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 <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>
67 /* check if request's xid is equal to last one or not*/
68 static inline int req_xid_is_last(struct ptlrpc_request *req)
70 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
71 return (req->rq_xid == lcd->lcd_last_xid ||
72 req->rq_xid == lcd->lcd_last_close_xid);
77 /* file data for open files on MDS */
78 struct mdt_file_data {
79 struct portals_handle mfd_handle; /* must be first */
80 __u64 mfd_mode; /* open mode provided by client */
81 cfs_list_t mfd_list; /* protected by med_open_lock */
82 __u64 mfd_xid; /* xid of the open request */
83 struct lustre_handle mfd_old_handle; /* old handle in replay case */
84 struct mdt_object *mfd_object; /* point to opened object */
87 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
88 #define CDT_NORETRY_ACTION (1ULL << 1)
89 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
90 #define CDT_POLICY_SHIFT_COUNT 2
91 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
94 /* when adding a new policy, do not forget to update
95 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
97 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
99 enum cdt_states { CDT_STOPPED = 0,
105 /* when multiple lock are needed, the lock order is
113 struct ptlrpc_thread cdt_thread; /**< coordinator thread */
114 struct lu_env cdt_env; /**< coordinator lustre
116 struct lu_context cdt_session; /** session for lu_ucred */
117 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
118 __u64 cdt_policy; /**< policy flags */
119 enum cdt_states cdt_state; /**< state */
120 atomic_t cdt_compound_id; /**< compound id
122 __u64 cdt_last_cookie; /**< last cookie
124 struct mutex cdt_llog_lock; /**< protect llog
126 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
127 struct rw_semaphore cdt_request_lock; /**< protect request
129 struct mutex cdt_restore_lock; /**< protect restore
131 cfs_time_t cdt_loop_period; /**< llog scan period */
132 cfs_time_t cdt_grace_delay; /**< request grace
134 cfs_time_t cdt_active_req_timeout; /**< request timeout */
135 __u32 cdt_default_archive_id; /**< archive id used
138 __u64 cdt_max_requests; /**< max count of started
140 atomic_t cdt_request_count; /**< current count of
141 * started requests */
142 struct list_head cdt_requests; /**< list of started
144 struct list_head cdt_agents; /**< list of register
146 struct list_head cdt_restore_hdl; /**< list of restore lock
148 /* Bitmasks indexed by the HSMA_XXX constants. */
149 __u64 cdt_user_request_mask;
150 __u64 cdt_group_request_mask;
151 __u64 cdt_other_request_mask;
154 /* mdt state flag bits */
155 #define MDT_FL_CFGLOG 0
156 #define MDT_FL_SYNCED 1
160 struct lu_device mdt_lu_dev;
161 struct seq_server_site mdt_seq_site;
162 /* DLM name-space for meta-data locks maintained by this server */
163 struct ldlm_namespace *mdt_namespace;
164 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
165 struct ptlrpc_client *mdt_ldlm_client;
166 /* underlying device */
167 struct obd_export *mdt_child_exp;
168 struct md_device *mdt_child;
169 struct dt_device *mdt_bottom;
170 struct obd_export *mdt_bottom_exp;
172 struct lu_target mdt_lut;
174 * Options bit-fields.
177 unsigned int mo_user_xattr:1,
182 /* mdt state flags */
183 unsigned long mdt_state;
184 /* lock to protect IOepoch */
185 spinlock_t mdt_ioepoch_lock;
188 /* transaction callbacks */
189 struct dt_txn_callback mdt_txn_cb;
191 /* these values should be updated from lov if necessary.
192 * or should be placed somewhere else. */
194 int mdt_max_cookiesize;
198 struct upcall_cache *mdt_identity_cache;
200 /* capability keys */
201 unsigned long mdt_capa_timeout;
203 struct dt_object *mdt_ck_obj;
204 unsigned long mdt_ck_timeout;
205 unsigned long mdt_ck_expiry;
206 struct timer_list mdt_ck_timer;
207 struct ptlrpc_thread mdt_ck_thread;
208 struct lustre_capa_key mdt_capa_keys[2];
209 unsigned int mdt_capa_conf:1,
211 /* Enable remote dir on non-MDT0 */
212 mdt_enable_remote_dir:1;
214 gid_t mdt_enable_remote_dir_gid;
215 /* statfs optimization: we cache a bit */
216 struct obd_statfs mdt_osfs;
218 spinlock_t mdt_osfs_lock;
221 uid_t mdt_squash_uid;
222 gid_t mdt_squash_gid;
223 cfs_list_t mdt_nosquash_nids;
224 char *mdt_nosquash_str;
225 int mdt_nosquash_strlen;
226 struct rw_semaphore mdt_squash_sem;
228 struct rename_stats mdt_rename_stats;
229 struct lu_fid mdt_md_root_fid;
231 /* connection to quota master */
232 struct obd_export *mdt_qmt_exp;
233 /* quota master device associated with this MDT */
234 struct lu_device *mdt_qmt_dev;
236 struct coordinator mdt_coordinator;
239 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
240 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
241 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
242 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
243 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
244 #define MDT_COS_DEFAULT (0)
247 struct lu_object_header mot_header;
248 struct lu_object mot_obj;
251 int mot_ioepoch_count;
253 /* Lock to protect object's IO epoch. */
254 struct mutex mot_ioepoch_mutex;
255 /* Lock to protect create_data */
256 struct mutex mot_lov_mutex;
257 /* Lock to protect lease open.
258 * Lease open acquires write lock; normal open acquires read lock */
259 struct rw_semaphore mot_open_sem;
260 atomic_t mot_lease_count;
261 atomic_t mot_open_count;
264 enum mdt_object_flags {
265 /** SOM attributes are changed. */
266 MOF_SOM_CHANGE = (1 << 0),
268 * The SOM recovery state for mdt object.
269 * This state is an in-memory equivalent of an absent SOM EA, used
270 * instead of invalidating SOM EA while IOEpoch is still opened when
271 * a client eviction occurs or a client fails to obtain SOM attributes.
272 * It indicates that the last IOEpoch holder will need to obtain SOM
273 * attributes under [0;EOF] extent lock to flush all the client's
274 * cached of evicted from MDS clients (but not necessary evicted from
275 * OST) before taking ost attributes.
277 MOF_SOM_RECOV = (1 << 1),
278 /** File has been just created. */
279 MOF_SOM_CREATED = (1 << 2),
280 /** lov object has been created. */
281 MOF_LOV_CREATED = (1 << 3),
284 struct mdt_lock_handle {
285 /* Lock type, reg for cross-ref use or pdo lock. */
289 struct lustre_handle mlh_reg_lh;
290 ldlm_mode_t mlh_reg_mode;
293 struct lustre_handle mlh_pdo_lh;
294 ldlm_mode_t mlh_pdo_mode;
295 unsigned int mlh_pdo_hash;
297 /* Remote regular lock */
298 struct lustre_handle mlh_rreg_lh;
299 ldlm_mode_t mlh_rreg_mode;
303 MDT_LH_PARENT, /* parent lockh */
304 MDT_LH_CHILD, /* child lockh */
305 MDT_LH_OLD, /* old lockh for rename */
306 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
307 MDT_LH_NEW, /* new lockh for rename */
308 MDT_LH_RMT, /* used for return lh to caller */
309 MDT_LH_LOCAL, /* local lock never return to client */
318 struct mdt_reint_record {
319 mdt_reint_t rr_opcode;
320 const struct lustre_handle *rr_handle;
321 const struct lu_fid *rr_fid1;
322 const struct lu_fid *rr_fid2;
327 const void *rr_eadata;
330 const struct llog_cookie *rr_logcookies;
334 enum mdt_reint_flag {
335 MRF_OPEN_TRUNC = 1 << 0,
338 struct mdt_thread_info;
340 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
344 tx_exec_func_t exec_fn;
345 tx_exec_func_t undo_fn;
346 struct dt_object *object;
348 struct update_reply *reply;
353 const struct dt_rec *rec;
354 const struct dt_key *key;
369 struct dt_allocation_hint hint;
370 struct dt_object_format dof;
378 struct ost_body *body;
383 #define TX_MAX_OPS 10
384 struct thandle_exec_args {
385 struct thandle *ta_handle;
386 struct dt_device *ta_dev;
388 struct tx_arg ta_args[TX_MAX_OPS];
389 int ta_argno; /* used args */
393 * Common data shared by mdt-level handlers. This is allocated per-thread to
394 * reduce stack consumption.
396 struct mdt_thread_info {
399 * The following members will be filled explicitly
400 * with specific data in mdt_thread_info_init().
402 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
403 * request handling may migrate from one server thread to another.
405 struct req_capsule *mti_pill;
407 /* although we have export in req, there are cases when it is not
408 * available, e.g. closing files upon export destroy */
409 struct obd_export *mti_exp;
411 * A couple of lock handles.
413 struct mdt_lock_handle mti_lh[MDT_LH_NR];
415 struct mdt_device *mti_mdt;
416 const struct lu_env *mti_env;
417 /* XXX: temporary flag to have healthy mti during OUT calls
418 * to be removed upon moving MDT to the unified target code */
421 /* transaction number of current request */
427 * The following members will be filled expilictly
428 * with zero in mdt_thread_info_init(). These members may be used
435 struct md_attr mti_attr;
437 * Body for "habeo corpus" operations.
439 const struct mdt_body *mti_body;
441 * Host object. This is released at the end of mdt_handler().
443 struct mdt_object *mti_object;
445 * Lock request for "habeo clavis" operations.
447 const struct ldlm_request *mti_dlm_req;
449 __u32 mti_has_trans:1, /* has txn already? */
452 /* opdata for mdt_reint_open(), has the same as
453 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
454 * value onto disk for recovery when mdt_trans_stop_cb() is called.
460 * The following members will be filled explicitly
461 * with zero in mdt_reint_unpack(), because they are only used
462 * by reint requests (including mdt_reint_open()).
466 * reint record. contains information for reint operations.
468 struct mdt_reint_record mti_rr;
470 /** md objects included in operation */
471 struct mdt_object *mti_mos;
472 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
474 * Operation specification (currently create and lookup)
476 struct md_op_spec mti_spec;
480 * The following members will _NOT_ be initialized at all.
481 * DO NOT expect them to contain any valid value.
482 * They should be initialized explicitly by the user themselves.
485 /* XXX: If something is in a union, make sure they do not conflict */
487 struct lu_fid mti_tmp_fid1;
488 struct lu_fid mti_tmp_fid2;
489 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
490 * mdt_rename_lock() */
491 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
494 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
495 char ns_name[48]; /* for mdt_init0() */
496 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
497 struct obd_statfs osfs; /* for mdt_statfs() */
499 /* for mdt_readpage() */
500 struct lu_rdpg mti_rdpg;
501 /* for mdt_sendpage() */
502 struct l_wait_info mti_wait_info;
506 struct md_som_data data;
509 struct dt_object_format mti_update_dof;
510 struct update_reply *mti_update_reply;
511 struct update *mti_update;
512 int mti_update_reply_index;
513 struct obdo mti_obdo;
514 struct dt_object *mti_dt_object;
518 /* IO epoch related stuff. */
519 struct mdt_ioepoch *mti_ioepoch;
520 __u64 mti_replayepoch;
523 struct lu_buf mti_buf;
524 struct lu_buf mti_big_buf;
525 struct lustre_capa_key mti_capa_key;
527 /* Ops object filename */
528 struct lu_name mti_name;
529 /* per-thread values, can be re-used */
532 /* big_lmm buffer was used and must be used in reply */
533 int mti_big_lmm_used;
534 /* should be enough to fit lustre_mdt_attrs */
535 char mti_xattr_buf[128];
536 struct thandle_exec_args mti_handle;
537 struct ldlm_enqueue_info mti_einfo;
540 extern struct lu_context_key mdt_thread_key;
542 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
544 struct mdt_thread_info *mti;
546 lu_env_refill((void *)env);
547 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
552 /* ptlrpc request handler for MDT. All handlers are
553 * grouped into several slices - struct mdt_opc_slice,
554 * and stored in an array - mdt_handlers[].
557 /* The name of this handler. */
559 /* Fail id for this handler, checked at the beginning of this handler*/
561 /* Operation code for this handler */
563 /* flags are listed in enum mdt_handler_flags below. */
565 /* The actual handler function to execute. */
566 int (*mh_act)(struct mdt_thread_info *info);
567 /* Request format for this request. */
568 const struct req_format *mh_fmt;
571 struct mdt_opc_slice {
574 struct mdt_handler *mos_hs;
577 struct cdt_req_progress {
578 struct mutex crp_lock; /**< protect tree */
579 struct interval_node *crp_root; /**< tree to track extent
581 struct interval_node **crp_node; /**< buffer for tree nodes
582 * vector of fixed size
584 int crp_cnt; /**< # of used nodes */
585 int crp_max; /**< # of allocated nodes */
588 struct cdt_agent_req {
589 cfs_list_t car_request_list; /**< to chain all the req. */
590 atomic_t car_refcount; /**< reference counter */
591 __u64 car_compound_id; /**< compound id */
592 __u64 car_flags; /**< request original flags */
593 struct obd_uuid car_uuid; /**< agent doing the req. */
594 __u32 car_archive_id; /**< archive id */
595 int car_canceled; /**< request was canceled */
596 cfs_time_t car_req_start; /**< start time */
597 cfs_time_t car_req_update; /**< last update time */
598 struct hsm_action_item *car_hai; /**< req. to the agent */
599 struct cdt_req_progress car_progress; /**< track data mvt
602 extern struct kmem_cache *mdt_hsm_car_kmem;
605 cfs_list_t ha_list; /**< to chain the agents */
606 struct obd_uuid ha_uuid; /**< agent uuid */
607 __u32 *ha_archive_id; /**< archive id */
608 int ha_archive_cnt; /**< number of archive entries
609 * 0 means any archive */
610 atomic_t ha_requests; /**< current request count */
611 atomic_t ha_success; /**< number of successful
613 atomic_t ha_failure; /**< number of failed actions */
616 struct cdt_restore_handle {
617 cfs_list_t crh_list; /**< to chain the handle */
618 struct lu_fid crh_fid; /**< fid of the object */
619 struct ldlm_extent crh_extent; /**< extent of the restore */
620 struct mdt_lock_handle crh_lh; /**< lock handle */
622 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
624 static inline const struct md_device_operations *
625 mdt_child_ops(struct mdt_device * m)
627 LASSERT(m->mdt_child);
628 return m->mdt_child->md_ops;
631 static inline struct md_object *mdt_object_child(struct mdt_object *o)
634 return lu2md(lu_object_next(&o->mot_obj));
637 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
639 return info->mti_pill ? info->mti_pill->rc_req : NULL;
642 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
644 LASSERT(info->mti_exp);
645 return exp_connect_flags(info->mti_exp);
648 static inline void mdt_object_get(const struct lu_env *env,
649 struct mdt_object *o)
652 lu_object_get(&o->mot_obj);
656 static inline void mdt_object_put(const struct lu_env *env,
657 struct mdt_object *o)
660 lu_object_put(env, &o->mot_obj);
664 static inline int mdt_object_exists(const struct mdt_object *o)
666 return lu_object_exists(&o->mot_obj);
669 static inline int mdt_object_remote(const struct mdt_object *o)
671 return lu_object_remote(&o->mot_obj);
674 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
676 return lu_object_fid(&o->mot_obj);
679 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
681 return mdt->mdt_lu_dev.ld_site;
684 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
686 return &mdt->mdt_seq_site;
689 static inline void mdt_export_evict(struct obd_export *exp)
691 class_fail_export(exp);
692 class_export_put(exp);
695 /* Here we use LVB_TYPE to check dne client, because it is
696 * also landed on 2.4. */
697 static inline int mdt_is_dne_client(struct obd_export *exp)
699 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
702 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
703 void mdt_set_disposition(struct mdt_thread_info *info,
704 struct ldlm_reply *rep, int flag);
705 void mdt_clear_disposition(struct mdt_thread_info *info,
706 struct ldlm_reply *rep, int flag);
708 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
709 ldlm_mode_t lm, const char *name,
712 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
715 int mdt_lock_setup(struct mdt_thread_info *info,
716 struct mdt_object *o,
717 struct mdt_lock_handle *lh);
719 int mdt_object_lock(struct mdt_thread_info *,
721 struct mdt_lock_handle *,
724 int mdt_object_lock_try(struct mdt_thread_info *,
726 struct mdt_lock_handle *,
729 void mdt_object_unlock(struct mdt_thread_info *,
731 struct mdt_lock_handle *,
734 struct mdt_object *mdt_object_new(const struct lu_env *,
736 const struct lu_fid *);
737 struct mdt_object *mdt_object_find(const struct lu_env *,
739 const struct lu_fid *);
740 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
741 const struct lu_fid *,
742 struct mdt_lock_handle *,
744 void mdt_object_unlock_put(struct mdt_thread_info *,
746 struct mdt_lock_handle *,
749 void mdt_client_compatibility(struct mdt_thread_info *info);
751 int mdt_remote_object_lock(struct mdt_thread_info *mti,
752 struct mdt_object *o, struct lustre_handle *lh,
753 ldlm_mode_t mode, __u64 ibits);
754 int mdt_close_unpack(struct mdt_thread_info *info);
755 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
756 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
757 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
758 const struct lu_attr *attr, const struct lu_fid *fid);
760 int mdt_getxattr(struct mdt_thread_info *info);
761 int mdt_reint_setxattr(struct mdt_thread_info *info,
762 struct mdt_lock_handle *lh);
764 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
765 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
767 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
768 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
769 struct mdt_lock_handle *lhc);
771 extern void target_recovery_fini(struct obd_device *obd);
772 extern void target_recovery_init(struct lu_target *lut,
773 svc_handler_t handler);
774 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
775 struct obd_device *, struct lustre_sb_info *lsi);
776 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
778 int mdt_export_stats_init(struct obd_device *obd,
779 struct obd_export *exp,
782 int mdt_lock_new_child(struct mdt_thread_info *info,
783 struct mdt_object *o,
784 struct mdt_lock_handle *child_lockh);
786 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
789 int mdt_reint_open(struct mdt_thread_info *info,
790 struct mdt_lock_handle *lhc);
792 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
793 const struct lustre_handle *handle,
797 MDT_IOEPOCH_CLOSED = 0,
798 MDT_IOEPOCH_OPENED = 1,
799 MDT_IOEPOCH_GETATTR = 2,
807 int mdt_get_info(struct tgt_session_info *tsi);
808 int mdt_attr_get_complex(struct mdt_thread_info *info,
809 struct mdt_object *o, struct md_attr *ma);
810 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
812 int mdt_object_is_som_enabled(struct mdt_object *mo);
813 int mdt_write_get(struct mdt_object *o);
814 void mdt_write_put(struct mdt_object *o);
815 int mdt_write_read(struct mdt_object *o);
816 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
817 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
818 void mdt_mfd_free(struct mdt_file_data *mfd);
819 int mdt_close(struct tgt_session_info *tsi);
820 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
821 struct md_attr *ma, int flags);
822 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
824 int mdt_done_writing(struct tgt_session_info *tsi);
825 int mdt_fix_reply(struct mdt_thread_info *info);
826 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
827 const struct md_attr *);
828 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
830 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
831 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
832 const void *area, ssize_t len);
834 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
836 int mdt_check_ucred(struct mdt_thread_info *);
837 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
838 int mdt_init_ucred_reint(struct mdt_thread_info *);
839 void mdt_exit_ucred(struct mdt_thread_info *);
840 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
841 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
842 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
844 void mdt_thread_info_init(struct ptlrpc_request *req,
845 struct mdt_thread_info *mti);
846 void mdt_thread_info_fini(struct mdt_thread_info *mti);
847 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
849 extern struct lprocfs_vars lprocfs_mds_module_vars[];
850 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
852 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
853 const struct md_hsm *mh);
856 int mdt_init_idmap(struct tgt_session_info *tsi);
857 void mdt_cleanup_idmap(struct mdt_export_data *);
858 int mdt_handle_idmap(struct tgt_session_info *tsi);
859 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
860 struct ptlrpc_user_desc *);
861 void mdt_body_reverse_idmap(struct mdt_thread_info *,
863 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
864 struct mdt_remote_perm *);
865 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
867 static inline struct mdt_device *mdt_dev(struct lu_device *d)
869 return container_of0(d, struct mdt_device, mdt_lu_dev);
872 static inline struct mdt_object *mdt_obj(struct lu_object *o)
874 return container_of0(o, struct mdt_object, mot_obj);
877 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
879 struct lu_object *lo;
880 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
882 lo = lu_object_locate(mo->mot_obj.lo_header,
883 mdt->mdt_bottom->dd_lu_dev.ld_type);
888 /* mdt/mdt_identity.c */
889 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
891 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
893 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
895 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
897 void mdt_flush_identity(struct upcall_cache *, int);
899 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
901 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
904 int mdt_hsm_state_get(struct tgt_session_info *tsi);
905 int mdt_hsm_state_set(struct tgt_session_info *tsi);
906 int mdt_hsm_action(struct tgt_session_info *tsi);
907 int mdt_hsm_progress(struct tgt_session_info *tsi);
908 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
909 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
910 int mdt_hsm_request(struct tgt_session_info *tsi);
912 /* mdt/mdt_hsm_cdt_actions.c */
913 extern const struct file_operations mdt_hsm_actions_fops;
914 void dump_llog_agent_req_rec(const char *prefix,
915 const struct llog_agent_req_rec *larr);
916 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
917 llog_cb_t cb, void *data);
918 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
919 __u64 compound_id, __u32 archive_id,
920 __u64 flags, struct hsm_action_item *hai);
921 int mdt_agent_record_update(const struct lu_env *env,
922 struct mdt_device *mdt, __u64 *cookies,
923 int cookies_count, enum agent_req_status status);
924 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
925 struct llog_handle *llh,
926 struct llog_agent_req_rec *larr);
928 /* mdt/mdt_hsm_cdt_agent.c */
929 extern const struct file_operations mdt_hsm_agent_fops;
930 int mdt_hsm_agent_register(struct mdt_thread_info *info,
931 const struct obd_uuid *uuid,
932 int nr_archives, __u32 *archive_num);
933 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
934 const struct obd_uuid *uuid,
936 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
937 const struct obd_uuid *uuid);
938 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
939 int succ_rq, int fail_rq, int new_rq,
940 const struct obd_uuid *uuid);
941 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
942 struct obd_uuid *uuid);
943 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
945 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
946 struct hsm_progress_kernel *pgs);
947 /* mdt/mdt_hsm_cdt_client.c */
948 int mdt_hsm_add_actions(struct mdt_thread_info *info,
949 struct hsm_action_list *hal, __u64 *compound_id);
950 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
951 struct hsm_action_list *hal);
952 int mdt_hsm_get_running(struct mdt_thread_info *mti,
953 struct hsm_action_list *hal);
954 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
955 const struct lu_fid *fid);
956 /* mdt/mdt_hsm_cdt_requests.c */
957 extern const struct file_operations mdt_hsm_active_requests_fops;
958 void dump_requests(char *prefix, struct coordinator *cdt);
959 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
960 __u64 flags, struct obd_uuid *uuid,
961 struct hsm_action_item *hai);
962 void mdt_cdt_free_request(struct cdt_agent_req *car);
963 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
964 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
966 const struct lu_fid *fid);
967 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
968 void mdt_cdt_get_request(struct cdt_agent_req *car);
969 void mdt_cdt_put_request(struct cdt_agent_req *car);
970 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
971 const struct hsm_progress_kernel *pgs);
972 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
973 /* mdt/mdt_coordinator.c */
974 void mdt_hsm_dump_hal(int level, const char *prefix,
975 struct hsm_action_list *hal);
976 /* coordinator management */
977 int mdt_hsm_cdt_init(struct mdt_device *mdt);
978 int mdt_hsm_cdt_start(struct mdt_device *mdt);
979 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
980 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
981 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
983 /* coordinator control /proc interface */
984 int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
985 unsigned long count, void *data);
986 int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
987 int count, int *eof, void *data);
988 int hsm_cdt_procfs_init(struct mdt_device *mdt);
989 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
990 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
992 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
993 const struct lu_fid *fid,
995 /* actions/request helpers */
996 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
997 struct hsm_action_list *hal, struct obd_uuid *uuid);
998 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
999 const int hal_an, const __u64 rq_flags,
1000 const struct md_hsm *hsm);
1001 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1002 struct hsm_progress_kernel *pgs,
1003 const int update_record);
1005 extern struct lu_context_key mdt_thread_key;
1007 /* debug issues helper starts here*/
1008 static inline int mdt_fail_write(const struct lu_env *env,
1009 struct dt_device *dd, int id)
1011 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1012 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1014 return dd->dd_ops->dt_ro(env, dd);
1015 /* We set FAIL_ONCE because we never "un-fail" a device */
1021 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1023 return &req->rq_export->exp_mdt_data;
1026 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
1028 return mdt_dev(exp->exp_obd->obd_lu_dev);
1031 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1032 struct mdt_lock_handle *lhc);
1033 static inline int mdt_check_resent(struct mdt_thread_info *info,
1034 mdt_reconstruct_t reconstruct,
1035 struct mdt_lock_handle *lhc)
1037 struct ptlrpc_request *req = mdt_info_req(info);
1040 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1041 if (req_xid_is_last(req)) {
1042 reconstruct(info, lhc);
1045 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
1046 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
1051 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1052 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1054 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1056 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1059 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1061 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1062 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1063 struct lustre_handle *lh,
1065 ldlm_policy_data_t *policy,
1066 const struct ldlm_res_id *res_id,
1067 __u64 flags, const __u64 *client_cookie)
1071 LASSERT(ns != NULL);
1072 LASSERT(lh != NULL);
1074 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1075 mode, &flags, mdt_blocking_ast,
1076 ldlm_completion_ast, NULL, NULL, 0,
1077 LVB_T_NONE, client_cookie, lh);
1078 return rc == ELDLM_OK ? 0 : -EIO;
1081 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1084 ldlm_lock_decref(lh, mode);
1087 extern mdl_mode_t mdt_mdl_lock_modes[];
1088 extern ldlm_mode_t mdt_dlm_lock_modes[];
1090 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1092 LASSERT(IS_PO2(mode));
1093 return mdt_mdl_lock_modes[mode];
1096 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1098 LASSERT(IS_PO2(mode));
1099 return mdt_dlm_lock_modes[mode];
1103 extern struct ldlm_valblock_ops mdt_lvbo;
1105 static inline struct lu_name *mdt_name(const struct lu_env *env,
1106 char *name, int namelen)
1108 struct lu_name *lname;
1109 struct mdt_thread_info *mti;
1111 LASSERT(namelen > 0);
1112 /* trailing '\0' in buffer */
1113 LASSERT(name[namelen] == '\0');
1115 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
1116 lname = &mti->mti_name;
1117 lname->ln_name = name;
1118 lname->ln_namelen = namelen;
1122 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
1123 struct lu_name *slname)
1128 tlname->ln_name = slname->ln_name;
1129 tlname->ln_namelen = slname->ln_namelen;
1133 void mdt_enable_cos(struct mdt_device *, int);
1134 int mdt_cos_is_enabled(struct mdt_device *);
1152 LPROC_MDT_SAMEDIR_RENAME,
1153 LPROC_MDT_CROSSDIR_RENAME,
1156 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1157 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1158 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1159 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1160 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1161 void mdt_procfs_fini(struct mdt_device *mdt);
1162 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1163 struct mdt_device *mdt,
1164 struct ptlrpc_request *req,
1165 struct mdt_object *src, struct mdt_object *tgt);
1168 int mdt_ck_thread_start(struct mdt_device *mdt);
1169 void mdt_ck_thread_stop(struct mdt_device *mdt);
1170 void mdt_ck_timer_callback(unsigned long castmeharder);
1171 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1172 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1173 const struct lu_fid *fid, struct lustre_capa *capa);
1174 void mdt_dump_capainfo(struct mdt_thread_info *info);
1176 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1178 return mdt->mdt_lu_dev.ld_obd;
1181 extern const struct lu_device_operations mdt_lu_ops;
1183 static inline char *mdt_obd_name(struct mdt_device *mdt)
1185 return mdt->mdt_lu_dev.ld_obd->obd_name;
1188 int mds_mod_init(void);
1189 void mds_mod_exit(void);
1191 #endif /* __KERNEL__ */