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 /* Coordinator states. Keep the cdt_transition table in sync. */
96 enum cdt_states { CDT_STOPPED = 0,
105 static inline char *cdt_mdt_state2str(int state)
123 /* when multiple lock are needed, the lock order is
131 wait_queue_head_t cdt_waitq; /**< cdt wait queue */
132 unsigned int cdt_flags; /**< cdt event flags */
133 struct lu_env cdt_env; /**< coordinator lustre
135 struct lu_context cdt_session; /** session for lu_ucred */
136 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
137 __u64 cdt_policy; /**< policy flags */
138 enum cdt_states cdt_state; /**< state */
139 spinlock_t cdt_state_lock; /**< cdt_state lock */
140 atomic_t cdt_compound_id; /**< compound id
142 __u64 cdt_last_cookie; /**< last cookie
144 struct mutex cdt_llog_lock; /**< protect llog
146 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
147 struct rw_semaphore cdt_request_lock; /**< protect request
149 struct mutex cdt_restore_lock; /**< protect restore
151 cfs_time_t cdt_loop_period; /**< llog scan period */
152 cfs_time_t cdt_grace_delay; /**< request grace
154 cfs_time_t cdt_active_req_timeout; /**< request timeout */
155 __u32 cdt_default_archive_id; /**< archive id used
158 __u64 cdt_max_requests; /**< max count of started
160 atomic_t cdt_request_count; /**< current count of
161 * started requests */
162 /* started requests (struct cdt_agent_req:car_cookie_hash)
163 * indexed by cookie */
164 struct cfs_hash *cdt_request_cookie_hash;
165 /* started requests (struct cdt_agent_req:car_request_list) */
166 struct list_head cdt_request_list;
167 struct list_head cdt_agents; /**< list of register
169 struct list_head cdt_restore_hdl; /**< list of restore lock
171 /* Bitmasks indexed by the HSMA_XXX constants. */
172 __u64 cdt_user_request_mask;
173 __u64 cdt_group_request_mask;
174 __u64 cdt_other_request_mask;
177 /* mdt state flag bits */
178 #define MDT_FL_CFGLOG 0
179 #define MDT_FL_SYNCED 1
183 struct lu_device mdt_lu_dev;
184 struct seq_server_site mdt_seq_site;
185 /* DLM name-space for meta-data locks maintained by this server */
186 struct ldlm_namespace *mdt_namespace;
187 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
188 struct ptlrpc_client *mdt_ldlm_client;
189 /* underlying device */
190 struct obd_export *mdt_child_exp;
191 struct md_device *mdt_child;
192 struct dt_device *mdt_bottom;
193 struct obd_export *mdt_bottom_exp;
195 struct lu_target mdt_lut;
197 * Options bit-fields.
200 unsigned int mo_user_xattr:1,
205 /* mdt state flags */
206 unsigned long mdt_state;
208 /* transaction callbacks */
209 struct dt_txn_callback mdt_txn_cb;
211 /* these values should be updated from lov if necessary.
212 * or should be placed somewhere else. */
217 struct upcall_cache *mdt_identity_cache;
219 unsigned int mdt_capa_conf:1,
220 /* Enable remote dir on non-MDT0 */
221 mdt_enable_remote_dir:1,
224 gid_t mdt_enable_remote_dir_gid;
226 /* lock for osfs and md_root */
229 /* statfs optimization: we cache a bit */
230 struct obd_statfs mdt_osfs;
234 struct root_squash_info mdt_squash;
236 struct rename_stats mdt_rename_stats;
237 struct lu_fid mdt_md_root_fid;
239 /* connection to quota master */
240 struct obd_export *mdt_qmt_exp;
241 /* quota master device associated with this MDT */
242 struct lu_device *mdt_qmt_dev;
244 struct coordinator mdt_coordinator;
246 /* inter-MDT connection count */
247 atomic_t mdt_mds_mds_conns;
249 /* MDT device async commit count, used for debug and sanity test */
250 atomic_t mdt_async_commit_count;
252 struct mdt_object *mdt_md_root;
255 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
256 #define MDT_COS_DEFAULT (0)
259 struct lu_object_header mot_header;
260 struct lu_object mot_obj;
261 unsigned int mot_lov_created:1, /* lov object created */
262 mot_cache_attr:1; /* enable remote object
265 spinlock_t mot_write_lock;
266 /* Lock to protect create_data */
267 struct mutex mot_lov_mutex;
268 /* Lock to protect lease open.
269 * Lease open acquires write lock; normal open acquires read lock */
270 struct rw_semaphore mot_open_sem;
271 atomic_t mot_lease_count;
272 atomic_t mot_open_count;
275 struct mdt_lock_handle {
276 /* Lock type, reg for cross-ref use or pdo lock. */
280 struct lustre_handle mlh_reg_lh;
281 enum ldlm_mode mlh_reg_mode;
284 struct lustre_handle mlh_pdo_lh;
285 enum ldlm_mode mlh_pdo_mode;
286 unsigned int mlh_pdo_hash;
288 /* Remote regular lock */
289 struct lustre_handle mlh_rreg_lh;
290 enum ldlm_mode mlh_rreg_mode;
294 MDT_LH_PARENT, /* parent lockh */
295 MDT_LH_CHILD, /* child lockh */
296 MDT_LH_OLD, /* old lockh for rename */
297 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
298 MDT_LH_NEW, /* new lockh for rename */
299 MDT_LH_RMT, /* used for return lh to caller */
300 MDT_LH_LOCAL, /* local lock never return to client */
309 /* Special magical errno for communicaiton between mdt_reint_open()
310 * and mdt_intent_reint() which means return the lock to the client
311 * for subsequent cross ref open. Previously we used plain -EREMOTE
312 * but other functions called in that path might return it too and
313 * confuse us. This is not returned to the client. See LU-5370. */
314 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
316 struct mdt_reint_record {
317 mdt_reint_t rr_opcode;
318 const struct lustre_handle *rr_handle;
319 const struct lu_fid *rr_fid1;
320 const struct lu_fid *rr_fid2;
321 struct lu_name rr_name;
322 struct lu_name rr_tgt_name;
328 enum mdt_reint_flag {
329 MRF_OPEN_TRUNC = 1 << 0,
333 * Common data shared by mdt-level handlers. This is allocated per-thread to
334 * reduce stack consumption.
336 struct mdt_thread_info {
339 * The following members will be filled explicitly
340 * with specific data in mdt_thread_info_init().
342 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
343 * request handling may migrate from one server thread to another.
345 struct req_capsule *mti_pill;
347 /* although we have export in req, there are cases when it is not
348 * available, e.g. closing files upon export destroy */
349 struct obd_export *mti_exp;
351 * A couple of lock handles.
353 struct mdt_lock_handle mti_lh[MDT_LH_NR];
355 struct mdt_device *mti_mdt;
356 const struct lu_env *mti_env;
358 /* transaction number of current request */
364 * The following members will be filled expilictly
365 * with zero in mdt_thread_info_init(). These members may be used
372 struct md_attr mti_attr;
374 * Body for "habeo corpus" operations.
376 const struct mdt_body *mti_body;
378 * Host object. This is released at the end of mdt_handler().
380 struct mdt_object *mti_object;
382 * Lock request for "habeo clavis" operations.
384 const struct ldlm_request *mti_dlm_req;
386 __u32 mti_has_trans:1, /* has txn already? */
388 /* big_lmm buffer was used and must be used in reply */
392 /* opdata for mdt_reint_open(), has the same as
393 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
394 * value onto disk for recovery when mdt_trans_stop_cb() is called.
400 * The following members will be filled explicitly
401 * with zero in mdt_reint_unpack(), because they are only used
402 * by reint requests (including mdt_reint_open()).
406 * reint record. contains information for reint operations.
408 struct mdt_reint_record mti_rr;
410 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
412 * Operation specification (currently create and lookup)
414 struct md_op_spec mti_spec;
418 * The following members will _NOT_ be initialized at all.
419 * DO NOT expect them to contain any valid value.
420 * They should be initialized explicitly by the user themselves.
423 /* XXX: If something is in a union, make sure they do not conflict */
424 struct lu_fid mti_tmp_fid1;
425 struct lu_fid mti_tmp_fid2;
426 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
427 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
429 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
430 char ns_name[48];/* for mdt_init0() */
431 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
432 struct obd_statfs osfs; /* for mdt_statfs() */
434 /* for mdt_readpage() */
435 struct lu_rdpg mti_rdpg;
436 /* for mdt_sendpage() */
437 struct l_wait_info mti_wait_info;
444 struct lustre_handle mti_close_handle;
446 struct lu_buf mti_buf;
447 struct lu_buf mti_big_buf;
449 /* Ops object filename */
450 struct lu_name mti_name;
451 /* per-thread values, can be re-used, may be vmalloc'd */
456 /* should be enough to fit lustre_mdt_attrs */
457 char mti_xattr_buf[128];
458 struct ldlm_enqueue_info mti_einfo;
459 struct tg_reply_data *mti_reply_data;
462 extern struct lu_context_key mdt_thread_key;
464 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
466 struct mdt_thread_info *mti;
468 lu_env_refill((void *)env);
469 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
474 struct cdt_req_progress {
475 struct mutex crp_lock; /**< protect tree */
476 struct interval_node *crp_root; /**< tree to track extent
478 struct interval_node **crp_node; /**< buffer for tree nodes
479 * vector of fixed size
481 int crp_cnt; /**< # of used nodes */
482 int crp_max; /**< # of allocated nodes */
485 struct cdt_agent_req {
486 struct hlist_node car_cookie_hash; /**< find req by cookie */
487 struct list_head car_request_list; /**< to chain all the req. */
488 atomic_t car_refcount; /**< reference counter */
489 __u64 car_compound_id; /**< compound id */
490 __u64 car_flags; /**< request original flags */
491 struct obd_uuid car_uuid; /**< agent doing the req. */
492 __u32 car_archive_id; /**< archive id */
493 int car_canceled; /**< request was canceled */
494 cfs_time_t car_req_start; /**< start time */
495 cfs_time_t car_req_update; /**< last update time */
496 struct hsm_action_item *car_hai; /**< req. to the agent */
497 struct cdt_req_progress car_progress; /**< track data mvt
500 extern struct kmem_cache *mdt_hsm_car_kmem;
503 struct list_head ha_list; /**< to chain the agents */
504 struct obd_uuid ha_uuid; /**< agent uuid */
505 __u32 *ha_archive_id; /**< archive id */
506 int ha_archive_cnt; /**< number of archive entries
507 * 0 means any archive */
508 atomic_t ha_requests; /**< current request count */
509 atomic_t ha_success; /**< number of successful
511 atomic_t ha_failure; /**< number of failed actions */
514 struct cdt_restore_handle {
515 struct list_head crh_list; /**< to chain the handle */
516 struct lu_fid crh_fid; /**< fid of the object */
517 struct ldlm_extent crh_extent; /**< extent of the restore */
518 struct mdt_lock_handle crh_lh; /**< lock handle */
520 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
522 static inline const struct md_device_operations *
523 mdt_child_ops(struct mdt_device * m)
525 LASSERT(m->mdt_child);
526 return m->mdt_child->md_ops;
529 static inline struct md_object *mdt_object_child(struct mdt_object *o)
532 return lu2md(lu_object_next(&o->mot_obj));
535 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
537 return info->mti_pill ? info->mti_pill->rc_req : NULL;
540 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
542 LASSERT(info->mti_exp);
543 return exp_connect_flags(info->mti_exp);
546 static inline void mdt_object_get(const struct lu_env *env,
547 struct mdt_object *o)
550 lu_object_get(&o->mot_obj);
554 static inline void mdt_object_put(const struct lu_env *env,
555 struct mdt_object *o)
558 lu_object_put(env, &o->mot_obj);
562 static inline int mdt_object_exists(const struct mdt_object *o)
564 return lu_object_exists(&o->mot_obj);
567 static inline int mdt_object_remote(const struct mdt_object *o)
569 return lu_object_remote(&o->mot_obj);
572 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
574 return lu_object_fid(&o->mot_obj);
577 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
579 return mdt->mdt_lu_dev.ld_site;
582 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
584 return &mdt->mdt_seq_site;
587 static inline void mdt_export_evict(struct obd_export *exp)
589 class_fail_export(exp);
592 /* Here we use LVB_TYPE to check dne client, because it is
593 * also landed on 2.4. */
594 static inline bool mdt_is_dne_client(struct obd_export *exp)
596 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
599 static inline bool mdt_is_striped_client(struct obd_export *exp)
601 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
604 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
605 void mdt_set_disposition(struct mdt_thread_info *info,
606 struct ldlm_reply *rep, __u64 op_flag);
607 void mdt_clear_disposition(struct mdt_thread_info *info,
608 struct ldlm_reply *rep, __u64 op_flag);
610 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
611 const struct lu_name *lname);
613 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
615 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
616 struct mdt_lock_handle *lh);
618 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
619 struct mdt_lock_handle *lhc);
621 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *mo,
622 struct mdt_lock_handle *lh, __u64 ibits);
624 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
625 struct mdt_lock_handle *lh, __u64 ibits,
628 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
629 struct mdt_lock_handle *lh, __u64 ibits);
631 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
632 struct mdt_object *o, struct mdt_lock_handle *lh,
633 __u64 ibits, bool cos_incompat);
635 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
636 struct mdt_lock_handle *lh, int decref);
638 struct mdt_object *mdt_object_new(const struct lu_env *env,
640 const struct lu_fid *);
641 struct mdt_object *mdt_object_find(const struct lu_env *,
643 const struct lu_fid *);
644 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
645 const struct lu_fid *,
646 struct mdt_lock_handle *,
648 void mdt_object_unlock_put(struct mdt_thread_info *,
650 struct mdt_lock_handle *,
653 void mdt_client_compatibility(struct mdt_thread_info *info);
655 int mdt_remote_object_lock(struct mdt_thread_info *mti,
656 struct mdt_object *o, const struct lu_fid *fid,
657 struct lustre_handle *lh,
658 enum ldlm_mode mode, __u64 ibits, bool nonblock,
661 enum mdt_name_flags {
665 int mdt_name_unpack(struct req_capsule *pill,
666 const struct req_msg_field *field,
668 enum mdt_name_flags flags);
669 int mdt_close_unpack(struct mdt_thread_info *info);
670 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
671 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata);
672 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
673 #ifdef CONFIG_FS_POSIX_ACL
674 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
675 struct mdt_object *o, struct lu_nodemap *nodemap);
677 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
678 const struct lu_attr *attr, const struct lu_fid *fid);
680 int mdt_getxattr(struct mdt_thread_info *info);
681 int mdt_reint_setxattr(struct mdt_thread_info *info,
682 struct mdt_lock_handle *lh);
684 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
685 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
687 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
688 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
689 struct mdt_lock_handle *lhc);
691 extern void target_recovery_fini(struct obd_device *obd);
692 extern void target_recovery_init(struct lu_target *lut,
693 svc_handler_t handler);
694 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
695 struct obd_device *, struct lustre_sb_info *lsi);
696 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
698 int mdt_export_stats_init(struct obd_device *obd,
699 struct obd_export *exp,
702 int mdt_lock_new_child(struct mdt_thread_info *info,
703 struct mdt_object *o,
704 struct mdt_lock_handle *child_lockh);
706 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
709 int mdt_reint_open(struct mdt_thread_info *info,
710 struct mdt_lock_handle *lhc);
712 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
713 const struct lustre_handle *handle,
716 int mdt_get_info(struct tgt_session_info *tsi);
717 int mdt_attr_get_complex(struct mdt_thread_info *info,
718 struct mdt_object *o, struct md_attr *ma);
719 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
721 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
722 struct md_attr *ma, const char *name);
723 int mdt_write_get(struct mdt_object *o);
724 void mdt_write_put(struct mdt_object *o);
725 int mdt_write_read(struct mdt_object *o);
726 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
727 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
728 void mdt_mfd_free(struct mdt_file_data *mfd);
729 int mdt_close(struct tgt_session_info *tsi);
730 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
732 int mdt_fix_reply(struct mdt_thread_info *info);
733 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
734 const struct md_attr *);
735 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
737 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
738 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
739 const void *area, ssize_t len);
741 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
742 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
744 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
745 int mdt_check_ucred(struct mdt_thread_info *);
746 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
747 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *, struct mdt_body *);
748 int mdt_init_ucred_reint(struct mdt_thread_info *);
749 void mdt_exit_ucred(struct mdt_thread_info *);
750 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
751 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
752 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
754 void mdt_thread_info_init(struct ptlrpc_request *req,
755 struct mdt_thread_info *mti);
756 void mdt_thread_info_fini(struct mdt_thread_info *mti);
757 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
759 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
760 const struct md_hsm *mh);
762 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
763 void *data, int flag);
764 int mdt_links_read(struct mdt_thread_info *info,
765 struct mdt_object *mdt_obj,
766 struct linkea_data *ldata);
767 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
768 struct mdt_body *repbody);
770 static inline struct mdt_device *mdt_dev(struct lu_device *d)
772 return container_of0(d, struct mdt_device, mdt_lu_dev);
775 static inline struct mdt_object *mdt_obj(struct lu_object *o)
777 return container_of0(o, struct mdt_object, mot_obj);
780 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
782 struct lu_object *lo;
783 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
785 lo = lu_object_locate(mo->mot_obj.lo_header,
786 mdt->mdt_bottom->dd_lu_dev.ld_type);
791 static inline bool agent_req_in_final_state(enum agent_req_status ars)
793 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
794 (ars == ARS_CANCELED));
797 /* mdt/mdt_identity.c */
798 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
800 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
802 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
804 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
806 void mdt_flush_identity(struct upcall_cache *, int);
808 __u32 mdt_identity_get_perm(struct md_identity *, lnet_nid_t);
810 /* mdt/mdt_recovery.c */
811 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
814 int mdt_hsm_state_get(struct tgt_session_info *tsi);
815 int mdt_hsm_state_set(struct tgt_session_info *tsi);
816 int mdt_hsm_action(struct tgt_session_info *tsi);
817 int mdt_hsm_progress(struct tgt_session_info *tsi);
818 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
819 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
820 int mdt_hsm_request(struct tgt_session_info *tsi);
822 /* mdt/mdt_hsm_cdt_actions.c */
823 extern const struct file_operations mdt_hsm_actions_fops;
824 void dump_llog_agent_req_rec(const char *prefix,
825 const struct llog_agent_req_rec *larr);
826 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
827 llog_cb_t cb, void *data);
828 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
829 __u64 compound_id, __u32 archive_id,
830 __u64 flags, struct hsm_action_item *hai);
831 int mdt_agent_record_update(const struct lu_env *env,
832 struct mdt_device *mdt, __u64 *cookies,
833 int cookies_count, enum agent_req_status status);
835 /* mdt/mdt_hsm_cdt_agent.c */
836 extern const struct file_operations mdt_hsm_agent_fops;
837 int mdt_hsm_agent_register(struct mdt_thread_info *info,
838 const struct obd_uuid *uuid,
839 int nr_archives, __u32 *archive_num);
840 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
841 const struct obd_uuid *uuid,
843 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
844 const struct obd_uuid *uuid);
845 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
846 int succ_rq, int fail_rq, int new_rq,
847 const struct obd_uuid *uuid);
848 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
849 struct obd_uuid *uuid);
850 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
852 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
853 struct hsm_progress_kernel *pgs);
854 /* mdt/mdt_hsm_cdt_client.c */
855 int mdt_hsm_add_actions(struct mdt_thread_info *info,
856 struct hsm_action_list *hal);
857 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
858 struct hsm_action_list *hal);
859 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
860 const struct lu_fid *fid);
861 /* mdt/mdt_hsm_cdt_requests.c */
862 extern struct cfs_hash_ops cdt_request_cookie_hash_ops;
863 extern const struct file_operations mdt_hsm_active_requests_fops;
864 void dump_requests(char *prefix, struct coordinator *cdt);
865 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
866 __u64 flags, struct obd_uuid *uuid,
867 struct hsm_action_item *hai);
868 void mdt_cdt_free_request(struct cdt_agent_req *car);
869 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
870 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt, u64 cookie);
871 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
872 void mdt_cdt_get_request(struct cdt_agent_req *car);
873 void mdt_cdt_put_request(struct cdt_agent_req *car);
874 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
875 const struct hsm_progress_kernel *pgs);
876 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
877 /* mdt/mdt_coordinator.c */
878 void mdt_hsm_dump_hal(int level, const char *prefix,
879 struct hsm_action_list *hal);
880 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
881 const struct lu_fid *fid);
882 /* coordinator management */
883 int mdt_hsm_cdt_init(struct mdt_device *mdt);
884 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
885 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
886 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
888 /* coordinator control /proc interface */
889 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file,
890 const char __user *buffer,
891 size_t count, loff_t *off);
892 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
893 int hsm_cdt_procfs_init(struct mdt_device *mdt);
894 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
895 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
897 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
898 const struct lu_fid *fid,
900 /* actions/request helpers */
901 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
902 struct hsm_action_list *hal, struct obd_uuid *uuid);
903 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
904 u32 archive_id, u64 rq_flags,
905 const struct md_hsm *hsm);
906 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
907 struct hsm_progress_kernel *pgs,
908 const int update_record);
910 int mdt_close_swap_layouts(struct mdt_thread_info *info,
911 struct mdt_object *o, struct md_attr *ma);
913 extern struct lu_context_key mdt_thread_key;
915 /* debug issues helper starts here*/
916 static inline int mdt_fail_write(const struct lu_env *env,
917 struct dt_device *dd, int id)
919 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
920 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
922 return dt_ro(env, dd);
923 /* We set FAIL_ONCE because we never "un-fail" a device */
929 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
931 return &req->rq_export->exp_mdt_data;
934 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
936 return mdt_dev(exp->exp_obd->obd_lu_dev);
939 static inline bool mdt_rdonly(struct obd_export *exp)
941 if (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
942 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly)
947 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
948 struct mdt_lock_handle *lhc);
949 static inline int mdt_check_resent(struct mdt_thread_info *info,
950 mdt_reconstruct_t reconstruct,
951 struct mdt_lock_handle *lhc)
953 struct ptlrpc_request *req = mdt_info_req(info);
957 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
958 OBD_ALLOC_PTR(info->mti_reply_data);
959 if (info->mti_reply_data == NULL)
962 if (req_can_reconstruct(req, info->mti_reply_data)) {
963 reconstruct(info, lhc);
967 "no reply data found for RESENT req");
970 OBD_FREE_PTR(info->mti_reply_data);
971 info->mti_reply_data = NULL;
976 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
977 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
979 static inline int is_identity_get_disabled(struct upcall_cache *cache)
981 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
984 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
986 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
987 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
988 struct lustre_handle *lh, enum ldlm_mode mode,
989 union ldlm_policy_data *policy,
990 const struct ldlm_res_id *res_id,
991 __u64 flags, const __u64 *client_cookie)
998 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
999 mode, &flags, mdt_blocking_ast,
1000 ldlm_completion_ast, NULL, NULL, 0,
1001 LVB_T_NONE, client_cookie, lh);
1002 return rc == ELDLM_OK ? 0 : -EIO;
1005 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1007 ldlm_lock_decref(lh, mode);
1010 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
1012 return mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL;
1015 extern mdl_mode_t mdt_mdl_lock_modes[];
1016 extern enum ldlm_mode mdt_dlm_lock_modes[];
1018 /* LCK_MINMODE which is zero returns false for is_power_of_2 */
1020 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
1022 LASSERT(mode == LCK_MINMODE || is_power_of_2(mode));
1023 return mdt_mdl_lock_modes[mode];
1026 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1028 LASSERT(mode == MDL_MINMODE || is_power_of_2(mode));
1029 return mdt_dlm_lock_modes[mode];
1033 extern struct ldlm_valblock_ops mdt_lvbo;
1035 void mdt_enable_cos(struct mdt_device *, int);
1036 int mdt_cos_is_enabled(struct mdt_device *);
1054 LPROC_MDT_SAMEDIR_RENAME,
1055 LPROC_MDT_CROSSDIR_RENAME,
1058 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1059 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1060 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1061 void mdt_procfs_fini(struct mdt_device *mdt);
1063 /* lustre/mdt_mdt_lproc.c */
1064 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1066 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1067 struct mdt_device *mdt,
1068 struct ptlrpc_request *req,
1069 struct mdt_object *src, struct mdt_object *tgt);
1071 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1073 return mdt->mdt_lu_dev.ld_obd;
1076 extern const struct lu_device_operations mdt_lu_ops;
1078 static inline char *mdt_obd_name(struct mdt_device *mdt)
1080 return mdt->mdt_lu_dev.ld_obd->obd_name;
1083 int mds_mod_init(void);
1084 void mds_mod_exit(void);
1086 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1088 struct obd_export *exp = req->rq_export;
1091 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1092 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1097 #endif /* _MDT_INTERNAL_H */