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_delay; /**< request grace
134 cfs_time_t cdt_timeout; /**< request timeout */
135 __u32 cdt_default_archive_id; /** archive id used
138 __u64 cdt_max_request; /**< 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
150 /* mdt state flag bits */
151 #define MDT_FL_CFGLOG 0
152 #define MDT_FL_SYNCED 1
156 struct lu_device mdt_lu_dev;
157 struct seq_server_site mdt_seq_site;
158 /* DLM name-space for meta-data locks maintained by this server */
159 struct ldlm_namespace *mdt_namespace;
160 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
161 struct ptlrpc_client *mdt_ldlm_client;
162 /* underlying device */
163 struct obd_export *mdt_child_exp;
164 struct md_device *mdt_child;
165 struct dt_device *mdt_bottom;
166 struct obd_export *mdt_bottom_exp;
168 struct lu_target mdt_lut;
170 * Options bit-fields.
173 unsigned int mo_user_xattr:1,
181 /* mdt state flags */
182 unsigned long mdt_state;
183 /* lock to protect IOepoch */
184 spinlock_t mdt_ioepoch_lock;
187 /* transaction callbacks */
188 struct dt_txn_callback mdt_txn_cb;
190 /* these values should be updated from lov if necessary.
191 * or should be placed somewhere else. */
193 int mdt_max_cookiesize;
197 struct upcall_cache *mdt_identity_cache;
200 rwlock_t mdt_sptlrpc_lock;
201 struct sptlrpc_rule_set mdt_sptlrpc_rset;
203 /* capability keys */
204 unsigned long mdt_capa_timeout;
206 struct dt_object *mdt_ck_obj;
207 unsigned long mdt_ck_timeout;
208 unsigned long mdt_ck_expiry;
209 struct timer_list mdt_ck_timer;
210 struct ptlrpc_thread mdt_ck_thread;
211 struct lustre_capa_key mdt_capa_keys[2];
212 unsigned int mdt_capa_conf:1,
214 /* Enable remote dir on non-MDT0 */
215 mdt_enable_remote_dir:1;
217 gid_t mdt_enable_remote_dir_gid;
218 /* statfs optimization: we cache a bit */
219 struct obd_statfs mdt_osfs;
221 spinlock_t mdt_osfs_lock;
224 uid_t mdt_squash_uid;
225 gid_t mdt_squash_gid;
226 cfs_list_t mdt_nosquash_nids;
227 char *mdt_nosquash_str;
228 int mdt_nosquash_strlen;
229 struct rw_semaphore mdt_squash_sem;
232 struct rename_stats mdt_rename_stats;
233 struct lu_fid mdt_md_root_fid;
235 /* connection to quota master */
236 struct obd_export *mdt_qmt_exp;
237 /* quota master device associated with this MDT */
238 struct lu_device *mdt_qmt_dev;
240 struct coordinator mdt_coordinator;
243 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
244 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
245 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
246 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
247 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
248 #define MDT_COS_DEFAULT (0)
251 struct lu_object_header mot_header;
252 struct lu_object mot_obj;
255 int mot_ioepoch_count;
257 /* Lock to protect object's IO epoch. */
258 struct mutex mot_ioepoch_mutex;
259 /* Lock to protect create_data */
260 struct mutex mot_lov_mutex;
261 /* Lock to protect lease open.
262 * Lease open acquires write lock; normal open acquires read lock */
263 struct rw_semaphore mot_open_sem;
264 atomic_t mot_lease_count;
265 atomic_t mot_open_count;
268 enum mdt_object_flags {
269 /** SOM attributes are changed. */
270 MOF_SOM_CHANGE = (1 << 0),
272 * The SOM recovery state for mdt object.
273 * This state is an in-memory equivalent of an absent SOM EA, used
274 * instead of invalidating SOM EA while IOEpoch is still opened when
275 * a client eviction occurs or a client fails to obtain SOM attributes.
276 * It indicates that the last IOEpoch holder will need to obtain SOM
277 * attributes under [0;EOF] extent lock to flush all the client's
278 * cached of evicted from MDS clients (but not necessary evicted from
279 * OST) before taking ost attributes.
281 MOF_SOM_RECOV = (1 << 1),
282 /** File has been just created. */
283 MOF_SOM_CREATED = (1 << 2),
284 /** lov object has been created. */
285 MOF_LOV_CREATED = (1 << 3),
288 struct mdt_lock_handle {
289 /* Lock type, reg for cross-ref use or pdo lock. */
293 struct lustre_handle mlh_reg_lh;
294 ldlm_mode_t mlh_reg_mode;
297 struct lustre_handle mlh_pdo_lh;
298 ldlm_mode_t mlh_pdo_mode;
299 unsigned int mlh_pdo_hash;
301 /* Remote regular lock */
302 struct lustre_handle mlh_rreg_lh;
303 ldlm_mode_t mlh_rreg_mode;
307 MDT_LH_PARENT, /* parent lockh */
308 MDT_LH_CHILD, /* child lockh */
309 MDT_LH_OLD, /* old lockh for rename */
310 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
311 MDT_LH_NEW, /* new lockh for rename */
312 MDT_LH_RMT, /* used for return lh to caller */
313 MDT_LH_LOCAL, /* local lock never return to client */
322 struct mdt_reint_record {
323 mdt_reint_t rr_opcode;
324 const struct lustre_handle *rr_handle;
325 const struct lu_fid *rr_fid1;
326 const struct lu_fid *rr_fid2;
331 const void *rr_eadata;
334 const struct llog_cookie *rr_logcookies;
338 enum mdt_reint_flag {
339 MRF_OPEN_TRUNC = 1 << 0,
342 struct mdt_thread_info;
344 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
348 tx_exec_func_t exec_fn;
349 tx_exec_func_t undo_fn;
350 struct dt_object *object;
352 struct update_reply *reply;
357 const struct dt_rec *rec;
358 const struct dt_key *key;
373 struct dt_allocation_hint hint;
374 struct dt_object_format dof;
382 struct ost_body *body;
387 #define TX_MAX_OPS 10
388 struct thandle_exec_args {
389 struct thandle *ta_handle;
390 struct dt_device *ta_dev;
392 struct tx_arg ta_args[TX_MAX_OPS];
393 int ta_argno; /* used args */
397 * Common data shared by mdt-level handlers. This is allocated per-thread to
398 * reduce stack consumption.
400 struct mdt_thread_info {
403 * The following members will be filled explicitly
404 * with specific data in mdt_thread_info_init().
406 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
407 * request handling may migrate from one server thread to another.
409 struct req_capsule *mti_pill;
411 /* although we have export in req, there are cases when it is not
412 * available, e.g. closing files upon export destroy */
413 struct obd_export *mti_exp;
415 * A couple of lock handles.
417 struct mdt_lock_handle mti_lh[MDT_LH_NR];
419 struct mdt_device *mti_mdt;
420 const struct lu_env *mti_env;
421 /* XXX: temporary flag to have healthy mti during OUT calls
422 * to be removed upon moving MDT to the unified target code */
426 * Additional fail id that can be set by handler. Passed to
427 * target_send_reply().
431 /* transaction number of current request */
437 * The following members will be filled expilictly
438 * with zero in mdt_thread_info_init(). These members may be used
445 struct md_attr mti_attr;
447 * Body for "habeo corpus" operations.
449 const struct mdt_body *mti_body;
451 * Host object. This is released at the end of mdt_handler().
453 struct mdt_object *mti_object;
455 * Lock request for "habeo clavis" operations.
457 const struct ldlm_request *mti_dlm_req;
459 __u32 mti_has_trans:1, /* has txn already? */
462 /* opdata for mdt_reint_open(), has the same as
463 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
464 * value onto disk for recovery when mdt_trans_stop_cb() is called.
470 * The following members will be filled explicitly
471 * with zero in mdt_reint_unpack(), because they are only used
472 * by reint requests (including mdt_reint_open()).
476 * reint record. contains information for reint operations.
478 struct mdt_reint_record mti_rr;
480 /** md objects included in operation */
481 struct mdt_object *mti_mos;
482 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
484 * Operation specification (currently create and lookup)
486 struct md_op_spec mti_spec;
490 * The following members will _NOT_ be initialized at all.
491 * DO NOT expect them to contain any valid value.
492 * They should be initialized explicitly by the user themselves.
495 /* XXX: If something is in a union, make sure they do not conflict */
497 struct lu_fid mti_tmp_fid1;
498 struct lu_fid mti_tmp_fid2;
499 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
500 * mdt_rename_lock() */
501 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
504 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
505 char ns_name[48]; /* for mdt_init0() */
506 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
507 struct obd_statfs osfs; /* for mdt_statfs() */
509 /* for mdt_readpage() */
510 struct lu_rdpg mti_rdpg;
511 /* for mdt_sendpage() */
512 struct l_wait_info mti_wait_info;
516 struct md_som_data data;
519 struct dt_object_format mti_update_dof;
520 struct update_reply *mti_update_reply;
521 struct update *mti_update;
522 int mti_update_reply_index;
523 struct obdo mti_obdo;
524 struct dt_object *mti_dt_object;
528 /* IO epoch related stuff. */
529 struct mdt_ioepoch *mti_ioepoch;
530 __u64 mti_replayepoch;
533 struct lu_buf mti_buf;
534 struct lu_buf mti_big_buf;
535 struct lustre_capa_key mti_capa_key;
537 /* Ops object filename */
538 struct lu_name mti_name;
539 /* per-thread values, can be re-used */
542 /* big_lmm buffer was used and must be used in reply */
543 int mti_big_lmm_used;
544 /* should be enough to fit lustre_mdt_attrs */
545 char mti_xattr_buf[128];
546 struct thandle_exec_args mti_handle;
547 struct ldlm_enqueue_info mti_einfo;
550 /* ptlrpc request handler for MDT. All handlers are
551 * grouped into several slices - struct mdt_opc_slice,
552 * and stored in an array - mdt_handlers[].
555 /* The name of this handler. */
557 /* Fail id for this handler, checked at the beginning of this handler*/
559 /* Operation code for this handler */
561 /* flags are listed in enum mdt_handler_flags below. */
563 /* The actual handler function to execute. */
564 int (*mh_act)(struct mdt_thread_info *info);
565 /* Request format for this request. */
566 const struct req_format *mh_fmt;
569 struct mdt_opc_slice {
572 struct mdt_handler *mos_hs;
575 struct cdt_req_progress {
576 struct mutex crp_lock; /**< protect tree */
577 struct interval_node *crp_root; /**< tree to track extent
579 struct interval_node **crp_node; /**< buffer for tree nodes
580 * vector of fixed size
582 int crp_cnt; /**< # of used nodes */
583 int crp_max; /**< # of allocated nodes */
586 struct cdt_agent_req {
587 cfs_list_t car_request_list; /**< to chain all the req. */
588 atomic_t car_refcount; /**< reference counter */
589 __u64 car_compound_id; /**< compound id */
590 __u64 car_flags; /**< request original flags */
591 struct obd_uuid car_uuid; /**< agent doing the req. */
592 __u32 car_archive_id; /**< archive id */
593 int car_canceled; /**< request was canceled */
594 cfs_time_t car_req_start; /**< start time */
595 cfs_time_t car_req_update; /**< last update time */
596 struct hsm_action_item *car_hai; /**< req. to the agent */
597 struct cdt_req_progress car_progress; /**< track data mvt
600 extern struct kmem_cache *mdt_hsm_car_kmem;
603 cfs_list_t ha_list; /**< to chain the agents */
604 struct obd_uuid ha_uuid; /**< agent uuid */
605 __u32 *ha_archive_id; /**< archive id */
606 int ha_archive_cnt; /**< number of archive entries
607 * 0 means any archive */
608 atomic_t ha_requests; /**< current request count */
609 atomic_t ha_success; /**< number of successful
611 atomic_t ha_failure; /**< number of failed actions */
614 struct cdt_restore_handle {
615 cfs_list_t crh_list; /**< to chain the handle */
616 struct lu_fid crh_fid; /**< fid of the object */
617 struct ldlm_extent crh_extent; /**< extent of the restore */
618 struct mdt_lock_handle crh_lh; /**< lock handle */
620 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
622 static inline const struct md_device_operations *
623 mdt_child_ops(struct mdt_device * m)
625 LASSERT(m->mdt_child);
626 return m->mdt_child->md_ops;
629 static inline struct md_object *mdt_object_child(struct mdt_object *o)
632 return lu2md(lu_object_next(&o->mot_obj));
635 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
637 return info->mti_pill ? info->mti_pill->rc_req : NULL;
640 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
642 LASSERT(info->mti_exp);
643 return exp_connect_flags(info->mti_exp);
646 static inline void mdt_object_get(const struct lu_env *env,
647 struct mdt_object *o)
650 lu_object_get(&o->mot_obj);
654 static inline void mdt_object_put(const struct lu_env *env,
655 struct mdt_object *o)
658 lu_object_put(env, &o->mot_obj);
662 static inline int mdt_object_exists(const struct mdt_object *o)
664 return lu_object_exists(&o->mot_obj);
667 static inline int mdt_object_remote(const struct mdt_object *o)
669 return lu_object_remote(&o->mot_obj);
672 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
674 return lu_object_fid(&o->mot_obj);
677 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
679 return mdt->mdt_lu_dev.ld_site;
682 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
684 return &mdt->mdt_seq_site;
687 static inline void mdt_export_evict(struct obd_export *exp)
689 class_fail_export(exp);
690 class_export_put(exp);
693 /* Here we use LVB_TYPE to check dne client, because it is
694 * also landed on 2.4. */
695 static inline int mdt_is_dne_client(struct obd_export *exp)
697 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
700 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
701 void mdt_set_disposition(struct mdt_thread_info *info,
702 struct ldlm_reply *rep, int flag);
703 void mdt_clear_disposition(struct mdt_thread_info *info,
704 struct ldlm_reply *rep, int flag);
706 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
707 ldlm_mode_t lm, const char *name,
710 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
713 int mdt_lock_setup(struct mdt_thread_info *info,
714 struct mdt_object *o,
715 struct mdt_lock_handle *lh);
717 int mdt_object_lock(struct mdt_thread_info *,
719 struct mdt_lock_handle *,
722 int mdt_object_lock_try(struct mdt_thread_info *,
724 struct mdt_lock_handle *,
727 void mdt_object_unlock(struct mdt_thread_info *,
729 struct mdt_lock_handle *,
732 struct mdt_object *mdt_object_new(const struct lu_env *,
734 const struct lu_fid *);
735 struct mdt_object *mdt_object_find(const struct lu_env *,
737 const struct lu_fid *);
738 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
739 const struct lu_fid *,
740 struct mdt_lock_handle *,
742 void mdt_object_unlock_put(struct mdt_thread_info *,
744 struct mdt_lock_handle *,
747 void mdt_client_compatibility(struct mdt_thread_info *info);
749 int mdt_remote_object_lock(struct mdt_thread_info *mti,
750 struct mdt_object *o, struct lustre_handle *lh,
751 ldlm_mode_t mode, __u64 ibits);
752 int mdt_close_unpack(struct mdt_thread_info *info);
753 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
754 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
755 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
756 const struct lu_attr *attr, const struct lu_fid *fid);
758 int mdt_getxattr(struct mdt_thread_info *info);
759 int mdt_reint_setxattr(struct mdt_thread_info *info,
760 struct mdt_lock_handle *lh);
762 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
763 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
765 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
766 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
767 struct mdt_lock_handle *lhc);
769 extern void target_recovery_fini(struct obd_device *obd);
770 extern void target_recovery_init(struct lu_target *lut,
771 svc_handler_t handler);
772 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
773 struct obd_device *, struct lustre_sb_info *lsi);
774 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
776 int mdt_export_stats_init(struct obd_device *obd,
777 struct obd_export *exp,
780 int mdt_pin(struct mdt_thread_info* info);
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_attr_get_complex(struct mdt_thread_info *info,
808 struct mdt_object *o, struct md_attr *ma);
809 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
811 int mdt_object_is_som_enabled(struct mdt_object *mo);
812 int mdt_write_get(struct mdt_object *o);
813 void mdt_write_put(struct mdt_object *o);
814 int mdt_write_read(struct mdt_object *o);
815 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
816 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
817 void mdt_mfd_free(struct mdt_file_data *mfd);
818 int mdt_close(struct mdt_thread_info *info);
819 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
820 struct md_attr *ma, int flags);
821 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
823 int mdt_done_writing(struct mdt_thread_info *info);
824 int mdt_fix_reply(struct mdt_thread_info *info);
825 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
826 const struct md_attr *);
827 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
829 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
830 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
831 const void *area, ssize_t len);
833 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
835 int mdt_check_ucred(struct mdt_thread_info *);
836 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
837 int mdt_init_ucred_reint(struct mdt_thread_info *);
838 void mdt_exit_ucred(struct mdt_thread_info *);
839 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
840 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
841 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
843 int mdt_handle_common(struct ptlrpc_request *req,
844 struct mdt_opc_slice *supported);
845 int mdt_connect(struct mdt_thread_info *info);
846 int mdt_disconnect(struct mdt_thread_info *info);
847 int mdt_set_info(struct mdt_thread_info *info);
848 int mdt_get_info(struct mdt_thread_info *info);
849 int mdt_getstatus(struct mdt_thread_info *info);
850 int mdt_getattr(struct mdt_thread_info *info);
851 int mdt_getattr_name(struct mdt_thread_info *info);
852 int mdt_statfs(struct mdt_thread_info *info);
853 int mdt_reint(struct mdt_thread_info *info);
854 int mdt_sync(struct mdt_thread_info *info);
855 int mdt_is_subdir(struct mdt_thread_info *info);
856 int mdt_obd_ping(struct mdt_thread_info *info);
857 int mdt_obd_log_cancel(struct mdt_thread_info *info);
858 int mdt_obd_qc_callback(struct mdt_thread_info *info);
859 int mdt_enqueue(struct mdt_thread_info *info);
860 int mdt_convert(struct mdt_thread_info *info);
861 int mdt_bl_callback(struct mdt_thread_info *info);
862 int mdt_cp_callback(struct mdt_thread_info *info);
863 int mdt_llog_create(struct mdt_thread_info *info);
864 int mdt_llog_destroy(struct mdt_thread_info *info);
865 int mdt_llog_read_header(struct mdt_thread_info *info);
866 int mdt_llog_next_block(struct mdt_thread_info *info);
867 int mdt_llog_prev_block(struct mdt_thread_info *info);
868 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
869 int mdt_readpage(struct mdt_thread_info *info);
870 int mdt_obd_idx_read(struct mdt_thread_info *info);
871 int mdt_tgt_connect(struct tgt_session_info *tsi);
872 void mdt_thread_info_init(struct ptlrpc_request *req,
873 struct mdt_thread_info *mti);
874 void mdt_thread_info_fini(struct mdt_thread_info *mti);
876 extern struct mdt_opc_slice mdt_regular_handlers[];
877 extern struct mdt_opc_slice mdt_seq_handlers[];
878 extern struct mdt_opc_slice mdt_fld_handlers[];
880 int mdt_quotacheck(struct mdt_thread_info *info);
881 int mdt_quotactl(struct mdt_thread_info *info);
882 int mdt_quota_dqacq(struct mdt_thread_info *info);
883 int mdt_swap_layouts(struct mdt_thread_info *info);
885 extern struct lprocfs_vars lprocfs_mds_module_vars[];
886 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
888 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
889 const struct md_hsm *mh);
891 struct mdt_handler *mdt_handler_find(__u32 opc,
892 struct mdt_opc_slice *supported);
894 int mdt_init_sec_level(struct mdt_thread_info *);
895 int mdt_init_idmap(struct mdt_thread_info *);
896 void mdt_cleanup_idmap(struct mdt_export_data *);
897 int mdt_handle_idmap(struct mdt_thread_info *);
898 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
899 struct ptlrpc_user_desc *);
900 void mdt_body_reverse_idmap(struct mdt_thread_info *,
902 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
903 struct mdt_remote_perm *);
904 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
906 static inline struct mdt_device *mdt_dev(struct lu_device *d)
908 return container_of0(d, struct mdt_device, mdt_lu_dev);
911 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
913 struct lu_object *lo;
914 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
916 lo = lu_object_locate(mo->mot_obj.lo_header,
917 mdt->mdt_bottom->dd_lu_dev.ld_type);
922 /* mdt/mdt_identity.c */
923 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
925 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
927 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
929 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
931 void mdt_flush_identity(struct upcall_cache *, int);
933 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
935 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
938 int mdt_hsm_state_get(struct mdt_thread_info *info);
939 int mdt_hsm_state_set(struct mdt_thread_info *info);
940 int mdt_hsm_action(struct mdt_thread_info *info);
941 int mdt_hsm_progress(struct mdt_thread_info *info);
942 int mdt_hsm_ct_register(struct mdt_thread_info *info);
943 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
944 int mdt_hsm_request(struct mdt_thread_info *info);
945 /* mdt/mdt_hsm_cdt_actions.c */
946 extern const struct file_operations mdt_agent_actions_fops;
947 void dump_llog_agent_req_rec(const char *prefix,
948 const struct llog_agent_req_rec *larr);
949 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
950 llog_cb_t cb, void *data);
951 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
952 __u64 compound_id, __u32 archive_id,
953 __u64 flags, struct hsm_action_item *hai);
954 int mdt_agent_record_update(const struct lu_env *env,
955 struct mdt_device *mdt, __u64 *cookies,
956 int cookies_count, enum agent_req_status status);
957 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
958 struct llog_handle *llh,
959 struct llog_agent_req_rec *larr);
961 /* mdt/mdt_hsm_cdt_agent.c */
962 extern const struct file_operations mdt_hsm_agent_fops;
963 int mdt_hsm_agent_register(struct mdt_thread_info *info,
964 const struct obd_uuid *uuid,
965 int nr_archives, __u32 *archive_num);
966 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
967 const struct obd_uuid *uuid,
969 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
970 const struct obd_uuid *uuid);
971 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
972 int succ_rq, int fail_rq, int new_rq,
973 const struct obd_uuid *uuid);
974 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
975 struct obd_uuid *uuid);
976 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
978 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
979 struct hsm_progress_kernel *pgs);
980 /* mdt/mdt_hsm_cdt_client.c */
981 int mdt_hsm_add_actions(struct mdt_thread_info *info,
982 struct hsm_action_list *hal, __u64 *compound_id);
983 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
984 struct hsm_action_list *hal);
985 int mdt_hsm_get_running(struct mdt_thread_info *mti,
986 struct hsm_action_list *hal);
987 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
988 const struct lu_fid *fid);
989 /* mdt/mdt_hsm_cdt_requests.c */
990 extern const struct file_operations mdt_hsm_request_fops;
991 void dump_requests(char *prefix, struct coordinator *cdt);
992 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
993 __u64 flags, struct obd_uuid *uuid,
994 struct hsm_action_item *hai);
995 void mdt_cdt_free_request(struct cdt_agent_req *car);
996 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
997 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
999 const struct lu_fid *fid);
1000 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
1001 void mdt_cdt_get_request(struct cdt_agent_req *car);
1002 void mdt_cdt_put_request(struct cdt_agent_req *car);
1003 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
1004 const struct hsm_progress_kernel *pgs);
1005 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
1006 /* mdt/mdt_coordinator.c */
1007 void mdt_hsm_dump_hal(int level, const char *prefix,
1008 struct hsm_action_list *hal);
1009 /* coordinator management */
1010 int mdt_hsm_cdt_init(struct mdt_device *mdt);
1011 int mdt_hsm_cdt_start(struct mdt_device *mdt);
1012 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
1013 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
1014 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
1016 /* coordinator control /proc interface */
1017 int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
1018 unsigned long count, void *data);
1019 int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
1020 int count, int *eof, void *data);
1021 int hsm_cdt_procfs_init(struct mdt_device *mdt);
1022 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
1023 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
1024 /* md_hsm helpers */
1025 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1026 const struct lu_fid *fid,
1027 struct md_hsm *hsm);
1028 /* actions/request helpers */
1029 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1030 struct hsm_action_list *hal, struct obd_uuid *uuid);
1031 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1032 const int hal_an, const __u64 rq_flags,
1033 const struct md_hsm *hsm);
1034 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1035 struct hsm_progress_kernel *pgs,
1036 const int update_record);
1038 extern struct lu_context_key mdt_thread_key;
1039 /* debug issues helper starts here*/
1040 static inline int mdt_fail_write(const struct lu_env *env,
1041 struct dt_device *dd, int id)
1043 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1044 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1046 return dd->dd_ops->dt_ro(env, dd);
1047 /* We set FAIL_ONCE because we never "un-fail" a device */
1053 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1055 return &req->rq_export->exp_mdt_data;
1058 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1059 struct mdt_lock_handle *lhc);
1060 static inline int mdt_check_resent(struct mdt_thread_info *info,
1061 mdt_reconstruct_t reconstruct,
1062 struct mdt_lock_handle *lhc)
1064 struct ptlrpc_request *req = mdt_info_req(info);
1067 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1068 if (req_xid_is_last(req)) {
1069 reconstruct(info, lhc);
1072 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
1073 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
1078 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1079 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1081 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1083 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1086 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1088 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1089 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1090 struct lustre_handle *lh,
1092 ldlm_policy_data_t *policy,
1093 const struct ldlm_res_id *res_id,
1094 __u64 flags, const __u64 *client_cookie)
1098 LASSERT(ns != NULL);
1099 LASSERT(lh != NULL);
1101 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1102 mode, &flags, mdt_blocking_ast,
1103 ldlm_completion_ast, NULL, NULL, 0,
1104 LVB_T_NONE, client_cookie, lh);
1105 return rc == ELDLM_OK ? 0 : -EIO;
1108 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1111 ldlm_lock_decref(lh, mode);
1114 extern mdl_mode_t mdt_mdl_lock_modes[];
1115 extern ldlm_mode_t mdt_dlm_lock_modes[];
1117 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1119 LASSERT(IS_PO2(mode));
1120 return mdt_mdl_lock_modes[mode];
1123 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1125 LASSERT(IS_PO2(mode));
1126 return mdt_dlm_lock_modes[mode];
1130 extern struct ldlm_valblock_ops mdt_lvbo;
1132 static inline struct lu_name *mdt_name(const struct lu_env *env,
1133 char *name, int namelen)
1135 struct lu_name *lname;
1136 struct mdt_thread_info *mti;
1138 LASSERT(namelen > 0);
1139 /* trailing '\0' in buffer */
1140 LASSERT(name[namelen] == '\0');
1142 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
1143 lname = &mti->mti_name;
1144 lname->ln_name = name;
1145 lname->ln_namelen = namelen;
1149 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
1150 struct lu_name *slname)
1155 tlname->ln_name = slname->ln_name;
1156 tlname->ln_namelen = slname->ln_namelen;
1160 void mdt_enable_cos(struct mdt_device *, int);
1161 int mdt_cos_is_enabled(struct mdt_device *);
1179 LPROC_MDT_SAMEDIR_RENAME,
1180 LPROC_MDT_CROSSDIR_RENAME,
1183 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1184 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1185 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1186 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1187 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1188 void mdt_procfs_fini(struct mdt_device *mdt);
1189 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1190 struct mdt_device *mdt,
1191 struct ptlrpc_request *req,
1192 struct mdt_object *src, struct mdt_object *tgt);
1195 int mdt_ck_thread_start(struct mdt_device *mdt);
1196 void mdt_ck_thread_stop(struct mdt_device *mdt);
1197 void mdt_ck_timer_callback(unsigned long castmeharder);
1198 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1199 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1200 const struct lu_fid *fid, struct lustre_capa *capa);
1201 void mdt_dump_capainfo(struct mdt_thread_info *info);
1203 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1205 return mdt->mdt_lu_dev.ld_obd;
1208 extern const struct lu_device_operations mdt_lu_ops;
1210 static inline int lu_device_is_mdt(struct lu_device *d)
1212 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1215 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1217 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1218 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1219 return container_of0(d, struct mdt_device, mdt_lu_dev);
1222 static inline char *mdt_obd_name(struct mdt_device *mdt)
1224 return mdt->mdt_lu_dev.ld_obd->obd_name;
1227 int mds_mod_init(void);
1228 void mds_mod_exit(void);
1230 #endif /* __KERNEL__ */