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, 2013, 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
51 #if defined(__KERNEL__)
53 #include <libcfs/lucache.h>
54 #include <lustre_net.h>
55 #include <lustre/lustre_idl.h>
56 #include <obd_class.h>
57 #include <lustre_disk.h>
58 #include <lu_target.h>
59 #include <md_object.h>
60 #include <lustre_fid.h>
61 #include <lustre_fld.h>
62 #include <lustre_req_layout.h>
63 #include <lustre_sec.h>
64 #include <lustre_idmap.h>
65 #include <lustre_eacl.h>
66 #include <lustre_quota.h>
67 #include <lustre_linkea.h>
69 /* check if request's xid is equal to last one or not*/
70 static inline int req_xid_is_last(struct ptlrpc_request *req)
72 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
73 return (req->rq_xid == lcd->lcd_last_xid ||
74 req->rq_xid == lcd->lcd_last_close_xid);
79 /* file data for open files on MDS */
80 struct mdt_file_data {
81 struct portals_handle mfd_handle; /* must be first */
82 __u64 mfd_mode; /* open mode provided by client */
83 cfs_list_t mfd_list; /* protected by med_open_lock */
84 __u64 mfd_xid; /* xid of the open request */
85 struct lustre_handle mfd_old_handle; /* old handle in replay case */
86 struct mdt_object *mfd_object; /* point to opened object */
89 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
90 #define CDT_NORETRY_ACTION (1ULL << 1)
91 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
92 #define CDT_POLICY_SHIFT_COUNT 2
93 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
96 /* when adding a new policy, do not forget to update
97 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
99 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
101 enum cdt_states { CDT_STOPPED = 0,
107 /* when multiple lock are needed, the lock order is
115 struct ptlrpc_thread cdt_thread; /**< coordinator thread */
116 struct lu_env cdt_env; /**< coordinator lustre
118 struct lu_context cdt_session; /** session for lu_ucred */
119 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
120 __u64 cdt_policy; /**< policy flags */
121 enum cdt_states cdt_state; /**< state */
122 atomic_t cdt_compound_id; /**< compound id
124 __u64 cdt_last_cookie; /**< last cookie
126 struct mutex cdt_llog_lock; /**< protect llog
128 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
129 struct rw_semaphore cdt_request_lock; /**< protect request
131 struct mutex cdt_restore_lock; /**< protect restore
133 cfs_time_t cdt_loop_period; /**< llog scan period */
134 cfs_time_t cdt_grace_delay; /**< request grace
136 cfs_time_t cdt_active_req_timeout; /**< request timeout */
137 __u32 cdt_default_archive_id; /**< archive id used
140 __u64 cdt_max_requests; /**< max count of started
142 atomic_t cdt_request_count; /**< current count of
143 * started requests */
144 struct list_head cdt_requests; /**< list of started
146 struct list_head cdt_agents; /**< list of register
148 struct list_head cdt_restore_hdl; /**< list of restore lock
150 /* Bitmasks indexed by the HSMA_XXX constants. */
151 __u64 cdt_user_request_mask;
152 __u64 cdt_group_request_mask;
153 __u64 cdt_other_request_mask;
156 /* mdt state flag bits */
157 #define MDT_FL_CFGLOG 0
158 #define MDT_FL_SYNCED 1
162 struct lu_device mdt_lu_dev;
163 struct seq_server_site mdt_seq_site;
164 /* DLM name-space for meta-data locks maintained by this server */
165 struct ldlm_namespace *mdt_namespace;
166 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
167 struct ptlrpc_client *mdt_ldlm_client;
168 /* underlying device */
169 struct obd_export *mdt_child_exp;
170 struct md_device *mdt_child;
171 struct dt_device *mdt_bottom;
172 struct obd_export *mdt_bottom_exp;
174 struct lu_target mdt_lut;
176 * Options bit-fields.
179 unsigned int mo_user_xattr:1,
184 /* mdt state flags */
185 unsigned long mdt_state;
186 /* lock to protect IOepoch */
187 spinlock_t mdt_ioepoch_lock;
190 /* transaction callbacks */
191 struct dt_txn_callback mdt_txn_cb;
193 /* these values should be updated from lov if necessary.
194 * or should be placed somewhere else. */
199 struct upcall_cache *mdt_identity_cache;
201 /* capability keys */
202 unsigned long mdt_capa_timeout;
204 struct dt_object *mdt_ck_obj;
205 unsigned long mdt_ck_timeout;
206 unsigned long mdt_ck_expiry;
207 struct timer_list mdt_ck_timer;
208 struct ptlrpc_thread mdt_ck_thread;
209 struct lustre_capa_key mdt_capa_keys[2];
210 unsigned int mdt_capa_conf:1,
212 /* Enable remote dir on non-MDT0 */
213 mdt_enable_remote_dir:1;
215 gid_t mdt_enable_remote_dir_gid;
216 /* statfs optimization: we cache a bit */
217 struct obd_statfs mdt_osfs;
219 spinlock_t mdt_osfs_lock;
222 struct root_squash_info mdt_squash;
224 struct rename_stats mdt_rename_stats;
225 struct lu_fid mdt_md_root_fid;
227 /* connection to quota master */
228 struct obd_export *mdt_qmt_exp;
229 /* quota master device associated with this MDT */
230 struct lu_device *mdt_qmt_dev;
232 struct coordinator mdt_coordinator;
235 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
236 #define MDT_COS_DEFAULT (0)
239 struct lu_object_header mot_header;
240 struct lu_object mot_obj;
243 int mot_ioepoch_count;
245 /* Lock to protect object's IO epoch. */
246 struct mutex mot_ioepoch_mutex;
247 /* Lock to protect create_data */
248 struct mutex mot_lov_mutex;
249 /* Lock to protect lease open.
250 * Lease open acquires write lock; normal open acquires read lock */
251 struct rw_semaphore mot_open_sem;
252 atomic_t mot_lease_count;
253 atomic_t mot_open_count;
256 enum mdt_object_flags {
257 /** SOM attributes are changed. */
258 MOF_SOM_CHANGE = (1 << 0),
260 * The SOM recovery state for mdt object.
261 * This state is an in-memory equivalent of an absent SOM EA, used
262 * instead of invalidating SOM EA while IOEpoch is still opened when
263 * a client eviction occurs or a client fails to obtain SOM attributes.
264 * It indicates that the last IOEpoch holder will need to obtain SOM
265 * attributes under [0;EOF] extent lock to flush all the client's
266 * cached of evicted from MDS clients (but not necessary evicted from
267 * OST) before taking ost attributes.
269 MOF_SOM_RECOV = (1 << 1),
270 /** File has been just created. */
271 MOF_SOM_CREATED = (1 << 2),
272 /** lov object has been created. */
273 MOF_LOV_CREATED = (1 << 3),
276 struct mdt_lock_handle {
277 /* Lock type, reg for cross-ref use or pdo lock. */
281 struct lustre_handle mlh_reg_lh;
282 ldlm_mode_t mlh_reg_mode;
285 struct lustre_handle mlh_pdo_lh;
286 ldlm_mode_t mlh_pdo_mode;
287 unsigned int mlh_pdo_hash;
289 /* Remote regular lock */
290 struct lustre_handle mlh_rreg_lh;
291 ldlm_mode_t mlh_rreg_mode;
295 MDT_LH_PARENT, /* parent lockh */
296 MDT_LH_CHILD, /* child lockh */
297 MDT_LH_OLD, /* old lockh for rename */
298 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
299 MDT_LH_NEW, /* new lockh for rename */
300 MDT_LH_RMT, /* used for return lh to caller */
301 MDT_LH_LOCAL, /* local lock never return to client */
310 struct mdt_reint_record {
311 mdt_reint_t rr_opcode;
312 const struct lustre_handle *rr_handle;
313 const struct lu_fid *rr_fid1;
314 const struct lu_fid *rr_fid2;
315 struct lu_name rr_name;
316 struct lu_name rr_tgt_name;
317 const void *rr_eadata;
322 enum mdt_reint_flag {
323 MRF_OPEN_TRUNC = 1 << 0,
327 * Common data shared by mdt-level handlers. This is allocated per-thread to
328 * reduce stack consumption.
330 struct mdt_thread_info {
333 * The following members will be filled explicitly
334 * with specific data in mdt_thread_info_init().
336 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
337 * request handling may migrate from one server thread to another.
339 struct req_capsule *mti_pill;
341 /* although we have export in req, there are cases when it is not
342 * available, e.g. closing files upon export destroy */
343 struct obd_export *mti_exp;
345 * A couple of lock handles.
347 struct mdt_lock_handle mti_lh[MDT_LH_NR];
349 struct mdt_device *mti_mdt;
350 const struct lu_env *mti_env;
352 /* transaction number of current request */
358 * The following members will be filled expilictly
359 * with zero in mdt_thread_info_init(). These members may be used
366 struct md_attr mti_attr;
368 * Body for "habeo corpus" operations.
370 const struct mdt_body *mti_body;
372 * Host object. This is released at the end of mdt_handler().
374 struct mdt_object *mti_object;
376 * Lock request for "habeo clavis" operations.
378 const struct ldlm_request *mti_dlm_req;
380 __u32 mti_has_trans:1, /* has txn already? */
383 /* opdata for mdt_reint_open(), has the same as
384 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
385 * value onto disk for recovery when mdt_trans_stop_cb() is called.
391 * The following members will be filled explicitly
392 * with zero in mdt_reint_unpack(), because they are only used
393 * by reint requests (including mdt_reint_open()).
397 * reint record. contains information for reint operations.
399 struct mdt_reint_record mti_rr;
401 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
403 * Operation specification (currently create and lookup)
405 struct md_op_spec mti_spec;
409 * The following members will _NOT_ be initialized at all.
410 * DO NOT expect them to contain any valid value.
411 * They should be initialized explicitly by the user themselves.
414 /* XXX: If something is in a union, make sure they do not conflict */
416 struct lu_fid mti_tmp_fid1;
417 struct lu_fid mti_tmp_fid2;
418 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
419 * mdt_rename_lock() */
420 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
423 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
424 char ns_name[48]; /* for mdt_init0() */
425 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
426 struct obd_statfs osfs; /* for mdt_statfs() */
428 /* for mdt_readpage() */
429 struct lu_rdpg mti_rdpg;
430 /* for mdt_sendpage() */
431 struct l_wait_info mti_wait_info;
435 struct md_som_data data;
439 /* IO epoch related stuff. */
440 struct mdt_ioepoch *mti_ioepoch;
441 __u64 mti_replayepoch;
444 struct lu_buf mti_buf;
445 struct lu_buf mti_big_buf;
446 struct lustre_capa_key mti_capa_key;
448 /* Ops object filename */
449 struct lu_name mti_name;
450 /* per-thread values, can be re-used */
453 /* big_lmm buffer was used and must be used in reply */
454 int mti_big_lmm_used;
455 /* should be enough to fit lustre_mdt_attrs */
456 char mti_xattr_buf[128];
457 struct ldlm_enqueue_info mti_einfo;
460 extern struct lu_context_key mdt_thread_key;
462 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
464 struct mdt_thread_info *mti;
466 lu_env_refill((void *)env);
467 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
472 struct cdt_req_progress {
473 struct mutex crp_lock; /**< protect tree */
474 struct interval_node *crp_root; /**< tree to track extent
476 struct interval_node **crp_node; /**< buffer for tree nodes
477 * vector of fixed size
479 int crp_cnt; /**< # of used nodes */
480 int crp_max; /**< # of allocated nodes */
483 struct cdt_agent_req {
484 cfs_list_t car_request_list; /**< to chain all the req. */
485 atomic_t car_refcount; /**< reference counter */
486 __u64 car_compound_id; /**< compound id */
487 __u64 car_flags; /**< request original flags */
488 struct obd_uuid car_uuid; /**< agent doing the req. */
489 __u32 car_archive_id; /**< archive id */
490 int car_canceled; /**< request was canceled */
491 cfs_time_t car_req_start; /**< start time */
492 cfs_time_t car_req_update; /**< last update time */
493 struct hsm_action_item *car_hai; /**< req. to the agent */
494 struct cdt_req_progress car_progress; /**< track data mvt
497 extern struct kmem_cache *mdt_hsm_car_kmem;
500 cfs_list_t ha_list; /**< to chain the agents */
501 struct obd_uuid ha_uuid; /**< agent uuid */
502 __u32 *ha_archive_id; /**< archive id */
503 int ha_archive_cnt; /**< number of archive entries
504 * 0 means any archive */
505 atomic_t ha_requests; /**< current request count */
506 atomic_t ha_success; /**< number of successful
508 atomic_t ha_failure; /**< number of failed actions */
511 struct cdt_restore_handle {
512 cfs_list_t crh_list; /**< to chain the handle */
513 struct lu_fid crh_fid; /**< fid of the object */
514 struct ldlm_extent crh_extent; /**< extent of the restore */
515 struct mdt_lock_handle crh_lh; /**< lock handle */
517 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
519 static inline const struct md_device_operations *
520 mdt_child_ops(struct mdt_device * m)
522 LASSERT(m->mdt_child);
523 return m->mdt_child->md_ops;
526 static inline struct md_object *mdt_object_child(struct mdt_object *o)
529 return lu2md(lu_object_next(&o->mot_obj));
532 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
534 return info->mti_pill ? info->mti_pill->rc_req : NULL;
537 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
539 LASSERT(info->mti_exp);
540 return exp_connect_flags(info->mti_exp);
543 static inline void mdt_object_get(const struct lu_env *env,
544 struct mdt_object *o)
547 lu_object_get(&o->mot_obj);
551 static inline void mdt_object_put(const struct lu_env *env,
552 struct mdt_object *o)
555 lu_object_put(env, &o->mot_obj);
559 static inline int mdt_object_exists(const struct mdt_object *o)
561 return lu_object_exists(&o->mot_obj);
564 static inline int mdt_object_remote(const struct mdt_object *o)
566 return lu_object_remote(&o->mot_obj);
569 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
571 return lu_object_fid(&o->mot_obj);
574 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
576 return mdt->mdt_lu_dev.ld_site;
579 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
581 return &mdt->mdt_seq_site;
584 static inline void mdt_export_evict(struct obd_export *exp)
586 class_fail_export(exp);
587 class_export_put(exp);
590 /* Here we use LVB_TYPE to check dne client, because it is
591 * also landed on 2.4. */
592 static inline bool mdt_is_dne_client(struct obd_export *exp)
594 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
597 /* Here we use OBD_CONNECT_DISP_STRIPE to check the client who support
598 * striped directory, because it is also landed on 2.6 */
599 static inline bool mdt_is_striped_client(struct obd_export *exp)
601 return exp_connect_flags(exp) & OBD_CONNECT_DISP_STRIPE;
604 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
605 void mdt_set_disposition(struct mdt_thread_info *info,
606 struct ldlm_reply *rep, int flag);
607 void mdt_clear_disposition(struct mdt_thread_info *info,
608 struct ldlm_reply *rep, int flag);
610 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
611 ldlm_mode_t lock_mode,
612 const struct lu_name *lname);
614 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
617 int mdt_lock_setup(struct mdt_thread_info *info,
618 struct mdt_object *o,
619 struct mdt_lock_handle *lh);
621 int mdt_check_resent_lock(struct mdt_thread_info *info,
622 struct mdt_object *mo,
623 struct mdt_lock_handle *lhc);
625 int mdt_object_lock(struct mdt_thread_info *,
627 struct mdt_lock_handle *,
630 int mdt_object_lock_try(struct mdt_thread_info *,
632 struct mdt_lock_handle *,
635 void mdt_object_unlock(struct mdt_thread_info *,
637 struct mdt_lock_handle *,
640 struct mdt_object *mdt_object_new(const struct lu_env *,
642 const struct lu_fid *);
643 struct mdt_object *mdt_object_find(const struct lu_env *,
645 const struct lu_fid *);
646 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
647 const struct lu_fid *,
648 struct mdt_lock_handle *,
650 void mdt_object_unlock_put(struct mdt_thread_info *,
652 struct mdt_lock_handle *,
655 void mdt_client_compatibility(struct mdt_thread_info *info);
657 int mdt_remote_object_lock(struct mdt_thread_info *mti,
658 struct mdt_object *o, const struct lu_fid *fid,
659 struct lustre_handle *lh,
660 ldlm_mode_t mode, __u64 ibits);
662 enum mdt_name_flags {
666 int mdt_name_unpack(struct req_capsule *pill,
667 const struct req_msg_field *field,
669 enum mdt_name_flags flags);
670 int mdt_close_unpack(struct mdt_thread_info *info);
671 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
672 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
673 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
674 const struct lu_attr *attr, const struct lu_fid *fid);
676 int mdt_getxattr(struct mdt_thread_info *info);
677 int mdt_reint_setxattr(struct mdt_thread_info *info,
678 struct mdt_lock_handle *lh);
680 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
681 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
683 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
684 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
685 struct mdt_lock_handle *lhc);
687 extern void target_recovery_fini(struct obd_device *obd);
688 extern void target_recovery_init(struct lu_target *lut,
689 svc_handler_t handler);
690 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
691 struct obd_device *, struct lustre_sb_info *lsi);
692 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
694 int mdt_export_stats_init(struct obd_device *obd,
695 struct obd_export *exp,
698 int mdt_lock_new_child(struct mdt_thread_info *info,
699 struct mdt_object *o,
700 struct mdt_lock_handle *child_lockh);
702 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
705 int mdt_reint_open(struct mdt_thread_info *info,
706 struct mdt_lock_handle *lhc);
708 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
709 const struct lustre_handle *handle,
713 MDT_IOEPOCH_CLOSED = 0,
714 MDT_IOEPOCH_OPENED = 1,
715 MDT_IOEPOCH_GETATTR = 2,
723 int mdt_get_info(struct tgt_session_info *tsi);
724 int mdt_attr_get_complex(struct mdt_thread_info *info,
725 struct mdt_object *o, struct md_attr *ma);
726 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
728 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
729 struct md_attr *ma, const char *name);
730 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
732 int mdt_object_is_som_enabled(struct mdt_object *mo);
733 int mdt_write_get(struct mdt_object *o);
734 void mdt_write_put(struct mdt_object *o);
735 int mdt_write_read(struct mdt_object *o);
736 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
737 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
738 void mdt_mfd_free(struct mdt_file_data *mfd);
739 int mdt_close(struct tgt_session_info *tsi);
740 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
741 struct md_attr *ma, int flags);
742 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
744 int mdt_done_writing(struct tgt_session_info *tsi);
745 int mdt_fix_reply(struct mdt_thread_info *info);
746 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
747 const struct md_attr *);
748 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
750 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
751 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
752 const void *area, ssize_t len);
754 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
755 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
757 int mdt_check_ucred(struct mdt_thread_info *);
758 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
759 int mdt_init_ucred_reint(struct mdt_thread_info *);
760 void mdt_exit_ucred(struct mdt_thread_info *);
761 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
762 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
763 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
765 void mdt_thread_info_init(struct ptlrpc_request *req,
766 struct mdt_thread_info *mti);
767 void mdt_thread_info_fini(struct mdt_thread_info *mti);
768 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
770 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
771 const struct md_hsm *mh);
773 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
774 void *data, int flag);
775 int mdt_links_read(struct mdt_thread_info *info,
776 struct mdt_object *mdt_obj,
777 struct linkea_data *ldata);
779 int mdt_init_idmap(struct tgt_session_info *tsi);
780 void mdt_cleanup_idmap(struct mdt_export_data *);
781 int mdt_handle_idmap(struct tgt_session_info *tsi);
782 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
783 struct ptlrpc_user_desc *);
784 void mdt_body_reverse_idmap(struct mdt_thread_info *,
786 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
787 struct mdt_remote_perm *);
788 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
790 static inline struct mdt_device *mdt_dev(struct lu_device *d)
792 return container_of0(d, struct mdt_device, mdt_lu_dev);
795 static inline struct mdt_object *mdt_obj(struct lu_object *o)
797 return container_of0(o, struct mdt_object, mot_obj);
800 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
802 struct lu_object *lo;
803 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
805 lo = lu_object_locate(mo->mot_obj.lo_header,
806 mdt->mdt_bottom->dd_lu_dev.ld_type);
811 /* mdt/mdt_identity.c */
812 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
814 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
816 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
818 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
820 void mdt_flush_identity(struct upcall_cache *, int);
822 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
824 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
827 int mdt_hsm_state_get(struct tgt_session_info *tsi);
828 int mdt_hsm_state_set(struct tgt_session_info *tsi);
829 int mdt_hsm_action(struct tgt_session_info *tsi);
830 int mdt_hsm_progress(struct tgt_session_info *tsi);
831 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
832 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
833 int mdt_hsm_request(struct tgt_session_info *tsi);
835 /* mdt/mdt_hsm_cdt_actions.c */
836 extern const struct file_operations mdt_hsm_actions_fops;
837 void dump_llog_agent_req_rec(const char *prefix,
838 const struct llog_agent_req_rec *larr);
839 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
840 llog_cb_t cb, void *data);
841 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
842 __u64 compound_id, __u32 archive_id,
843 __u64 flags, struct hsm_action_item *hai);
844 int mdt_agent_record_update(const struct lu_env *env,
845 struct mdt_device *mdt, __u64 *cookies,
846 int cookies_count, enum agent_req_status status);
847 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
848 struct llog_handle *llh,
849 struct llog_agent_req_rec *larr);
851 /* mdt/mdt_hsm_cdt_agent.c */
852 extern const struct file_operations mdt_hsm_agent_fops;
853 int mdt_hsm_agent_register(struct mdt_thread_info *info,
854 const struct obd_uuid *uuid,
855 int nr_archives, __u32 *archive_num);
856 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
857 const struct obd_uuid *uuid,
859 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
860 const struct obd_uuid *uuid);
861 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
862 int succ_rq, int fail_rq, int new_rq,
863 const struct obd_uuid *uuid);
864 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
865 struct obd_uuid *uuid);
866 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
868 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
869 struct hsm_progress_kernel *pgs);
870 /* mdt/mdt_hsm_cdt_client.c */
871 int mdt_hsm_add_actions(struct mdt_thread_info *info,
872 struct hsm_action_list *hal, __u64 *compound_id);
873 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
874 struct hsm_action_list *hal);
875 int mdt_hsm_get_running(struct mdt_thread_info *mti,
876 struct hsm_action_list *hal);
877 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
878 const struct lu_fid *fid);
879 /* mdt/mdt_hsm_cdt_requests.c */
880 extern const struct file_operations mdt_hsm_active_requests_fops;
881 void dump_requests(char *prefix, struct coordinator *cdt);
882 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
883 __u64 flags, struct obd_uuid *uuid,
884 struct hsm_action_item *hai);
885 void mdt_cdt_free_request(struct cdt_agent_req *car);
886 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
887 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
889 const struct lu_fid *fid);
890 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
891 void mdt_cdt_get_request(struct cdt_agent_req *car);
892 void mdt_cdt_put_request(struct cdt_agent_req *car);
893 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
894 const struct hsm_progress_kernel *pgs);
895 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
896 /* mdt/mdt_coordinator.c */
897 void mdt_hsm_dump_hal(int level, const char *prefix,
898 struct hsm_action_list *hal);
899 /* coordinator management */
900 int mdt_hsm_cdt_init(struct mdt_device *mdt);
901 int mdt_hsm_cdt_start(struct mdt_device *mdt);
902 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
903 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
904 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
906 /* coordinator control /proc interface */
907 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file, const char *buffer,
908 size_t count, loff_t *off);
909 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
910 int hsm_cdt_procfs_init(struct mdt_device *mdt);
911 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
912 struct lprocfs_seq_vars *hsm_cdt_get_proc_vars(void);
914 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
915 const struct lu_fid *fid,
917 /* actions/request helpers */
918 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
919 struct hsm_action_list *hal, struct obd_uuid *uuid);
920 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
921 const int hal_an, const __u64 rq_flags,
922 const struct md_hsm *hsm);
923 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
924 struct hsm_progress_kernel *pgs,
925 const int update_record);
927 extern struct lu_context_key mdt_thread_key;
929 /* debug issues helper starts here*/
930 static inline int mdt_fail_write(const struct lu_env *env,
931 struct dt_device *dd, int id)
933 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
934 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
936 return dd->dd_ops->dt_ro(env, dd);
937 /* We set FAIL_ONCE because we never "un-fail" a device */
943 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
945 return &req->rq_export->exp_mdt_data;
948 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
950 return mdt_dev(exp->exp_obd->obd_lu_dev);
953 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
954 struct mdt_lock_handle *lhc);
955 static inline int mdt_check_resent(struct mdt_thread_info *info,
956 mdt_reconstruct_t reconstruct,
957 struct mdt_lock_handle *lhc)
959 struct ptlrpc_request *req = mdt_info_req(info);
962 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
963 if (req_xid_is_last(req)) {
964 reconstruct(info, lhc);
967 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
968 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
973 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
974 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
976 static inline int is_identity_get_disabled(struct upcall_cache *cache)
978 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
981 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
983 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
984 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
985 struct lustre_handle *lh,
987 ldlm_policy_data_t *policy,
988 const struct ldlm_res_id *res_id,
989 __u64 flags, const __u64 *client_cookie)
996 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
997 mode, &flags, mdt_blocking_ast,
998 ldlm_completion_ast, NULL, NULL, 0,
999 LVB_T_NONE, client_cookie, lh);
1000 return rc == ELDLM_OK ? 0 : -EIO;
1003 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1006 ldlm_lock_decref(lh, mode);
1009 extern mdl_mode_t mdt_mdl_lock_modes[];
1010 extern ldlm_mode_t mdt_dlm_lock_modes[];
1012 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1014 LASSERT(IS_PO2(mode));
1015 return mdt_mdl_lock_modes[mode];
1018 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1020 LASSERT(IS_PO2(mode));
1021 return mdt_dlm_lock_modes[mode];
1025 extern struct ldlm_valblock_ops mdt_lvbo;
1027 void mdt_enable_cos(struct mdt_device *, int);
1028 int mdt_cos_is_enabled(struct mdt_device *);
1046 LPROC_MDT_SAMEDIR_RENAME,
1047 LPROC_MDT_CROSSDIR_RENAME,
1050 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1051 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1052 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1053 void mdt_procfs_fini(struct mdt_device *mdt);
1055 /* lustre/mdt_mdt_lproc.c */
1056 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1058 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1059 struct mdt_device *mdt,
1060 struct ptlrpc_request *req,
1061 struct mdt_object *src, struct mdt_object *tgt);
1064 int mdt_ck_thread_start(struct mdt_device *mdt);
1065 void mdt_ck_thread_stop(struct mdt_device *mdt);
1066 void mdt_ck_timer_callback(unsigned long castmeharder);
1067 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1068 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1069 const struct lu_fid *fid, struct lustre_capa *capa);
1070 void mdt_dump_capainfo(struct mdt_thread_info *info);
1072 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1074 return mdt->mdt_lu_dev.ld_obd;
1077 extern const struct lu_device_operations mdt_lu_ops;
1079 static inline char *mdt_obd_name(struct mdt_device *mdt)
1081 return mdt->mdt_lu_dev.ld_obd->obd_name;
1084 int mds_mod_init(void);
1085 void mds_mod_exit(void);
1087 #endif /* __KERNEL__ */