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 0x0000000000000001ULL
88 #define CDT_NORETRY_ACTION 0x0000000000000002ULL
89 #define CDT_POLICY_MASK CDT_NONBLOCKING_RESTORE | \
92 /* when adding a new policy, do not forget to update
93 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
95 #define CDT_DEFAULT_POLICY 0x0000000000000000ULL
97 enum cdt_states { CDT_STOPPED = 0,
103 /* when multiple lock are needed, the lock order is
111 struct ptlrpc_thread *cdt_thread; /**< coordinator thread */
112 struct lu_env cdt_env; /**< coordinator lustre
114 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
115 __u64 cdt_policy; /**< flags to defined
117 enum cdt_states cdt_state; /**< state */
118 cfs_atomic_t cdt_compound_id; /**< compound id counter */
119 __u64 cdt_last_cookie; /**< last cookie allocated */
120 struct semaphore cdt_counter_lock; /**< protect request
122 struct semaphore cdt_llog_lock; /**< protect llog access */
123 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
124 struct rw_semaphore cdt_request_lock; /**< protect request list */
125 struct semaphore cdt_restore_lock; /**< protect restore list */
126 cfs_time_t cdt_loop_period; /**< llog scan period */
127 cfs_time_t cdt_delay; /**< request grace delay */
128 cfs_time_t cdt_timeout; /**< request timeout */
129 __u64 cdt_max_request; /**< max count of started
131 __u64 cdt_request_count; /**< current count of
132 * started requests */
133 cfs_list_t cdt_requests; /**< list of started
135 cfs_list_t cdt_agents; /**< list of register
137 cfs_list_t cdt_restore_hdl; /**< list of restore lock
141 /* mdt state flag bits */
142 #define MDT_FL_CFGLOG 0
143 #define MDT_FL_SYNCED 1
147 struct lu_device mdt_lu_dev;
148 struct seq_server_site mdt_seq_site;
149 /* DLM name-space for meta-data locks maintained by this server */
150 struct ldlm_namespace *mdt_namespace;
151 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
152 struct ptlrpc_client *mdt_ldlm_client;
153 /* underlying device */
154 struct obd_export *mdt_child_exp;
155 struct md_device *mdt_child;
156 struct dt_device *mdt_bottom;
157 struct obd_export *mdt_bottom_exp;
159 struct lu_target mdt_lut;
161 * Options bit-fields.
164 unsigned int mo_user_xattr:1,
172 /* mdt state flags */
173 unsigned long mdt_state;
174 /* lock to protect IOepoch */
175 spinlock_t mdt_ioepoch_lock;
178 /* transaction callbacks */
179 struct dt_txn_callback mdt_txn_cb;
181 /* these values should be updated from lov if necessary.
182 * or should be placed somewhere else. */
184 int mdt_max_cookiesize;
188 struct upcall_cache *mdt_identity_cache;
191 rwlock_t mdt_sptlrpc_lock;
192 struct sptlrpc_rule_set mdt_sptlrpc_rset;
194 /* capability keys */
195 unsigned long mdt_capa_timeout;
197 struct dt_object *mdt_ck_obj;
198 unsigned long mdt_ck_timeout;
199 unsigned long mdt_ck_expiry;
200 cfs_timer_t mdt_ck_timer;
201 struct ptlrpc_thread mdt_ck_thread;
202 struct lustre_capa_key mdt_capa_keys[2];
203 unsigned int mdt_capa_conf:1,
205 /* Enable remote dir on non-MDT0 */
206 mdt_enable_remote_dir:1;
208 gid_t mdt_enable_remote_dir_gid;
209 /* statfs optimization: we cache a bit */
210 struct obd_statfs mdt_osfs;
212 spinlock_t mdt_osfs_lock;
215 uid_t mdt_squash_uid;
216 gid_t mdt_squash_gid;
217 cfs_list_t mdt_nosquash_nids;
218 char *mdt_nosquash_str;
219 int mdt_nosquash_strlen;
220 struct rw_semaphore mdt_squash_sem;
223 struct rename_stats mdt_rename_stats;
224 struct lu_fid mdt_md_root_fid;
226 /* connection to quota master */
227 struct obd_export *mdt_qmt_exp;
228 /* quota master device associated with this MDT */
229 struct lu_device *mdt_qmt_dev;
231 struct coordinator mdt_coordinator;
234 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
235 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
236 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
237 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
238 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
239 #define MDT_COS_DEFAULT (0)
242 struct lu_object_header mot_header;
243 struct lu_object mot_obj;
246 int mot_ioepoch_count;
248 /* Lock to protect object's IO epoch. */
249 struct mutex mot_ioepoch_mutex;
250 /* Lock to protect create_data */
251 struct mutex mot_lov_mutex;
252 /* Lock to protect lease open.
253 * Lease open acquires write lock; normal open acquires read lock */
254 struct rw_semaphore mot_open_sem;
255 atomic_t mot_lease_count;
256 atomic_t mot_open_count;
257 /* A lock to protect EA data from racing setxattr and getxattrall */
258 struct rw_semaphore mot_xattr_sem;
261 enum mdt_object_flags {
262 /** SOM attributes are changed. */
263 MOF_SOM_CHANGE = (1 << 0),
265 * The SOM recovery state for mdt object.
266 * This state is an in-memory equivalent of an absent SOM EA, used
267 * instead of invalidating SOM EA while IOEpoch is still opened when
268 * a client eviction occurs or a client fails to obtain SOM attributes.
269 * It indicates that the last IOEpoch holder will need to obtain SOM
270 * attributes under [0;EOF] extent lock to flush all the client's
271 * cached of evicted from MDS clients (but not necessary evicted from
272 * OST) before taking ost attributes.
274 MOF_SOM_RECOV = (1 << 1),
275 /** File has been just created. */
276 MOF_SOM_CREATED = (1 << 2),
277 /** lov object has been created. */
278 MOF_LOV_CREATED = (1 << 3),
281 struct mdt_lock_handle {
282 /* Lock type, reg for cross-ref use or pdo lock. */
286 struct lustre_handle mlh_reg_lh;
287 ldlm_mode_t mlh_reg_mode;
290 struct lustre_handle mlh_pdo_lh;
291 ldlm_mode_t mlh_pdo_mode;
292 unsigned int mlh_pdo_hash;
294 /* Remote regular lock */
295 struct lustre_handle mlh_rreg_lh;
296 ldlm_mode_t mlh_rreg_mode;
300 MDT_LH_PARENT, /* parent lockh */
301 MDT_LH_CHILD, /* child lockh */
302 MDT_LH_OLD, /* old lockh for rename */
303 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
304 MDT_LH_NEW, /* new lockh for rename */
305 MDT_LH_RMT, /* used for return lh to caller */
306 MDT_LH_LOCAL, /* local lock never return to client */
315 struct mdt_reint_record {
316 mdt_reint_t rr_opcode;
317 const struct lustre_handle *rr_handle;
318 const struct lu_fid *rr_fid1;
319 const struct lu_fid *rr_fid2;
324 const void *rr_eadata;
327 const struct llog_cookie *rr_logcookies;
331 enum mdt_reint_flag {
332 MRF_OPEN_TRUNC = 1 << 0,
335 struct mdt_thread_info;
337 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
341 tx_exec_func_t exec_fn;
342 tx_exec_func_t undo_fn;
343 struct dt_object *object;
345 struct update_reply *reply;
350 const struct dt_rec *rec;
351 const struct dt_key *key;
366 struct dt_allocation_hint hint;
367 struct dt_object_format dof;
375 struct ost_body *body;
380 #define TX_MAX_OPS 10
381 struct thandle_exec_args {
382 struct thandle *ta_handle;
383 struct dt_device *ta_dev;
385 struct tx_arg ta_args[TX_MAX_OPS];
386 int ta_argno; /* used args */
390 * Common data shared by mdt-level handlers. This is allocated per-thread to
391 * reduce stack consumption.
393 struct mdt_thread_info {
396 * The following members will be filled explicitly
397 * with specific data in mdt_thread_info_init().
399 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
400 * request handling may migrate from one server thread to another.
402 struct req_capsule *mti_pill;
404 /* although we have export in req, there are cases when it is not
405 * available, e.g. closing files upon export destroy */
406 struct obd_export *mti_exp;
408 * A couple of lock handles.
410 struct mdt_lock_handle mti_lh[MDT_LH_NR];
412 struct mdt_device *mti_mdt;
413 const struct lu_env *mti_env;
414 /* XXX: temporary flag to have healthy mti during OUT calls
415 * to be removed upon moving MDT to the unified target code */
419 * Additional fail id that can be set by handler. Passed to
420 * target_send_reply().
424 /* transaction number of current request */
430 * The following members will be filled expilictly
431 * with zero in mdt_thread_info_init(). These members may be used
438 struct md_attr mti_attr;
440 * Body for "habeo corpus" operations.
442 const struct mdt_body *mti_body;
444 * Host object. This is released at the end of mdt_handler().
446 struct mdt_object *mti_object;
448 * Lock request for "habeo clavis" operations.
450 const struct ldlm_request *mti_dlm_req;
452 __u32 mti_has_trans:1, /* has txn already? */
455 /* opdata for mdt_reint_open(), has the same as
456 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
457 * value onto disk for recovery when mdt_trans_stop_cb() is called.
463 * The following members will be filled explicitly
464 * with zero in mdt_reint_unpack(), because they are only used
465 * by reint requests (including mdt_reint_open()).
469 * reint record. contains information for reint operations.
471 struct mdt_reint_record mti_rr;
473 /** md objects included in operation */
474 struct mdt_object *mti_mos;
475 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
477 * Operation specification (currently create and lookup)
479 struct md_op_spec mti_spec;
483 * The following members will _NOT_ be initialized at all.
484 * DO NOT expect them to contain any valid value.
485 * They should be initialized explicitly by the user themselves.
488 /* XXX: If something is in a union, make sure they do not conflict */
490 struct lu_fid mti_tmp_fid1;
491 struct lu_fid mti_tmp_fid2;
492 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
493 * mdt_rename_lock() */
494 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
497 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
498 char ns_name[48]; /* for mdt_init0() */
499 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
500 struct obd_statfs osfs; /* for mdt_statfs() */
502 /* for mdt_readpage() */
503 struct lu_rdpg mti_rdpg;
504 /* for mdt_sendpage() */
505 struct l_wait_info mti_wait_info;
509 struct md_som_data data;
512 struct dt_object_format mti_update_dof;
513 struct update_reply *mti_update_reply;
514 struct update *mti_update;
515 int mti_update_reply_index;
516 struct obdo mti_obdo;
517 struct dt_object *mti_dt_object;
521 /* IO epoch related stuff. */
522 struct mdt_ioepoch *mti_ioepoch;
523 __u64 mti_replayepoch;
526 struct lu_buf mti_buf;
527 struct lu_buf mti_big_buf;
528 struct lustre_capa_key mti_capa_key;
530 /* Ops object filename */
531 struct lu_name mti_name;
532 /* per-thread values, can be re-used */
535 /* big_lmm buffer was used and must be used in reply */
536 int mti_big_lmm_used;
537 /* should be enough to fit lustre_mdt_attrs */
538 char mti_xattr_buf[128];
539 struct thandle_exec_args mti_handle;
540 struct ldlm_enqueue_info mti_einfo;
543 /* ptlrpc request handler for MDT. All handlers are
544 * grouped into several slices - struct mdt_opc_slice,
545 * and stored in an array - mdt_handlers[].
548 /* The name of this handler. */
550 /* Fail id for this handler, checked at the beginning of this handler*/
552 /* Operation code for this handler */
554 /* flags are listed in enum mdt_handler_flags below. */
556 /* The actual handler function to execute. */
557 int (*mh_act)(struct mdt_thread_info *info);
558 /* Request format for this request. */
559 const struct req_format *mh_fmt;
562 struct mdt_opc_slice {
565 struct mdt_handler *mos_hs;
568 struct cdt_req_progress {
569 struct mutex crp_lock; /**< protect tree */
570 struct interval_node *crp_root; /**< tree to track extent
572 struct interval_node **crp_node; /**< buffer for tree nodes
573 * vector of fixed size
575 int crp_cnt; /**< # of used nodes */
576 int crp_max; /**< # of allocated nodes */
579 struct cdt_agent_req {
580 cfs_list_t car_request_list; /**< to chain all the req. */
581 cfs_atomic_t car_refcount; /**< reference counter */
582 __u64 car_compound_id; /**< compound id */
583 __u64 car_flags; /**< request original flags */
584 struct obd_uuid car_uuid; /**< agent doing the req. */
585 __u32 car_archive_id; /**< archive id */
586 int car_canceled; /**< request was canceled */
587 cfs_time_t car_req_start; /**< start time */
588 cfs_time_t car_req_update; /**< last update time */
589 struct hsm_action_item *car_hai; /**< req. to the agent */
590 struct cdt_req_progress car_progress; /**< track data mvt
595 cfs_list_t ha_list; /**< to chain the agents */
596 struct obd_uuid ha_uuid; /**< agent uuid */
597 __u32 *ha_archive_id; /**< archive id */
598 int ha_archive_cnt; /**< number of archive entries
599 * 0 means any archive */
600 cfs_atomic_t ha_requests; /**< current request count */
601 cfs_atomic_t ha_success; /**< number of successful
603 cfs_atomic_t ha_failure; /**< number of failed actions */
606 struct cdt_restore_handle {
607 cfs_list_t crh_list; /**< to chain the handle */
608 struct lu_fid crh_fid; /**< fid of the object */
609 struct ldlm_extent crh_extent; /**< extent of the restore */
610 struct mdt_lock_handle crh_lh; /**< lock handle */
613 static inline const struct md_device_operations *
614 mdt_child_ops(struct mdt_device * m)
616 LASSERT(m->mdt_child);
617 return m->mdt_child->md_ops;
620 static inline struct md_object *mdt_object_child(struct mdt_object *o)
623 return lu2md(lu_object_next(&o->mot_obj));
626 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
628 return info->mti_pill ? info->mti_pill->rc_req : NULL;
631 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
633 LASSERT(info->mti_exp);
634 return exp_connect_flags(info->mti_exp);
637 static inline void mdt_object_get(const struct lu_env *env,
638 struct mdt_object *o)
641 lu_object_get(&o->mot_obj);
645 static inline void mdt_object_put(const struct lu_env *env,
646 struct mdt_object *o)
649 lu_object_put(env, &o->mot_obj);
653 static inline int mdt_object_exists(const struct mdt_object *o)
655 return lu_object_exists(&o->mot_obj);
658 static inline int mdt_object_remote(const struct mdt_object *o)
660 return lu_object_remote(&o->mot_obj);
663 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
665 return lu_object_fid(&o->mot_obj);
668 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
670 return mdt->mdt_lu_dev.ld_site;
673 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
675 return &mdt->mdt_seq_site;
678 static inline void mdt_export_evict(struct obd_export *exp)
680 class_fail_export(exp);
681 class_export_put(exp);
684 /* Here we use LVB_TYPE to check dne client, because it is
685 * also landed on 2.4. */
686 static inline int mdt_is_dne_client(struct obd_export *exp)
688 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
691 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
692 void mdt_set_disposition(struct mdt_thread_info *info,
693 struct ldlm_reply *rep, int flag);
694 void mdt_clear_disposition(struct mdt_thread_info *info,
695 struct ldlm_reply *rep, int flag);
697 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
698 ldlm_mode_t lm, const char *name,
701 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
704 int mdt_lock_setup(struct mdt_thread_info *info,
705 struct mdt_object *o,
706 struct mdt_lock_handle *lh);
708 int mdt_object_lock(struct mdt_thread_info *,
710 struct mdt_lock_handle *,
713 int mdt_object_lock_try(struct mdt_thread_info *,
715 struct mdt_lock_handle *,
718 void mdt_object_unlock(struct mdt_thread_info *,
720 struct mdt_lock_handle *,
723 struct mdt_object *mdt_object_new(const struct lu_env *,
725 const struct lu_fid *);
726 struct mdt_object *mdt_object_find(const struct lu_env *,
728 const struct lu_fid *);
729 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
730 const struct lu_fid *,
731 struct mdt_lock_handle *,
733 void mdt_object_unlock_put(struct mdt_thread_info *,
735 struct mdt_lock_handle *,
738 void mdt_client_compatibility(struct mdt_thread_info *info);
740 int mdt_remote_object_lock(struct mdt_thread_info *mti,
741 struct mdt_object *o, struct lustre_handle *lh,
742 ldlm_mode_t mode, __u64 ibits);
743 int mdt_close_unpack(struct mdt_thread_info *info);
744 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
745 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
746 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
747 const struct lu_attr *attr, const struct lu_fid *fid);
749 int mdt_getxattr(struct mdt_thread_info *info);
750 int mdt_reint_setxattr(struct mdt_thread_info *info,
751 struct mdt_lock_handle *lh);
753 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
754 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
756 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
757 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
758 struct mdt_lock_handle *lhc);
760 extern void target_recovery_fini(struct obd_device *obd);
761 extern void target_recovery_init(struct lu_target *lut,
762 svc_handler_t handler);
763 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
764 struct obd_device *, struct lustre_sb_info *lsi);
765 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
767 int mdt_export_stats_init(struct obd_device *obd,
768 struct obd_export *exp,
771 int mdt_pin(struct mdt_thread_info* info);
773 int mdt_lock_new_child(struct mdt_thread_info *info,
774 struct mdt_object *o,
775 struct mdt_lock_handle *child_lockh);
777 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
780 int mdt_reint_open(struct mdt_thread_info *info,
781 struct mdt_lock_handle *lhc);
783 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
784 const struct lustre_handle *handle,
788 MDT_IOEPOCH_CLOSED = 0,
789 MDT_IOEPOCH_OPENED = 1,
790 MDT_IOEPOCH_GETATTR = 2,
798 int mdt_attr_get_complex(struct mdt_thread_info *info,
799 struct mdt_object *o, struct md_attr *ma);
800 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
802 int mdt_object_is_som_enabled(struct mdt_object *mo);
803 int mdt_write_get(struct mdt_object *o);
804 void mdt_write_put(struct mdt_object *o);
805 int mdt_write_read(struct mdt_object *o);
806 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
807 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
808 void mdt_mfd_free(struct mdt_file_data *mfd);
809 int mdt_close(struct mdt_thread_info *info);
810 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
811 struct md_attr *ma, int flags);
812 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
814 int mdt_done_writing(struct mdt_thread_info *info);
815 int mdt_fix_reply(struct mdt_thread_info *info);
816 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
817 const struct md_attr *);
818 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
820 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
821 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
822 const void *area, ssize_t len);
824 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
826 int mdt_check_ucred(struct mdt_thread_info *);
827 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
828 int mdt_init_ucred_reint(struct mdt_thread_info *);
829 void mdt_exit_ucred(struct mdt_thread_info *);
830 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
831 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
832 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
834 int mdt_handle_common(struct ptlrpc_request *req,
835 struct mdt_opc_slice *supported);
836 int mdt_connect(struct mdt_thread_info *info);
837 int mdt_disconnect(struct mdt_thread_info *info);
838 int mdt_set_info(struct mdt_thread_info *info);
839 int mdt_get_info(struct mdt_thread_info *info);
840 int mdt_getstatus(struct mdt_thread_info *info);
841 int mdt_getattr(struct mdt_thread_info *info);
842 int mdt_getattr_name(struct mdt_thread_info *info);
843 int mdt_statfs(struct mdt_thread_info *info);
844 int mdt_reint(struct mdt_thread_info *info);
845 int mdt_sync(struct mdt_thread_info *info);
846 int mdt_is_subdir(struct mdt_thread_info *info);
847 int mdt_obd_ping(struct mdt_thread_info *info);
848 int mdt_obd_log_cancel(struct mdt_thread_info *info);
849 int mdt_obd_qc_callback(struct mdt_thread_info *info);
850 int mdt_enqueue(struct mdt_thread_info *info);
851 int mdt_convert(struct mdt_thread_info *info);
852 int mdt_bl_callback(struct mdt_thread_info *info);
853 int mdt_cp_callback(struct mdt_thread_info *info);
854 int mdt_llog_create(struct mdt_thread_info *info);
855 int mdt_llog_destroy(struct mdt_thread_info *info);
856 int mdt_llog_read_header(struct mdt_thread_info *info);
857 int mdt_llog_next_block(struct mdt_thread_info *info);
858 int mdt_llog_prev_block(struct mdt_thread_info *info);
859 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
860 int mdt_readpage(struct mdt_thread_info *info);
861 int mdt_obd_idx_read(struct mdt_thread_info *info);
862 int mdt_tgt_connect(struct tgt_session_info *tsi);
863 void mdt_thread_info_init(struct ptlrpc_request *req,
864 struct mdt_thread_info *mti);
865 void mdt_thread_info_fini(struct mdt_thread_info *mti);
867 extern struct mdt_opc_slice mdt_regular_handlers[];
868 extern struct mdt_opc_slice mdt_seq_handlers[];
869 extern struct mdt_opc_slice mdt_fld_handlers[];
871 int mdt_quotacheck(struct mdt_thread_info *info);
872 int mdt_quotactl(struct mdt_thread_info *info);
873 int mdt_quota_dqacq(struct mdt_thread_info *info);
874 int mdt_swap_layouts(struct mdt_thread_info *info);
876 extern struct lprocfs_vars lprocfs_mds_module_vars[];
877 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
879 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
882 struct mdt_handler *mdt_handler_find(__u32 opc,
883 struct mdt_opc_slice *supported);
885 int mdt_init_sec_level(struct mdt_thread_info *);
886 int mdt_init_idmap(struct mdt_thread_info *);
887 void mdt_cleanup_idmap(struct mdt_export_data *);
888 int mdt_handle_idmap(struct mdt_thread_info *);
889 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
890 struct ptlrpc_user_desc *);
891 void mdt_body_reverse_idmap(struct mdt_thread_info *,
893 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
894 struct mdt_remote_perm *);
895 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
897 static inline struct mdt_device *mdt_dev(struct lu_device *d)
899 return container_of0(d, struct mdt_device, mdt_lu_dev);
902 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
904 struct lu_object *lo;
905 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
907 lo = lu_object_locate(mo->mot_obj.lo_header,
908 mdt->mdt_bottom->dd_lu_dev.ld_type);
913 /* mdt/mdt_identity.c */
914 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
916 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
918 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
920 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
922 void mdt_flush_identity(struct upcall_cache *, int);
924 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
926 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
929 int mdt_hsm_state_get(struct mdt_thread_info *info);
930 int mdt_hsm_state_set(struct mdt_thread_info *info);
931 int mdt_hsm_action(struct mdt_thread_info *info);
932 int mdt_hsm_progress(struct mdt_thread_info *info);
933 int mdt_hsm_ct_register(struct mdt_thread_info *info);
934 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
935 int mdt_hsm_request(struct mdt_thread_info *info);
936 /* mdt/mdt_hsm_cdt_actions.c */
937 extern const struct file_operations mdt_agent_actions_fops;
938 void dump_llog_agent_req_rec(char *prefix, struct llog_agent_req_rec *larr);
939 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
940 llog_cb_t cb, void *data);
941 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
942 __u64 compound_id, __u32 archive_id,
943 __u64 flags, struct hsm_action_item *hai);
944 int mdt_agent_record_update(const struct lu_env *env,
945 struct mdt_device *mdt, __u64 *cookies,
946 int cookies_count, enum agent_req_status status);
947 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
948 struct llog_handle *llh,
949 struct llog_agent_req_rec *larr);
951 /* mdt/mdt_hsm_cdt_agent.c */
952 extern const struct file_operations mdt_hsm_agent_fops;
953 int mdt_hsm_agent_register(struct mdt_thread_info *info,
954 const struct obd_uuid *uuid,
955 int nr_archives, __u32 *archive_num);
956 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
957 const struct obd_uuid *uuid,
959 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
960 const struct obd_uuid *uuid);
961 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
962 int succ_rq, int fail_rq, int new_rq,
963 const struct obd_uuid *uuid);
964 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
965 struct obd_uuid *uuid);
966 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
968 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
969 struct hsm_progress_kernel *pgs);
970 /* mdt/mdt_hsm_cdt_client.c */
971 int mdt_hsm_add_actions(struct mdt_thread_info *info,
972 struct hsm_action_list *hal, __u64 *compound_id);
973 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
974 struct hsm_action_list *hal);
975 int mdt_hsm_get_running(struct mdt_thread_info *mti,
976 struct hsm_action_list *hal);
977 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
978 const struct lu_fid *fid);
980 /* mdt/mdt_hsm_cdt_requests.c */
981 extern const struct file_operations mdt_hsm_request_fops;
982 void dump_requests(char *prefix, struct coordinator *cdt);
983 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
984 __u64 flags, struct obd_uuid *uuid,
985 struct hsm_action_item *hai);
986 void mdt_cdt_free_request(struct cdt_agent_req *car);
987 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
988 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
990 const struct lu_fid *fid);
991 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
992 void mdt_cdt_get_request(struct cdt_agent_req *car);
993 void mdt_cdt_put_request(struct cdt_agent_req *car);
994 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
995 struct hsm_progress_kernel *pgs);
996 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
998 /* fake functions, will be remove with patch LU-3343 */
999 static inline struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1002 struct mdt_lock_handle *lh)
1004 return ERR_PTR(-EINVAL);
1006 static inline bool mdt_hsm_is_action_compat(struct hsm_action_item *hai,
1007 int hal_an, __u64 rq_flags,
1012 static inline int mdt_hsm_cdt_init(struct mdt_device *mdt)
1014 struct coordinator *cdt = &mdt->mdt_coordinator;
1016 /* minimal init before final patch landing */
1017 sema_init(&cdt->cdt_llog_lock, 1);
1018 init_rwsem(&cdt->cdt_agent_lock);
1019 init_rwsem(&cdt->cdt_request_lock);
1020 sema_init(&cdt->cdt_restore_lock, 1);
1022 CFS_INIT_LIST_HEAD(&cdt->cdt_requests);
1023 CFS_INIT_LIST_HEAD(&cdt->cdt_agents);
1024 CFS_INIT_LIST_HEAD(&cdt->cdt_restore_hdl);
1026 cdt->cdt_state = CDT_STOPPED;
1029 static inline int mdt_hsm_cdt_start(struct mdt_device *mdt)
1033 static inline int mdt_hsm_cdt_stop(struct mdt_device *mdt)
1037 static inline int mdt_hsm_cdt_fini(struct mdt_device *mdt)
1041 static inline int mdt_hsm_cdt_wakeup(struct mdt_device *mdt)
1045 static inline int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1046 struct hsm_action_list *hal,
1047 const struct obd_uuid *uuid)
1051 static inline int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1052 struct hsm_progress_kernel *pgs,
1057 /* end of fake functions */
1059 extern struct lu_context_key mdt_thread_key;
1060 /* debug issues helper starts here*/
1061 static inline int mdt_fail_write(const struct lu_env *env,
1062 struct dt_device *dd, int id)
1064 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1065 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1067 return dd->dd_ops->dt_ro(env, dd);
1068 /* We set FAIL_ONCE because we never "un-fail" a device */
1074 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1076 return &req->rq_export->exp_mdt_data;
1079 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1080 struct mdt_lock_handle *lhc);
1081 static inline int mdt_check_resent(struct mdt_thread_info *info,
1082 mdt_reconstruct_t reconstruct,
1083 struct mdt_lock_handle *lhc)
1085 struct ptlrpc_request *req = mdt_info_req(info);
1088 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1089 if (req_xid_is_last(req)) {
1090 reconstruct(info, lhc);
1093 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
1094 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
1099 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1100 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1102 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1104 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1107 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1109 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1110 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1111 struct lustre_handle *lh,
1113 ldlm_policy_data_t *policy,
1114 const struct ldlm_res_id *res_id,
1115 __u64 flags, const __u64 *client_cookie)
1119 LASSERT(ns != NULL);
1120 LASSERT(lh != NULL);
1122 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1123 mode, &flags, mdt_blocking_ast,
1124 ldlm_completion_ast, NULL, NULL, 0,
1125 LVB_T_NONE, client_cookie, lh);
1126 return rc == ELDLM_OK ? 0 : -EIO;
1129 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1132 ldlm_lock_decref(lh, mode);
1135 extern mdl_mode_t mdt_mdl_lock_modes[];
1136 extern ldlm_mode_t mdt_dlm_lock_modes[];
1138 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1140 LASSERT(IS_PO2(mode));
1141 return mdt_mdl_lock_modes[mode];
1144 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1146 LASSERT(IS_PO2(mode));
1147 return mdt_dlm_lock_modes[mode];
1151 extern struct ldlm_valblock_ops mdt_lvbo;
1153 static inline struct lu_name *mdt_name(const struct lu_env *env,
1154 char *name, int namelen)
1156 struct lu_name *lname;
1157 struct mdt_thread_info *mti;
1159 LASSERT(namelen > 0);
1160 /* trailing '\0' in buffer */
1161 LASSERT(name[namelen] == '\0');
1163 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
1164 lname = &mti->mti_name;
1165 lname->ln_name = name;
1166 lname->ln_namelen = namelen;
1170 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
1171 struct lu_name *slname)
1176 tlname->ln_name = slname->ln_name;
1177 tlname->ln_namelen = slname->ln_namelen;
1181 void mdt_enable_cos(struct mdt_device *, int);
1182 int mdt_cos_is_enabled(struct mdt_device *);
1200 LPROC_MDT_SAMEDIR_RENAME,
1201 LPROC_MDT_CROSSDIR_RENAME,
1204 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1205 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1206 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1207 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1208 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1209 void mdt_procfs_fini(struct mdt_device *mdt);
1210 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1211 struct mdt_device *mdt,
1212 struct ptlrpc_request *req,
1213 struct mdt_object *src, struct mdt_object *tgt);
1216 int mdt_ck_thread_start(struct mdt_device *mdt);
1217 void mdt_ck_thread_stop(struct mdt_device *mdt);
1218 void mdt_ck_timer_callback(unsigned long castmeharder);
1219 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1221 static inline void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1222 const struct lu_fid *fid,
1223 struct lustre_capa *capa)
1225 struct md_capainfo *ci;
1227 LASSERT(offset >= 0 && offset < MD_CAPAINFO_MAX);
1228 if (!info->mti_mdt->mdt_opts.mo_mds_capa ||
1229 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
1232 ci = md_capainfo(info->mti_env);
1234 ci->mc_fid[offset] = *fid;
1235 ci->mc_capa[offset] = capa;
1238 static inline void mdt_dump_capainfo(struct mdt_thread_info *info)
1240 struct md_capainfo *ci = md_capainfo(info->mti_env);
1245 for (i = 0; i < MD_CAPAINFO_MAX; i++) {
1246 if (!ci->mc_capa[i]) {
1247 CERROR("no capa for index %d "DFID"\n",
1248 i, PFID(&ci->mc_fid[i]));
1251 if (ci->mc_capa[i] == BYPASS_CAPA) {
1252 CERROR("bypass for index %d "DFID"\n",
1253 i, PFID(&ci->mc_fid[i]));
1256 DEBUG_CAPA(D_ERROR, ci->mc_capa[i], "index %d", i);
1260 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1262 return mdt->mdt_lu_dev.ld_obd;
1265 extern const struct lu_device_operations mdt_lu_ops;
1267 static inline int lu_device_is_mdt(struct lu_device *d)
1269 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1272 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1274 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1275 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1276 return container_of0(d, struct mdt_device, mdt_lu_dev);
1279 static inline char *mdt_obd_name(struct mdt_device *mdt)
1281 return mdt->mdt_lu_dev.ld_obd->obd_name;
1284 int mds_mod_init(void);
1285 void mds_mod_exit(void);
1287 #endif /* __KERNEL__ */