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,
185 /* mdt state flags */
186 unsigned long mdt_state;
187 /* lock to protect IOepoch */
188 spinlock_t mdt_ioepoch_lock;
191 /* transaction callbacks */
192 struct dt_txn_callback mdt_txn_cb;
194 /* these values should be updated from lov if necessary.
195 * or should be placed somewhere else. */
200 struct upcall_cache *mdt_identity_cache;
203 rwlock_t mdt_sptlrpc_lock;
204 struct sptlrpc_rule_set mdt_sptlrpc_rset;
206 /* capability keys */
207 unsigned long mdt_capa_timeout;
209 struct dt_object *mdt_ck_obj;
210 unsigned long mdt_ck_timeout;
211 unsigned long mdt_ck_expiry;
212 struct timer_list mdt_ck_timer;
213 struct ptlrpc_thread mdt_ck_thread;
214 struct lustre_capa_key mdt_capa_keys[2];
215 unsigned int mdt_capa_conf:1,
217 /* Enable remote dir on non-MDT0 */
218 mdt_enable_remote_dir:1;
220 gid_t mdt_enable_remote_dir_gid;
221 /* statfs optimization: we cache a bit */
222 struct obd_statfs mdt_osfs;
224 spinlock_t mdt_osfs_lock;
227 uid_t mdt_squash_uid;
228 gid_t mdt_squash_gid;
229 cfs_list_t mdt_nosquash_nids;
230 char *mdt_nosquash_str;
231 int mdt_nosquash_strlen;
232 struct rw_semaphore mdt_squash_sem;
235 struct rename_stats mdt_rename_stats;
236 struct lu_fid mdt_md_root_fid;
238 /* connection to quota master */
239 struct obd_export *mdt_qmt_exp;
240 /* quota master device associated with this MDT */
241 struct lu_device *mdt_qmt_dev;
243 struct coordinator mdt_coordinator;
246 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
247 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
248 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
249 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
250 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
251 #define MDT_COS_DEFAULT (0)
254 struct lu_object_header mot_header;
255 struct lu_object mot_obj;
258 int mot_ioepoch_count;
260 /* Lock to protect object's IO epoch. */
261 struct mutex mot_ioepoch_mutex;
262 /* Lock to protect create_data */
263 struct mutex mot_lov_mutex;
264 /* Lock to protect lease open.
265 * Lease open acquires write lock; normal open acquires read lock */
266 struct rw_semaphore mot_open_sem;
267 atomic_t mot_lease_count;
268 atomic_t mot_open_count;
271 enum mdt_object_flags {
272 /** SOM attributes are changed. */
273 MOF_SOM_CHANGE = (1 << 0),
275 * The SOM recovery state for mdt object.
276 * This state is an in-memory equivalent of an absent SOM EA, used
277 * instead of invalidating SOM EA while IOEpoch is still opened when
278 * a client eviction occurs or a client fails to obtain SOM attributes.
279 * It indicates that the last IOEpoch holder will need to obtain SOM
280 * attributes under [0;EOF] extent lock to flush all the client's
281 * cached of evicted from MDS clients (but not necessary evicted from
282 * OST) before taking ost attributes.
284 MOF_SOM_RECOV = (1 << 1),
285 /** File has been just created. */
286 MOF_SOM_CREATED = (1 << 2),
287 /** lov object has been created. */
288 MOF_LOV_CREATED = (1 << 3),
291 struct mdt_lock_handle {
292 /* Lock type, reg for cross-ref use or pdo lock. */
296 struct lustre_handle mlh_reg_lh;
297 ldlm_mode_t mlh_reg_mode;
300 struct lustre_handle mlh_pdo_lh;
301 ldlm_mode_t mlh_pdo_mode;
302 unsigned int mlh_pdo_hash;
304 /* Remote regular lock */
305 struct lustre_handle mlh_rreg_lh;
306 ldlm_mode_t mlh_rreg_mode;
310 MDT_LH_PARENT, /* parent lockh */
311 MDT_LH_CHILD, /* child lockh */
312 MDT_LH_OLD, /* old lockh for rename */
313 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
314 MDT_LH_NEW, /* new lockh for rename */
315 MDT_LH_RMT, /* used for return lh to caller */
316 MDT_LH_LOCAL, /* local lock never return to client */
325 struct mdt_reint_record {
326 mdt_reint_t rr_opcode;
327 const struct lustre_handle *rr_handle;
328 const struct lu_fid *rr_fid1;
329 const struct lu_fid *rr_fid2;
334 const void *rr_eadata;
337 const struct llog_cookie *rr_logcookies;
341 enum mdt_reint_flag {
342 MRF_OPEN_TRUNC = 1 << 0,
345 struct mdt_thread_info;
347 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
351 tx_exec_func_t exec_fn;
352 tx_exec_func_t undo_fn;
353 struct dt_object *object;
355 struct update_reply *reply;
360 const struct dt_rec *rec;
361 const struct dt_key *key;
376 struct dt_allocation_hint hint;
377 struct dt_object_format dof;
385 struct ost_body *body;
390 #define TX_MAX_OPS 10
391 struct thandle_exec_args {
392 struct thandle *ta_handle;
393 struct dt_device *ta_dev;
395 struct tx_arg ta_args[TX_MAX_OPS];
396 int ta_argno; /* used args */
400 * Common data shared by mdt-level handlers. This is allocated per-thread to
401 * reduce stack consumption.
403 struct mdt_thread_info {
406 * The following members will be filled explicitly
407 * with specific data in mdt_thread_info_init().
409 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
410 * request handling may migrate from one server thread to another.
412 struct req_capsule *mti_pill;
414 /* although we have export in req, there are cases when it is not
415 * available, e.g. closing files upon export destroy */
416 struct obd_export *mti_exp;
418 * A couple of lock handles.
420 struct mdt_lock_handle mti_lh[MDT_LH_NR];
422 struct mdt_device *mti_mdt;
423 const struct lu_env *mti_env;
424 /* XXX: temporary flag to have healthy mti during OUT calls
425 * to be removed upon moving MDT to the unified target code */
429 * Additional fail id that can be set by handler. Passed to
430 * target_send_reply().
434 /* transaction number of current request */
440 * The following members will be filled expilictly
441 * with zero in mdt_thread_info_init(). These members may be used
448 struct md_attr mti_attr;
450 * Body for "habeo corpus" operations.
452 const struct mdt_body *mti_body;
454 * Host object. This is released at the end of mdt_handler().
456 struct mdt_object *mti_object;
458 * Lock request for "habeo clavis" operations.
460 const struct ldlm_request *mti_dlm_req;
462 __u32 mti_has_trans:1, /* has txn already? */
465 /* opdata for mdt_reint_open(), has the same as
466 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
467 * value onto disk for recovery when mdt_trans_stop_cb() is called.
473 * The following members will be filled explicitly
474 * with zero in mdt_reint_unpack(), because they are only used
475 * by reint requests (including mdt_reint_open()).
479 * reint record. contains information for reint operations.
481 struct mdt_reint_record mti_rr;
483 /** md objects included in operation */
484 struct mdt_object *mti_mos;
485 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
487 * Operation specification (currently create and lookup)
489 struct md_op_spec mti_spec;
493 * The following members will _NOT_ be initialized at all.
494 * DO NOT expect them to contain any valid value.
495 * They should be initialized explicitly by the user themselves.
498 /* XXX: If something is in a union, make sure they do not conflict */
500 struct lu_fid mti_tmp_fid1;
501 struct lu_fid mti_tmp_fid2;
502 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
503 * mdt_rename_lock() */
504 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
507 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
508 char ns_name[48]; /* for mdt_init0() */
509 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
510 struct obd_statfs osfs; /* for mdt_statfs() */
512 /* for mdt_readpage() */
513 struct lu_rdpg mti_rdpg;
514 /* for mdt_sendpage() */
515 struct l_wait_info mti_wait_info;
519 struct md_som_data data;
522 struct dt_object_format mti_update_dof;
523 struct update_reply *mti_update_reply;
524 struct update *mti_update;
525 int mti_update_reply_index;
526 struct obdo mti_obdo;
527 struct dt_object *mti_dt_object;
531 /* IO epoch related stuff. */
532 struct mdt_ioepoch *mti_ioepoch;
533 __u64 mti_replayepoch;
536 struct lu_buf mti_buf;
537 struct lu_buf mti_big_buf;
538 struct lustre_capa_key mti_capa_key;
540 /* Ops object filename */
541 struct lu_name mti_name;
542 /* per-thread values, can be re-used */
545 /* big_lmm buffer was used and must be used in reply */
546 int mti_big_lmm_used;
547 /* should be enough to fit lustre_mdt_attrs */
548 char mti_xattr_buf[128];
549 struct thandle_exec_args mti_handle;
550 struct ldlm_enqueue_info mti_einfo;
553 /* ptlrpc request handler for MDT. All handlers are
554 * grouped into several slices - struct mdt_opc_slice,
555 * and stored in an array - mdt_handlers[].
558 /* The name of this handler. */
560 /* Fail id for this handler, checked at the beginning of this handler*/
562 /* Operation code for this handler */
564 /* flags are listed in enum mdt_handler_flags below. */
566 /* The actual handler function to execute. */
567 int (*mh_act)(struct mdt_thread_info *info);
568 /* Request format for this request. */
569 const struct req_format *mh_fmt;
572 struct mdt_opc_slice {
575 struct mdt_handler *mos_hs;
578 struct cdt_req_progress {
579 struct mutex crp_lock; /**< protect tree */
580 struct interval_node *crp_root; /**< tree to track extent
582 struct interval_node **crp_node; /**< buffer for tree nodes
583 * vector of fixed size
585 int crp_cnt; /**< # of used nodes */
586 int crp_max; /**< # of allocated nodes */
589 struct cdt_agent_req {
590 cfs_list_t car_request_list; /**< to chain all the req. */
591 atomic_t car_refcount; /**< reference counter */
592 __u64 car_compound_id; /**< compound id */
593 __u64 car_flags; /**< request original flags */
594 struct obd_uuid car_uuid; /**< agent doing the req. */
595 __u32 car_archive_id; /**< archive id */
596 int car_canceled; /**< request was canceled */
597 cfs_time_t car_req_start; /**< start time */
598 cfs_time_t car_req_update; /**< last update time */
599 struct hsm_action_item *car_hai; /**< req. to the agent */
600 struct cdt_req_progress car_progress; /**< track data mvt
603 extern struct kmem_cache *mdt_hsm_car_kmem;
606 cfs_list_t ha_list; /**< to chain the agents */
607 struct obd_uuid ha_uuid; /**< agent uuid */
608 __u32 *ha_archive_id; /**< archive id */
609 int ha_archive_cnt; /**< number of archive entries
610 * 0 means any archive */
611 atomic_t ha_requests; /**< current request count */
612 atomic_t ha_success; /**< number of successful
614 atomic_t ha_failure; /**< number of failed actions */
617 struct cdt_restore_handle {
618 cfs_list_t crh_list; /**< to chain the handle */
619 struct lu_fid crh_fid; /**< fid of the object */
620 struct ldlm_extent crh_extent; /**< extent of the restore */
621 struct mdt_lock_handle crh_lh; /**< lock handle */
623 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
625 static inline const struct md_device_operations *
626 mdt_child_ops(struct mdt_device * m)
628 LASSERT(m->mdt_child);
629 return m->mdt_child->md_ops;
632 static inline struct md_object *mdt_object_child(struct mdt_object *o)
635 return lu2md(lu_object_next(&o->mot_obj));
638 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
640 return info->mti_pill ? info->mti_pill->rc_req : NULL;
643 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
645 LASSERT(info->mti_exp);
646 return exp_connect_flags(info->mti_exp);
649 static inline void mdt_object_get(const struct lu_env *env,
650 struct mdt_object *o)
653 lu_object_get(&o->mot_obj);
657 static inline void mdt_object_put(const struct lu_env *env,
658 struct mdt_object *o)
661 lu_object_put(env, &o->mot_obj);
665 static inline int mdt_object_exists(const struct mdt_object *o)
667 return lu_object_exists(&o->mot_obj);
670 static inline int mdt_object_remote(const struct mdt_object *o)
672 return lu_object_remote(&o->mot_obj);
675 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
677 return lu_object_fid(&o->mot_obj);
680 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
682 return mdt->mdt_lu_dev.ld_site;
685 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
687 return &mdt->mdt_seq_site;
690 static inline void mdt_export_evict(struct obd_export *exp)
692 class_fail_export(exp);
693 class_export_put(exp);
696 /* Here we use LVB_TYPE to check dne client, because it is
697 * also landed on 2.4. */
698 static inline int mdt_is_dne_client(struct obd_export *exp)
700 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
703 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
704 void mdt_set_disposition(struct mdt_thread_info *info,
705 struct ldlm_reply *rep, int flag);
706 void mdt_clear_disposition(struct mdt_thread_info *info,
707 struct ldlm_reply *rep, int flag);
709 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
710 ldlm_mode_t lm, const char *name,
713 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
716 int mdt_lock_setup(struct mdt_thread_info *info,
717 struct mdt_object *o,
718 struct mdt_lock_handle *lh);
720 int mdt_check_resent_lock(struct mdt_thread_info *info,
721 struct mdt_object *mo,
722 struct mdt_lock_handle *lhc);
724 int mdt_object_lock(struct mdt_thread_info *,
726 struct mdt_lock_handle *,
729 int mdt_object_lock_try(struct mdt_thread_info *,
731 struct mdt_lock_handle *,
734 void mdt_object_unlock(struct mdt_thread_info *,
736 struct mdt_lock_handle *,
739 struct mdt_object *mdt_object_new(const struct lu_env *,
741 const struct lu_fid *);
742 struct mdt_object *mdt_object_find(const struct lu_env *,
744 const struct lu_fid *);
745 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
746 const struct lu_fid *,
747 struct mdt_lock_handle *,
749 void mdt_object_unlock_put(struct mdt_thread_info *,
751 struct mdt_lock_handle *,
754 void mdt_client_compatibility(struct mdt_thread_info *info);
756 int mdt_remote_object_lock(struct mdt_thread_info *mti,
757 struct mdt_object *o, struct lustre_handle *lh,
758 ldlm_mode_t mode, __u64 ibits);
759 int mdt_close_unpack(struct mdt_thread_info *info);
760 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
761 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
762 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
763 const struct lu_attr *attr, const struct lu_fid *fid);
765 int mdt_getxattr(struct mdt_thread_info *info);
766 int mdt_reint_setxattr(struct mdt_thread_info *info,
767 struct mdt_lock_handle *lh);
769 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
770 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
772 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
773 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
774 struct mdt_lock_handle *lhc);
776 extern void target_recovery_fini(struct obd_device *obd);
777 extern void target_recovery_init(struct lu_target *lut,
778 svc_handler_t handler);
779 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
780 struct obd_device *, struct lustre_sb_info *lsi);
781 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
783 int mdt_export_stats_init(struct obd_device *obd,
784 struct obd_export *exp,
787 int mdt_pin(struct mdt_thread_info* info);
789 int mdt_lock_new_child(struct mdt_thread_info *info,
790 struct mdt_object *o,
791 struct mdt_lock_handle *child_lockh);
793 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
796 int mdt_reint_open(struct mdt_thread_info *info,
797 struct mdt_lock_handle *lhc);
799 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
800 const struct lustre_handle *handle,
804 MDT_IOEPOCH_CLOSED = 0,
805 MDT_IOEPOCH_OPENED = 1,
806 MDT_IOEPOCH_GETATTR = 2,
814 int mdt_attr_get_complex(struct mdt_thread_info *info,
815 struct mdt_object *o, struct md_attr *ma);
816 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
818 int mdt_object_is_som_enabled(struct mdt_object *mo);
819 int mdt_write_get(struct mdt_object *o);
820 void mdt_write_put(struct mdt_object *o);
821 int mdt_write_read(struct mdt_object *o);
822 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
823 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
824 void mdt_mfd_free(struct mdt_file_data *mfd);
825 int mdt_close(struct mdt_thread_info *info);
826 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
827 struct md_attr *ma, int flags);
828 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
830 int mdt_done_writing(struct mdt_thread_info *info);
831 int mdt_fix_reply(struct mdt_thread_info *info);
832 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
833 const struct md_attr *);
834 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
836 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
837 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
838 const void *area, ssize_t len);
840 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
842 int mdt_check_ucred(struct mdt_thread_info *);
843 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
844 int mdt_init_ucred_reint(struct mdt_thread_info *);
845 void mdt_exit_ucred(struct mdt_thread_info *);
846 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
847 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
848 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
850 int mdt_handle_common(struct ptlrpc_request *req,
851 struct mdt_opc_slice *supported);
852 int mdt_connect(struct mdt_thread_info *info);
853 int mdt_disconnect(struct mdt_thread_info *info);
854 int mdt_set_info(struct mdt_thread_info *info);
855 int mdt_get_info(struct mdt_thread_info *info);
856 int mdt_getstatus(struct mdt_thread_info *info);
857 int mdt_getattr(struct mdt_thread_info *info);
858 int mdt_getattr_name(struct mdt_thread_info *info);
859 int mdt_statfs(struct mdt_thread_info *info);
860 int mdt_reint(struct mdt_thread_info *info);
861 int mdt_sync(struct mdt_thread_info *info);
862 int mdt_is_subdir(struct mdt_thread_info *info);
863 int mdt_obd_ping(struct mdt_thread_info *info);
864 int mdt_obd_log_cancel(struct mdt_thread_info *info);
865 int mdt_obd_qc_callback(struct mdt_thread_info *info);
866 int mdt_enqueue(struct mdt_thread_info *info);
867 int mdt_convert(struct mdt_thread_info *info);
868 int mdt_bl_callback(struct mdt_thread_info *info);
869 int mdt_cp_callback(struct mdt_thread_info *info);
870 int mdt_llog_create(struct mdt_thread_info *info);
871 int mdt_llog_destroy(struct mdt_thread_info *info);
872 int mdt_llog_read_header(struct mdt_thread_info *info);
873 int mdt_llog_next_block(struct mdt_thread_info *info);
874 int mdt_llog_prev_block(struct mdt_thread_info *info);
875 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
876 int mdt_readpage(struct mdt_thread_info *info);
877 int mdt_obd_idx_read(struct mdt_thread_info *info);
878 int mdt_tgt_connect(struct tgt_session_info *tsi);
879 void mdt_thread_info_init(struct ptlrpc_request *req,
880 struct mdt_thread_info *mti);
881 void mdt_thread_info_fini(struct mdt_thread_info *mti);
883 extern struct mdt_opc_slice mdt_regular_handlers[];
884 extern struct mdt_opc_slice mdt_seq_handlers[];
885 extern struct mdt_opc_slice mdt_fld_handlers[];
887 int mdt_quotacheck(struct mdt_thread_info *info);
888 int mdt_quotactl(struct mdt_thread_info *info);
889 int mdt_quota_dqacq(struct mdt_thread_info *info);
890 int mdt_swap_layouts(struct mdt_thread_info *info);
892 extern struct lprocfs_vars lprocfs_mds_module_vars[];
893 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
895 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
896 const struct md_hsm *mh);
898 struct mdt_handler *mdt_handler_find(__u32 opc,
899 struct mdt_opc_slice *supported);
901 int mdt_init_sec_level(struct mdt_thread_info *);
902 int mdt_init_idmap(struct mdt_thread_info *);
903 void mdt_cleanup_idmap(struct mdt_export_data *);
904 int mdt_handle_idmap(struct mdt_thread_info *);
905 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
906 struct ptlrpc_user_desc *);
907 void mdt_body_reverse_idmap(struct mdt_thread_info *,
909 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
910 struct mdt_remote_perm *);
911 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
913 static inline struct mdt_device *mdt_dev(struct lu_device *d)
915 return container_of0(d, struct mdt_device, mdt_lu_dev);
918 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
920 struct lu_object *lo;
921 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
923 lo = lu_object_locate(mo->mot_obj.lo_header,
924 mdt->mdt_bottom->dd_lu_dev.ld_type);
929 /* mdt/mdt_identity.c */
930 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
932 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
934 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
936 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
938 void mdt_flush_identity(struct upcall_cache *, int);
940 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
942 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
945 int mdt_hsm_state_get(struct mdt_thread_info *info);
946 int mdt_hsm_state_set(struct mdt_thread_info *info);
947 int mdt_hsm_action(struct mdt_thread_info *info);
948 int mdt_hsm_progress(struct mdt_thread_info *info);
949 int mdt_hsm_ct_register(struct mdt_thread_info *info);
950 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
951 int mdt_hsm_request(struct mdt_thread_info *info);
952 /* mdt/mdt_hsm_cdt_actions.c */
953 extern const struct file_operations mdt_hsm_actions_fops;
954 void dump_llog_agent_req_rec(const char *prefix,
955 const struct llog_agent_req_rec *larr);
956 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
957 llog_cb_t cb, void *data);
958 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
959 __u64 compound_id, __u32 archive_id,
960 __u64 flags, struct hsm_action_item *hai);
961 int mdt_agent_record_update(const struct lu_env *env,
962 struct mdt_device *mdt, __u64 *cookies,
963 int cookies_count, enum agent_req_status status);
964 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
965 struct llog_handle *llh,
966 struct llog_agent_req_rec *larr);
968 /* mdt/mdt_hsm_cdt_agent.c */
969 extern const struct file_operations mdt_hsm_agent_fops;
970 int mdt_hsm_agent_register(struct mdt_thread_info *info,
971 const struct obd_uuid *uuid,
972 int nr_archives, __u32 *archive_num);
973 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
974 const struct obd_uuid *uuid,
976 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
977 const struct obd_uuid *uuid);
978 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
979 int succ_rq, int fail_rq, int new_rq,
980 const struct obd_uuid *uuid);
981 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
982 struct obd_uuid *uuid);
983 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
985 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
986 struct hsm_progress_kernel *pgs);
987 /* mdt/mdt_hsm_cdt_client.c */
988 int mdt_hsm_add_actions(struct mdt_thread_info *info,
989 struct hsm_action_list *hal, __u64 *compound_id);
990 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
991 struct hsm_action_list *hal);
992 int mdt_hsm_get_running(struct mdt_thread_info *mti,
993 struct hsm_action_list *hal);
994 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
995 const struct lu_fid *fid);
996 /* mdt/mdt_hsm_cdt_requests.c */
997 extern const struct file_operations mdt_hsm_active_requests_fops;
998 void dump_requests(char *prefix, struct coordinator *cdt);
999 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
1000 __u64 flags, struct obd_uuid *uuid,
1001 struct hsm_action_item *hai);
1002 void mdt_cdt_free_request(struct cdt_agent_req *car);
1003 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
1004 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
1006 const struct lu_fid *fid);
1007 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
1008 void mdt_cdt_get_request(struct cdt_agent_req *car);
1009 void mdt_cdt_put_request(struct cdt_agent_req *car);
1010 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
1011 const struct hsm_progress_kernel *pgs);
1012 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
1013 /* mdt/mdt_coordinator.c */
1014 void mdt_hsm_dump_hal(int level, const char *prefix,
1015 struct hsm_action_list *hal);
1016 /* coordinator management */
1017 int mdt_hsm_cdt_init(struct mdt_device *mdt);
1018 int mdt_hsm_cdt_start(struct mdt_device *mdt);
1019 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
1020 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
1021 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
1023 /* coordinator control /proc interface */
1024 int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
1025 unsigned long count, void *data);
1026 int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
1027 int count, int *eof, void *data);
1028 int hsm_cdt_procfs_init(struct mdt_device *mdt);
1029 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
1030 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
1031 /* md_hsm helpers */
1032 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1033 const struct lu_fid *fid,
1034 struct md_hsm *hsm);
1035 /* actions/request helpers */
1036 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1037 struct hsm_action_list *hal, struct obd_uuid *uuid);
1038 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1039 const int hal_an, const __u64 rq_flags,
1040 const struct md_hsm *hsm);
1041 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1042 struct hsm_progress_kernel *pgs,
1043 const int update_record);
1045 extern struct lu_context_key mdt_thread_key;
1046 /* debug issues helper starts here*/
1047 static inline int mdt_fail_write(const struct lu_env *env,
1048 struct dt_device *dd, int id)
1050 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1051 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1053 return dd->dd_ops->dt_ro(env, dd);
1054 /* We set FAIL_ONCE because we never "un-fail" a device */
1060 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1062 return &req->rq_export->exp_mdt_data;
1065 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1066 struct mdt_lock_handle *lhc);
1067 static inline int mdt_check_resent(struct mdt_thread_info *info,
1068 mdt_reconstruct_t reconstruct,
1069 struct mdt_lock_handle *lhc)
1071 struct ptlrpc_request *req = mdt_info_req(info);
1074 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1075 if (req_xid_is_last(req)) {
1076 reconstruct(info, lhc);
1079 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
1080 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
1085 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1086 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1088 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1090 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1093 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1095 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1096 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1097 struct lustre_handle *lh,
1099 ldlm_policy_data_t *policy,
1100 const struct ldlm_res_id *res_id,
1101 __u64 flags, const __u64 *client_cookie)
1105 LASSERT(ns != NULL);
1106 LASSERT(lh != NULL);
1108 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1109 mode, &flags, mdt_blocking_ast,
1110 ldlm_completion_ast, NULL, NULL, 0,
1111 LVB_T_NONE, client_cookie, lh);
1112 return rc == ELDLM_OK ? 0 : -EIO;
1115 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1118 ldlm_lock_decref(lh, mode);
1121 extern mdl_mode_t mdt_mdl_lock_modes[];
1122 extern ldlm_mode_t mdt_dlm_lock_modes[];
1124 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1126 LASSERT(IS_PO2(mode));
1127 return mdt_mdl_lock_modes[mode];
1130 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1132 LASSERT(IS_PO2(mode));
1133 return mdt_dlm_lock_modes[mode];
1137 extern struct ldlm_valblock_ops mdt_lvbo;
1139 static inline struct lu_name *mdt_name(const struct lu_env *env,
1140 char *name, int namelen)
1142 struct lu_name *lname;
1143 struct mdt_thread_info *mti;
1145 LASSERT(namelen > 0);
1146 /* trailing '\0' in buffer */
1147 LASSERT(name[namelen] == '\0');
1149 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
1150 lname = &mti->mti_name;
1151 lname->ln_name = name;
1152 lname->ln_namelen = namelen;
1156 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
1157 struct lu_name *slname)
1162 tlname->ln_name = slname->ln_name;
1163 tlname->ln_namelen = slname->ln_namelen;
1167 void mdt_enable_cos(struct mdt_device *, int);
1168 int mdt_cos_is_enabled(struct mdt_device *);
1186 LPROC_MDT_SAMEDIR_RENAME,
1187 LPROC_MDT_CROSSDIR_RENAME,
1190 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1191 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1192 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1193 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1194 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1195 void mdt_procfs_fini(struct mdt_device *mdt);
1197 /* lustre/mdt_mdt_lproc.c */
1198 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1200 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1201 struct mdt_device *mdt,
1202 struct ptlrpc_request *req,
1203 struct mdt_object *src, struct mdt_object *tgt);
1206 int mdt_ck_thread_start(struct mdt_device *mdt);
1207 void mdt_ck_thread_stop(struct mdt_device *mdt);
1208 void mdt_ck_timer_callback(unsigned long castmeharder);
1209 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1210 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1211 const struct lu_fid *fid, struct lustre_capa *capa);
1212 void mdt_dump_capainfo(struct mdt_thread_info *info);
1214 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1216 return mdt->mdt_lu_dev.ld_obd;
1219 extern const struct lu_device_operations mdt_lu_ops;
1221 static inline int lu_device_is_mdt(struct lu_device *d)
1223 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1226 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1228 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1229 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1230 return container_of0(d, struct mdt_device, mdt_lu_dev);
1233 static inline char *mdt_obd_name(struct mdt_device *mdt)
1235 return mdt->mdt_lu_dev.ld_obd->obd_name;
1238 int mds_mod_init(void);
1239 void mds_mod_exit(void);
1241 #endif /* __KERNEL__ */