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__)
54 * struct ptlrpc_client
56 #include <lustre_net.h>
59 * struct obd_connect_data
60 * struct lustre_handle
62 #include <lustre/lustre_idl.h>
63 #include <lustre_disk.h>
64 #include <lu_target.h>
65 #include <md_object.h>
66 #include <lustre_fid.h>
67 #include <lustre_fld.h>
68 #include <lustre_mdt.h>
69 #include <lustre_req_layout.h>
70 #include <lustre_sec.h>
72 #include <lustre_idmap.h>
73 #include <lustre_eacl.h>
74 #include <lustre_fsfilt.h>
75 #include <lustre_quota.h>
77 /* check if request's xid is equal to last one or not*/
78 static inline int req_xid_is_last(struct ptlrpc_request *req)
80 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
81 return (req->rq_xid == lcd->lcd_last_xid ||
82 req->rq_xid == lcd->lcd_last_close_xid);
86 /* file data for open files on MDS */
87 struct mdt_file_data {
88 struct portals_handle mfd_handle; /* must be first */
89 int mfd_mode; /* open mode provided by client */
90 cfs_list_t mfd_list; /* protected by med_open_lock */
91 __u64 mfd_xid; /* xid of the open request */
92 struct lustre_handle mfd_old_handle; /* old handle in replay case */
93 struct mdt_object *mfd_object; /* point to opened object */
96 /* mdt state flag bits */
97 #define MDT_FL_CFGLOG 0
98 #define MDT_FL_SYNCED 1
102 struct md_device mdt_md_dev;
103 struct seq_server_site mdt_seq_site;
104 struct ptlrpc_service *mdt_regular_service;
105 struct ptlrpc_service *mdt_readpage_service;
106 struct ptlrpc_service *mdt_out_service;
107 struct ptlrpc_service *mdt_setattr_service;
108 struct ptlrpc_service *mdt_mdsc_service;
109 struct ptlrpc_service *mdt_mdss_service;
110 struct ptlrpc_service *mdt_dtss_service;
111 struct ptlrpc_service *mdt_fld_service;
112 /* DLM name-space for meta-data locks maintained by this server */
113 struct ldlm_namespace *mdt_namespace;
114 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
115 struct ptlrpc_client *mdt_ldlm_client;
116 /* underlying device */
117 struct obd_export *mdt_child_exp;
118 struct md_device *mdt_child;
119 struct dt_device *mdt_bottom;
120 struct obd_export *mdt_bottom_exp;
122 struct lu_target mdt_lut;
124 * Options bit-fields.
127 unsigned int mo_user_xattr:1,
134 /* mdt state flags */
135 unsigned long mdt_state;
136 /* lock to protect IOepoch */
137 spinlock_t mdt_ioepoch_lock;
140 /* transaction callbacks */
141 struct dt_txn_callback mdt_txn_cb;
143 /* these values should be updated from lov if necessary.
144 * or should be placed somewhere else. */
146 int mdt_max_cookiesize;
148 struct upcall_cache *mdt_identity_cache;
151 rwlock_t mdt_sptlrpc_lock;
152 struct sptlrpc_rule_set mdt_sptlrpc_rset;
154 /* capability keys */
155 unsigned long mdt_capa_timeout;
157 struct dt_object *mdt_ck_obj;
158 unsigned long mdt_ck_timeout;
159 unsigned long mdt_ck_expiry;
160 cfs_timer_t mdt_ck_timer;
161 struct ptlrpc_thread mdt_ck_thread;
162 struct lustre_capa_key mdt_capa_keys[2];
163 unsigned int mdt_capa_conf:1,
165 /* Enable remote dir on non-MDT0 */
166 mdt_enable_remote_dir:1;
168 gid_t mdt_enable_remote_dir_gid;
169 /* statfs optimization: we cache a bit */
170 struct obd_statfs mdt_osfs;
172 spinlock_t mdt_osfs_lock;
175 uid_t mdt_squash_uid;
176 gid_t mdt_squash_gid;
177 cfs_list_t mdt_nosquash_nids;
178 char *mdt_nosquash_str;
179 int mdt_nosquash_strlen;
180 struct rw_semaphore mdt_squash_sem;
183 struct rename_stats mdt_rename_stats;
184 struct lu_fid mdt_md_root_fid;
186 /* connection to quota master */
187 struct obd_export *mdt_qmt_exp;
188 /* quota master device associated with this MDT */
189 struct lu_device *mdt_qmt_dev;
192 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
193 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
194 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
195 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
196 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
197 #define MDT_COS_DEFAULT (0)
200 struct lu_object_header mot_header;
201 struct md_object mot_obj;
204 int mot_ioepoch_count;
206 /* Lock to protect object's IO epoch. */
207 struct mutex mot_ioepoch_mutex;
208 /* Lock to protect create_data */
209 struct mutex mot_lov_mutex;
212 enum mdt_object_flags {
213 /** SOM attributes are changed. */
214 MOF_SOM_CHANGE = (1 << 0),
216 * The SOM recovery state for mdt object.
217 * This state is an in-memory equivalent of an absent SOM EA, used
218 * instead of invalidating SOM EA while IOEpoch is still opened when
219 * a client eviction occurs or a client fails to obtain SOM attributes.
220 * It indicates that the last IOEpoch holder will need to obtain SOM
221 * attributes under [0;EOF] extent lock to flush all the client's
222 * cached of evicted from MDS clients (but not necessary evicted from
223 * OST) before taking ost attributes.
225 MOF_SOM_RECOV = (1 << 1),
226 /** File has been just created. */
227 MOF_SOM_CREATED = (1 << 2),
228 /** lov object has been created. */
229 MOF_LOV_CREATED = (1 << 3),
232 struct mdt_lock_handle {
233 /* Lock type, reg for cross-ref use or pdo lock. */
237 struct lustre_handle mlh_reg_lh;
238 ldlm_mode_t mlh_reg_mode;
241 struct lustre_handle mlh_pdo_lh;
242 ldlm_mode_t mlh_pdo_mode;
243 unsigned int mlh_pdo_hash;
245 /* Remote regular lock */
246 struct lustre_handle mlh_rreg_lh;
247 ldlm_mode_t mlh_rreg_mode;
251 MDT_LH_PARENT, /* parent lockh */
252 MDT_LH_CHILD, /* child lockh */
253 MDT_LH_OLD, /* old lockh for rename */
254 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
255 MDT_LH_NEW, /* new lockh for rename */
256 MDT_LH_RMT, /* used for return lh to caller */
265 struct mdt_reint_record {
266 mdt_reint_t rr_opcode;
267 const struct lustre_handle *rr_handle;
268 const struct lu_fid *rr_fid1;
269 const struct lu_fid *rr_fid2;
274 const void *rr_eadata;
277 const struct llog_cookie *rr_logcookies;
281 enum mdt_reint_flag {
282 MRF_OPEN_TRUNC = 1 << 0,
285 struct mdt_thread_info;
287 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
291 tx_exec_func_t exec_fn;
292 tx_exec_func_t undo_fn;
293 struct dt_object *object;
295 struct update_reply *reply;
300 const struct dt_rec *rec;
301 const struct dt_key *key;
316 struct dt_allocation_hint hint;
317 struct dt_object_format dof;
325 struct ost_body *body;
330 #define TX_MAX_OPS 10
331 struct thandle_exec_args {
332 struct thandle *ta_handle;
333 struct dt_device *ta_dev;
335 struct tx_arg ta_args[TX_MAX_OPS];
336 int ta_argno; /* used args */
340 * Common data shared by mdt-level handlers. This is allocated per-thread to
341 * reduce stack consumption.
343 struct mdt_thread_info {
346 * The following members will be filled explicitly
347 * with specific data in mdt_thread_info_init().
349 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
350 * request handling may migrate from one server thread to another.
352 struct req_capsule *mti_pill;
354 /* although we have export in req, there are cases when it is not
355 * available, e.g. closing files upon export destroy */
356 struct obd_export *mti_exp;
358 * A couple of lock handles.
360 struct mdt_lock_handle mti_lh[MDT_LH_NR];
362 struct mdt_device *mti_mdt;
363 const struct lu_env *mti_env;
366 * Additional fail id that can be set by handler. Passed to
367 * target_send_reply().
371 /* transaction number of current request */
377 * The following members will be filled expilictly
378 * with zero in mdt_thread_info_init(). These members may be used
385 struct md_attr mti_attr;
387 * Body for "habeo corpus" operations.
389 const struct mdt_body *mti_body;
391 * Host object. This is released at the end of mdt_handler().
393 struct mdt_object *mti_object;
395 * Lock request for "habeo clavis" operations.
397 const struct ldlm_request *mti_dlm_req;
399 __u32 mti_has_trans:1, /* has txn already? */
402 /* opdata for mdt_reint_open(), has the same as
403 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
404 * value onto disk for recovery when mdt_trans_stop_cb() is called.
410 * The following members will be filled explicitly
411 * with zero in mdt_reint_unpack(), because they are only used
412 * by reint requests (including mdt_reint_open()).
416 * reint record. contains information for reint operations.
418 struct mdt_reint_record mti_rr;
420 /** md objects included in operation */
421 struct mdt_object *mti_mos;
422 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
424 * Operation specification (currently create and lookup)
426 struct md_op_spec mti_spec;
430 * The following members will _NOT_ be initialized at all.
431 * DO NOT expect them to contain any valid value.
432 * They should be initialized explicitly by the user themselves.
435 /* XXX: If something is in a union, make sure they do not conflict */
437 struct lu_fid mti_tmp_fid1;
438 struct lu_fid mti_tmp_fid2;
439 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
440 * mdt_rename_lock() */
441 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
444 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
445 char ns_name[48]; /* for mdt_init0() */
446 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
447 struct obd_statfs osfs; /* for mdt_statfs() */
449 /* for mdt_readpage() */
450 struct lu_rdpg mti_rdpg;
451 /* for mdt_sendpage() */
452 struct l_wait_info mti_wait_info;
456 struct md_som_data data;
459 struct dt_object_format mti_update_dof;
460 struct update_reply *mti_update_reply;
461 struct update *mti_update;
462 int mti_update_reply_index;
463 struct obdo mti_obdo;
464 struct dt_object *mti_dt_object;
468 /* IO epoch related stuff. */
469 struct mdt_ioepoch *mti_ioepoch;
470 __u64 mti_replayepoch;
473 struct lu_buf mti_buf;
474 struct lu_buf mti_big_buf;
475 struct lustre_capa_key mti_capa_key;
477 /* Ops object filename */
478 struct lu_name mti_name;
479 /* per-thread values, can be re-used */
482 /* big_lmm buffer was used and must be used in reply */
483 int mti_big_lmm_used;
484 /* should be enough to fit lustre_mdt_attrs */
485 char mti_xattr_buf[128];
486 struct thandle_exec_args mti_handle;
487 struct ldlm_enqueue_info mti_einfo;
490 /* ptlrpc request handler for MDT. All handlers are
491 * grouped into several slices - struct mdt_opc_slice,
492 * and stored in an array - mdt_handlers[].
495 /* The name of this handler. */
497 /* Fail id for this handler, checked at the beginning of this handler*/
499 /* Operation code for this handler */
501 /* flags are listed in enum mdt_handler_flags below. */
503 /* The actual handler function to execute. */
504 int (*mh_act)(struct mdt_thread_info *info);
505 /* Request format for this request. */
506 const struct req_format *mh_fmt;
509 enum mdt_handler_flags {
511 * struct mdt_body is passed in the incoming message, and object
512 * identified by this fid exists on disk.
514 * "habeo corpus" == "I have a body"
516 HABEO_CORPUS = (1 << 0),
518 * struct ldlm_request is passed in the incoming message.
520 * "habeo clavis" == "I have a key"
522 HABEO_CLAVIS = (1 << 1),
524 * this request has fixed reply format, so that reply message can be
525 * packed by generic code.
527 * "habeo refero" == "I have a reply"
529 HABEO_REFERO = (1 << 2),
531 * this request will modify something, so check whether the filesystem
532 * is readonly or not, then return -EROFS to client asap if necessary.
534 * "mutabor" == "I shall modify"
539 struct mdt_opc_slice {
542 struct mdt_handler *mos_hs;
545 static inline const struct md_device_operations *
546 mdt_child_ops(struct mdt_device * m)
548 LASSERT(m->mdt_child);
549 return m->mdt_child->md_ops;
552 static inline struct md_object *mdt_object_child(struct mdt_object *o)
555 return lu2md(lu_object_next(&o->mot_obj.mo_lu));
558 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
560 return info->mti_pill ? info->mti_pill->rc_req : NULL;
563 static inline int req_is_replay(struct ptlrpc_request *req)
565 LASSERT(req->rq_reqmsg);
566 return !!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
569 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
571 LASSERT(info->mti_exp);
572 return exp_connect_flags(info->mti_exp);
575 static inline void mdt_object_get(const struct lu_env *env,
576 struct mdt_object *o)
579 lu_object_get(&o->mot_obj.mo_lu);
583 static inline void mdt_object_put(const struct lu_env *env,
584 struct mdt_object *o)
587 lu_object_put(env, &o->mot_obj.mo_lu);
591 static inline int mdt_object_exists(const struct mdt_object *o)
593 return lu_object_exists(&o->mot_obj.mo_lu);
596 static inline int mdt_object_remote(const struct mdt_object *o)
598 return lu_object_remote(&o->mot_obj.mo_lu);
601 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
603 return lu_object_fid(&o->mot_obj.mo_lu);
606 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
608 return mdt->mdt_md_dev.md_lu_dev.ld_site;
611 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
613 return &mdt->mdt_seq_site;
616 static inline void mdt_export_evict(struct obd_export *exp)
618 class_fail_export(exp);
619 class_export_put(exp);
622 /* Here we use LVB_TYPE to check dne client, because it is
623 * also landed on 2.4. */
624 static inline int mdt_is_dne_client(struct obd_export *exp)
626 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
629 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
630 void mdt_set_disposition(struct mdt_thread_info *info,
631 struct ldlm_reply *rep, int flag);
632 void mdt_clear_disposition(struct mdt_thread_info *info,
633 struct ldlm_reply *rep, int flag);
635 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
636 ldlm_mode_t lm, const char *name,
639 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
642 int mdt_lock_setup(struct mdt_thread_info *info,
643 struct mdt_object *o,
644 struct mdt_lock_handle *lh);
646 int mdt_object_lock(struct mdt_thread_info *,
648 struct mdt_lock_handle *,
651 int mdt_object_lock_try(struct mdt_thread_info *,
653 struct mdt_lock_handle *,
656 void mdt_object_unlock(struct mdt_thread_info *,
658 struct mdt_lock_handle *,
661 struct mdt_object *mdt_object_new(const struct lu_env *,
663 const struct lu_fid *);
664 struct mdt_object *mdt_object_find(const struct lu_env *,
666 const struct lu_fid *);
667 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
668 const struct lu_fid *,
669 struct mdt_lock_handle *,
671 void mdt_object_unlock_put(struct mdt_thread_info *,
673 struct mdt_lock_handle *,
676 void mdt_client_compatibility(struct mdt_thread_info *info);
678 int mdt_remote_object_lock(struct mdt_thread_info *mti,
679 struct mdt_object *o, struct lustre_handle *lh,
680 ldlm_mode_t mode, __u64 ibits);
681 int mdt_close_unpack(struct mdt_thread_info *info);
682 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
683 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
684 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
685 const struct lu_attr *attr, const struct lu_fid *fid);
687 int mdt_getxattr(struct mdt_thread_info *info);
688 int mdt_reint_setxattr(struct mdt_thread_info *info,
689 struct mdt_lock_handle *lh);
691 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
692 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
694 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
695 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
696 struct mdt_lock_handle *lhc);
698 extern void target_recovery_fini(struct obd_device *obd);
699 extern void target_recovery_init(struct lu_target *lut,
700 svc_handler_t handler);
701 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
702 struct obd_device *, struct lustre_sb_info *lsi);
703 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
705 int mdt_export_stats_init(struct obd_device *obd,
706 struct obd_export *exp,
709 int mdt_pin(struct mdt_thread_info* info);
711 int mdt_lock_new_child(struct mdt_thread_info *info,
712 struct mdt_object *o,
713 struct mdt_lock_handle *child_lockh);
715 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
718 int mdt_reint_open(struct mdt_thread_info *info,
719 struct mdt_lock_handle *lhc);
721 struct mdt_file_data *mdt_handle2mfd(struct mdt_thread_info *,
722 const struct lustre_handle *);
725 MDT_IOEPOCH_CLOSED = 0,
726 MDT_IOEPOCH_OPENED = 1,
727 MDT_IOEPOCH_GETATTR = 2,
735 int mdt_attr_get_complex(struct mdt_thread_info *info,
736 struct mdt_object *o, struct md_attr *ma);
737 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
739 int mdt_object_is_som_enabled(struct mdt_object *mo);
740 int mdt_write_get(struct mdt_object *o);
741 void mdt_write_put(struct mdt_object *o);
742 int mdt_write_read(struct mdt_object *o);
743 struct mdt_file_data *mdt_mfd_new(void);
744 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
745 void mdt_mfd_free(struct mdt_file_data *mfd);
746 int mdt_close(struct mdt_thread_info *info);
747 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
748 struct md_attr *ma, int flags);
749 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
751 int mdt_done_writing(struct mdt_thread_info *info);
752 int mdt_fix_reply(struct mdt_thread_info *info);
753 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
754 const struct md_attr *);
755 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
757 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
758 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
759 const void *area, ssize_t len);
761 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
763 int mdt_check_ucred(struct mdt_thread_info *);
764 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
765 int mdt_init_ucred_reint(struct mdt_thread_info *);
766 void mdt_exit_ucred(struct mdt_thread_info *);
767 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
768 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
769 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
771 int mdt_handle_common(struct ptlrpc_request *req,
772 struct mdt_opc_slice *supported);
773 int mdt_connect(struct mdt_thread_info *info);
774 int mdt_disconnect(struct mdt_thread_info *info);
775 int mdt_set_info(struct mdt_thread_info *info);
776 int mdt_get_info(struct mdt_thread_info *info);
777 int mdt_getstatus(struct mdt_thread_info *info);
778 int mdt_getattr(struct mdt_thread_info *info);
779 int mdt_getattr_name(struct mdt_thread_info *info);
780 int mdt_statfs(struct mdt_thread_info *info);
781 int mdt_reint(struct mdt_thread_info *info);
782 int mdt_sync(struct mdt_thread_info *info);
783 int mdt_is_subdir(struct mdt_thread_info *info);
784 int mdt_obd_ping(struct mdt_thread_info *info);
785 int mdt_obd_log_cancel(struct mdt_thread_info *info);
786 int mdt_obd_qc_callback(struct mdt_thread_info *info);
787 int mdt_enqueue(struct mdt_thread_info *info);
788 int mdt_convert(struct mdt_thread_info *info);
789 int mdt_bl_callback(struct mdt_thread_info *info);
790 int mdt_cp_callback(struct mdt_thread_info *info);
791 int mdt_llog_create(struct mdt_thread_info *info);
792 int mdt_llog_destroy(struct mdt_thread_info *info);
793 int mdt_llog_read_header(struct mdt_thread_info *info);
794 int mdt_llog_next_block(struct mdt_thread_info *info);
795 int mdt_llog_prev_block(struct mdt_thread_info *info);
796 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
797 int mdt_readpage(struct mdt_thread_info *info);
798 int mdt_obd_idx_read(struct mdt_thread_info *info);
800 extern struct mdt_opc_slice mdt_regular_handlers[];
801 extern struct mdt_opc_slice mdt_seq_handlers[];
802 extern struct mdt_opc_slice mdt_fld_handlers[];
804 int mdt_quotacheck(struct mdt_thread_info *info);
805 int mdt_quotactl(struct mdt_thread_info *info);
806 int mdt_quota_dqacq(struct mdt_thread_info *info);
807 int mdt_swap_layouts(struct mdt_thread_info *info);
809 extern struct lprocfs_vars lprocfs_mds_module_vars[];
810 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
812 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
815 struct mdt_handler *mdt_handler_find(__u32 opc,
816 struct mdt_opc_slice *supported);
818 int mdt_init_sec_level(struct mdt_thread_info *);
819 int mdt_init_idmap(struct mdt_thread_info *);
820 void mdt_cleanup_idmap(struct mdt_export_data *);
821 int mdt_handle_idmap(struct mdt_thread_info *);
822 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
823 struct ptlrpc_user_desc *);
824 void mdt_body_reverse_idmap(struct mdt_thread_info *,
826 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
827 struct mdt_remote_perm *);
828 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
830 static inline struct mdt_device *mdt_dev(struct lu_device *d)
832 // LASSERT(lu_device_is_mdt(d));
833 return container_of0(d, struct mdt_device, mdt_md_dev.md_lu_dev);
836 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
838 struct lu_object *lo;
839 struct mdt_device *mdt = mdt_dev(mo->mot_obj.mo_lu.lo_dev);
841 lo = lu_object_locate(mo->mot_obj.mo_lu.lo_header,
842 mdt->mdt_bottom->dd_lu_dev.ld_type);
846 /* mdt/mdt_identity.c */
847 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
849 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
851 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
853 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
855 void mdt_flush_identity(struct upcall_cache *, int);
857 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
859 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
862 int mdt_hsm_state_get(struct mdt_thread_info *info);
863 int mdt_hsm_state_set(struct mdt_thread_info *info);
864 int mdt_hsm_action(struct mdt_thread_info *info);
865 int mdt_hsm_progress(struct mdt_thread_info *info);
866 int mdt_hsm_ct_register(struct mdt_thread_info *info);
867 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
868 int mdt_hsm_request(struct mdt_thread_info *info);
870 extern struct lu_context_key mdt_thread_key;
871 /* debug issues helper starts here*/
872 static inline int mdt_fail_write(const struct lu_env *env,
873 struct dt_device *dd, int id)
875 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
876 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
878 return dd->dd_ops->dt_ro(env, dd);
879 /* We set FAIL_ONCE because we never "un-fail" a device */
885 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
887 return &req->rq_export->exp_mdt_data;
890 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
891 struct mdt_lock_handle *lhc);
892 static inline int mdt_check_resent(struct mdt_thread_info *info,
893 mdt_reconstruct_t reconstruct,
894 struct mdt_lock_handle *lhc)
896 struct ptlrpc_request *req = mdt_info_req(info);
899 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
900 if (req_xid_is_last(req)) {
901 reconstruct(info, lhc);
904 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
905 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
910 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
911 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
913 static inline int is_identity_get_disabled(struct upcall_cache *cache)
915 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
918 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
920 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
921 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
922 struct lustre_handle *lh,
924 ldlm_policy_data_t *policy,
925 const struct ldlm_res_id *res_id,
926 __u64 flags, const __u64 *client_cookie)
933 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
934 mode, &flags, mdt_blocking_ast,
935 ldlm_completion_ast, NULL, NULL, 0,
936 LVB_T_NONE, client_cookie, lh);
937 return rc == ELDLM_OK ? 0 : -EIO;
940 static inline void mdt_fid_unlock(struct lustre_handle *lh,
943 ldlm_lock_decref(lh, mode);
946 extern mdl_mode_t mdt_mdl_lock_modes[];
947 extern ldlm_mode_t mdt_dlm_lock_modes[];
949 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
951 LASSERT(IS_PO2(mode));
952 return mdt_mdl_lock_modes[mode];
955 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
957 LASSERT(IS_PO2(mode));
958 return mdt_dlm_lock_modes[mode];
962 extern struct ldlm_valblock_ops mdt_lvbo;
964 static inline struct lu_name *mdt_name(const struct lu_env *env,
965 char *name, int namelen)
967 struct lu_name *lname;
968 struct mdt_thread_info *mti;
970 LASSERT(namelen > 0);
971 /* trailing '\0' in buffer */
972 LASSERT(name[namelen] == '\0');
974 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
975 lname = &mti->mti_name;
976 lname->ln_name = name;
977 lname->ln_namelen = namelen;
981 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
982 struct lu_name *slname)
987 tlname->ln_name = slname->ln_name;
988 tlname->ln_namelen = slname->ln_namelen;
992 void mdt_enable_cos(struct mdt_device *, int);
993 int mdt_cos_is_enabled(struct mdt_device *);
994 int mdt_hsm_copytool_send(struct obd_export *exp);
1012 LPROC_MDT_SAMEDIR_RENAME,
1013 LPROC_MDT_CROSSDIR_RENAME,
1016 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1017 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1018 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1019 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1020 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1021 void mdt_procfs_fini(struct mdt_device *mdt);
1022 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1023 struct mdt_device *mdt,
1024 struct ptlrpc_request *req,
1025 struct mdt_object *src, struct mdt_object *tgt);
1028 int mdt_ck_thread_start(struct mdt_device *mdt);
1029 void mdt_ck_thread_stop(struct mdt_device *mdt);
1030 void mdt_ck_timer_callback(unsigned long castmeharder);
1031 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1033 static inline void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1034 const struct lu_fid *fid,
1035 struct lustre_capa *capa)
1037 struct md_capainfo *ci;
1039 LASSERT(offset >= 0 && offset < MD_CAPAINFO_MAX);
1040 if (!info->mti_mdt->mdt_opts.mo_mds_capa ||
1041 !(exp_connect_flags(info->mti_exp) & OBD_CONNECT_MDS_CAPA))
1044 ci = md_capainfo(info->mti_env);
1046 ci->mc_fid[offset] = *fid;
1047 ci->mc_capa[offset] = capa;
1050 static inline void mdt_dump_capainfo(struct mdt_thread_info *info)
1052 struct md_capainfo *ci = md_capainfo(info->mti_env);
1057 for (i = 0; i < MD_CAPAINFO_MAX; i++) {
1058 if (!ci->mc_capa[i]) {
1059 CERROR("no capa for index %d "DFID"\n",
1060 i, PFID(&ci->mc_fid[i]));
1063 if (ci->mc_capa[i] == BYPASS_CAPA) {
1064 CERROR("bypass for index %d "DFID"\n",
1065 i, PFID(&ci->mc_fid[i]));
1068 DEBUG_CAPA(D_ERROR, ci->mc_capa[i], "index %d", i);
1072 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1074 return mdt->mdt_md_dev.md_lu_dev.ld_obd;
1077 extern const struct lu_device_operations mdt_lu_ops;
1079 static inline int lu_device_is_mdt(struct lu_device *d)
1081 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1084 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1086 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1087 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1088 return container_of0(d, struct mdt_device, mdt_md_dev.md_lu_dev);
1091 static inline char *mdt_obd_name(struct mdt_device *mdt)
1093 return mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name;
1096 int mds_mod_init(void);
1097 void mds_mod_exit(void);
1099 /* Update handlers */
1100 int out_handle(struct mdt_thread_info *info);
1102 #define out_tx_create(info, obj, attr, fid, dof, th, reply, idx) \
1103 __out_tx_create(info, obj, attr, fid, dof, th, reply, idx, \
1106 #define out_tx_attr_set(info, obj, attr, th, reply, idx) \
1107 __out_tx_attr_set(info, obj, attr, th, reply, idx, \
1110 #define out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx) \
1111 __out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx, \
1114 #define out_tx_ref_add(info, obj, th, reply, idx) \
1115 __out_tx_ref_add(info, obj, th, reply, idx, __FILE__, __LINE__)
1117 #define out_tx_ref_del(info, obj, th, reply, idx) \
1118 __out_tx_ref_del(info, obj, th, reply, idx, __FILE__, __LINE__)
1120 #define out_tx_index_insert(info, obj, th, name, fid, reply, idx) \
1121 __out_tx_index_insert(info, obj, th, name, fid, reply, idx, \
1124 #define out_tx_index_delete(info, obj, th, name, reply, idx) \
1125 __out_tx_index_delete(info, obj, th, name, reply, idx, \
1128 #define out_tx_destroy(info, obj, th, reply, idx) \
1129 __out_tx_destroy(info, obj, th, reply, idx, __FILE__, __LINE__)
1131 #endif /* __KERNEL__ */