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
48 #include <upcall_cache.h>
49 #include <lustre_net.h>
50 #include <lustre/lustre_idl.h>
51 #include <obd_class.h>
52 #include <lustre_disk.h>
53 #include <lu_target.h>
54 #include <md_object.h>
55 #include <lustre_fid.h>
56 #include <lustre_fld.h>
57 #include <lustre_req_layout.h>
58 #include <lustre_sec.h>
59 #include <lustre_idmap.h>
60 #include <lustre_eacl.h>
61 #include <lustre_quota.h>
62 #include <lustre_linkea.h>
66 /* file data for open files on MDS */
67 struct mdt_file_data {
68 /** portals handle must be first */
69 struct portals_handle mfd_handle;
70 /** open mode provided by client */
72 /** protected by med_open_lock */
73 struct list_head mfd_list;
74 /** xid of the open request */
76 /** old handle in replay case */
77 struct lustre_handle mfd_old_handle;
78 /** point to opened object */
79 struct mdt_object *mfd_object;
82 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
83 #define CDT_NORETRY_ACTION (1ULL << 1)
84 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
85 #define CDT_POLICY_SHIFT_COUNT 2
86 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
89 /* when adding a new policy, do not forget to update
90 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
92 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
94 enum cdt_states { CDT_STOPPED = 0,
100 /* when multiple lock are needed, the lock order is
108 wait_queue_head_t cdt_waitq; /**< cdt wait queue */
109 unsigned int cdt_flags; /**< cdt event flags */
110 struct lu_env cdt_env; /**< coordinator lustre
112 struct lu_context cdt_session; /** session for lu_ucred */
113 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
114 __u64 cdt_policy; /**< policy flags */
115 enum cdt_states cdt_state; /**< state */
116 atomic_t cdt_compound_id; /**< compound id
118 __u64 cdt_last_cookie; /**< last cookie
120 struct mutex cdt_llog_lock; /**< protect llog
122 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
123 struct rw_semaphore cdt_request_lock; /**< protect request
125 struct mutex cdt_restore_lock; /**< protect restore
127 cfs_time_t cdt_loop_period; /**< llog scan period */
128 cfs_time_t cdt_grace_delay; /**< request grace
130 cfs_time_t cdt_active_req_timeout; /**< request timeout */
131 __u32 cdt_default_archive_id; /**< archive id used
134 __u64 cdt_max_requests; /**< max count of started
136 atomic_t cdt_request_count; /**< current count of
137 * started requests */
138 struct list_head cdt_requests; /**< list of started
140 struct list_head cdt_agents; /**< list of register
142 struct list_head cdt_restore_hdl; /**< list of restore lock
144 /* Bitmasks indexed by the HSMA_XXX constants. */
145 __u64 cdt_user_request_mask;
146 __u64 cdt_group_request_mask;
147 __u64 cdt_other_request_mask;
150 /* mdt state flag bits */
151 #define MDT_FL_CFGLOG 0
152 #define MDT_FL_SYNCED 1
156 struct lu_device mdt_lu_dev;
157 struct seq_server_site mdt_seq_site;
158 /* DLM name-space for meta-data locks maintained by this server */
159 struct ldlm_namespace *mdt_namespace;
160 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
161 struct ptlrpc_client *mdt_ldlm_client;
162 /* underlying device */
163 struct obd_export *mdt_child_exp;
164 struct md_device *mdt_child;
165 struct dt_device *mdt_bottom;
166 struct obd_export *mdt_bottom_exp;
168 struct lu_target mdt_lut;
170 * Options bit-fields.
173 unsigned int mo_user_xattr:1,
179 /* mdt state flags */
180 unsigned long mdt_state;
182 /* transaction callbacks */
183 struct dt_txn_callback mdt_txn_cb;
185 /* these values should be updated from lov if necessary.
186 * or should be placed somewhere else. */
191 struct upcall_cache *mdt_identity_cache;
193 unsigned int mdt_capa_conf:1,
194 /* Enable remote dir on non-MDT0 */
195 mdt_enable_remote_dir:1,
198 gid_t mdt_enable_remote_dir_gid;
200 /* lock for osfs and md_root */
203 /* statfs optimization: we cache a bit */
204 struct obd_statfs mdt_osfs;
208 struct root_squash_info mdt_squash;
210 struct rename_stats mdt_rename_stats;
211 struct lu_fid mdt_md_root_fid;
213 /* connection to quota master */
214 struct obd_export *mdt_qmt_exp;
215 /* quota master device associated with this MDT */
216 struct lu_device *mdt_qmt_dev;
218 struct coordinator mdt_coordinator;
220 /* inter-MDT connection count */
221 atomic_t mdt_mds_mds_conns;
223 /* MDT device async commit count, used for debug and sanity test */
224 atomic_t mdt_async_commit_count;
226 struct mdt_object *mdt_md_root;
229 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
230 #define MDT_COS_DEFAULT (0)
233 struct lu_object_header mot_header;
234 struct lu_object mot_obj;
235 unsigned int mot_lov_created:1, /* lov object created */
236 mot_cache_attr:1; /* enable remote object
239 spinlock_t mot_write_lock;
240 /* Lock to protect create_data */
241 struct mutex mot_lov_mutex;
242 /* Lock to protect lease open.
243 * Lease open acquires write lock; normal open acquires read lock */
244 struct rw_semaphore mot_open_sem;
245 atomic_t mot_lease_count;
246 atomic_t mot_open_count;
249 struct mdt_lock_handle {
250 /* Lock type, reg for cross-ref use or pdo lock. */
254 struct lustre_handle mlh_reg_lh;
255 enum ldlm_mode mlh_reg_mode;
258 struct lustre_handle mlh_pdo_lh;
259 enum ldlm_mode mlh_pdo_mode;
260 unsigned int mlh_pdo_hash;
262 /* Remote regular lock */
263 struct lustre_handle mlh_rreg_lh;
264 enum ldlm_mode mlh_rreg_mode;
268 MDT_LH_PARENT, /* parent lockh */
269 MDT_LH_CHILD, /* child lockh */
270 MDT_LH_OLD, /* old lockh for rename */
271 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
272 MDT_LH_NEW, /* new lockh for rename */
273 MDT_LH_RMT, /* used for return lh to caller */
274 MDT_LH_LOCAL, /* local lock never return to client */
283 /* Special magical errno for communicaiton between mdt_reint_open()
284 * and mdt_intent_reint() which means return the lock to the client
285 * for subsequent cross ref open. Previously we used plain -EREMOTE
286 * but other functions called in that path might return it too and
287 * confuse us. This is not returned to the client. See LU-5370. */
288 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
290 struct mdt_reint_record {
291 mdt_reint_t rr_opcode;
292 const struct lustre_handle *rr_handle;
293 const struct lu_fid *rr_fid1;
294 const struct lu_fid *rr_fid2;
295 struct lu_name rr_name;
296 struct lu_name rr_tgt_name;
302 enum mdt_reint_flag {
303 MRF_OPEN_TRUNC = 1 << 0,
307 * Common data shared by mdt-level handlers. This is allocated per-thread to
308 * reduce stack consumption.
310 struct mdt_thread_info {
313 * The following members will be filled explicitly
314 * with specific data in mdt_thread_info_init().
316 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
317 * request handling may migrate from one server thread to another.
319 struct req_capsule *mti_pill;
321 /* although we have export in req, there are cases when it is not
322 * available, e.g. closing files upon export destroy */
323 struct obd_export *mti_exp;
325 * A couple of lock handles.
327 struct mdt_lock_handle mti_lh[MDT_LH_NR];
329 struct mdt_device *mti_mdt;
330 const struct lu_env *mti_env;
332 /* transaction number of current request */
338 * The following members will be filled expilictly
339 * with zero in mdt_thread_info_init(). These members may be used
346 struct md_attr mti_attr;
348 * Body for "habeo corpus" operations.
350 const struct mdt_body *mti_body;
352 * Host object. This is released at the end of mdt_handler().
354 struct mdt_object *mti_object;
356 * Lock request for "habeo clavis" operations.
358 const struct ldlm_request *mti_dlm_req;
360 __u32 mti_has_trans:1, /* has txn already? */
363 /* opdata for mdt_reint_open(), has the same as
364 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
365 * value onto disk for recovery when mdt_trans_stop_cb() is called.
371 * The following members will be filled explicitly
372 * with zero in mdt_reint_unpack(), because they are only used
373 * by reint requests (including mdt_reint_open()).
377 * reint record. contains information for reint operations.
379 struct mdt_reint_record mti_rr;
381 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
383 * Operation specification (currently create and lookup)
385 struct md_op_spec mti_spec;
389 * The following members will _NOT_ be initialized at all.
390 * DO NOT expect them to contain any valid value.
391 * They should be initialized explicitly by the user themselves.
394 /* XXX: If something is in a union, make sure they do not conflict */
395 struct lu_fid mti_tmp_fid1;
396 struct lu_fid mti_tmp_fid2;
397 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
398 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
400 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
401 char ns_name[48];/* for mdt_init0() */
402 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
403 struct obd_statfs osfs; /* for mdt_statfs() */
405 /* for mdt_readpage() */
406 struct lu_rdpg mti_rdpg;
407 /* for mdt_sendpage() */
408 struct l_wait_info mti_wait_info;
415 struct lustre_handle mti_close_handle;
417 struct lu_buf mti_buf;
418 struct lu_buf mti_big_buf;
420 /* Ops object filename */
421 struct lu_name mti_name;
422 /* per-thread values, can be re-used, may be vmalloc'd */
425 /* big_lmm buffer was used and must be used in reply */
426 int mti_big_lmm_used;
427 /* should be enough to fit lustre_mdt_attrs */
428 char mti_xattr_buf[128];
429 struct ldlm_enqueue_info mti_einfo;
430 struct tg_reply_data *mti_reply_data;
433 extern struct lu_context_key mdt_thread_key;
435 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
437 struct mdt_thread_info *mti;
439 lu_env_refill((void *)env);
440 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
445 struct cdt_req_progress {
446 struct mutex crp_lock; /**< protect tree */
447 struct interval_node *crp_root; /**< tree to track extent
449 struct interval_node **crp_node; /**< buffer for tree nodes
450 * vector of fixed size
452 int crp_cnt; /**< # of used nodes */
453 int crp_max; /**< # of allocated nodes */
456 struct cdt_agent_req {
457 struct list_head car_request_list; /**< to chain all the req. */
458 atomic_t car_refcount; /**< reference counter */
459 __u64 car_compound_id; /**< compound id */
460 __u64 car_flags; /**< request original flags */
461 struct obd_uuid car_uuid; /**< agent doing the req. */
462 __u32 car_archive_id; /**< archive id */
463 int car_canceled; /**< request was canceled */
464 cfs_time_t car_req_start; /**< start time */
465 cfs_time_t car_req_update; /**< last update time */
466 struct hsm_action_item *car_hai; /**< req. to the agent */
467 struct cdt_req_progress car_progress; /**< track data mvt
470 extern struct kmem_cache *mdt_hsm_car_kmem;
473 struct list_head ha_list; /**< to chain the agents */
474 struct obd_uuid ha_uuid; /**< agent uuid */
475 __u32 *ha_archive_id; /**< archive id */
476 int ha_archive_cnt; /**< number of archive entries
477 * 0 means any archive */
478 atomic_t ha_requests; /**< current request count */
479 atomic_t ha_success; /**< number of successful
481 atomic_t ha_failure; /**< number of failed actions */
484 struct cdt_restore_handle {
485 struct list_head crh_list; /**< to chain the handle */
486 struct lu_fid crh_fid; /**< fid of the object */
487 struct ldlm_extent crh_extent; /**< extent of the restore */
488 struct mdt_lock_handle crh_lh; /**< lock handle */
490 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
492 static inline const struct md_device_operations *
493 mdt_child_ops(struct mdt_device * m)
495 LASSERT(m->mdt_child);
496 return m->mdt_child->md_ops;
499 static inline struct md_object *mdt_object_child(struct mdt_object *o)
502 return lu2md(lu_object_next(&o->mot_obj));
505 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
507 return info->mti_pill ? info->mti_pill->rc_req : NULL;
510 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
512 LASSERT(info->mti_exp);
513 return exp_connect_flags(info->mti_exp);
516 static inline void mdt_object_get(const struct lu_env *env,
517 struct mdt_object *o)
520 lu_object_get(&o->mot_obj);
524 static inline void mdt_object_put(const struct lu_env *env,
525 struct mdt_object *o)
528 lu_object_put(env, &o->mot_obj);
532 static inline int mdt_object_exists(const struct mdt_object *o)
534 return lu_object_exists(&o->mot_obj);
537 static inline int mdt_object_remote(const struct mdt_object *o)
539 return lu_object_remote(&o->mot_obj);
542 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
544 return lu_object_fid(&o->mot_obj);
547 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
549 return mdt->mdt_lu_dev.ld_site;
552 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
554 return &mdt->mdt_seq_site;
557 static inline void mdt_export_evict(struct obd_export *exp)
559 class_fail_export(exp);
562 /* Here we use LVB_TYPE to check dne client, because it is
563 * also landed on 2.4. */
564 static inline bool mdt_is_dne_client(struct obd_export *exp)
566 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
569 static inline bool mdt_is_striped_client(struct obd_export *exp)
571 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
574 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
575 void mdt_set_disposition(struct mdt_thread_info *info,
576 struct ldlm_reply *rep, __u64 op_flag);
577 void mdt_clear_disposition(struct mdt_thread_info *info,
578 struct ldlm_reply *rep, __u64 op_flag);
580 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
581 const struct lu_name *lname);
583 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
585 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
586 struct mdt_lock_handle *lh);
588 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
589 struct mdt_lock_handle *lhc);
591 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *mo,
592 struct mdt_lock_handle *lh, __u64 ibits);
594 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
595 struct mdt_lock_handle *lh, __u64 ibits,
598 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
599 struct mdt_lock_handle *lh, __u64 ibits);
601 int mdt_reint_object_lock_try(struct mdt_thread_info *info,
602 struct mdt_object *o, struct mdt_lock_handle *lh,
603 __u64 ibits, bool cos_incompat);
605 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
606 struct mdt_lock_handle *lh, int decref);
608 struct mdt_object *mdt_object_new(const struct lu_env *env,
610 const struct lu_fid *);
611 struct mdt_object *mdt_object_find(const struct lu_env *,
613 const struct lu_fid *);
614 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
615 const struct lu_fid *,
616 struct mdt_lock_handle *,
618 void mdt_object_unlock_put(struct mdt_thread_info *,
620 struct mdt_lock_handle *,
623 void mdt_client_compatibility(struct mdt_thread_info *info);
625 int mdt_remote_object_lock(struct mdt_thread_info *mti,
626 struct mdt_object *o, const struct lu_fid *fid,
627 struct lustre_handle *lh,
628 enum ldlm_mode mode, __u64 ibits, bool nonblock,
631 enum mdt_name_flags {
635 int mdt_name_unpack(struct req_capsule *pill,
636 const struct req_msg_field *field,
638 enum mdt_name_flags flags);
639 int mdt_close_unpack(struct mdt_thread_info *info);
640 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
641 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
642 #ifdef CONFIG_FS_POSIX_ACL
643 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
644 struct mdt_object *o, struct lu_nodemap *nodemap);
646 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
647 const struct lu_attr *attr, const struct lu_fid *fid);
649 int mdt_getxattr(struct mdt_thread_info *info);
650 int mdt_reint_setxattr(struct mdt_thread_info *info,
651 struct mdt_lock_handle *lh);
653 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
654 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
656 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
657 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
658 struct mdt_lock_handle *lhc);
660 extern void target_recovery_fini(struct obd_device *obd);
661 extern void target_recovery_init(struct lu_target *lut,
662 svc_handler_t handler);
663 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
664 struct obd_device *, struct lustre_sb_info *lsi);
665 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
667 int mdt_export_stats_init(struct obd_device *obd,
668 struct obd_export *exp,
671 int mdt_lock_new_child(struct mdt_thread_info *info,
672 struct mdt_object *o,
673 struct mdt_lock_handle *child_lockh);
675 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
678 int mdt_reint_open(struct mdt_thread_info *info,
679 struct mdt_lock_handle *lhc);
681 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
682 const struct lustre_handle *handle,
685 int mdt_get_info(struct tgt_session_info *tsi);
686 int mdt_attr_get_complex(struct mdt_thread_info *info,
687 struct mdt_object *o, struct md_attr *ma);
688 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
690 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
691 struct md_attr *ma, const char *name);
692 int mdt_write_get(struct mdt_object *o);
693 void mdt_write_put(struct mdt_object *o);
694 int mdt_write_read(struct mdt_object *o);
695 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
696 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
697 void mdt_mfd_free(struct mdt_file_data *mfd);
698 int mdt_close(struct tgt_session_info *tsi);
699 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
701 int mdt_fix_reply(struct mdt_thread_info *info);
702 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
703 const struct md_attr *);
704 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
706 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
707 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
708 const void *area, ssize_t len);
710 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
711 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
713 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
714 int mdt_check_ucred(struct mdt_thread_info *);
715 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
716 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *, struct mdt_body *);
717 int mdt_init_ucred_reint(struct mdt_thread_info *);
718 void mdt_exit_ucred(struct mdt_thread_info *);
719 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
720 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
721 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
723 void mdt_thread_info_init(struct ptlrpc_request *req,
724 struct mdt_thread_info *mti);
725 void mdt_thread_info_fini(struct mdt_thread_info *mti);
726 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
728 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
729 const struct md_hsm *mh);
731 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
732 void *data, int flag);
733 int mdt_links_read(struct mdt_thread_info *info,
734 struct mdt_object *mdt_obj,
735 struct linkea_data *ldata);
736 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
737 struct mdt_body *repbody);
739 static inline struct mdt_device *mdt_dev(struct lu_device *d)
741 return container_of0(d, struct mdt_device, mdt_lu_dev);
744 static inline struct mdt_object *mdt_obj(struct lu_object *o)
746 return container_of0(o, struct mdt_object, mot_obj);
749 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
751 struct lu_object *lo;
752 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
754 lo = lu_object_locate(mo->mot_obj.lo_header,
755 mdt->mdt_bottom->dd_lu_dev.ld_type);
760 static inline bool agent_req_in_final_state(enum agent_req_status ars)
762 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
763 (ars == ARS_CANCELED));
766 /* mdt/mdt_identity.c */
767 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
769 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
771 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
773 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
775 void mdt_flush_identity(struct upcall_cache *, int);
777 __u32 mdt_identity_get_perm(struct md_identity *, lnet_nid_t);
779 /* mdt/mdt_recovery.c */
780 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
783 int mdt_hsm_state_get(struct tgt_session_info *tsi);
784 int mdt_hsm_state_set(struct tgt_session_info *tsi);
785 int mdt_hsm_action(struct tgt_session_info *tsi);
786 int mdt_hsm_progress(struct tgt_session_info *tsi);
787 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
788 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
789 int mdt_hsm_request(struct tgt_session_info *tsi);
791 /* mdt/mdt_hsm_cdt_actions.c */
792 extern const struct file_operations mdt_hsm_actions_fops;
793 void dump_llog_agent_req_rec(const char *prefix,
794 const struct llog_agent_req_rec *larr);
795 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
796 llog_cb_t cb, void *data);
797 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
798 __u64 compound_id, __u32 archive_id,
799 __u64 flags, struct hsm_action_item *hai);
800 int mdt_agent_record_update(const struct lu_env *env,
801 struct mdt_device *mdt, __u64 *cookies,
802 int cookies_count, enum agent_req_status status);
804 /* mdt/mdt_hsm_cdt_agent.c */
805 extern const struct file_operations mdt_hsm_agent_fops;
806 int mdt_hsm_agent_register(struct mdt_thread_info *info,
807 const struct obd_uuid *uuid,
808 int nr_archives, __u32 *archive_num);
809 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
810 const struct obd_uuid *uuid,
812 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
813 const struct obd_uuid *uuid);
814 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
815 int succ_rq, int fail_rq, int new_rq,
816 const struct obd_uuid *uuid);
817 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
818 struct obd_uuid *uuid);
819 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
821 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
822 struct hsm_progress_kernel *pgs);
823 /* mdt/mdt_hsm_cdt_client.c */
824 int mdt_hsm_add_actions(struct mdt_thread_info *info,
825 struct hsm_action_list *hal);
826 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
827 struct hsm_action_list *hal);
828 int mdt_hsm_get_running(struct mdt_thread_info *mti,
829 struct hsm_action_list *hal);
830 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
831 const struct lu_fid *fid);
832 /* mdt/mdt_hsm_cdt_requests.c */
833 extern const struct file_operations mdt_hsm_active_requests_fops;
834 void dump_requests(char *prefix, struct coordinator *cdt);
835 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
836 __u64 flags, struct obd_uuid *uuid,
837 struct hsm_action_item *hai);
838 void mdt_cdt_free_request(struct cdt_agent_req *car);
839 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
840 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
842 const struct lu_fid *fid);
843 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
844 void mdt_cdt_get_request(struct cdt_agent_req *car);
845 void mdt_cdt_put_request(struct cdt_agent_req *car);
846 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
847 const struct hsm_progress_kernel *pgs);
848 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
849 /* mdt/mdt_coordinator.c */
850 void mdt_hsm_dump_hal(int level, const char *prefix,
851 struct hsm_action_list *hal);
852 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
853 const struct lu_fid *fid);
854 /* coordinator management */
855 int mdt_hsm_cdt_init(struct mdt_device *mdt);
856 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
857 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
858 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
860 /* coordinator control /proc interface */
861 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file,
862 const char __user *buffer,
863 size_t count, loff_t *off);
864 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
865 int hsm_cdt_procfs_init(struct mdt_device *mdt);
866 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
867 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
869 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
870 const struct lu_fid *fid,
872 /* actions/request helpers */
873 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
874 struct hsm_action_list *hal, struct obd_uuid *uuid);
875 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
876 const int hal_an, const __u64 rq_flags,
877 const struct md_hsm *hsm);
878 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
879 struct hsm_progress_kernel *pgs,
880 const int update_record);
882 int mdt_close_swap_layouts(struct mdt_thread_info *info,
883 struct mdt_object *o, struct md_attr *ma);
885 extern struct lu_context_key mdt_thread_key;
887 /* debug issues helper starts here*/
888 static inline int mdt_fail_write(const struct lu_env *env,
889 struct dt_device *dd, int id)
891 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
892 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
894 return dt_ro(env, dd);
895 /* We set FAIL_ONCE because we never "un-fail" a device */
901 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
903 return &req->rq_export->exp_mdt_data;
906 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
908 return mdt_dev(exp->exp_obd->obd_lu_dev);
911 static inline bool mdt_rdonly(struct obd_export *exp)
913 if (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
914 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly)
919 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
920 struct mdt_lock_handle *lhc);
921 static inline int mdt_check_resent(struct mdt_thread_info *info,
922 mdt_reconstruct_t reconstruct,
923 struct mdt_lock_handle *lhc)
925 struct ptlrpc_request *req = mdt_info_req(info);
929 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
930 OBD_ALLOC_PTR(info->mti_reply_data);
931 if (info->mti_reply_data == NULL)
934 if (req_can_reconstruct(req, info->mti_reply_data)) {
935 reconstruct(info, lhc);
939 "no reply data found for RESENT req");
942 OBD_FREE_PTR(info->mti_reply_data);
943 info->mti_reply_data = NULL;
948 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
949 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
951 static inline int is_identity_get_disabled(struct upcall_cache *cache)
953 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
956 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
958 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
959 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
960 struct lustre_handle *lh, enum ldlm_mode mode,
961 union ldlm_policy_data *policy,
962 const struct ldlm_res_id *res_id,
963 __u64 flags, const __u64 *client_cookie)
970 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
971 mode, &flags, mdt_blocking_ast,
972 ldlm_completion_ast, NULL, NULL, 0,
973 LVB_T_NONE, client_cookie, lh);
974 return rc == ELDLM_OK ? 0 : -EIO;
977 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
979 ldlm_lock_decref(lh, mode);
982 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
984 return mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL;
987 extern mdl_mode_t mdt_mdl_lock_modes[];
988 extern enum ldlm_mode mdt_dlm_lock_modes[];
990 /* LCK_MINMODE which is zero returns false for is_power_of_2 */
992 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
994 LASSERT(mode == LCK_MINMODE || is_power_of_2(mode));
995 return mdt_mdl_lock_modes[mode];
998 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1000 LASSERT(mode == MDL_MINMODE || is_power_of_2(mode));
1001 return mdt_dlm_lock_modes[mode];
1005 extern struct ldlm_valblock_ops mdt_lvbo;
1007 void mdt_enable_cos(struct mdt_device *, int);
1008 int mdt_cos_is_enabled(struct mdt_device *);
1026 LPROC_MDT_SAMEDIR_RENAME,
1027 LPROC_MDT_CROSSDIR_RENAME,
1030 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1031 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1032 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1033 void mdt_procfs_fini(struct mdt_device *mdt);
1035 /* lustre/mdt_mdt_lproc.c */
1036 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1038 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1039 struct mdt_device *mdt,
1040 struct ptlrpc_request *req,
1041 struct mdt_object *src, struct mdt_object *tgt);
1043 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1045 return mdt->mdt_lu_dev.ld_obd;
1048 extern const struct lu_device_operations mdt_lu_ops;
1050 static inline char *mdt_obd_name(struct mdt_device *mdt)
1052 return mdt->mdt_lu_dev.ld_obd->obd_name;
1055 int mds_mod_init(void);
1056 void mds_mod_exit(void);
1058 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1060 struct obd_export *exp = req->rq_export;
1063 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1064 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1069 #endif /* _MDT_INTERNAL_H */