4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_internal.h
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
48 #ifndef _MDT_INTERNAL_H
49 #define _MDT_INTERNAL_H
51 #if defined(__KERNEL__)
53 #include <libcfs/lucache.h>
54 #include <lustre_net.h>
55 #include <lustre/lustre_idl.h>
56 #include <obd_class.h>
57 #include <lustre_disk.h>
58 #include <lu_target.h>
59 #include <md_object.h>
60 #include <lustre_fid.h>
61 #include <lustre_fld.h>
62 #include <lustre_req_layout.h>
63 #include <lustre_sec.h>
64 #include <lustre_idmap.h>
65 #include <lustre_eacl.h>
66 #include <lustre_quota.h>
68 /* check if request's xid is equal to last one or not*/
69 static inline int req_xid_is_last(struct ptlrpc_request *req)
71 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
72 return (req->rq_xid == lcd->lcd_last_xid ||
73 req->rq_xid == lcd->lcd_last_close_xid);
78 /* file data for open files on MDS */
79 struct mdt_file_data {
80 struct portals_handle mfd_handle; /* must be first */
81 __u64 mfd_mode; /* open mode provided by client */
82 cfs_list_t mfd_list; /* protected by med_open_lock */
83 __u64 mfd_xid; /* xid of the open request */
84 struct lustre_handle mfd_old_handle; /* old handle in replay case */
85 struct mdt_object *mfd_object; /* point to opened object */
88 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
89 #define CDT_NORETRY_ACTION (1ULL << 1)
90 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
91 #define CDT_POLICY_SHIFT_COUNT 2
92 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
95 /* when adding a new policy, do not forget to update
96 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
98 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
100 enum cdt_states { CDT_STOPPED = 0,
106 /* when multiple lock are needed, the lock order is
114 struct ptlrpc_thread cdt_thread; /**< coordinator thread */
115 struct lu_env cdt_env; /**< coordinator lustre
117 struct lu_context cdt_session; /** session for lu_ucred */
118 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
119 __u64 cdt_policy; /**< policy flags */
120 enum cdt_states cdt_state; /**< state */
121 atomic_t cdt_compound_id; /**< compound id
123 __u64 cdt_last_cookie; /**< last cookie
125 struct mutex cdt_llog_lock; /**< protect llog
127 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
128 struct rw_semaphore cdt_request_lock; /**< protect request
130 struct mutex cdt_restore_lock; /**< protect restore
132 cfs_time_t cdt_loop_period; /**< llog scan period */
133 cfs_time_t cdt_grace_delay; /**< request grace
135 cfs_time_t cdt_active_req_timeout; /**< request timeout */
136 __u32 cdt_default_archive_id; /**< archive id used
139 __u64 cdt_max_requests; /**< max count of started
141 atomic_t cdt_request_count; /**< current count of
142 * started requests */
143 struct list_head cdt_requests; /**< list of started
145 struct list_head cdt_agents; /**< list of register
147 struct list_head cdt_restore_hdl; /**< list of restore lock
149 /* Bitmasks indexed by the HSMA_XXX constants. */
150 __u64 cdt_user_request_mask;
151 __u64 cdt_group_request_mask;
152 __u64 cdt_other_request_mask;
155 /* mdt state flag bits */
156 #define MDT_FL_CFGLOG 0
157 #define MDT_FL_SYNCED 1
161 struct lu_device mdt_lu_dev;
162 struct seq_server_site mdt_seq_site;
163 /* DLM name-space for meta-data locks maintained by this server */
164 struct ldlm_namespace *mdt_namespace;
165 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
166 struct ptlrpc_client *mdt_ldlm_client;
167 /* underlying device */
168 struct obd_export *mdt_child_exp;
169 struct md_device *mdt_child;
170 struct dt_device *mdt_bottom;
171 struct obd_export *mdt_bottom_exp;
173 struct lu_target mdt_lut;
175 * Options bit-fields.
178 unsigned int mo_user_xattr:1,
183 /* mdt state flags */
184 unsigned long mdt_state;
185 /* lock to protect IOepoch */
186 spinlock_t mdt_ioepoch_lock;
189 /* transaction callbacks */
190 struct dt_txn_callback mdt_txn_cb;
192 /* these values should be updated from lov if necessary.
193 * or should be placed somewhere else. */
195 int mdt_max_cookiesize;
199 struct upcall_cache *mdt_identity_cache;
201 /* capability keys */
202 unsigned long mdt_capa_timeout;
204 struct dt_object *mdt_ck_obj;
205 unsigned long mdt_ck_timeout;
206 unsigned long mdt_ck_expiry;
207 struct timer_list mdt_ck_timer;
208 struct ptlrpc_thread mdt_ck_thread;
209 struct lustre_capa_key mdt_capa_keys[2];
210 unsigned int mdt_capa_conf:1,
212 /* Enable remote dir on non-MDT0 */
213 mdt_enable_remote_dir:1;
215 gid_t mdt_enable_remote_dir_gid;
216 /* statfs optimization: we cache a bit */
217 struct obd_statfs mdt_osfs;
219 spinlock_t mdt_osfs_lock;
222 uid_t mdt_squash_uid;
223 gid_t mdt_squash_gid;
224 cfs_list_t mdt_nosquash_nids;
225 char *mdt_nosquash_str;
226 int mdt_nosquash_strlen;
227 struct rw_semaphore mdt_squash_sem;
229 struct rename_stats mdt_rename_stats;
230 struct lu_fid mdt_md_root_fid;
232 /* connection to quota master */
233 struct obd_export *mdt_qmt_exp;
234 /* quota master device associated with this MDT */
235 struct lu_device *mdt_qmt_dev;
237 struct coordinator mdt_coordinator;
240 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
241 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
242 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
243 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
244 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
245 #define MDT_COS_DEFAULT (0)
248 struct lu_object_header mot_header;
249 struct lu_object mot_obj;
252 int mot_ioepoch_count;
254 /* Lock to protect object's IO epoch. */
255 struct mutex mot_ioepoch_mutex;
256 /* Lock to protect create_data */
257 struct mutex mot_lov_mutex;
258 /* Lock to protect lease open.
259 * Lease open acquires write lock; normal open acquires read lock */
260 struct rw_semaphore mot_open_sem;
261 atomic_t mot_lease_count;
262 atomic_t mot_open_count;
265 enum mdt_object_flags {
266 /** SOM attributes are changed. */
267 MOF_SOM_CHANGE = (1 << 0),
269 * The SOM recovery state for mdt object.
270 * This state is an in-memory equivalent of an absent SOM EA, used
271 * instead of invalidating SOM EA while IOEpoch is still opened when
272 * a client eviction occurs or a client fails to obtain SOM attributes.
273 * It indicates that the last IOEpoch holder will need to obtain SOM
274 * attributes under [0;EOF] extent lock to flush all the client's
275 * cached of evicted from MDS clients (but not necessary evicted from
276 * OST) before taking ost attributes.
278 MOF_SOM_RECOV = (1 << 1),
279 /** File has been just created. */
280 MOF_SOM_CREATED = (1 << 2),
281 /** lov object has been created. */
282 MOF_LOV_CREATED = (1 << 3),
285 struct mdt_lock_handle {
286 /* Lock type, reg for cross-ref use or pdo lock. */
290 struct lustre_handle mlh_reg_lh;
291 ldlm_mode_t mlh_reg_mode;
294 struct lustre_handle mlh_pdo_lh;
295 ldlm_mode_t mlh_pdo_mode;
296 unsigned int mlh_pdo_hash;
298 /* Remote regular lock */
299 struct lustre_handle mlh_rreg_lh;
300 ldlm_mode_t mlh_rreg_mode;
304 MDT_LH_PARENT, /* parent lockh */
305 MDT_LH_CHILD, /* child lockh */
306 MDT_LH_OLD, /* old lockh for rename */
307 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
308 MDT_LH_NEW, /* new lockh for rename */
309 MDT_LH_RMT, /* used for return lh to caller */
310 MDT_LH_LOCAL, /* local lock never return to client */
319 struct mdt_reint_record {
320 mdt_reint_t rr_opcode;
321 const struct lustre_handle *rr_handle;
322 const struct lu_fid *rr_fid1;
323 const struct lu_fid *rr_fid2;
324 struct lu_name rr_name;
325 struct lu_name rr_tgt_name;
326 const void *rr_eadata;
331 enum mdt_reint_flag {
332 MRF_OPEN_TRUNC = 1 << 0,
336 * Common data shared by mdt-level handlers. This is allocated per-thread to
337 * reduce stack consumption.
339 struct mdt_thread_info {
342 * The following members will be filled explicitly
343 * with specific data in mdt_thread_info_init().
345 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
346 * request handling may migrate from one server thread to another.
348 struct req_capsule *mti_pill;
350 /* although we have export in req, there are cases when it is not
351 * available, e.g. closing files upon export destroy */
352 struct obd_export *mti_exp;
354 * A couple of lock handles.
356 struct mdt_lock_handle mti_lh[MDT_LH_NR];
358 struct mdt_device *mti_mdt;
359 const struct lu_env *mti_env;
361 /* transaction number of current request */
367 * The following members will be filled expilictly
368 * with zero in mdt_thread_info_init(). These members may be used
375 struct md_attr mti_attr;
377 * Body for "habeo corpus" operations.
379 const struct mdt_body *mti_body;
381 * Host object. This is released at the end of mdt_handler().
383 struct mdt_object *mti_object;
385 * Lock request for "habeo clavis" operations.
387 const struct ldlm_request *mti_dlm_req;
389 __u32 mti_has_trans:1, /* has txn already? */
392 /* opdata for mdt_reint_open(), has the same as
393 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
394 * value onto disk for recovery when mdt_trans_stop_cb() is called.
400 * The following members will be filled explicitly
401 * with zero in mdt_reint_unpack(), because they are only used
402 * by reint requests (including mdt_reint_open()).
406 * reint record. contains information for reint operations.
408 struct mdt_reint_record mti_rr;
410 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
412 * Operation specification (currently create and lookup)
414 struct md_op_spec mti_spec;
418 * The following members will _NOT_ be initialized at all.
419 * DO NOT expect them to contain any valid value.
420 * They should be initialized explicitly by the user themselves.
423 /* XXX: If something is in a union, make sure they do not conflict */
425 struct lu_fid mti_tmp_fid1;
426 struct lu_fid mti_tmp_fid2;
427 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
428 * mdt_rename_lock() */
429 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
432 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
433 char ns_name[48]; /* for mdt_init0() */
434 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
435 struct obd_statfs osfs; /* for mdt_statfs() */
437 /* for mdt_readpage() */
438 struct lu_rdpg mti_rdpg;
439 /* for mdt_sendpage() */
440 struct l_wait_info mti_wait_info;
444 struct md_som_data data;
448 /* IO epoch related stuff. */
449 struct mdt_ioepoch *mti_ioepoch;
450 __u64 mti_replayepoch;
453 struct lu_buf mti_buf;
454 struct lu_buf mti_big_buf;
455 struct lustre_capa_key mti_capa_key;
457 /* Ops object filename */
458 struct lu_name mti_name;
459 /* per-thread values, can be re-used */
462 /* big_lmm buffer was used and must be used in reply */
463 int mti_big_lmm_used;
464 /* should be enough to fit lustre_mdt_attrs */
465 char mti_xattr_buf[128];
466 struct ldlm_enqueue_info mti_einfo;
469 extern struct lu_context_key mdt_thread_key;
471 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
473 struct mdt_thread_info *mti;
475 lu_env_refill((void *)env);
476 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
481 /* ptlrpc request handler for MDT. All handlers are
482 * grouped into several slices - struct mdt_opc_slice,
483 * and stored in an array - mdt_handlers[].
486 /* The name of this handler. */
488 /* Fail id for this handler, checked at the beginning of this handler*/
490 /* Operation code for this handler */
492 /* flags are listed in enum mdt_handler_flags below. */
494 /* The actual handler function to execute. */
495 int (*mh_act)(struct mdt_thread_info *info);
496 /* Request format for this request. */
497 const struct req_format *mh_fmt;
500 struct mdt_opc_slice {
503 struct mdt_handler *mos_hs;
506 struct cdt_req_progress {
507 struct mutex crp_lock; /**< protect tree */
508 struct interval_node *crp_root; /**< tree to track extent
510 struct interval_node **crp_node; /**< buffer for tree nodes
511 * vector of fixed size
513 int crp_cnt; /**< # of used nodes */
514 int crp_max; /**< # of allocated nodes */
517 struct cdt_agent_req {
518 cfs_list_t car_request_list; /**< to chain all the req. */
519 atomic_t car_refcount; /**< reference counter */
520 __u64 car_compound_id; /**< compound id */
521 __u64 car_flags; /**< request original flags */
522 struct obd_uuid car_uuid; /**< agent doing the req. */
523 __u32 car_archive_id; /**< archive id */
524 int car_canceled; /**< request was canceled */
525 cfs_time_t car_req_start; /**< start time */
526 cfs_time_t car_req_update; /**< last update time */
527 struct hsm_action_item *car_hai; /**< req. to the agent */
528 struct cdt_req_progress car_progress; /**< track data mvt
531 extern struct kmem_cache *mdt_hsm_car_kmem;
534 cfs_list_t ha_list; /**< to chain the agents */
535 struct obd_uuid ha_uuid; /**< agent uuid */
536 __u32 *ha_archive_id; /**< archive id */
537 int ha_archive_cnt; /**< number of archive entries
538 * 0 means any archive */
539 atomic_t ha_requests; /**< current request count */
540 atomic_t ha_success; /**< number of successful
542 atomic_t ha_failure; /**< number of failed actions */
545 struct cdt_restore_handle {
546 cfs_list_t crh_list; /**< to chain the handle */
547 struct lu_fid crh_fid; /**< fid of the object */
548 struct ldlm_extent crh_extent; /**< extent of the restore */
549 struct mdt_lock_handle crh_lh; /**< lock handle */
551 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
553 static inline const struct md_device_operations *
554 mdt_child_ops(struct mdt_device * m)
556 LASSERT(m->mdt_child);
557 return m->mdt_child->md_ops;
560 static inline struct md_object *mdt_object_child(struct mdt_object *o)
563 return lu2md(lu_object_next(&o->mot_obj));
566 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
568 return info->mti_pill ? info->mti_pill->rc_req : NULL;
571 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
573 LASSERT(info->mti_exp);
574 return exp_connect_flags(info->mti_exp);
577 static inline void mdt_object_get(const struct lu_env *env,
578 struct mdt_object *o)
581 lu_object_get(&o->mot_obj);
585 static inline void mdt_object_put(const struct lu_env *env,
586 struct mdt_object *o)
589 lu_object_put(env, &o->mot_obj);
593 static inline int mdt_object_exists(const struct mdt_object *o)
595 return lu_object_exists(&o->mot_obj);
598 static inline int mdt_object_remote(const struct mdt_object *o)
600 return lu_object_remote(&o->mot_obj);
603 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
605 return lu_object_fid(&o->mot_obj);
608 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
610 return mdt->mdt_lu_dev.ld_site;
613 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
615 return &mdt->mdt_seq_site;
618 static inline void mdt_export_evict(struct obd_export *exp)
620 class_fail_export(exp);
621 class_export_put(exp);
624 /* Here we use LVB_TYPE to check dne client, because it is
625 * also landed on 2.4. */
626 static inline int mdt_is_dne_client(struct obd_export *exp)
628 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
631 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
632 void mdt_set_disposition(struct mdt_thread_info *info,
633 struct ldlm_reply *rep, int flag);
634 void mdt_clear_disposition(struct mdt_thread_info *info,
635 struct ldlm_reply *rep, int flag);
637 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
638 ldlm_mode_t lock_mode,
639 const struct lu_name *lname);
641 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
644 int mdt_lock_setup(struct mdt_thread_info *info,
645 struct mdt_object *o,
646 struct mdt_lock_handle *lh);
648 int mdt_object_lock(struct mdt_thread_info *,
650 struct mdt_lock_handle *,
653 int mdt_object_lock_try(struct mdt_thread_info *,
655 struct mdt_lock_handle *,
658 void mdt_object_unlock(struct mdt_thread_info *,
660 struct mdt_lock_handle *,
663 struct mdt_object *mdt_object_new(const struct lu_env *,
665 const struct lu_fid *);
666 struct mdt_object *mdt_object_find(const struct lu_env *,
668 const struct lu_fid *);
669 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
670 const struct lu_fid *,
671 struct mdt_lock_handle *,
673 void mdt_object_unlock_put(struct mdt_thread_info *,
675 struct mdt_lock_handle *,
678 void mdt_client_compatibility(struct mdt_thread_info *info);
680 int mdt_remote_object_lock(struct mdt_thread_info *mti,
681 struct mdt_object *o, struct lustre_handle *lh,
682 ldlm_mode_t mode, __u64 ibits);
684 enum mdt_name_flags {
688 int mdt_name_unpack(struct req_capsule *pill,
689 const struct req_msg_field *field,
691 enum mdt_name_flags flags);
692 int mdt_close_unpack(struct mdt_thread_info *info);
693 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
694 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
695 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
696 const struct lu_attr *attr, const struct lu_fid *fid);
698 int mdt_getxattr(struct mdt_thread_info *info);
699 int mdt_reint_setxattr(struct mdt_thread_info *info,
700 struct mdt_lock_handle *lh);
702 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
703 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
705 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
706 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
707 struct mdt_lock_handle *lhc);
709 extern void target_recovery_fini(struct obd_device *obd);
710 extern void target_recovery_init(struct lu_target *lut,
711 svc_handler_t handler);
712 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
713 struct obd_device *, struct lustre_sb_info *lsi);
714 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
716 int mdt_export_stats_init(struct obd_device *obd,
717 struct obd_export *exp,
720 int mdt_lock_new_child(struct mdt_thread_info *info,
721 struct mdt_object *o,
722 struct mdt_lock_handle *child_lockh);
724 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
727 int mdt_reint_open(struct mdt_thread_info *info,
728 struct mdt_lock_handle *lhc);
730 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
731 const struct lustre_handle *handle,
735 MDT_IOEPOCH_CLOSED = 0,
736 MDT_IOEPOCH_OPENED = 1,
737 MDT_IOEPOCH_GETATTR = 2,
745 int mdt_get_info(struct tgt_session_info *tsi);
746 int mdt_attr_get_complex(struct mdt_thread_info *info,
747 struct mdt_object *o, struct md_attr *ma);
748 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
750 int mdt_object_is_som_enabled(struct mdt_object *mo);
751 int mdt_write_get(struct mdt_object *o);
752 void mdt_write_put(struct mdt_object *o);
753 int mdt_write_read(struct mdt_object *o);
754 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
755 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
756 void mdt_mfd_free(struct mdt_file_data *mfd);
757 int mdt_close(struct tgt_session_info *tsi);
758 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
759 struct md_attr *ma, int flags);
760 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
762 int mdt_done_writing(struct tgt_session_info *tsi);
763 int mdt_fix_reply(struct mdt_thread_info *info);
764 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
765 const struct md_attr *);
766 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
768 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
769 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
770 const void *area, ssize_t len);
772 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
773 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
775 int mdt_check_ucred(struct mdt_thread_info *);
776 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
777 int mdt_init_ucred_reint(struct mdt_thread_info *);
778 void mdt_exit_ucred(struct mdt_thread_info *);
779 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
780 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
781 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
783 void mdt_thread_info_init(struct ptlrpc_request *req,
784 struct mdt_thread_info *mti);
785 void mdt_thread_info_fini(struct mdt_thread_info *mti);
786 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
788 extern struct lprocfs_vars lprocfs_mds_module_vars[];
789 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
791 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
792 const struct md_hsm *mh);
795 int mdt_init_idmap(struct tgt_session_info *tsi);
796 void mdt_cleanup_idmap(struct mdt_export_data *);
797 int mdt_handle_idmap(struct tgt_session_info *tsi);
798 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
799 struct ptlrpc_user_desc *);
800 void mdt_body_reverse_idmap(struct mdt_thread_info *,
802 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
803 struct mdt_remote_perm *);
804 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
806 static inline struct mdt_device *mdt_dev(struct lu_device *d)
808 return container_of0(d, struct mdt_device, mdt_lu_dev);
811 static inline struct mdt_object *mdt_obj(struct lu_object *o)
813 return container_of0(o, struct mdt_object, mot_obj);
816 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
818 struct lu_object *lo;
819 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
821 lo = lu_object_locate(mo->mot_obj.lo_header,
822 mdt->mdt_bottom->dd_lu_dev.ld_type);
827 /* mdt/mdt_identity.c */
828 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
830 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
832 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
834 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
836 void mdt_flush_identity(struct upcall_cache *, int);
838 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
840 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
843 int mdt_hsm_state_get(struct tgt_session_info *tsi);
844 int mdt_hsm_state_set(struct tgt_session_info *tsi);
845 int mdt_hsm_action(struct tgt_session_info *tsi);
846 int mdt_hsm_progress(struct tgt_session_info *tsi);
847 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
848 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
849 int mdt_hsm_request(struct tgt_session_info *tsi);
851 /* mdt/mdt_hsm_cdt_actions.c */
852 extern const struct file_operations mdt_hsm_actions_fops;
853 void dump_llog_agent_req_rec(const char *prefix,
854 const struct llog_agent_req_rec *larr);
855 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
856 llog_cb_t cb, void *data);
857 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
858 __u64 compound_id, __u32 archive_id,
859 __u64 flags, struct hsm_action_item *hai);
860 int mdt_agent_record_update(const struct lu_env *env,
861 struct mdt_device *mdt, __u64 *cookies,
862 int cookies_count, enum agent_req_status status);
863 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
864 struct llog_handle *llh,
865 struct llog_agent_req_rec *larr);
867 /* mdt/mdt_hsm_cdt_agent.c */
868 extern const struct file_operations mdt_hsm_agent_fops;
869 int mdt_hsm_agent_register(struct mdt_thread_info *info,
870 const struct obd_uuid *uuid,
871 int nr_archives, __u32 *archive_num);
872 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
873 const struct obd_uuid *uuid,
875 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
876 const struct obd_uuid *uuid);
877 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
878 int succ_rq, int fail_rq, int new_rq,
879 const struct obd_uuid *uuid);
880 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
881 struct obd_uuid *uuid);
882 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
884 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
885 struct hsm_progress_kernel *pgs);
886 /* mdt/mdt_hsm_cdt_client.c */
887 int mdt_hsm_add_actions(struct mdt_thread_info *info,
888 struct hsm_action_list *hal, __u64 *compound_id);
889 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
890 struct hsm_action_list *hal);
891 int mdt_hsm_get_running(struct mdt_thread_info *mti,
892 struct hsm_action_list *hal);
893 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
894 const struct lu_fid *fid);
895 /* mdt/mdt_hsm_cdt_requests.c */
896 extern const struct file_operations mdt_hsm_active_requests_fops;
897 void dump_requests(char *prefix, struct coordinator *cdt);
898 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
899 __u64 flags, struct obd_uuid *uuid,
900 struct hsm_action_item *hai);
901 void mdt_cdt_free_request(struct cdt_agent_req *car);
902 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
903 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
905 const struct lu_fid *fid);
906 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
907 void mdt_cdt_get_request(struct cdt_agent_req *car);
908 void mdt_cdt_put_request(struct cdt_agent_req *car);
909 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
910 const struct hsm_progress_kernel *pgs);
911 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
912 /* mdt/mdt_coordinator.c */
913 void mdt_hsm_dump_hal(int level, const char *prefix,
914 struct hsm_action_list *hal);
915 /* coordinator management */
916 int mdt_hsm_cdt_init(struct mdt_device *mdt);
917 int mdt_hsm_cdt_start(struct mdt_device *mdt);
918 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
919 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
920 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
922 /* coordinator control /proc interface */
923 int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
924 unsigned long count, void *data);
925 int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
926 int count, int *eof, void *data);
927 int hsm_cdt_procfs_init(struct mdt_device *mdt);
928 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
929 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
931 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
932 const struct lu_fid *fid,
934 /* actions/request helpers */
935 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
936 struct hsm_action_list *hal, struct obd_uuid *uuid);
937 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
938 const int hal_an, const __u64 rq_flags,
939 const struct md_hsm *hsm);
940 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
941 struct hsm_progress_kernel *pgs,
942 const int update_record);
944 extern struct lu_context_key mdt_thread_key;
946 /* debug issues helper starts here*/
947 static inline int mdt_fail_write(const struct lu_env *env,
948 struct dt_device *dd, int id)
950 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
951 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
953 return dd->dd_ops->dt_ro(env, dd);
954 /* We set FAIL_ONCE because we never "un-fail" a device */
960 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
962 return &req->rq_export->exp_mdt_data;
965 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
967 return mdt_dev(exp->exp_obd->obd_lu_dev);
970 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
971 struct mdt_lock_handle *lhc);
972 static inline int mdt_check_resent(struct mdt_thread_info *info,
973 mdt_reconstruct_t reconstruct,
974 struct mdt_lock_handle *lhc)
976 struct ptlrpc_request *req = mdt_info_req(info);
979 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
980 if (req_xid_is_last(req)) {
981 reconstruct(info, lhc);
984 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
985 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
990 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
991 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
993 static inline int is_identity_get_disabled(struct upcall_cache *cache)
995 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
998 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1000 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1001 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1002 struct lustre_handle *lh,
1004 ldlm_policy_data_t *policy,
1005 const struct ldlm_res_id *res_id,
1006 __u64 flags, const __u64 *client_cookie)
1010 LASSERT(ns != NULL);
1011 LASSERT(lh != NULL);
1013 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1014 mode, &flags, mdt_blocking_ast,
1015 ldlm_completion_ast, NULL, NULL, 0,
1016 LVB_T_NONE, client_cookie, lh);
1017 return rc == ELDLM_OK ? 0 : -EIO;
1020 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1023 ldlm_lock_decref(lh, mode);
1026 extern mdl_mode_t mdt_mdl_lock_modes[];
1027 extern ldlm_mode_t mdt_dlm_lock_modes[];
1029 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1031 LASSERT(IS_PO2(mode));
1032 return mdt_mdl_lock_modes[mode];
1035 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1037 LASSERT(IS_PO2(mode));
1038 return mdt_dlm_lock_modes[mode];
1042 extern struct ldlm_valblock_ops mdt_lvbo;
1044 void mdt_enable_cos(struct mdt_device *, int);
1045 int mdt_cos_is_enabled(struct mdt_device *);
1063 LPROC_MDT_SAMEDIR_RENAME,
1064 LPROC_MDT_CROSSDIR_RENAME,
1067 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1068 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1069 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1070 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1071 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1072 void mdt_procfs_fini(struct mdt_device *mdt);
1074 /* lustre/mdt_mdt_lproc.c */
1075 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1077 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1078 struct mdt_device *mdt,
1079 struct ptlrpc_request *req,
1080 struct mdt_object *src, struct mdt_object *tgt);
1083 int mdt_ck_thread_start(struct mdt_device *mdt);
1084 void mdt_ck_thread_stop(struct mdt_device *mdt);
1085 void mdt_ck_timer_callback(unsigned long castmeharder);
1086 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1087 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1088 const struct lu_fid *fid, struct lustre_capa *capa);
1089 void mdt_dump_capainfo(struct mdt_thread_info *info);
1091 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1093 return mdt->mdt_lu_dev.ld_obd;
1096 extern const struct lu_device_operations mdt_lu_ops;
1098 static inline char *mdt_obd_name(struct mdt_device *mdt)
1100 return mdt->mdt_lu_dev.ld_obd->obd_name;
1103 int mds_mod_init(void);
1104 void mds_mod_exit(void);
1106 #endif /* __KERNEL__ */