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 <obd_class.h>
54 #include <lustre_net.h>
55 #include <lustre/lustre_idl.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_mdt.h>
62 #include <lustre_req_layout.h>
63 #include <lustre_sec.h>
65 #include <lustre_idmap.h>
66 #include <lustre_eacl.h>
67 #include <lustre_quota.h>
69 /* check if request's xid is equal to last one or not*/
70 static inline int req_xid_is_last(struct ptlrpc_request *req)
72 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
73 return (req->rq_xid == lcd->lcd_last_xid ||
74 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 int 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 /* mdt state flag bits */
89 #define MDT_FL_CFGLOG 0
90 #define MDT_FL_SYNCED 1
94 struct md_device mdt_md_dev;
95 struct seq_server_site mdt_seq_site;
96 struct ptlrpc_service *mdt_regular_service;
97 struct ptlrpc_service *mdt_readpage_service;
98 struct ptlrpc_service *mdt_out_service;
99 struct ptlrpc_service *mdt_setattr_service;
100 struct ptlrpc_service *mdt_mdsc_service;
101 struct ptlrpc_service *mdt_mdss_service;
102 struct ptlrpc_service *mdt_dtss_service;
103 struct ptlrpc_service *mdt_fld_service;
104 /* DLM name-space for meta-data locks maintained by this server */
105 struct ldlm_namespace *mdt_namespace;
106 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
107 struct ptlrpc_client *mdt_ldlm_client;
108 /* underlying device */
109 struct obd_export *mdt_child_exp;
110 struct md_device *mdt_child;
111 struct dt_device *mdt_bottom;
112 struct obd_export *mdt_bottom_exp;
114 struct lu_target mdt_lut;
116 * Options bit-fields.
119 unsigned int mo_user_xattr:1,
126 /* mdt state flags */
127 unsigned long mdt_state;
128 /* lock to protect IOepoch */
129 spinlock_t mdt_ioepoch_lock;
132 /* transaction callbacks */
133 struct dt_txn_callback mdt_txn_cb;
135 /* these values should be updated from lov if necessary.
136 * or should be placed somewhere else. */
138 int mdt_max_cookiesize;
140 struct upcall_cache *mdt_identity_cache;
143 rwlock_t mdt_sptlrpc_lock;
144 struct sptlrpc_rule_set mdt_sptlrpc_rset;
146 /* capability keys */
147 unsigned long mdt_capa_timeout;
149 struct dt_object *mdt_ck_obj;
150 unsigned long mdt_ck_timeout;
151 unsigned long mdt_ck_expiry;
152 cfs_timer_t mdt_ck_timer;
153 struct ptlrpc_thread mdt_ck_thread;
154 struct lustre_capa_key mdt_capa_keys[2];
155 unsigned int mdt_capa_conf:1,
157 /* Enable remote dir on non-MDT0 */
158 mdt_enable_remote_dir:1;
160 gid_t mdt_enable_remote_dir_gid;
161 /* statfs optimization: we cache a bit */
162 struct obd_statfs mdt_osfs;
164 spinlock_t mdt_osfs_lock;
167 uid_t mdt_squash_uid;
168 gid_t mdt_squash_gid;
169 cfs_list_t mdt_nosquash_nids;
170 char *mdt_nosquash_str;
171 int mdt_nosquash_strlen;
172 struct rw_semaphore mdt_squash_sem;
175 struct rename_stats mdt_rename_stats;
176 struct lu_fid mdt_md_root_fid;
178 /* connection to quota master */
179 struct obd_export *mdt_qmt_exp;
180 /* quota master device associated with this MDT */
181 struct lu_device *mdt_qmt_dev;
184 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
185 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
186 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
187 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
188 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
189 #define MDT_COS_DEFAULT (0)
192 struct lu_object_header mot_header;
193 struct md_object mot_obj;
196 int mot_ioepoch_count;
198 /* Lock to protect object's IO epoch. */
199 struct mutex mot_ioepoch_mutex;
200 /* Lock to protect create_data */
201 struct mutex mot_lov_mutex;
204 enum mdt_object_flags {
205 /** SOM attributes are changed. */
206 MOF_SOM_CHANGE = (1 << 0),
208 * The SOM recovery state for mdt object.
209 * This state is an in-memory equivalent of an absent SOM EA, used
210 * instead of invalidating SOM EA while IOEpoch is still opened when
211 * a client eviction occurs or a client fails to obtain SOM attributes.
212 * It indicates that the last IOEpoch holder will need to obtain SOM
213 * attributes under [0;EOF] extent lock to flush all the client's
214 * cached of evicted from MDS clients (but not necessary evicted from
215 * OST) before taking ost attributes.
217 MOF_SOM_RECOV = (1 << 1),
218 /** File has been just created. */
219 MOF_SOM_CREATED = (1 << 2),
220 /** lov object has been created. */
221 MOF_LOV_CREATED = (1 << 3),
224 struct mdt_lock_handle {
225 /* Lock type, reg for cross-ref use or pdo lock. */
229 struct lustre_handle mlh_reg_lh;
230 ldlm_mode_t mlh_reg_mode;
233 struct lustre_handle mlh_pdo_lh;
234 ldlm_mode_t mlh_pdo_mode;
235 unsigned int mlh_pdo_hash;
237 /* Remote regular lock */
238 struct lustre_handle mlh_rreg_lh;
239 ldlm_mode_t mlh_rreg_mode;
243 MDT_LH_PARENT, /* parent lockh */
244 MDT_LH_CHILD, /* child lockh */
245 MDT_LH_OLD, /* old lockh for rename */
246 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
247 MDT_LH_NEW, /* new lockh for rename */
248 MDT_LH_RMT, /* used for return lh to caller */
257 struct mdt_reint_record {
258 mdt_reint_t rr_opcode;
259 const struct lustre_handle *rr_handle;
260 const struct lu_fid *rr_fid1;
261 const struct lu_fid *rr_fid2;
266 const void *rr_eadata;
269 const struct llog_cookie *rr_logcookies;
273 enum mdt_reint_flag {
274 MRF_OPEN_TRUNC = 1 << 0,
277 struct mdt_thread_info;
279 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
283 tx_exec_func_t exec_fn;
284 tx_exec_func_t undo_fn;
285 struct dt_object *object;
287 struct update_reply *reply;
292 const struct dt_rec *rec;
293 const struct dt_key *key;
308 struct dt_allocation_hint hint;
309 struct dt_object_format dof;
317 struct ost_body *body;
322 #define TX_MAX_OPS 10
323 struct thandle_exec_args {
324 struct thandle *ta_handle;
325 struct dt_device *ta_dev;
327 struct tx_arg ta_args[TX_MAX_OPS];
328 int ta_argno; /* used args */
332 * Common data shared by mdt-level handlers. This is allocated per-thread to
333 * reduce stack consumption.
335 struct mdt_thread_info {
338 * The following members will be filled explicitly
339 * with specific data in mdt_thread_info_init().
341 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
342 * request handling may migrate from one server thread to another.
344 struct req_capsule *mti_pill;
346 /* although we have export in req, there are cases when it is not
347 * available, e.g. closing files upon export destroy */
348 struct obd_export *mti_exp;
350 * A couple of lock handles.
352 struct mdt_lock_handle mti_lh[MDT_LH_NR];
354 struct mdt_device *mti_mdt;
355 const struct lu_env *mti_env;
358 * Additional fail id that can be set by handler. Passed to
359 * target_send_reply().
363 /* transaction number of current request */
369 * The following members will be filled expilictly
370 * with zero in mdt_thread_info_init(). These members may be used
377 struct md_attr mti_attr;
379 * Body for "habeo corpus" operations.
381 const struct mdt_body *mti_body;
383 * Host object. This is released at the end of mdt_handler().
385 struct mdt_object *mti_object;
387 * Lock request for "habeo clavis" operations.
389 const struct ldlm_request *mti_dlm_req;
391 __u32 mti_has_trans:1, /* has txn already? */
394 /* opdata for mdt_reint_open(), has the same as
395 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
396 * value onto disk for recovery when mdt_trans_stop_cb() is called.
402 * The following members will be filled explicitly
403 * with zero in mdt_reint_unpack(), because they are only used
404 * by reint requests (including mdt_reint_open()).
408 * reint record. contains information for reint operations.
410 struct mdt_reint_record mti_rr;
412 /** md objects included in operation */
413 struct mdt_object *mti_mos;
414 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
416 * Operation specification (currently create and lookup)
418 struct md_op_spec mti_spec;
422 * The following members will _NOT_ be initialized at all.
423 * DO NOT expect them to contain any valid value.
424 * They should be initialized explicitly by the user themselves.
427 /* XXX: If something is in a union, make sure they do not conflict */
429 struct lu_fid mti_tmp_fid1;
430 struct lu_fid mti_tmp_fid2;
431 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
432 * mdt_rename_lock() */
433 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
436 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
437 char ns_name[48]; /* for mdt_init0() */
438 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
439 struct obd_statfs osfs; /* for mdt_statfs() */
441 /* for mdt_readpage() */
442 struct lu_rdpg mti_rdpg;
443 /* for mdt_sendpage() */
444 struct l_wait_info mti_wait_info;
448 struct md_som_data data;
451 struct dt_object_format mti_update_dof;
452 struct update_reply *mti_update_reply;
453 struct update *mti_update;
454 int mti_update_reply_index;
455 struct obdo mti_obdo;
456 struct dt_object *mti_dt_object;
460 /* IO epoch related stuff. */
461 struct mdt_ioepoch *mti_ioepoch;
462 __u64 mti_replayepoch;
465 struct lu_buf mti_buf;
466 struct lu_buf mti_big_buf;
467 struct lustre_capa_key mti_capa_key;
469 /* Ops object filename */
470 struct lu_name mti_name;
471 /* per-thread values, can be re-used */
474 /* big_lmm buffer was used and must be used in reply */
475 int mti_big_lmm_used;
476 /* should be enough to fit lustre_mdt_attrs */
477 char mti_xattr_buf[128];
478 struct thandle_exec_args mti_handle;
479 struct ldlm_enqueue_info mti_einfo;
482 /* ptlrpc request handler for MDT. All handlers are
483 * grouped into several slices - struct mdt_opc_slice,
484 * and stored in an array - mdt_handlers[].
487 /* The name of this handler. */
489 /* Fail id for this handler, checked at the beginning of this handler*/
491 /* Operation code for this handler */
493 /* flags are listed in enum mdt_handler_flags below. */
495 /* The actual handler function to execute. */
496 int (*mh_act)(struct mdt_thread_info *info);
497 /* Request format for this request. */
498 const struct req_format *mh_fmt;
501 enum mdt_handler_flags {
503 * struct mdt_body is passed in the incoming message, and object
504 * identified by this fid exists on disk.
506 * "habeo corpus" == "I have a body"
508 HABEO_CORPUS = (1 << 0),
510 * struct ldlm_request is passed in the incoming message.
512 * "habeo clavis" == "I have a key"
514 HABEO_CLAVIS = (1 << 1),
516 * this request has fixed reply format, so that reply message can be
517 * packed by generic code.
519 * "habeo refero" == "I have a reply"
521 HABEO_REFERO = (1 << 2),
523 * this request will modify something, so check whether the filesystem
524 * is readonly or not, then return -EROFS to client asap if necessary.
526 * "mutabor" == "I shall modify"
531 struct mdt_opc_slice {
534 struct mdt_handler *mos_hs;
537 static inline const struct md_device_operations *
538 mdt_child_ops(struct mdt_device * m)
540 LASSERT(m->mdt_child);
541 return m->mdt_child->md_ops;
544 static inline struct md_object *mdt_object_child(struct mdt_object *o)
547 return lu2md(lu_object_next(&o->mot_obj.mo_lu));
550 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
552 return info->mti_pill ? info->mti_pill->rc_req : NULL;
555 static inline int req_is_replay(struct ptlrpc_request *req)
557 LASSERT(req->rq_reqmsg);
558 return !!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
561 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
563 LASSERT(info->mti_exp);
564 return exp_connect_flags(info->mti_exp);
567 static inline void mdt_object_get(const struct lu_env *env,
568 struct mdt_object *o)
571 lu_object_get(&o->mot_obj.mo_lu);
575 static inline void mdt_object_put(const struct lu_env *env,
576 struct mdt_object *o)
579 lu_object_put(env, &o->mot_obj.mo_lu);
583 static inline int mdt_object_exists(const struct mdt_object *o)
585 return lu_object_exists(&o->mot_obj.mo_lu);
588 static inline int mdt_object_remote(const struct mdt_object *o)
590 return lu_object_remote(&o->mot_obj.mo_lu);
593 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
595 return lu_object_fid(&o->mot_obj.mo_lu);
598 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
600 return mdt->mdt_md_dev.md_lu_dev.ld_site;
603 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
605 return &mdt->mdt_seq_site;
608 static inline void mdt_export_evict(struct obd_export *exp)
610 class_fail_export(exp);
611 class_export_put(exp);
614 /* Here we use LVB_TYPE to check dne client, because it is
615 * also landed on 2.4. */
616 static inline int mdt_is_dne_client(struct obd_export *exp)
618 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
621 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
622 void mdt_set_disposition(struct mdt_thread_info *info,
623 struct ldlm_reply *rep, int flag);
624 void mdt_clear_disposition(struct mdt_thread_info *info,
625 struct ldlm_reply *rep, int flag);
627 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
628 ldlm_mode_t lm, const char *name,
631 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
634 int mdt_lock_setup(struct mdt_thread_info *info,
635 struct mdt_object *o,
636 struct mdt_lock_handle *lh);
638 int mdt_object_lock(struct mdt_thread_info *,
640 struct mdt_lock_handle *,
643 int mdt_object_lock_try(struct mdt_thread_info *,
645 struct mdt_lock_handle *,
648 void mdt_object_unlock(struct mdt_thread_info *,
650 struct mdt_lock_handle *,
653 struct mdt_object *mdt_object_new(const struct lu_env *,
655 const struct lu_fid *);
656 struct mdt_object *mdt_object_find(const struct lu_env *,
658 const struct lu_fid *);
659 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
660 const struct lu_fid *,
661 struct mdt_lock_handle *,
663 void mdt_object_unlock_put(struct mdt_thread_info *,
665 struct mdt_lock_handle *,
668 void mdt_client_compatibility(struct mdt_thread_info *info);
670 int mdt_remote_object_lock(struct mdt_thread_info *mti,
671 struct mdt_object *o, struct lustre_handle *lh,
672 ldlm_mode_t mode, __u64 ibits);
673 int mdt_close_unpack(struct mdt_thread_info *info);
674 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
675 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
676 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
677 const struct lu_attr *attr, const struct lu_fid *fid);
679 int mdt_getxattr(struct mdt_thread_info *info);
680 int mdt_reint_setxattr(struct mdt_thread_info *info,
681 struct mdt_lock_handle *lh);
683 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
684 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
686 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
687 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
688 struct mdt_lock_handle *lhc);
690 extern void target_recovery_fini(struct obd_device *obd);
691 extern void target_recovery_init(struct lu_target *lut,
692 svc_handler_t handler);
693 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
694 struct obd_device *, struct lustre_sb_info *lsi);
695 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
697 int mdt_export_stats_init(struct obd_device *obd,
698 struct obd_export *exp,
701 int mdt_pin(struct mdt_thread_info* info);
703 int mdt_lock_new_child(struct mdt_thread_info *info,
704 struct mdt_object *o,
705 struct mdt_lock_handle *child_lockh);
707 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
710 int mdt_reint_open(struct mdt_thread_info *info,
711 struct mdt_lock_handle *lhc);
713 struct mdt_file_data *mdt_handle2mfd(struct mdt_thread_info *,
714 const struct lustre_handle *);
717 MDT_IOEPOCH_CLOSED = 0,
718 MDT_IOEPOCH_OPENED = 1,
719 MDT_IOEPOCH_GETATTR = 2,
727 int mdt_attr_get_complex(struct mdt_thread_info *info,
728 struct mdt_object *o, struct md_attr *ma);
729 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
731 int mdt_object_is_som_enabled(struct mdt_object *mo);
732 int mdt_write_get(struct mdt_object *o);
733 void mdt_write_put(struct mdt_object *o);
734 int mdt_write_read(struct mdt_object *o);
735 struct mdt_file_data *mdt_mfd_new(void);
736 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
737 void mdt_mfd_free(struct mdt_file_data *mfd);
738 int mdt_close(struct mdt_thread_info *info);
739 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
740 struct md_attr *ma, int flags);
741 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
743 int mdt_done_writing(struct mdt_thread_info *info);
744 int mdt_fix_reply(struct mdt_thread_info *info);
745 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
746 const struct md_attr *);
747 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
749 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
750 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
751 const void *area, ssize_t len);
753 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
755 int mdt_check_ucred(struct mdt_thread_info *);
756 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
757 int mdt_init_ucred_reint(struct mdt_thread_info *);
758 void mdt_exit_ucred(struct mdt_thread_info *);
759 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
760 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
761 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
763 int mdt_handle_common(struct ptlrpc_request *req,
764 struct mdt_opc_slice *supported);
765 int mdt_connect(struct mdt_thread_info *info);
766 int mdt_disconnect(struct mdt_thread_info *info);
767 int mdt_set_info(struct mdt_thread_info *info);
768 int mdt_get_info(struct mdt_thread_info *info);
769 int mdt_getstatus(struct mdt_thread_info *info);
770 int mdt_getattr(struct mdt_thread_info *info);
771 int mdt_getattr_name(struct mdt_thread_info *info);
772 int mdt_statfs(struct mdt_thread_info *info);
773 int mdt_reint(struct mdt_thread_info *info);
774 int mdt_sync(struct mdt_thread_info *info);
775 int mdt_is_subdir(struct mdt_thread_info *info);
776 int mdt_obd_ping(struct mdt_thread_info *info);
777 int mdt_obd_log_cancel(struct mdt_thread_info *info);
778 int mdt_obd_qc_callback(struct mdt_thread_info *info);
779 int mdt_enqueue(struct mdt_thread_info *info);
780 int mdt_convert(struct mdt_thread_info *info);
781 int mdt_bl_callback(struct mdt_thread_info *info);
782 int mdt_cp_callback(struct mdt_thread_info *info);
783 int mdt_llog_create(struct mdt_thread_info *info);
784 int mdt_llog_destroy(struct mdt_thread_info *info);
785 int mdt_llog_read_header(struct mdt_thread_info *info);
786 int mdt_llog_next_block(struct mdt_thread_info *info);
787 int mdt_llog_prev_block(struct mdt_thread_info *info);
788 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
789 int mdt_readpage(struct mdt_thread_info *info);
790 int mdt_obd_idx_read(struct mdt_thread_info *info);
792 extern struct mdt_opc_slice mdt_regular_handlers[];
793 extern struct mdt_opc_slice mdt_seq_handlers[];
794 extern struct mdt_opc_slice mdt_fld_handlers[];
796 int mdt_quotacheck(struct mdt_thread_info *info);
797 int mdt_quotactl(struct mdt_thread_info *info);
798 int mdt_quota_dqacq(struct mdt_thread_info *info);
799 int mdt_swap_layouts(struct mdt_thread_info *info);
801 extern struct lprocfs_vars lprocfs_mds_module_vars[];
802 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
804 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
807 struct mdt_handler *mdt_handler_find(__u32 opc,
808 struct mdt_opc_slice *supported);
810 int mdt_init_sec_level(struct mdt_thread_info *);
811 int mdt_init_idmap(struct mdt_thread_info *);
812 void mdt_cleanup_idmap(struct mdt_export_data *);
813 int mdt_handle_idmap(struct mdt_thread_info *);
814 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
815 struct ptlrpc_user_desc *);
816 void mdt_body_reverse_idmap(struct mdt_thread_info *,
818 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
819 struct mdt_remote_perm *);
820 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
822 static inline struct mdt_device *mdt_dev(struct lu_device *d)
824 // LASSERT(lu_device_is_mdt(d));
825 return container_of0(d, struct mdt_device, mdt_md_dev.md_lu_dev);
828 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
830 struct lu_object *lo;
831 struct mdt_device *mdt = mdt_dev(mo->mot_obj.mo_lu.lo_dev);
833 lo = lu_object_locate(mo->mot_obj.mo_lu.lo_header,
834 mdt->mdt_bottom->dd_lu_dev.ld_type);
838 /* mdt/mdt_identity.c */
839 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
841 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
843 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
845 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
847 void mdt_flush_identity(struct upcall_cache *, int);
849 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
851 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
854 int mdt_hsm_state_get(struct mdt_thread_info *info);
855 int mdt_hsm_state_set(struct mdt_thread_info *info);
856 int mdt_hsm_action(struct mdt_thread_info *info);
857 int mdt_hsm_progress(struct mdt_thread_info *info);
858 int mdt_hsm_ct_register(struct mdt_thread_info *info);
859 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
860 int mdt_hsm_request(struct mdt_thread_info *info);
862 extern struct lu_context_key mdt_thread_key;
863 /* debug issues helper starts here*/
864 static inline int mdt_fail_write(const struct lu_env *env,
865 struct dt_device *dd, int id)
867 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
868 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
870 return dd->dd_ops->dt_ro(env, dd);
871 /* We set FAIL_ONCE because we never "un-fail" a device */
877 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
879 return &req->rq_export->exp_mdt_data;
882 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
883 struct mdt_lock_handle *lhc);
884 static inline int mdt_check_resent(struct mdt_thread_info *info,
885 mdt_reconstruct_t reconstruct,
886 struct mdt_lock_handle *lhc)
888 struct ptlrpc_request *req = mdt_info_req(info);
891 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
892 if (req_xid_is_last(req)) {
893 reconstruct(info, lhc);
896 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
897 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
902 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
903 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
905 static inline int is_identity_get_disabled(struct upcall_cache *cache)
907 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
910 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
912 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
913 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
914 struct lustre_handle *lh,
916 ldlm_policy_data_t *policy,
917 const struct ldlm_res_id *res_id,
918 __u64 flags, const __u64 *client_cookie)
925 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
926 mode, &flags, mdt_blocking_ast,
927 ldlm_completion_ast, NULL, NULL, 0,
928 LVB_T_NONE, client_cookie, lh);
929 return rc == ELDLM_OK ? 0 : -EIO;
932 static inline void mdt_fid_unlock(struct lustre_handle *lh,
935 ldlm_lock_decref(lh, mode);
938 extern mdl_mode_t mdt_mdl_lock_modes[];
939 extern ldlm_mode_t mdt_dlm_lock_modes[];
941 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
943 LASSERT(IS_PO2(mode));
944 return mdt_mdl_lock_modes[mode];
947 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
949 LASSERT(IS_PO2(mode));
950 return mdt_dlm_lock_modes[mode];
954 extern struct ldlm_valblock_ops mdt_lvbo;
956 static inline struct lu_name *mdt_name(const struct lu_env *env,
957 char *name, int namelen)
959 struct lu_name *lname;
960 struct mdt_thread_info *mti;
962 LASSERT(namelen > 0);
963 /* trailing '\0' in buffer */
964 LASSERT(name[namelen] == '\0');
966 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
967 lname = &mti->mti_name;
968 lname->ln_name = name;
969 lname->ln_namelen = namelen;
973 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
974 struct lu_name *slname)
979 tlname->ln_name = slname->ln_name;
980 tlname->ln_namelen = slname->ln_namelen;
984 void mdt_enable_cos(struct mdt_device *, int);
985 int mdt_cos_is_enabled(struct mdt_device *);
986 int mdt_hsm_copytool_send(struct obd_export *exp);
1004 LPROC_MDT_SAMEDIR_RENAME,
1005 LPROC_MDT_CROSSDIR_RENAME,
1008 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1009 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1010 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1011 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1012 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1013 void mdt_procfs_fini(struct mdt_device *mdt);
1014 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1015 struct mdt_device *mdt,
1016 struct ptlrpc_request *req,
1017 struct mdt_object *src, struct mdt_object *tgt);
1020 int mdt_ck_thread_start(struct mdt_device *mdt);
1021 void mdt_ck_thread_stop(struct mdt_device *mdt);
1022 void mdt_ck_timer_callback(unsigned long castmeharder);
1023 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1025 static inline void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1026 const struct lu_fid *fid,
1027 struct lustre_capa *capa)
1029 struct md_capainfo *ci;
1031 LASSERT(offset >= 0 && offset < MD_CAPAINFO_MAX);
1032 if (!info->mti_mdt->mdt_opts.mo_mds_capa ||
1033 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
1036 ci = md_capainfo(info->mti_env);
1038 ci->mc_fid[offset] = *fid;
1039 ci->mc_capa[offset] = capa;
1042 static inline void mdt_dump_capainfo(struct mdt_thread_info *info)
1044 struct md_capainfo *ci = md_capainfo(info->mti_env);
1049 for (i = 0; i < MD_CAPAINFO_MAX; i++) {
1050 if (!ci->mc_capa[i]) {
1051 CERROR("no capa for index %d "DFID"\n",
1052 i, PFID(&ci->mc_fid[i]));
1055 if (ci->mc_capa[i] == BYPASS_CAPA) {
1056 CERROR("bypass for index %d "DFID"\n",
1057 i, PFID(&ci->mc_fid[i]));
1060 DEBUG_CAPA(D_ERROR, ci->mc_capa[i], "index %d", i);
1064 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1066 return mdt->mdt_md_dev.md_lu_dev.ld_obd;
1069 extern const struct lu_device_operations mdt_lu_ops;
1071 static inline int lu_device_is_mdt(struct lu_device *d)
1073 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1076 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1078 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1079 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1080 return container_of0(d, struct mdt_device, mdt_md_dev.md_lu_dev);
1083 static inline char *mdt_obd_name(struct mdt_device *mdt)
1085 return mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name;
1088 int mds_mod_init(void);
1089 void mds_mod_exit(void);
1091 /* Update handlers */
1092 int out_handle(struct mdt_thread_info *info);
1094 #define out_tx_create(info, obj, attr, fid, dof, th, reply, idx) \
1095 __out_tx_create(info, obj, attr, fid, dof, th, reply, idx, \
1098 #define out_tx_attr_set(info, obj, attr, th, reply, idx) \
1099 __out_tx_attr_set(info, obj, attr, th, reply, idx, \
1102 #define out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx) \
1103 __out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx, \
1106 #define out_tx_ref_add(info, obj, th, reply, idx) \
1107 __out_tx_ref_add(info, obj, th, reply, idx, __FILE__, __LINE__)
1109 #define out_tx_ref_del(info, obj, th, reply, idx) \
1110 __out_tx_ref_del(info, obj, th, reply, idx, __FILE__, __LINE__)
1112 #define out_tx_index_insert(info, obj, th, name, fid, reply, idx) \
1113 __out_tx_index_insert(info, obj, th, name, fid, reply, idx, \
1116 #define out_tx_index_delete(info, obj, th, name, reply, idx) \
1117 __out_tx_index_delete(info, obj, th, name, reply, idx, \
1120 #define out_tx_destroy(info, obj, th, reply, idx) \
1121 __out_tx_destroy(info, obj, th, reply, idx, __FILE__, __LINE__)
1123 #endif /* __KERNEL__ */