protect restore
128 cfs_time_t cdt_loop_period; /**< llog scan period */
129 cfs_time_t cdt_grace_delay; /**< request grace
131 cfs_time_t cdt_active_req_timeout; /**< request timeout */
132 __u32 cdt_default_archive_id; /**< archive id used
135 __u64 cdt_max_requests; /**< max count of started
137 atomic_t cdt_request_count; /**< current count of
138 * started requests */
139 /* started requests (struct cdt_agent_req:car_cookie_hash)
140 * indexed by cookie */
141 struct cfs_hash *cdt_request_cookie_hash;
142 /* started requests (struct cdt_agent_req:car_request_list) */
143 struct list_head cdt_request_list;
144 struct list_head cdt_agents; /**< list of register
146 struct list_head cdt_restore_hdl; /**< list of restore lock
148 /* Bitmasks indexed by the HSMA_XXX constants. */
149 __u64 cdt_user_request_mask;
150 __u64 cdt_group_request_mask;
151 __u64 cdt_other_request_mask;
154 /* mdt state flag bits */
155 #define MDT_FL_CFGLOG 0
156 #define MDT_FL_SYNCED 1
160 struct lu_device mdt_lu_dev;
161 struct seq_server_site mdt_seq_site;
162 /* DLM name-space for meta-data locks maintained by this server */
163 struct ldlm_namespace *mdt_namespace;
164 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
165 struct ptlrpc_client *mdt_ldlm_client;
166 /* underlying device */
167 struct obd_export *mdt_child_exp;
168 struct md_device *mdt_child;
169 struct dt_device *mdt_bottom;
170 struct obd_export *mdt_bottom_exp;
172 struct lu_target mdt_lut;
174 * Options bit-fields.
177 unsigned int mo_user_xattr:1,
183 /* mdt state flags */
184 unsigned long mdt_state;
186 /* transaction callbacks */
187 struct dt_txn_callback mdt_txn_cb;
189 /* these values should be updated from lov if necessary.
190 * or should be placed somewhere else. */
195 struct upcall_cache *mdt_identity_cache;
197 unsigned int mdt_capa_conf:1,
198 /* Enable remote dir on non-MDT0 */
199 mdt_enable_remote_dir:1,
202 gid_t mdt_enable_remote_dir_gid;
204 /* lock for osfs and md_root */
207 /* statfs optimization: we cache a bit */
208 struct obd_statfs mdt_osfs;
212 struct root_squash_info mdt_squash;
214 struct rename_stats mdt_rename_stats;
215 struct lu_fid mdt_md_root_fid;
217 /* connection to quota master */
218 struct obd_export *mdt_qmt_exp;
219 /* quota master device associated with this MDT */
220 struct lu_device *mdt_qmt_dev;
222 struct coordinator mdt_coordinator;
224 /* inter-MDT connection count */
225 atomic_t mdt_mds_mds_conns;
227 /* MDT device async commit count, used for debug and sanity test */
228 atomic_t mdt_async_commit_count;
230 struct mdt_object *mdt_md_root;
233 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
234 #define MDT_COS_DEFAULT (0)
237 struct lu_object_header mot_header;
238 struct lu_object mot_obj;
239 unsigned int mot_lov_created:1, /* lov object created */
240 mot_cache_attr:1; /* enable remote object
243 spinlock_t mot_write_lock;
244 /* Lock to protect create_data */
245 struct mutex mot_lov_mutex;
246 /* Lock to protect lease open.
247 * Lease open acquires write lock; normal open acquires read lock */
248 struct rw_semaphore mot_open_sem;
249 atomic_t mot_lease_count;
250 atomic_t mot_open_count;
253 struct mdt_lock_handle {
254 /* Lock type, reg for cross-ref use or pdo lock. */
258 struct lustre_handle mlh_reg_lh;
259 enum ldlm_mode mlh_reg_mode;
262 struct lustre_handle mlh_pdo_lh;
263 enum ldlm_mode mlh_pdo_mode;
264 unsigned int mlh_pdo_hash;
266 /* Remote regular lock */
267 struct lustre_handle mlh_rreg_lh;
268 enum ldlm_mode mlh_rreg_mode;
272 MDT_LH_PARENT, /* parent lockh */
273 MDT_LH_CHILD, /* child lockh */
274 MDT_LH_OLD, /* old lockh for rename */
275 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
276 MDT_LH_NEW, /* new lockh for rename */
277 MDT_LH_RMT, /* used for return lh to caller */
278 MDT_LH_LOCAL, /* local lock never return to client */
287 /* Special magical errno for communicaiton between mdt_reint_open()
288 * and mdt_intent_reint() which means return the lock to the client
289 * for subsequent cross ref open. Previously we used plain -EREMOTE
290 * but other functions called in that path might return it too and
291 * confuse us. This is not returned to the client. See LU-5370. */
292 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
294 struct mdt_reint_record {
295 mdt_reint_t rr_opcode;
296 const struct lustre_handle *rr_handle;
297 const struct lu_fid *rr_fid1;
298 const struct lu_fid *rr_fid2;
299 struct lu_name rr_name;
300 struct lu_name rr_tgt_name;
306 enum mdt_reint_flag {
307 MRF_OPEN_TRUNC = 1 << 0,
311 * Common data shared by mdt-level handlers. This is allocated per-thread to
312 * reduce stack consumption.
314 struct mdt_thread_info {
317 * The following members will be filled explicitly
318 * with specific data in mdt_thread_info_init().
320 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
321 * request handling may migrate from one server thread to another.
323 struct req_capsule *mti_pill;
325 /* although we have export in req, there are cases when it is not
326 * available, e.g. closing files upon export destroy */
327 struct obd_export *mti_exp;
329 * A couple of lock handles.
331 struct mdt_lock_handle mti_lh[MDT_LH_NR];
333 struct mdt_device *mti_mdt;
334 const struct lu_env *mti_env;
336 /* transaction number of current request */
342 * The following members will be filled expilictly
343 * with zero in mdt_thread_info_init(). These members may be used
350 struct md_attr mti_attr;
352 * Body for "habeo corpus" operations.
354 const struct mdt_body *mti_body;
356 * Host object. This is released at the end of mdt_handler().
358 struct mdt_object *mti_object;
360 * Lock request for "habeo clavis" operations.
362 const struct ldlm_request *mti_dlm_req;
364 __u32 mti_has_trans:1, /* has txn already? */
366 /* big_lmm buffer was used and must be used in reply */
370 /* opdata for mdt_reint_open(), has the same as
371 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
372 * value onto disk for recovery when mdt_trans_stop_cb() is called.
378 * The following members will be filled explicitly
379 * with zero in mdt_reint_unpack(), because they are only used
380 * by reint requests (including mdt_reint_open()).
384 * reint record. contains information for reint operations.
386 struct mdt_reint_record mti_rr;
388 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
390 * Operation specification (currently create and lookup)
392 struct md_op_spec mti_spec;
396 * The following members will _NOT_ be initialized at all.
397 * DO NOT expect them to contain any valid value.
398 * They should be initialized explicitly by the user themselves.
401 /* XXX: If something is in a union, make sure they do not conflict */
402 struct lu_fid mti_tmp_fid1;
403 struct lu_fid mti_tmp_fid2;
404 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
405 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
407 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
408 char ns_name[48];/* for mdt_init0() */
409 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
410 struct obd_statfs osfs; /* for mdt_statfs() */
412 /* for mdt_readpage() */
413 struct lu_rdpg mti_rdpg;
414 /* for mdt_sendpage() */
415 struct l_wait_info mti_wait_info;
422 struct lustre_handle mti_close_handle;
424 struct lu_buf mti_buf;
425 struct lu_buf mti_big_buf;
427 /* Ops object filename */
428 struct lu_name mti_name;
429 /* per-thread values, can be re-used, may be vmalloc'd */
434 /* should be enough to fit lustre_mdt_attrs */
435 char mti_xattr_buf[128];
436 struct ldlm_enqueue_info mti_einfo;
437 struct tg_reply_data *mti_reply_data;
440 extern struct lu_context_key mdt_thread_key;
442 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
444 struct mdt_thread_info *mti;
446 lu_env_refill((void *)env);
447 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
452 struct cdt_req_progress {
453 struct mutex crp_lock; /**< protect tree */
454 struct interval_node *crp_root; /**< tree to track extent
456 struct interval_node **crp_node; /**< buffer for tree nodes
457 * vector of fixed size
459 int crp_cnt; /**< # of used nodes */
460 int crp_max; /**< # of allocated nodes */
463 struct cdt_agent_req {
464 struct hlist_node car_cookie_hash; /**< find req by cookie */
465 struct list_head car_request_list; /**< to chain all the req. */
466 atomic_t car_refcount; /**< reference counter */
467 __u64 car_compound_id; /**< compound id */
468 __u64 car_flags; /**< request original flags */
469 struct obd_uuid car_uuid; /**< agent doing the req. */
470 __u32 car_archive_id; /**< archive id */
471 int car_canceled; /**< request was canceled */
472 cfs_time_t car_req_start; /**< start time */
473 cfs_time_t car_req_update; /**< last update time */
474 struct hsm_action_item *car_hai; /**< req. to the agent */
475 struct cdt_req_progress car_progress; /**< track data mvt
478 extern struct kmem_cache *mdt_hsm_car_kmem;
481 struct list_head ha_list; /**< to chain the agents */
482 struct obd_uuid ha_uuid; /**< agent uuid */
483 __u32 *ha_archive_id; /**< archive id */
484 int ha_archive_cnt; /**< number of archive entries
485 * 0 means any archive */
486 atomic_t ha_requests; /**< current request count */
487 atomic_t ha_success; /**< number of successful
489 atomic_t ha_failure; /**< number of failed actions */
492 struct cdt_restore_handle {
493 struct list_head crh_list; /**< to chain the handle */
494 struct lu_fid crh_fid; /**< fid of the object */
495 struct ldlm_extent crh_extent; /**< extent of the restore */
496 struct mdt_lock_handle crh_lh; /**< lock handle */
498 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
500 static inline const struct md_device_operations *
501 mdt_child_ops(struct mdt_device * m)
503 LASSERT(m->mdt_child);
504 return m->mdt_child->md_ops;
507 static inline struct md_object *mdt_object_child(struct mdt_object *o)
510 return lu2md(lu_object_next(&o->mot_obj));
513 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
515 return info->mti_pill ? info->mti_pill->rc_req : NULL;
518 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
520 LASSERT(info->mti_exp);
521 return exp_connect_flags(info->mti_exp);
524 static inline void mdt_object_get(const struct lu_env *env,
525 struct mdt_object *o)
528 lu_object_get(&o->mot_obj);
532 static inline void mdt_object_put(const struct lu_env *env,
533 struct mdt_object *o)
536 lu_object_put(env, &o->mot_obj);
540 static inline int mdt_object_exists(const struct mdt_object *o)
542 return lu_object_exists(&o->mot_obj);
545 static inline int mdt_object_remote(const struct mdt_object *o)
547 return lu_object_remote(&o->mot_obj);
550 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
552 return lu_object_fid(&o->mot_obj);
555 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
557 return mdt->mdt_lu_dev.ld_site;
560 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
562 return &mdt->mdt_seq_site;
565 static inline void mdt_export_evict(struct obd_export *exp)
567 class_fail_export(exp);
570 /* Here we use LVB_TYPE to check dne client, because it is
571 * also landed on 2.4. */
572 static inline bool mdt_is_dne_client(struct obd_export *exp)
574 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
577 static inline bool mdt_is_striped_client(struct obd_export *exp)
579 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
582 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
583 void mdt_set_disposition(struct mdt_thread_info *info,
584 struct ldlm_reply *rep, __u64 op_flag);
585 void mdt_clear_disposition(struct mdt_thread_info *info,
586 struct ldlm_reply *rep, __u64 op_flag);
588 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
589 const struct lu_name *lname);
591 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
593 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
594 struct mdt_lock_handle *lh);
596 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
597 struct mdt_lock_handle *lhc);
599 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *mo,
600 struct mdt_lock_handle *lh, __u64 ibits);
602 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
603 struct mdt_lock_handle *lh, __u64 ibits,
606 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
607 struct mdt_lock_handle *lh, __u64 ibits);
609 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
610 struct mdt_object *o, struct mdt_lock_handle *lh,
611 __u64 ibits, bool cos_incompat);
613 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
614 struct mdt_lock_handle *lh, int decref);
616 struct mdt_object *mdt_object_new(const struct lu_env *env,
618 const struct lu_fid *);
619 struct mdt_object *mdt_object_find(const struct lu_env *,
621 const struct lu_fid *);
622 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
623 const struct lu_fid *,
624 struct mdt_lock_handle *,
626 void mdt_object_unlock_put(struct mdt_thread_info *,
628 struct mdt_lock_handle *,
631 void mdt_client_compatibility(struct mdt_thread_info *info);
633 int mdt_remote_object_lock(struct mdt_thread_info *mti,
634 struct mdt_object *o, const struct lu_fid *fid,
635 struct lustre_handle *lh,
636 enum ldlm_mode mode, __u64 ibits, bool nonblock,
639 enum mdt_name_flags {
643 int mdt_name_unpack(struct req_capsule *pill,
644 const struct req_msg_field *field,
646 enum mdt_name_flags flags);
647 int mdt_close_unpack(struct mdt_thread_info *info);
648 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
649 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata);
650 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
651 #ifdef CONFIG_FS_POSIX_ACL
652 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
653 struct mdt_object *o, struct lu_nodemap *nodemap);
655 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
656 const struct lu_attr *attr, const struct lu_fid *fid);
658 int mdt_getxattr(struct mdt_thread_info *info);
659 int mdt_reint_setxattr(struct mdt_thread_info *info,
660 struct mdt_lock_handle *lh);
662 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
663 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
665 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
666 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
667 struct mdt_lock_handle *lhc);
669 extern void target_recovery_fini(struct obd_device *obd);
670 extern void target_recovery_init(struct lu_target *lut,
671 svc_handler_t handler);
672 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
673 struct obd_device *, struct lustre_sb_info *lsi);
674 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
676 int mdt_export_stats_init(struct obd_device *obd,
677 struct obd_export *exp,
680 int mdt_lock_new_child(struct mdt_thread_info *info,
681 struct mdt_object *o,
682 struct mdt_lock_handle *child_lockh);
684 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
687 int mdt_reint_open(struct mdt_thread_info *info,
688 struct mdt_lock_handle *lhc);
690 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
691 const struct lustre_handle *handle,
694 int mdt_get_info(struct tgt_session_info *tsi);
695 int mdt_attr_get_complex(struct mdt_thread_info *info,
696 struct mdt_object *o, struct md_attr *ma);
697 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
699 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
700 struct md_attr *ma, const char *name);
701 int mdt_write_get(struct mdt_object *o);
702 void mdt_write_put(struct mdt_object *o);
703 int mdt_write_read(struct mdt_object *o);
704 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
705 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
706 void mdt_mfd_free(struct mdt_file_data *mfd);
707 int mdt_close(struct tgt_session_info *tsi);
708 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
710 int mdt_fix_reply(struct mdt_thread_info *info);
711 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
712 const struct md_attr *);
713 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
715 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
716 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
717 const void *area, ssize_t len);
719 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
720 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
722 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
723 int mdt_check_ucred(struct mdt_thread_info *);
724 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
725 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *, struct mdt_body *);
726 int mdt_init_ucred_reint(struct mdt_thread_info *);
727 void mdt_exit_ucred(struct mdt_thread_info *);
728 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
729 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
730 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
732 void mdt_thread_info_init(struct ptlrpc_request *req,
733 struct mdt_thread_info *mti);
734 void mdt_thread_info_fini(struct mdt_thread_info *mti);
735 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
737 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
738 const struct md_hsm *mh);
740 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
741 void *data, int flag);
742 int mdt_links_read(struct mdt_thread_info *info,
743 struct mdt_object *mdt_obj,
744 struct linkea_data *ldata);
745 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
746 struct mdt_body *repbody);
748 static inline struct mdt_device *mdt_dev(struct lu_device *d)
750 return container_of0(d, struct mdt_device, mdt_lu_dev);
753 static inline struct mdt_object *mdt_obj(struct lu_object *o)
755 return container_of0(o, struct mdt_object, mot_obj);
758 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
760 struct lu_object *lo;
761 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
763 lo = lu_object_locate(mo->mot_obj.lo_header,
764 mdt->mdt_bottom->dd_lu_dev.ld_type);
769 static inline bool agent_req_in_final_state(enum agent_req_status ars)
771 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
772 (ars == ARS_CANCELED));
775 /* mdt/mdt_identity.c */
776 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
778 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
780 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
782 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
784 void mdt_flush_identity(struct upcall_cache *, int);
786 __u32 mdt_identity_get_perm(struct md_identity *, lnet_nid_t);
788 /* mdt/mdt_recovery.c */
789 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
792 int mdt_hsm_state_get(struct tgt_session_info *tsi);
793 int mdt_hsm_state_set(struct tgt_session_info *tsi);
794 int mdt_hsm_action(struct tgt_session_info *tsi);
795 int mdt_hsm_progress(struct tgt_session_info *tsi);
796 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
797 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
798 int mdt_hsm_request(struct tgt_session_info *tsi);
800 /* mdt/mdt_hsm_cdt_actions.c */
801 extern const struct file_operations mdt_hsm_actions_fops;
802 void dump_llog_agent_req_rec(const char *prefix,
803 const struct llog_agent_req_rec *larr);
804 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
805 llog_cb_t cb, void *data);
806 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
807 __u64 compound_id, __u32 archive_id,
808 __u64 flags, struct hsm_action_item *hai);
809 int mdt_agent_record_update(const struct lu_env *env,
810 struct mdt_device *mdt, __u64 *cookies,
811 int cookies_count, enum agent_req_status status);
813 /* mdt/mdt_hsm_cdt_agent.c */
814 extern const struct file_operations mdt_hsm_agent_fops;
815 int mdt_hsm_agent_register(struct mdt_thread_info *info,
816 const struct obd_uuid *uuid,
817 int nr_archives, __u32 *archive_num);
818 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
819 const struct obd_uuid *uuid,
821 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
822 const struct obd_uuid *uuid);
823 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
824 int succ_rq, int fail_rq, int new_rq,
825 const struct obd_uuid *uuid);
826 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
827 struct obd_uuid *uuid);
828 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
830 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
831 struct hsm_progress_kernel *pgs);
832 /* mdt/mdt_hsm_cdt_client.c */
833 int mdt_hsm_add_actions(struct mdt_thread_info *info,
834 struct hsm_action_list *hal);
835 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
836 struct hsm_action_list *hal);
837 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
838 const struct lu_fid *fid);
839 /* mdt/mdt_hsm_cdt_requests.c */
840 extern struct cfs_hash_ops cdt_request_cookie_hash_ops;
841 extern const struct file_operations mdt_hsm_active_requests_fops;
842 void dump_requests(char *prefix, struct coordinator *cdt);
843 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
844 __u64 flags, struct obd_uuid *uuid,
845 struct hsm_action_item *hai);
846 void mdt_cdt_free_request(struct cdt_agent_req *car);
847 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
848 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt, u64 cookie);
849 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
850 void mdt_cdt_get_request(struct cdt_agent_req *car);
851 void mdt_cdt_put_request(struct cdt_agent_req *car);
852 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
853 const struct hsm_progress_kernel *pgs);
854 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
855 /* mdt/mdt_coordinator.c */
856 void mdt_hsm_dump_hal(int level, const char *prefix,
857 struct hsm_action_list *hal);
858 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
859 const struct lu_fid *fid);
860 /* coordinator management */
861 int mdt_hsm_cdt_init(struct mdt_device *mdt);
862 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
863 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
864 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
866 /* coordinator control /proc interface */
867 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file,
868 const char __user *buffer,
869 size_t count, loff_t *off);
870 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
871 int hsm_cdt_procfs_init(struct mdt_device *mdt);
872 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
873 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
875 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
876 const struct lu_fid *fid,
878 /* actions/request helpers */
879 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
880 struct hsm_action_list *hal, struct obd_uuid *uuid);
881 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
882 u32 archive_id, u64 rq_flags,
883 const struct md_hsm *hsm);
884 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
885 struct hsm_progress_kernel *pgs,
886 const int update_record);
888 int mdt_close_swap_layouts(struct mdt_thread_info *info,
889 struct mdt_object *o, struct md_attr *ma);
891 extern struct lu_context_key mdt_thread_key;
893 /* debug issues helper starts here*/
894 static inline int mdt_fail_write(const struct lu_env *env,
895 struct dt_device *dd, int id)
897 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
898 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
900 return dt_ro(env, dd);
901 /* We set FAIL_ONCE because we never "un-fail" a device */
907 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
909 return &req->rq_export->exp_mdt_data;
912 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
914 return mdt_dev(exp->exp_obd->obd_lu_dev);
917 static inline bool mdt_rdonly(struct obd_export *exp)
919 if (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
920 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly)
925 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
926 struct mdt_lock_handle *lhc);
927 static inline int mdt_check_resent(struct mdt_thread_info *info,
928 mdt_reconstruct_t reconstruct,
929 struct mdt_lock_handle *lhc)
931 struct ptlrpc_request *req = mdt_info_req(info);
935 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
936 OBD_ALLOC_PTR(info->mti_reply_data);
937 if (info->mti_reply_data == NULL)
940 if (req_can_reconstruct(req, info->mti_reply_data)) {
941 reconstruct(info, lhc);
945 "no reply data found for RESENT req");
948 OBD_FREE_PTR(info->mti_reply_data);
949 info->mti_reply_data = NULL;
954 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
955 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
957 static inline int is_identity_get_disabled(struct upcall_cache *cache)
959 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
962 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
964 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
965 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
966 struct lustre_handle *lh, enum ldlm_mode mode,
967 union ldlm_policy_data *policy,
968 const struct ldlm_res_id *res_id,
969 __u64 flags, const __u64 *client_cookie)
976 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
977 mode, &flags, mdt_blocking_ast,
978 ldlm_completion_ast, NULL, NULL, 0,
979 LVB_T_NONE, client_cookie, lh);
980 return rc == ELDLM_OK ? 0 : -EIO;
983 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
985 ldlm_lock_decref(lh, mode);
988 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
990 return mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL;
993 extern mdl_mode_t mdt_mdl_lock_modes[];
994 extern enum ldlm_mode mdt_dlm_lock_modes[];
996 /* LCK_MINMODE which is zero returns false for is_power_of_2 */
998 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
1000 LASSERT(mode == LCK_MINMODE || is_power_of_2(mode));
1001 return mdt_mdl_lock_modes[mode];
1004 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1006 LASSERT(mode == MDL_MINMODE || is_power_of_2(mode));
1007 return mdt_dlm_lock_modes[mode];
1011 extern struct ldlm_valblock_ops mdt_lvbo;
1013 void mdt_enable_cos(struct mdt_device *, int);
1014 int mdt_cos_is_enabled(struct mdt_device *);
1032 LPROC_MDT_SAMEDIR_RENAME,
1033 LPROC_MDT_CROSSDIR_RENAME,
1036 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1037 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1038 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1039 void mdt_procfs_fini(struct mdt_device *mdt);
1041 /* lustre/mdt_mdt_lproc.c */
1042 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1044 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1045 struct mdt_device *mdt,
1046 struct ptlrpc_request *req,
1047 struct mdt_object *src, struct mdt_object *tgt);
1049 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1051 return mdt->mdt_lu_dev.ld_obd;
1054 extern const struct lu_device_operations mdt_lu_ops;
1056 static inline char *mdt_obd_name(struct mdt_device *mdt)
1058 return mdt->mdt_lu_dev.ld_obd->obd_name;
1061 int mds_mod_init(void);
1062 void mds_mod_exit(void);
1064 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1066 struct obd_export *exp = req->rq_export;
1069 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1070 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1075 #endif /* _MDT_INTERNAL_H */