4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_internal.h
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
48 #ifndef _MDT_INTERNAL_H
49 #define _MDT_INTERNAL_H
52 #include <upcall_cache.h>
53 #include <lustre_net.h>
54 #include <lustre/lustre_idl.h>
55 #include <obd_class.h>
56 #include <lustre_disk.h>
57 #include <lu_target.h>
58 #include <md_object.h>
59 #include <lustre_fid.h>
60 #include <lustre_fld.h>
61 #include <lustre_req_layout.h>
62 #include <lustre_sec.h>
63 #include <lustre_idmap.h>
64 #include <lustre_eacl.h>
65 #include <lustre_quota.h>
66 #include <lustre_linkea.h>
70 /* file data for open files on MDS */
71 struct mdt_file_data {
72 /** portals handle must be first */
73 struct portals_handle mfd_handle;
74 /** open mode provided by client */
76 /** protected by med_open_lock */
77 struct list_head mfd_list;
78 /** xid of the open request */
80 /** old handle in replay case */
81 struct lustre_handle mfd_old_handle;
82 /** point to opened object */
83 struct mdt_object *mfd_object;
86 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
87 #define CDT_NORETRY_ACTION (1ULL << 1)
88 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
89 #define CDT_POLICY_SHIFT_COUNT 2
90 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
93 /* when adding a new policy, do not forget to update
94 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
96 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
98 enum cdt_states { CDT_STOPPED = 0,
104 /* when multiple lock are needed, the lock order is
112 struct ptlrpc_thread cdt_thread; /**< coordinator thread */
113 struct lu_env cdt_env; /**< coordinator lustre
115 struct lu_context cdt_session; /** session for lu_ucred */
116 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
117 __u64 cdt_policy; /**< policy flags */
118 enum cdt_states cdt_state; /**< state */
119 atomic_t cdt_compound_id; /**< compound id
121 __u64 cdt_last_cookie; /**< last cookie
123 struct mutex cdt_llog_lock; /**< protect llog
125 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
126 struct rw_semaphore cdt_request_lock; /**< protect request
128 struct mutex cdt_restore_lock; /**< protect restore
130 cfs_time_t cdt_loop_period; /**< llog scan period */
131 cfs_time_t cdt_grace_delay; /**< request grace
133 cfs_time_t cdt_active_req_timeout; /**< request timeout */
134 __u32 cdt_default_archive_id; /**< archive id used
137 __u64 cdt_max_requests; /**< max count of started
139 atomic_t cdt_request_count; /**< current count of
140 * started requests */
141 struct list_head cdt_requests; /**< list of started
143 struct list_head cdt_agents; /**< list of register
145 struct list_head cdt_restore_hdl; /**< list of restore lock
147 /* Bitmasks indexed by the HSMA_XXX constants. */
148 __u64 cdt_user_request_mask;
149 __u64 cdt_group_request_mask;
150 __u64 cdt_other_request_mask;
153 /* mdt state flag bits */
154 #define MDT_FL_CFGLOG 0
155 #define MDT_FL_SYNCED 1
159 struct lu_device mdt_lu_dev;
160 struct seq_server_site mdt_seq_site;
161 /* DLM name-space for meta-data locks maintained by this server */
162 struct ldlm_namespace *mdt_namespace;
163 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
164 struct ptlrpc_client *mdt_ldlm_client;
165 /* underlying device */
166 struct obd_export *mdt_child_exp;
167 struct md_device *mdt_child;
168 struct dt_device *mdt_bottom;
169 struct obd_export *mdt_bottom_exp;
171 struct lu_target mdt_lut;
173 * Options bit-fields.
176 unsigned int mo_user_xattr:1,
181 /* mdt state flags */
182 unsigned long mdt_state;
184 /* transaction callbacks */
185 struct dt_txn_callback mdt_txn_cb;
187 /* these values should be updated from lov if necessary.
188 * or should be placed somewhere else. */
193 struct upcall_cache *mdt_identity_cache;
195 unsigned int mdt_capa_conf:1,
196 /* Enable remote dir on non-MDT0 */
197 mdt_enable_remote_dir:1;
199 gid_t mdt_enable_remote_dir_gid;
200 /* statfs optimization: we cache a bit */
201 struct obd_statfs mdt_osfs;
203 spinlock_t mdt_osfs_lock;
206 struct root_squash_info mdt_squash;
208 struct rename_stats mdt_rename_stats;
209 struct lu_fid mdt_md_root_fid;
211 /* connection to quota master */
212 struct obd_export *mdt_qmt_exp;
213 /* quota master device associated with this MDT */
214 struct lu_device *mdt_qmt_dev;
216 struct coordinator mdt_coordinator;
219 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
220 #define MDT_COS_DEFAULT (0)
222 enum mdt_object_flags {
223 /** lov object has been created. */
224 MOF_LOV_CREATED = 1 << 0,
228 struct lu_object_header mot_header;
229 struct lu_object mot_obj;
230 enum mdt_object_flags mot_flags;
232 spinlock_t mot_write_lock;
233 /* Lock to protect create_data */
234 struct mutex mot_lov_mutex;
235 /* Lock to protect lease open.
236 * Lease open acquires write lock; normal open acquires read lock */
237 struct rw_semaphore mot_open_sem;
238 atomic_t mot_lease_count;
239 atomic_t mot_open_count;
242 struct mdt_lock_handle {
243 /* Lock type, reg for cross-ref use or pdo lock. */
247 struct lustre_handle mlh_reg_lh;
248 enum ldlm_mode mlh_reg_mode;
251 struct lustre_handle mlh_pdo_lh;
252 enum ldlm_mode mlh_pdo_mode;
253 unsigned int mlh_pdo_hash;
255 /* Remote regular lock */
256 struct lustre_handle mlh_rreg_lh;
257 enum ldlm_mode mlh_rreg_mode;
261 MDT_LH_PARENT, /* parent lockh */
262 MDT_LH_CHILD, /* child lockh */
263 MDT_LH_OLD, /* old lockh for rename */
264 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
265 MDT_LH_NEW, /* new lockh for rename */
266 MDT_LH_RMT, /* used for return lh to caller */
267 MDT_LH_LOCAL, /* local lock never return to client */
276 /* Special magical errno for communicaiton between mdt_reint_open()
277 * and mdt_intent_reint() which means return the lock to the client
278 * for subsequent cross ref open. Previously we used plain -EREMOTE
279 * but other functions called in that path might return it too and
280 * confuse us. This is not returned to the client. See LU-5370. */
281 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
283 struct mdt_reint_record {
284 mdt_reint_t rr_opcode;
285 const struct lustre_handle *rr_handle;
286 const struct lu_fid *rr_fid1;
287 const struct lu_fid *rr_fid2;
288 struct lu_name rr_name;
289 struct lu_name rr_tgt_name;
295 enum mdt_reint_flag {
296 MRF_OPEN_TRUNC = 1 << 0,
300 * Common data shared by mdt-level handlers. This is allocated per-thread to
301 * reduce stack consumption.
303 struct mdt_thread_info {
306 * The following members will be filled explicitly
307 * with specific data in mdt_thread_info_init().
309 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
310 * request handling may migrate from one server thread to another.
312 struct req_capsule *mti_pill;
314 /* although we have export in req, there are cases when it is not
315 * available, e.g. closing files upon export destroy */
316 struct obd_export *mti_exp;
318 * A couple of lock handles.
320 struct mdt_lock_handle mti_lh[MDT_LH_NR];
322 struct mdt_device *mti_mdt;
323 const struct lu_env *mti_env;
325 /* transaction number of current request */
331 * The following members will be filled expilictly
332 * with zero in mdt_thread_info_init(). These members may be used
339 struct md_attr mti_attr;
341 * Body for "habeo corpus" operations.
343 const struct mdt_body *mti_body;
345 * Host object. This is released at the end of mdt_handler().
347 struct mdt_object *mti_object;
349 * Lock request for "habeo clavis" operations.
351 const struct ldlm_request *mti_dlm_req;
353 __u32 mti_has_trans:1, /* has txn already? */
356 /* opdata for mdt_reint_open(), has the same as
357 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
358 * value onto disk for recovery when mdt_trans_stop_cb() is called.
364 * The following members will be filled explicitly
365 * with zero in mdt_reint_unpack(), because they are only used
366 * by reint requests (including mdt_reint_open()).
370 * reint record. contains information for reint operations.
372 struct mdt_reint_record mti_rr;
374 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
376 * Operation specification (currently create and lookup)
378 struct md_op_spec mti_spec;
382 * The following members will _NOT_ be initialized at all.
383 * DO NOT expect them to contain any valid value.
384 * They should be initialized explicitly by the user themselves.
387 /* XXX: If something is in a union, make sure they do not conflict */
388 struct lu_fid mti_tmp_fid1;
389 struct lu_fid mti_tmp_fid2;
390 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
391 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
393 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
394 char ns_name[48];/* for mdt_init0() */
395 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
396 struct obd_statfs osfs; /* for mdt_statfs() */
398 /* for mdt_readpage() */
399 struct lu_rdpg mti_rdpg;
400 /* for mdt_sendpage() */
401 struct l_wait_info mti_wait_info;
408 struct lustre_handle mti_close_handle;
410 struct lu_buf mti_buf;
411 struct lu_buf mti_big_buf;
413 /* Ops object filename */
414 struct lu_name mti_name;
415 /* per-thread values, can be re-used */
418 /* big_lmm buffer was used and must be used in reply */
419 int mti_big_lmm_used;
420 /* should be enough to fit lustre_mdt_attrs */
421 char mti_xattr_buf[128];
422 struct ldlm_enqueue_info mti_einfo;
423 struct tg_reply_data *mti_reply_data;
426 extern struct lu_context_key mdt_thread_key;
428 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
430 struct mdt_thread_info *mti;
432 lu_env_refill((void *)env);
433 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
438 struct cdt_req_progress {
439 struct mutex crp_lock; /**< protect tree */
440 struct interval_node *crp_root; /**< tree to track extent
442 struct interval_node **crp_node; /**< buffer for tree nodes
443 * vector of fixed size
445 int crp_cnt; /**< # of used nodes */
446 int crp_max; /**< # of allocated nodes */
449 struct cdt_agent_req {
450 struct list_head car_request_list; /**< to chain all the req. */
451 atomic_t car_refcount; /**< reference counter */
452 __u64 car_compound_id; /**< compound id */
453 __u64 car_flags; /**< request original flags */
454 struct obd_uuid car_uuid; /**< agent doing the req. */
455 __u32 car_archive_id; /**< archive id */
456 int car_canceled; /**< request was canceled */
457 cfs_time_t car_req_start; /**< start time */
458 cfs_time_t car_req_update; /**< last update time */
459 struct hsm_action_item *car_hai; /**< req. to the agent */
460 struct cdt_req_progress car_progress; /**< track data mvt
463 extern struct kmem_cache *mdt_hsm_car_kmem;
466 struct list_head ha_list; /**< to chain the agents */
467 struct obd_uuid ha_uuid; /**< agent uuid */
468 __u32 *ha_archive_id; /**< archive id */
469 int ha_archive_cnt; /**< number of archive entries
470 * 0 means any archive */
471 atomic_t ha_requests; /**< current request count */
472 atomic_t ha_success; /**< number of successful
474 atomic_t ha_failure; /**< number of failed actions */
477 struct cdt_restore_handle {
478 struct list_head crh_list; /**< to chain the handle */
479 struct lu_fid crh_fid; /**< fid of the object */
480 struct ldlm_extent crh_extent; /**< extent of the restore */
481 struct mdt_lock_handle crh_lh; /**< lock handle */
483 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
485 static inline const struct md_device_operations *
486 mdt_child_ops(struct mdt_device * m)
488 LASSERT(m->mdt_child);
489 return m->mdt_child->md_ops;
492 static inline struct md_object *mdt_object_child(struct mdt_object *o)
495 return lu2md(lu_object_next(&o->mot_obj));
498 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
500 return info->mti_pill ? info->mti_pill->rc_req : NULL;
503 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
505 LASSERT(info->mti_exp);
506 return exp_connect_flags(info->mti_exp);
509 static inline void mdt_object_get(const struct lu_env *env,
510 struct mdt_object *o)
513 lu_object_get(&o->mot_obj);
517 static inline void mdt_object_put(const struct lu_env *env,
518 struct mdt_object *o)
521 lu_object_put(env, &o->mot_obj);
525 static inline int mdt_object_exists(const struct mdt_object *o)
527 return lu_object_exists(&o->mot_obj);
530 static inline int mdt_object_remote(const struct mdt_object *o)
532 return lu_object_remote(&o->mot_obj);
535 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
537 return lu_object_fid(&o->mot_obj);
540 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
542 return mdt->mdt_lu_dev.ld_site;
545 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
547 return &mdt->mdt_seq_site;
550 static inline void mdt_export_evict(struct obd_export *exp)
552 class_fail_export(exp);
555 /* Here we use LVB_TYPE to check dne client, because it is
556 * also landed on 2.4. */
557 static inline bool mdt_is_dne_client(struct obd_export *exp)
559 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
562 static inline bool mdt_is_striped_client(struct obd_export *exp)
564 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
567 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
568 void mdt_set_disposition(struct mdt_thread_info *info,
569 struct ldlm_reply *rep, __u64 op_flag);
570 void mdt_clear_disposition(struct mdt_thread_info *info,
571 struct ldlm_reply *rep, __u64 op_flag);
573 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
574 const struct lu_name *lname);
576 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
578 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
579 struct mdt_lock_handle *lh);
581 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
582 struct mdt_lock_handle *lhc);
584 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *mo,
585 struct mdt_lock_handle *lh, __u64 ibits);
587 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
588 struct mdt_lock_handle *lh, __u64 ibits);
590 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
591 struct mdt_lock_handle *lh, int decref);
593 struct mdt_object *mdt_object_new(const struct lu_env *env,
595 const struct lu_fid *);
596 struct mdt_object *mdt_object_find(const struct lu_env *,
598 const struct lu_fid *);
599 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
600 const struct lu_fid *,
601 struct mdt_lock_handle *,
603 void mdt_object_unlock_put(struct mdt_thread_info *,
605 struct mdt_lock_handle *,
608 void mdt_client_compatibility(struct mdt_thread_info *info);
610 int mdt_remote_object_lock(struct mdt_thread_info *mti,
611 struct mdt_object *o, const struct lu_fid *fid,
612 struct lustre_handle *lh,
613 enum ldlm_mode mode, __u64 ibits, bool nonblock);
615 enum mdt_name_flags {
619 int mdt_name_unpack(struct req_capsule *pill,
620 const struct req_msg_field *field,
622 enum mdt_name_flags flags);
623 int mdt_close_unpack(struct mdt_thread_info *info);
624 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
625 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
626 #ifdef CONFIG_FS_POSIX_ACL
627 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
628 struct mdt_object *o, struct lu_nodemap *nodemap);
630 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
631 const struct lu_attr *attr, const struct lu_fid *fid);
633 int mdt_getxattr(struct mdt_thread_info *info);
634 int mdt_reint_setxattr(struct mdt_thread_info *info,
635 struct mdt_lock_handle *lh);
637 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
638 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
640 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
641 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
642 struct mdt_lock_handle *lhc);
644 extern void target_recovery_fini(struct obd_device *obd);
645 extern void target_recovery_init(struct lu_target *lut,
646 svc_handler_t handler);
647 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
648 struct obd_device *, struct lustre_sb_info *lsi);
649 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
651 int mdt_export_stats_init(struct obd_device *obd,
652 struct obd_export *exp,
655 int mdt_lock_new_child(struct mdt_thread_info *info,
656 struct mdt_object *o,
657 struct mdt_lock_handle *child_lockh);
659 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
662 int mdt_reint_open(struct mdt_thread_info *info,
663 struct mdt_lock_handle *lhc);
665 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
666 const struct lustre_handle *handle,
669 int mdt_get_info(struct tgt_session_info *tsi);
670 int mdt_attr_get_complex(struct mdt_thread_info *info,
671 struct mdt_object *o, struct md_attr *ma);
672 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
674 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
675 struct md_attr *ma, const char *name);
676 int mdt_write_get(struct mdt_object *o);
677 void mdt_write_put(struct mdt_object *o);
678 int mdt_write_read(struct mdt_object *o);
679 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
680 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
681 void mdt_mfd_free(struct mdt_file_data *mfd);
682 int mdt_close(struct tgt_session_info *tsi);
683 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
685 int mdt_fix_reply(struct mdt_thread_info *info);
686 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
687 const struct md_attr *);
688 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
690 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
691 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
692 const void *area, ssize_t len);
694 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
695 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
697 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
698 int mdt_check_ucred(struct mdt_thread_info *);
699 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
700 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *, struct mdt_body *);
701 int mdt_init_ucred_reint(struct mdt_thread_info *);
702 void mdt_exit_ucred(struct mdt_thread_info *);
703 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
704 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
705 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
707 void mdt_thread_info_init(struct ptlrpc_request *req,
708 struct mdt_thread_info *mti);
709 void mdt_thread_info_fini(struct mdt_thread_info *mti);
710 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
712 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
713 const struct md_hsm *mh);
715 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
716 void *data, int flag);
717 int mdt_links_read(struct mdt_thread_info *info,
718 struct mdt_object *mdt_obj,
719 struct linkea_data *ldata);
720 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
721 struct mdt_body *repbody);
723 int mdt_init_idmap(struct tgt_session_info *tsi);
724 void mdt_cleanup_idmap(struct mdt_export_data *);
725 int mdt_handle_idmap(struct tgt_session_info *tsi);
726 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
727 struct ptlrpc_user_desc *);
728 void mdt_body_reverse_idmap(struct mdt_thread_info *,
730 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
731 struct mdt_remote_perm *);
732 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
734 static inline struct mdt_device *mdt_dev(struct lu_device *d)
736 return container_of0(d, struct mdt_device, mdt_lu_dev);
739 static inline struct mdt_object *mdt_obj(struct lu_object *o)
741 return container_of0(o, struct mdt_object, mot_obj);
744 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
746 struct lu_object *lo;
747 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
749 lo = lu_object_locate(mo->mot_obj.lo_header,
750 mdt->mdt_bottom->dd_lu_dev.ld_type);
755 /* mdt/mdt_identity.c */
756 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
758 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
760 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
762 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
764 void mdt_flush_identity(struct upcall_cache *, int);
766 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
768 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
770 /* mdt/mdt_recovery.c */
771 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
774 int mdt_hsm_state_get(struct tgt_session_info *tsi);
775 int mdt_hsm_state_set(struct tgt_session_info *tsi);
776 int mdt_hsm_action(struct tgt_session_info *tsi);
777 int mdt_hsm_progress(struct tgt_session_info *tsi);
778 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
779 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
780 int mdt_hsm_request(struct tgt_session_info *tsi);
782 /* mdt/mdt_hsm_cdt_actions.c */
783 extern const struct file_operations mdt_hsm_actions_fops;
784 void dump_llog_agent_req_rec(const char *prefix,
785 const struct llog_agent_req_rec *larr);
786 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
787 llog_cb_t cb, void *data);
788 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
789 __u64 compound_id, __u32 archive_id,
790 __u64 flags, struct hsm_action_item *hai);
791 int mdt_agent_record_update(const struct lu_env *env,
792 struct mdt_device *mdt, __u64 *cookies,
793 int cookies_count, enum agent_req_status status);
794 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
795 struct llog_handle *llh,
796 struct llog_agent_req_rec *larr);
798 /* mdt/mdt_hsm_cdt_agent.c */
799 extern const struct file_operations mdt_hsm_agent_fops;
800 int mdt_hsm_agent_register(struct mdt_thread_info *info,
801 const struct obd_uuid *uuid,
802 int nr_archives, __u32 *archive_num);
803 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
804 const struct obd_uuid *uuid,
806 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
807 const struct obd_uuid *uuid);
808 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
809 int succ_rq, int fail_rq, int new_rq,
810 const struct obd_uuid *uuid);
811 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
812 struct obd_uuid *uuid);
813 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
815 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
816 struct hsm_progress_kernel *pgs);
817 /* mdt/mdt_hsm_cdt_client.c */
818 int mdt_hsm_add_actions(struct mdt_thread_info *info,
819 struct hsm_action_list *hal, __u64 *compound_id);
820 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
821 struct hsm_action_list *hal);
822 int mdt_hsm_get_running(struct mdt_thread_info *mti,
823 struct hsm_action_list *hal);
824 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
825 const struct lu_fid *fid);
826 /* mdt/mdt_hsm_cdt_requests.c */
827 extern const struct file_operations mdt_hsm_active_requests_fops;
828 void dump_requests(char *prefix, struct coordinator *cdt);
829 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
830 __u64 flags, struct obd_uuid *uuid,
831 struct hsm_action_item *hai);
832 void mdt_cdt_free_request(struct cdt_agent_req *car);
833 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
834 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
836 const struct lu_fid *fid);
837 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
838 void mdt_cdt_get_request(struct cdt_agent_req *car);
839 void mdt_cdt_put_request(struct cdt_agent_req *car);
840 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
841 const struct hsm_progress_kernel *pgs);
842 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
843 /* mdt/mdt_coordinator.c */
844 void mdt_hsm_dump_hal(int level, const char *prefix,
845 struct hsm_action_list *hal);
846 /* coordinator management */
847 int mdt_hsm_cdt_init(struct mdt_device *mdt);
848 int mdt_hsm_cdt_start(struct mdt_device *mdt);
849 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
850 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
851 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
853 /* coordinator control /proc interface */
854 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file, const char *buffer,
855 size_t count, loff_t *off);
856 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
857 int hsm_cdt_procfs_init(struct mdt_device *mdt);
858 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
859 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
861 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
862 const struct lu_fid *fid,
864 /* actions/request helpers */
865 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
866 struct hsm_action_list *hal, struct obd_uuid *uuid);
867 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
868 const int hal_an, const __u64 rq_flags,
869 const struct md_hsm *hsm);
870 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
871 struct hsm_progress_kernel *pgs,
872 const int update_record);
874 int mdt_close_swap_layouts(struct mdt_thread_info *info,
875 struct mdt_object *o, struct md_attr *ma);
877 extern struct lu_context_key mdt_thread_key;
879 /* debug issues helper starts here*/
880 static inline int mdt_fail_write(const struct lu_env *env,
881 struct dt_device *dd, int id)
883 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
884 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
886 return dd->dd_ops->dt_ro(env, dd);
887 /* We set FAIL_ONCE because we never "un-fail" a device */
893 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
895 return &req->rq_export->exp_mdt_data;
898 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
900 return mdt_dev(exp->exp_obd->obd_lu_dev);
903 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
904 struct mdt_lock_handle *lhc);
905 static inline int mdt_check_resent(struct mdt_thread_info *info,
906 mdt_reconstruct_t reconstruct,
907 struct mdt_lock_handle *lhc)
909 struct ptlrpc_request *req = mdt_info_req(info);
913 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
914 OBD_ALLOC_PTR(info->mti_reply_data);
915 if (info->mti_reply_data == NULL)
918 if (req_can_reconstruct(req, info->mti_reply_data)) {
919 reconstruct(info, lhc);
923 "no reply data found for RESENT req");
926 OBD_FREE_PTR(info->mti_reply_data);
927 info->mti_reply_data = NULL;
932 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
933 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
935 static inline int is_identity_get_disabled(struct upcall_cache *cache)
937 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
940 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
942 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
943 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
944 struct lustre_handle *lh, enum ldlm_mode mode,
945 union ldlm_policy_data *policy,
946 const struct ldlm_res_id *res_id,
947 __u64 flags, const __u64 *client_cookie)
954 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
955 mode, &flags, mdt_blocking_ast,
956 ldlm_completion_ast, NULL, NULL, 0,
957 LVB_T_NONE, client_cookie, lh);
958 return rc == ELDLM_OK ? 0 : -EIO;
961 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
963 ldlm_lock_decref(lh, mode);
966 extern mdl_mode_t mdt_mdl_lock_modes[];
967 extern enum ldlm_mode mdt_dlm_lock_modes[];
969 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
971 LASSERT(IS_PO2(mode));
972 return mdt_mdl_lock_modes[mode];
975 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
977 LASSERT(IS_PO2(mode));
978 return mdt_dlm_lock_modes[mode];
982 extern struct ldlm_valblock_ops mdt_lvbo;
984 void mdt_enable_cos(struct mdt_device *, int);
985 int mdt_cos_is_enabled(struct mdt_device *);
1003 LPROC_MDT_SAMEDIR_RENAME,
1004 LPROC_MDT_CROSSDIR_RENAME,
1007 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1008 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1009 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1010 void mdt_procfs_fini(struct mdt_device *mdt);
1012 /* lustre/mdt_mdt_lproc.c */
1013 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1015 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1016 struct mdt_device *mdt,
1017 struct ptlrpc_request *req,
1018 struct mdt_object *src, struct mdt_object *tgt);
1020 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1022 return mdt->mdt_lu_dev.ld_obd;
1025 extern const struct lu_device_operations mdt_lu_ops;
1027 static inline char *mdt_obd_name(struct mdt_device *mdt)
1029 return mdt->mdt_lu_dev.ld_obd->obd_name;
1032 int mds_mod_init(void);
1033 void mds_mod_exit(void);
1035 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1037 struct obd_export *exp = req->rq_export;
1040 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1041 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1046 #endif /* _MDT_INTERNAL_H */