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;
323 struct lu_name rr_name;
324 struct lu_name rr_tgt_name;
325 const void *rr_eadata;
330 enum mdt_reint_flag {
331 MRF_OPEN_TRUNC = 1 << 0,
335 * Common data shared by mdt-level handlers. This is allocated per-thread to
336 * reduce stack consumption.
338 struct mdt_thread_info {
341 * The following members will be filled explicitly
342 * with specific data in mdt_thread_info_init().
344 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
345 * request handling may migrate from one server thread to another.
347 struct req_capsule *mti_pill;
349 /* although we have export in req, there are cases when it is not
350 * available, e.g. closing files upon export destroy */
351 struct obd_export *mti_exp;
353 * A couple of lock handles.
355 struct mdt_lock_handle mti_lh[MDT_LH_NR];
357 struct mdt_device *mti_mdt;
358 const struct lu_env *mti_env;
360 /* transaction number of current request */
366 * The following members will be filled expilictly
367 * with zero in mdt_thread_info_init(). These members may be used
374 struct md_attr mti_attr;
376 * Body for "habeo corpus" operations.
378 const struct mdt_body *mti_body;
380 * Host object. This is released at the end of mdt_handler().
382 struct mdt_object *mti_object;
384 * Lock request for "habeo clavis" operations.
386 const struct ldlm_request *mti_dlm_req;
388 __u32 mti_has_trans:1, /* has txn already? */
391 /* opdata for mdt_reint_open(), has the same as
392 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
393 * value onto disk for recovery when mdt_trans_stop_cb() is called.
399 * The following members will be filled explicitly
400 * with zero in mdt_reint_unpack(), because they are only used
401 * by reint requests (including mdt_reint_open()).
405 * reint record. contains information for reint operations.
407 struct mdt_reint_record mti_rr;
409 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
411 * Operation specification (currently create and lookup)
413 struct md_op_spec mti_spec;
417 * The following members will _NOT_ be initialized at all.
418 * DO NOT expect them to contain any valid value.
419 * They should be initialized explicitly by the user themselves.
422 /* XXX: If something is in a union, make sure they do not conflict */
424 struct lu_fid mti_tmp_fid1;
425 struct lu_fid mti_tmp_fid2;
426 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
427 * mdt_rename_lock() */
428 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
431 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
432 char ns_name[48]; /* for mdt_init0() */
433 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
434 struct obd_statfs osfs; /* for mdt_statfs() */
436 /* for mdt_readpage() */
437 struct lu_rdpg mti_rdpg;
438 /* for mdt_sendpage() */
439 struct l_wait_info mti_wait_info;
443 struct md_som_data data;
447 /* IO epoch related stuff. */
448 struct mdt_ioepoch *mti_ioepoch;
449 __u64 mti_replayepoch;
452 struct lu_buf mti_buf;
453 struct lu_buf mti_big_buf;
454 struct lustre_capa_key mti_capa_key;
456 /* Ops object filename */
457 struct lu_name mti_name;
458 /* per-thread values, can be re-used */
461 /* big_lmm buffer was used and must be used in reply */
462 int mti_big_lmm_used;
463 /* should be enough to fit lustre_mdt_attrs */
464 char mti_xattr_buf[128];
465 struct ldlm_enqueue_info mti_einfo;
468 extern struct lu_context_key mdt_thread_key;
470 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
472 struct mdt_thread_info *mti;
474 lu_env_refill((void *)env);
475 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
480 /* ptlrpc request handler for MDT. All handlers are
481 * grouped into several slices - struct mdt_opc_slice,
482 * and stored in an array - mdt_handlers[].
485 /* The name of this handler. */
487 /* Fail id for this handler, checked at the beginning of this handler*/
489 /* Operation code for this handler */
491 /* flags are listed in enum mdt_handler_flags below. */
493 /* The actual handler function to execute. */
494 int (*mh_act)(struct mdt_thread_info *info);
495 /* Request format for this request. */
496 const struct req_format *mh_fmt;
499 struct mdt_opc_slice {
502 struct mdt_handler *mos_hs;
505 struct cdt_req_progress {
506 struct mutex crp_lock; /**< protect tree */
507 struct interval_node *crp_root; /**< tree to track extent
509 struct interval_node **crp_node; /**< buffer for tree nodes
510 * vector of fixed size
512 int crp_cnt; /**< # of used nodes */
513 int crp_max; /**< # of allocated nodes */
516 struct cdt_agent_req {
517 cfs_list_t car_request_list; /**< to chain all the req. */
518 atomic_t car_refcount; /**< reference counter */
519 __u64 car_compound_id; /**< compound id */
520 __u64 car_flags; /**< request original flags */
521 struct obd_uuid car_uuid; /**< agent doing the req. */
522 __u32 car_archive_id; /**< archive id */
523 int car_canceled; /**< request was canceled */
524 cfs_time_t car_req_start; /**< start time */
525 cfs_time_t car_req_update; /**< last update time */
526 struct hsm_action_item *car_hai; /**< req. to the agent */
527 struct cdt_req_progress car_progress; /**< track data mvt
530 extern struct kmem_cache *mdt_hsm_car_kmem;
533 cfs_list_t ha_list; /**< to chain the agents */
534 struct obd_uuid ha_uuid; /**< agent uuid */
535 __u32 *ha_archive_id; /**< archive id */
536 int ha_archive_cnt; /**< number of archive entries
537 * 0 means any archive */
538 atomic_t ha_requests; /**< current request count */
539 atomic_t ha_success; /**< number of successful
541 atomic_t ha_failure; /**< number of failed actions */
544 struct cdt_restore_handle {
545 cfs_list_t crh_list; /**< to chain the handle */
546 struct lu_fid crh_fid; /**< fid of the object */
547 struct ldlm_extent crh_extent; /**< extent of the restore */
548 struct mdt_lock_handle crh_lh; /**< lock handle */
550 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
552 static inline const struct md_device_operations *
553 mdt_child_ops(struct mdt_device * m)
555 LASSERT(m->mdt_child);
556 return m->mdt_child->md_ops;
559 static inline struct md_object *mdt_object_child(struct mdt_object *o)
562 return lu2md(lu_object_next(&o->mot_obj));
565 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
567 return info->mti_pill ? info->mti_pill->rc_req : NULL;
570 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
572 LASSERT(info->mti_exp);
573 return exp_connect_flags(info->mti_exp);
576 static inline void mdt_object_get(const struct lu_env *env,
577 struct mdt_object *o)
580 lu_object_get(&o->mot_obj);
584 static inline void mdt_object_put(const struct lu_env *env,
585 struct mdt_object *o)
588 lu_object_put(env, &o->mot_obj);
592 static inline int mdt_object_exists(const struct mdt_object *o)
594 return lu_object_exists(&o->mot_obj);
597 static inline int mdt_object_remote(const struct mdt_object *o)
599 return lu_object_remote(&o->mot_obj);
602 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
604 return lu_object_fid(&o->mot_obj);
607 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
609 return mdt->mdt_lu_dev.ld_site;
612 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
614 return &mdt->mdt_seq_site;
617 static inline void mdt_export_evict(struct obd_export *exp)
619 class_fail_export(exp);
620 class_export_put(exp);
623 /* Here we use LVB_TYPE to check dne client, because it is
624 * also landed on 2.4. */
625 static inline int mdt_is_dne_client(struct obd_export *exp)
627 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
630 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
631 void mdt_set_disposition(struct mdt_thread_info *info,
632 struct ldlm_reply *rep, int flag);
633 void mdt_clear_disposition(struct mdt_thread_info *info,
634 struct ldlm_reply *rep, int flag);
636 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
637 ldlm_mode_t lock_mode,
638 const struct lu_name *lname);
640 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
643 int mdt_lock_setup(struct mdt_thread_info *info,
644 struct mdt_object *o,
645 struct mdt_lock_handle *lh);
647 int mdt_object_lock(struct mdt_thread_info *,
649 struct mdt_lock_handle *,
652 int mdt_object_lock_try(struct mdt_thread_info *,
654 struct mdt_lock_handle *,
657 void mdt_object_unlock(struct mdt_thread_info *,
659 struct mdt_lock_handle *,
662 struct mdt_object *mdt_object_new(const struct lu_env *,
664 const struct lu_fid *);
665 struct mdt_object *mdt_object_find(const struct lu_env *,
667 const struct lu_fid *);
668 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
669 const struct lu_fid *,
670 struct mdt_lock_handle *,
672 void mdt_object_unlock_put(struct mdt_thread_info *,
674 struct mdt_lock_handle *,
677 void mdt_client_compatibility(struct mdt_thread_info *info);
679 int mdt_remote_object_lock(struct mdt_thread_info *mti,
680 struct mdt_object *o, struct lustre_handle *lh,
681 ldlm_mode_t mode, __u64 ibits);
683 enum mdt_name_flags {
687 int mdt_name_unpack(struct req_capsule *pill,
688 const struct req_msg_field *field,
690 enum mdt_name_flags flags);
691 int mdt_close_unpack(struct mdt_thread_info *info);
692 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
693 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
694 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
695 const struct lu_attr *attr, const struct lu_fid *fid);
697 int mdt_getxattr(struct mdt_thread_info *info);
698 int mdt_reint_setxattr(struct mdt_thread_info *info,
699 struct mdt_lock_handle *lh);
701 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
702 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
704 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
705 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
706 struct mdt_lock_handle *lhc);
708 extern void target_recovery_fini(struct obd_device *obd);
709 extern void target_recovery_init(struct lu_target *lut,
710 svc_handler_t handler);
711 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
712 struct obd_device *, struct lustre_sb_info *lsi);
713 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
715 int mdt_export_stats_init(struct obd_device *obd,
716 struct obd_export *exp,
719 int mdt_lock_new_child(struct mdt_thread_info *info,
720 struct mdt_object *o,
721 struct mdt_lock_handle *child_lockh);
723 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
726 int mdt_reint_open(struct mdt_thread_info *info,
727 struct mdt_lock_handle *lhc);
729 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
730 const struct lustre_handle *handle,
734 MDT_IOEPOCH_CLOSED = 0,
735 MDT_IOEPOCH_OPENED = 1,
736 MDT_IOEPOCH_GETATTR = 2,
744 int mdt_get_info(struct tgt_session_info *tsi);
745 int mdt_attr_get_complex(struct mdt_thread_info *info,
746 struct mdt_object *o, struct md_attr *ma);
747 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
749 int mdt_object_is_som_enabled(struct mdt_object *mo);
750 int mdt_write_get(struct mdt_object *o);
751 void mdt_write_put(struct mdt_object *o);
752 int mdt_write_read(struct mdt_object *o);
753 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
754 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
755 void mdt_mfd_free(struct mdt_file_data *mfd);
756 int mdt_close(struct tgt_session_info *tsi);
757 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
758 struct md_attr *ma, int flags);
759 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
761 int mdt_done_writing(struct tgt_session_info *tsi);
762 int mdt_fix_reply(struct mdt_thread_info *info);
763 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
764 const struct md_attr *);
765 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
767 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
768 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
769 const void *area, ssize_t len);
771 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
773 int mdt_check_ucred(struct mdt_thread_info *);
774 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
775 int mdt_init_ucred_reint(struct mdt_thread_info *);
776 void mdt_exit_ucred(struct mdt_thread_info *);
777 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
778 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
779 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
781 void mdt_thread_info_init(struct ptlrpc_request *req,
782 struct mdt_thread_info *mti);
783 void mdt_thread_info_fini(struct mdt_thread_info *mti);
784 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
786 extern struct lprocfs_vars lprocfs_mds_module_vars[];
787 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
789 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
790 const struct md_hsm *mh);
793 int mdt_init_idmap(struct tgt_session_info *tsi);
794 void mdt_cleanup_idmap(struct mdt_export_data *);
795 int mdt_handle_idmap(struct tgt_session_info *tsi);
796 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
797 struct ptlrpc_user_desc *);
798 void mdt_body_reverse_idmap(struct mdt_thread_info *,
800 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
801 struct mdt_remote_perm *);
802 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
804 static inline struct mdt_device *mdt_dev(struct lu_device *d)
806 return container_of0(d, struct mdt_device, mdt_lu_dev);
809 static inline struct mdt_object *mdt_obj(struct lu_object *o)
811 return container_of0(o, struct mdt_object, mot_obj);
814 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
816 struct lu_object *lo;
817 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
819 lo = lu_object_locate(mo->mot_obj.lo_header,
820 mdt->mdt_bottom->dd_lu_dev.ld_type);
825 /* mdt/mdt_identity.c */
826 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
828 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
830 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
832 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
834 void mdt_flush_identity(struct upcall_cache *, int);
836 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
838 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
841 int mdt_hsm_state_get(struct tgt_session_info *tsi);
842 int mdt_hsm_state_set(struct tgt_session_info *tsi);
843 int mdt_hsm_action(struct tgt_session_info *tsi);
844 int mdt_hsm_progress(struct tgt_session_info *tsi);
845 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
846 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
847 int mdt_hsm_request(struct tgt_session_info *tsi);
849 /* mdt/mdt_hsm_cdt_actions.c */
850 extern const struct file_operations mdt_hsm_actions_fops;
851 void dump_llog_agent_req_rec(const char *prefix,
852 const struct llog_agent_req_rec *larr);
853 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
854 llog_cb_t cb, void *data);
855 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
856 __u64 compound_id, __u32 archive_id,
857 __u64 flags, struct hsm_action_item *hai);
858 int mdt_agent_record_update(const struct lu_env *env,
859 struct mdt_device *mdt, __u64 *cookies,
860 int cookies_count, enum agent_req_status status);
861 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
862 struct llog_handle *llh,
863 struct llog_agent_req_rec *larr);
865 /* mdt/mdt_hsm_cdt_agent.c */
866 extern const struct file_operations mdt_hsm_agent_fops;
867 int mdt_hsm_agent_register(struct mdt_thread_info *info,
868 const struct obd_uuid *uuid,
869 int nr_archives, __u32 *archive_num);
870 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
871 const struct obd_uuid *uuid,
873 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
874 const struct obd_uuid *uuid);
875 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
876 int succ_rq, int fail_rq, int new_rq,
877 const struct obd_uuid *uuid);
878 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
879 struct obd_uuid *uuid);
880 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
882 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
883 struct hsm_progress_kernel *pgs);
884 /* mdt/mdt_hsm_cdt_client.c */
885 int mdt_hsm_add_actions(struct mdt_thread_info *info,
886 struct hsm_action_list *hal, __u64 *compound_id);
887 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
888 struct hsm_action_list *hal);
889 int mdt_hsm_get_running(struct mdt_thread_info *mti,
890 struct hsm_action_list *hal);
891 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
892 const struct lu_fid *fid);
893 /* mdt/mdt_hsm_cdt_requests.c */
894 extern const struct file_operations mdt_hsm_active_requests_fops;
895 void dump_requests(char *prefix, struct coordinator *cdt);
896 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
897 __u64 flags, struct obd_uuid *uuid,
898 struct hsm_action_item *hai);
899 void mdt_cdt_free_request(struct cdt_agent_req *car);
900 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
901 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
903 const struct lu_fid *fid);
904 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
905 void mdt_cdt_get_request(struct cdt_agent_req *car);
906 void mdt_cdt_put_request(struct cdt_agent_req *car);
907 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
908 const struct hsm_progress_kernel *pgs);
909 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
910 /* mdt/mdt_coordinator.c */
911 void mdt_hsm_dump_hal(int level, const char *prefix,
912 struct hsm_action_list *hal);
913 /* coordinator management */
914 int mdt_hsm_cdt_init(struct mdt_device *mdt);
915 int mdt_hsm_cdt_start(struct mdt_device *mdt);
916 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
917 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
918 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
920 /* coordinator control /proc interface */
921 int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
922 unsigned long count, void *data);
923 int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
924 int count, int *eof, void *data);
925 int hsm_cdt_procfs_init(struct mdt_device *mdt);
926 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
927 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
929 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
930 const struct lu_fid *fid,
932 /* actions/request helpers */
933 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
934 struct hsm_action_list *hal, struct obd_uuid *uuid);
935 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
936 const int hal_an, const __u64 rq_flags,
937 const struct md_hsm *hsm);
938 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
939 struct hsm_progress_kernel *pgs,
940 const int update_record);
942 extern struct lu_context_key mdt_thread_key;
944 /* debug issues helper starts here*/
945 static inline int mdt_fail_write(const struct lu_env *env,
946 struct dt_device *dd, int id)
948 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
949 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
951 return dd->dd_ops->dt_ro(env, dd);
952 /* We set FAIL_ONCE because we never "un-fail" a device */
958 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
960 return &req->rq_export->exp_mdt_data;
963 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
965 return mdt_dev(exp->exp_obd->obd_lu_dev);
968 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
969 struct mdt_lock_handle *lhc);
970 static inline int mdt_check_resent(struct mdt_thread_info *info,
971 mdt_reconstruct_t reconstruct,
972 struct mdt_lock_handle *lhc)
974 struct ptlrpc_request *req = mdt_info_req(info);
977 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
978 if (req_xid_is_last(req)) {
979 reconstruct(info, lhc);
982 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
983 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
988 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
989 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
991 static inline int is_identity_get_disabled(struct upcall_cache *cache)
993 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
996 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
998 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
999 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1000 struct lustre_handle *lh,
1002 ldlm_policy_data_t *policy,
1003 const struct ldlm_res_id *res_id,
1004 __u64 flags, const __u64 *client_cookie)
1008 LASSERT(ns != NULL);
1009 LASSERT(lh != NULL);
1011 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1012 mode, &flags, mdt_blocking_ast,
1013 ldlm_completion_ast, NULL, NULL, 0,
1014 LVB_T_NONE, client_cookie, lh);
1015 return rc == ELDLM_OK ? 0 : -EIO;
1018 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1021 ldlm_lock_decref(lh, mode);
1024 extern mdl_mode_t mdt_mdl_lock_modes[];
1025 extern ldlm_mode_t mdt_dlm_lock_modes[];
1027 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1029 LASSERT(IS_PO2(mode));
1030 return mdt_mdl_lock_modes[mode];
1033 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1035 LASSERT(IS_PO2(mode));
1036 return mdt_dlm_lock_modes[mode];
1040 extern struct ldlm_valblock_ops mdt_lvbo;
1042 void mdt_enable_cos(struct mdt_device *, int);
1043 int mdt_cos_is_enabled(struct mdt_device *);
1061 LPROC_MDT_SAMEDIR_RENAME,
1062 LPROC_MDT_CROSSDIR_RENAME,
1065 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1066 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1067 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1068 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1069 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1070 void mdt_procfs_fini(struct mdt_device *mdt);
1071 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1072 struct mdt_device *mdt,
1073 struct ptlrpc_request *req,
1074 struct mdt_object *src, struct mdt_object *tgt);
1077 int mdt_ck_thread_start(struct mdt_device *mdt);
1078 void mdt_ck_thread_stop(struct mdt_device *mdt);
1079 void mdt_ck_timer_callback(unsigned long castmeharder);
1080 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1081 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1082 const struct lu_fid *fid, struct lustre_capa *capa);
1083 void mdt_dump_capainfo(struct mdt_thread_info *info);
1085 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1087 return mdt->mdt_lu_dev.ld_obd;
1090 extern const struct lu_device_operations mdt_lu_ops;
1092 static inline char *mdt_obd_name(struct mdt_device *mdt)
1094 return mdt->mdt_lu_dev.ld_obd->obd_name;
1097 int mds_mod_init(void);
1098 void mds_mod_exit(void);
1100 #endif /* __KERNEL__ */