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_req_layout.h>
69 #include <lustre_sec.h>
71 #include <lustre_idmap.h>
72 #include <lustre_eacl.h>
73 #include <lustre_fsfilt.h>
74 #include <lustre_quota.h>
76 /* check if request's xid is equal to last one or not*/
77 static inline int req_xid_is_last(struct ptlrpc_request *req)
79 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
80 return (req->rq_xid == lcd->lcd_last_xid ||
81 req->rq_xid == lcd->lcd_last_close_xid);
85 /* file data for open files on MDS */
86 struct mdt_file_data {
87 struct portals_handle mfd_handle; /* must be first */
88 int mfd_mode; /* open mode provided by client */
89 cfs_list_t mfd_list; /* protected by med_open_lock */
90 __u64 mfd_xid; /* xid of the open request */
91 struct lustre_handle mfd_old_handle; /* old handle in replay case */
92 struct mdt_object *mfd_object; /* point to opened object */
95 /* mdt state flag bits */
96 #define MDT_FL_CFGLOG 0
97 #define MDT_FL_SYNCED 1
101 struct md_device mdt_md_dev;
102 struct seq_server_site mdt_seq_site;
103 struct ptlrpc_service *mdt_regular_service;
104 struct ptlrpc_service *mdt_readpage_service;
105 struct ptlrpc_service *mdt_out_service;
106 struct ptlrpc_service *mdt_setattr_service;
107 struct ptlrpc_service *mdt_mdsc_service;
108 struct ptlrpc_service *mdt_mdss_service;
109 struct ptlrpc_service *mdt_dtss_service;
110 struct ptlrpc_service *mdt_fld_service;
111 /* DLM name-space for meta-data locks maintained by this server */
112 struct ldlm_namespace *mdt_namespace;
113 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
114 struct ptlrpc_client *mdt_ldlm_client;
115 /* underlying device */
116 struct obd_export *mdt_child_exp;
117 struct md_device *mdt_child;
118 struct dt_device *mdt_bottom;
119 struct obd_export *mdt_bottom_exp;
121 struct lu_target mdt_lut;
123 * Options bit-fields.
126 signed int mo_user_xattr :1,
133 /* mdt state flags */
134 unsigned long mdt_state;
135 /* lock to protect IOepoch */
136 spinlock_t mdt_ioepoch_lock;
139 /* transaction callbacks */
140 struct dt_txn_callback mdt_txn_cb;
142 /* these values should be updated from lov if necessary.
143 * or should be placed somewhere else. */
145 int mdt_max_cookiesize;
147 struct upcall_cache *mdt_identity_cache;
150 rwlock_t mdt_sptlrpc_lock;
151 struct sptlrpc_rule_set mdt_sptlrpc_rset;
153 /* capability keys */
154 unsigned long mdt_capa_timeout;
156 struct dt_object *mdt_ck_obj;
157 unsigned long mdt_ck_timeout;
158 unsigned long mdt_ck_expiry;
159 cfs_timer_t mdt_ck_timer;
160 struct ptlrpc_thread mdt_ck_thread;
161 struct lustre_capa_key mdt_capa_keys[2];
162 unsigned int mdt_capa_conf:1,
164 mdt_enable_remote_dir:1;
166 /* statfs optimization: we cache a bit */
167 struct obd_statfs mdt_osfs;
169 spinlock_t mdt_osfs_lock;
172 uid_t mdt_squash_uid;
173 gid_t mdt_squash_gid;
174 cfs_list_t mdt_nosquash_nids;
175 char *mdt_nosquash_str;
176 int mdt_nosquash_strlen;
177 struct rw_semaphore mdt_squash_sem;
180 struct rename_stats mdt_rename_stats;
181 struct lu_fid mdt_md_root_fid;
183 /* connection to quota master */
184 struct obd_export *mdt_qmt_exp;
185 /* quota master device associated with this MDT */
186 struct lu_device *mdt_qmt_dev;
189 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
190 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
191 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
192 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
193 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
194 #define MDT_COS_DEFAULT (0)
197 struct lu_object_header mot_header;
198 struct md_object mot_obj;
201 int mot_ioepoch_count;
203 /* Lock to protect object's IO epoch. */
204 struct mutex mot_ioepoch_mutex;
205 /* Lock to protect create_data */
206 struct mutex mot_lov_mutex;
209 enum mdt_object_flags {
210 /** SOM attributes are changed. */
211 MOF_SOM_CHANGE = (1 << 0),
213 * The SOM recovery state for mdt object.
214 * This state is an in-memory equivalent of an absent SOM EA, used
215 * instead of invalidating SOM EA while IOEpoch is still opened when
216 * a client eviction occurs or a client fails to obtain SOM attributes.
217 * It indicates that the last IOEpoch holder will need to obtain SOM
218 * attributes under [0;EOF] extent lock to flush all the client's
219 * cached of evicted from MDS clients (but not necessary evicted from
220 * OST) before taking ost attributes.
222 MOF_SOM_RECOV = (1 << 1),
223 /** File has been just created. */
224 MOF_SOM_CREATED = (1 << 2),
225 /** lov object has been created. */
226 MOF_LOV_CREATED = (1 << 3),
229 struct mdt_lock_handle {
230 /* Lock type, reg for cross-ref use or pdo lock. */
234 struct lustre_handle mlh_reg_lh;
235 ldlm_mode_t mlh_reg_mode;
238 struct lustre_handle mlh_pdo_lh;
239 ldlm_mode_t mlh_pdo_mode;
240 unsigned int mlh_pdo_hash;
244 MDT_LH_PARENT, /* parent lockh */
245 MDT_LH_CHILD, /* child lockh */
246 MDT_LH_OLD, /* old lockh for rename */
247 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
248 MDT_LH_NEW, /* new lockh for rename */
249 MDT_LH_RMT, /* used for return lh to caller */
258 struct mdt_reint_record {
259 mdt_reint_t rr_opcode;
260 const struct lustre_handle *rr_handle;
261 const struct lu_fid *rr_fid1;
262 const struct lu_fid *rr_fid2;
267 const void *rr_eadata;
270 const struct llog_cookie *rr_logcookies;
274 enum mdt_reint_flag {
275 MRF_OPEN_TRUNC = 1 << 0,
278 struct mdt_thread_info;
280 typedef int (*tx_exec_func_t)(struct mdt_thread_info *, struct thandle *,
284 tx_exec_func_t exec_fn;
285 tx_exec_func_t undo_fn;
286 struct dt_object *object;
289 struct update_reply *reply;
293 const struct dt_rec *rec;
294 const struct dt_key *key;
309 struct dt_allocation_hint hint;
310 struct dt_object_format dof;
320 #define TX_MAX_OPS 10
321 struct thandle_exec_args {
322 struct thandle *ta_handle;
323 struct dt_device *ta_dev;
325 struct tx_arg ta_args[TX_MAX_OPS];
326 int ta_argno; /* used args */
330 * Common data shared by mdt-level handlers. This is allocated per-thread to
331 * reduce stack consumption.
333 struct mdt_thread_info {
336 * The following members will be filled explicitly
337 * with specific data in mdt_thread_info_init().
339 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
340 * request handling may migrate from one server thread to another.
342 struct req_capsule *mti_pill;
344 /* although we have export in req, there are cases when it is not
345 * available, e.g. closing files upon export destroy */
346 struct obd_export *mti_exp;
348 * A couple of lock handles.
350 struct mdt_lock_handle mti_lh[MDT_LH_NR];
352 struct mdt_device *mti_mdt;
353 const struct lu_env *mti_env;
356 * Additional fail id that can be set by handler. Passed to
357 * target_send_reply().
361 /* transaction number of current request */
367 * The following members will be filled expilictly
368 * with zero in mdt_thread_info_init(). These members may be used
375 struct md_attr mti_attr;
377 * Body for "habeo corpus" operations.
379 const struct mdt_body *mti_body;
381 * Host object. This is released at the end of mdt_handler().
383 struct mdt_object *mti_object;
385 * Lock request for "habeo clavis" operations.
387 const struct ldlm_request *mti_dlm_req;
389 __u32 mti_has_trans:1, /* has txn already? */
392 /* opdata for mdt_reint_open(), has the same as
393 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
394 * value onto disk for recovery when mdt_trans_stop_cb() is called.
400 * The following members will be filled explicitly
401 * with zero in mdt_reint_unpack(), because they are only used
402 * by reint requests (including mdt_reint_open()).
406 * reint record. contains information for reint operations.
408 struct mdt_reint_record mti_rr;
410 /** md objects included in operation */
411 struct mdt_object *mti_mos;
412 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
414 * Operation specification (currently create and lookup)
416 struct md_op_spec mti_spec;
420 * The following members will _NOT_ be initialized at all.
421 * DO NOT expect them to contain any valid value.
422 * They should be initialized explicitly by the user themselves.
425 /* XXX: If something is in a union, make sure they do not conflict */
427 struct lu_fid mti_tmp_fid1;
428 struct lu_fid mti_tmp_fid2;
429 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
430 * mdt_rename_lock() */
431 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
434 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
435 char ns_name[48]; /* for mdt_init0() */
436 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
437 struct obd_statfs osfs; /* for mdt_statfs() */
439 /* for mdt_readpage() */
440 struct lu_rdpg mti_rdpg;
441 /* for mdt_sendpage() */
442 struct l_wait_info mti_wait_info;
446 struct md_som_data data;
449 struct dt_object_format mti_update_dof;
450 struct update_reply *mti_update_reply;
451 struct update *mti_update;
452 int mti_update_reply_index;
453 struct obdo mti_obdo;
454 struct dt_object *mti_dt_object;
458 /* IO epoch related stuff. */
459 struct mdt_ioepoch *mti_ioepoch;
460 __u64 mti_replayepoch;
463 struct lu_buf mti_buf;
464 struct lustre_capa_key mti_capa_key;
466 /* Ops object filename */
467 struct lu_name mti_name;
468 /* per-thread values, can be re-used */
471 /* big_lmm buffer was used and must be used in reply */
472 int mti_big_lmm_used;
473 /* should be enough to fit lustre_mdt_attrs */
474 char mti_xattr_buf[128];
475 struct thandle_exec_args mti_handle;
476 struct ldlm_enqueue_info mti_einfo;
479 /* ptlrpc request handler for MDT. All handlers are
480 * grouped into several slices - struct mdt_opc_slice,
481 * and stored in an array - mdt_handlers[].
484 /* The name of this handler. */
486 /* Fail id for this handler, checked at the beginning of this handler*/
488 /* Operation code for this handler */
490 /* flags are listed in enum mdt_handler_flags below. */
492 /* The actual handler function to execute. */
493 int (*mh_act)(struct mdt_thread_info *info);
494 /* Request format for this request. */
495 const struct req_format *mh_fmt;
498 enum mdt_handler_flags {
500 * struct mdt_body is passed in the incoming message, and object
501 * identified by this fid exists on disk.
503 * "habeo corpus" == "I have a body"
505 HABEO_CORPUS = (1 << 0),
507 * struct ldlm_request is passed in the incoming message.
509 * "habeo clavis" == "I have a key"
511 HABEO_CLAVIS = (1 << 1),
513 * this request has fixed reply format, so that reply message can be
514 * packed by generic code.
516 * "habeo refero" == "I have a reply"
518 HABEO_REFERO = (1 << 2),
520 * this request will modify something, so check whether the filesystem
521 * is readonly or not, then return -EROFS to client asap if necessary.
523 * "mutabor" == "I shall modify"
528 struct mdt_opc_slice {
531 struct mdt_handler *mos_hs;
534 static inline const struct md_device_operations *
535 mdt_child_ops(struct mdt_device * m)
537 LASSERT(m->mdt_child);
538 return m->mdt_child->md_ops;
541 static inline struct md_object *mdt_object_child(struct mdt_object *o)
544 return lu2md(lu_object_next(&o->mot_obj.mo_lu));
547 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
549 return info->mti_pill ? info->mti_pill->rc_req : NULL;
552 static inline int req_is_replay(struct ptlrpc_request *req)
554 LASSERT(req->rq_reqmsg);
555 return !!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
558 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
560 LASSERT(info->mti_exp);
561 return info->mti_exp->exp_connect_flags;
564 static inline void mdt_object_get(const struct lu_env *env,
565 struct mdt_object *o)
568 lu_object_get(&o->mot_obj.mo_lu);
572 static inline void mdt_object_put(const struct lu_env *env,
573 struct mdt_object *o)
576 lu_object_put(env, &o->mot_obj.mo_lu);
580 static inline int mdt_object_exists(const struct mdt_object *o)
582 return lu_object_exists(&o->mot_obj.mo_lu);
585 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
587 return lu_object_fid(&o->mot_obj.mo_lu);
590 static inline int mdt_object_obf(const struct mdt_object *o)
592 return lu_fid_eq(mdt_object_fid(o), &LU_OBF_FID);
595 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
597 return mdt->mdt_md_dev.md_lu_dev.ld_site;
600 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
602 return &mdt->mdt_seq_site;
605 static inline void mdt_export_evict(struct obd_export *exp)
607 class_fail_export(exp);
608 class_export_put(exp);
611 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
612 void mdt_set_disposition(struct mdt_thread_info *info,
613 struct ldlm_reply *rep, int flag);
614 void mdt_clear_disposition(struct mdt_thread_info *info,
615 struct ldlm_reply *rep, int flag);
617 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
618 ldlm_mode_t lm, const char *name,
621 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
624 int mdt_lock_setup(struct mdt_thread_info *info,
625 struct mdt_object *o,
626 struct mdt_lock_handle *lh);
628 int mdt_object_lock(struct mdt_thread_info *,
630 struct mdt_lock_handle *,
633 int mdt_object_lock_try(struct mdt_thread_info *,
635 struct mdt_lock_handle *,
638 void mdt_object_unlock(struct mdt_thread_info *,
640 struct mdt_lock_handle *,
643 struct mdt_object *mdt_object_new(const struct lu_env *,
645 const struct lu_fid *);
646 struct mdt_object *mdt_object_find(const struct lu_env *,
648 const struct lu_fid *);
649 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
650 const struct lu_fid *,
651 struct mdt_lock_handle *,
653 void mdt_object_unlock_put(struct mdt_thread_info *,
655 struct mdt_lock_handle *,
658 void mdt_client_compatibility(struct mdt_thread_info *info);
660 int mdt_close_unpack(struct mdt_thread_info *info);
661 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
662 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
663 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
664 const struct lu_attr *attr, const struct lu_fid *fid);
666 int mdt_getxattr(struct mdt_thread_info *info);
667 int mdt_reint_setxattr(struct mdt_thread_info *info,
668 struct mdt_lock_handle *lh);
670 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
671 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
673 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
674 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
675 struct mdt_lock_handle *lhc);
677 extern void target_recovery_fini(struct obd_device *obd);
678 extern void target_recovery_init(struct lu_target *lut,
679 svc_handler_t handler);
680 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
681 struct obd_device *, struct lustre_sb_info *lsi);
682 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
684 int mdt_export_stats_init(struct obd_device *obd,
685 struct obd_export *exp,
688 int mdt_pin(struct mdt_thread_info* info);
690 int mdt_lock_new_child(struct mdt_thread_info *info,
691 struct mdt_object *o,
692 struct mdt_lock_handle *child_lockh);
694 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
697 int mdt_reint_open(struct mdt_thread_info *info,
698 struct mdt_lock_handle *lhc);
700 struct mdt_file_data *mdt_handle2mfd(struct mdt_thread_info *,
701 const struct lustre_handle *);
704 MDT_IOEPOCH_CLOSED = 0,
705 MDT_IOEPOCH_OPENED = 1,
706 MDT_IOEPOCH_GETATTR = 2,
714 int mdt_attr_get_complex(struct mdt_thread_info *info,
715 struct mdt_object *o, struct md_attr *ma);
716 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
718 int mdt_object_is_som_enabled(struct mdt_object *mo);
719 int mdt_write_get(struct mdt_object *o);
720 void mdt_write_put(struct mdt_object *o);
721 int mdt_write_read(struct mdt_object *o);
722 struct mdt_file_data *mdt_mfd_new(void);
723 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
724 void mdt_mfd_free(struct mdt_file_data *mfd);
725 int mdt_close(struct mdt_thread_info *info);
726 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
727 struct md_attr *ma, int flags);
728 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
730 int mdt_done_writing(struct mdt_thread_info *info);
731 int mdt_fix_reply(struct mdt_thread_info *info);
732 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
733 const struct md_attr *);
734 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
736 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
737 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
738 const void *area, ssize_t len);
740 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
742 int mdt_check_ucred(struct mdt_thread_info *);
743 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
744 int mdt_init_ucred_reint(struct mdt_thread_info *);
745 void mdt_exit_ucred(struct mdt_thread_info *);
746 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
747 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
748 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
750 int mdt_handle_common(struct ptlrpc_request *req,
751 struct mdt_opc_slice *supported);
752 int mdt_connect(struct mdt_thread_info *info);
753 int mdt_disconnect(struct mdt_thread_info *info);
754 int mdt_set_info(struct mdt_thread_info *info);
755 int mdt_get_info(struct mdt_thread_info *info);
756 int mdt_getstatus(struct mdt_thread_info *info);
757 int mdt_getattr(struct mdt_thread_info *info);
758 int mdt_getattr_name(struct mdt_thread_info *info);
759 int mdt_statfs(struct mdt_thread_info *info);
760 int mdt_reint(struct mdt_thread_info *info);
761 int mdt_sync(struct mdt_thread_info *info);
762 int mdt_is_subdir(struct mdt_thread_info *info);
763 int mdt_obd_ping(struct mdt_thread_info *info);
764 int mdt_obd_log_cancel(struct mdt_thread_info *info);
765 int mdt_obd_qc_callback(struct mdt_thread_info *info);
766 int mdt_enqueue(struct mdt_thread_info *info);
767 int mdt_convert(struct mdt_thread_info *info);
768 int mdt_bl_callback(struct mdt_thread_info *info);
769 int mdt_cp_callback(struct mdt_thread_info *info);
770 int mdt_llog_create(struct mdt_thread_info *info);
771 int mdt_llog_destroy(struct mdt_thread_info *info);
772 int mdt_llog_read_header(struct mdt_thread_info *info);
773 int mdt_llog_next_block(struct mdt_thread_info *info);
774 int mdt_llog_prev_block(struct mdt_thread_info *info);
775 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
776 int mdt_readpage(struct mdt_thread_info *info);
777 int mdt_obd_idx_read(struct mdt_thread_info *info);
779 extern struct mdt_opc_slice mdt_regular_handlers[];
780 extern struct mdt_opc_slice mdt_seq_handlers[];
781 extern struct mdt_opc_slice mdt_fld_handlers[];
783 int mdt_quotacheck(struct mdt_thread_info *info);
784 int mdt_quotactl(struct mdt_thread_info *info);
785 int mdt_quota_dqacq(struct mdt_thread_info *info);
787 extern struct lprocfs_vars lprocfs_mds_module_vars[];
788 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
790 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
793 struct mdt_handler *mdt_handler_find(__u32 opc,
794 struct mdt_opc_slice *supported);
796 int mdt_init_sec_level(struct mdt_thread_info *);
797 int mdt_init_idmap(struct mdt_thread_info *);
798 void mdt_cleanup_idmap(struct mdt_export_data *);
799 int mdt_handle_idmap(struct mdt_thread_info *);
800 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
801 struct ptlrpc_user_desc *);
802 void mdt_body_reverse_idmap(struct mdt_thread_info *,
804 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
805 struct mdt_remote_perm *);
806 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
808 static inline struct mdt_device *mdt_dev(struct lu_device *d)
810 // LASSERT(lu_device_is_mdt(d));
811 return container_of0(d, struct mdt_device, mdt_md_dev.md_lu_dev);
814 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
816 struct lu_object *lo;
817 struct mdt_device *mdt = mdt_dev(mo->mot_obj.mo_lu.lo_dev);
819 lo = lu_object_locate(mo->mot_obj.mo_lu.lo_header,
820 mdt->mdt_bottom->dd_lu_dev.ld_type);
824 /* mdt/mdt_identity.c */
825 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
827 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
829 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
831 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
833 void mdt_flush_identity(struct upcall_cache *, int);
835 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
837 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
840 int mdt_hsm_state_get(struct mdt_thread_info *info);
841 int mdt_hsm_state_set(struct mdt_thread_info *info);
842 int mdt_hsm_action(struct mdt_thread_info *info);
843 int mdt_hsm_progress(struct mdt_thread_info *info);
844 int mdt_hsm_ct_register(struct mdt_thread_info *info);
845 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
846 int mdt_hsm_request(struct mdt_thread_info *info);
848 extern struct lu_context_key mdt_thread_key;
849 /* debug issues helper starts here*/
850 static inline int mdt_fail_write(const struct lu_env *env,
851 struct dt_device *dd, int id)
853 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
854 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
856 return dd->dd_ops->dt_ro(env, dd);
857 /* We set FAIL_ONCE because we never "un-fail" a device */
863 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
865 return &req->rq_export->exp_mdt_data;
868 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
869 struct mdt_lock_handle *lhc);
870 static inline int mdt_check_resent(struct mdt_thread_info *info,
871 mdt_reconstruct_t reconstruct,
872 struct mdt_lock_handle *lhc)
874 struct ptlrpc_request *req = mdt_info_req(info);
877 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
878 if (req_xid_is_last(req)) {
879 reconstruct(info, lhc);
882 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
883 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
888 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
889 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
891 static inline int is_identity_get_disabled(struct upcall_cache *cache)
893 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
896 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
898 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
899 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
900 struct lustre_handle *lh,
902 ldlm_policy_data_t *policy,
903 const struct ldlm_res_id *res_id,
904 __u64 flags, const __u64 *client_cookie)
911 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
912 mode, &flags, mdt_blocking_ast,
913 ldlm_completion_ast, NULL, NULL, 0,
914 LVB_T_NONE, client_cookie, lh);
915 return rc == ELDLM_OK ? 0 : -EIO;
918 static inline void mdt_fid_unlock(struct lustre_handle *lh,
921 ldlm_lock_decref(lh, mode);
924 extern mdl_mode_t mdt_mdl_lock_modes[];
925 extern ldlm_mode_t mdt_dlm_lock_modes[];
927 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
929 LASSERT(IS_PO2(mode));
930 return mdt_mdl_lock_modes[mode];
933 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
935 LASSERT(IS_PO2(mode));
936 return mdt_dlm_lock_modes[mode];
940 extern struct ldlm_valblock_ops mdt_lvbo;
942 static inline struct lu_name *mdt_name(const struct lu_env *env,
943 char *name, int namelen)
945 struct lu_name *lname;
946 struct mdt_thread_info *mti;
948 LASSERT(namelen > 0);
949 /* trailing '\0' in buffer */
950 LASSERT(name[namelen] == '\0');
952 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
953 lname = &mti->mti_name;
954 lname->ln_name = name;
955 lname->ln_namelen = namelen;
959 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
960 struct lu_name *slname)
965 tlname->ln_name = slname->ln_name;
966 tlname->ln_namelen = slname->ln_namelen;
970 void mdt_enable_cos(struct mdt_device *, int);
971 int mdt_cos_is_enabled(struct mdt_device *);
972 int mdt_hsm_copytool_send(struct obd_export *exp);
990 LPROC_MDT_SAMEDIR_RENAME,
991 LPROC_MDT_CROSSDIR_RENAME,
994 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
995 void mdt_stats_counter_init(struct lprocfs_stats *stats);
996 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
997 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
998 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
999 int mdt_procfs_fini(struct mdt_device *mdt);
1000 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1001 struct mdt_device *mdt,
1002 struct ptlrpc_request *req,
1003 struct mdt_object *src, struct mdt_object *tgt);
1006 int mdt_ck_thread_start(struct mdt_device *mdt);
1007 void mdt_ck_thread_stop(struct mdt_device *mdt);
1008 void mdt_ck_timer_callback(unsigned long castmeharder);
1009 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1011 static inline void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1012 const struct lu_fid *fid,
1013 struct lustre_capa *capa)
1015 struct md_capainfo *ci;
1017 LASSERT(offset >= 0 && offset < MD_CAPAINFO_MAX);
1018 if (!info->mti_mdt->mdt_opts.mo_mds_capa ||
1019 !(info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA))
1022 ci = md_capainfo(info->mti_env);
1024 ci->mc_fid[offset] = *fid;
1025 ci->mc_capa[offset] = capa;
1028 static inline void mdt_dump_capainfo(struct mdt_thread_info *info)
1030 struct md_capainfo *ci = md_capainfo(info->mti_env);
1035 for (i = 0; i < MD_CAPAINFO_MAX; i++) {
1036 if (!ci->mc_capa[i]) {
1037 CERROR("no capa for index %d "DFID"\n",
1038 i, PFID(&ci->mc_fid[i]));
1041 if (ci->mc_capa[i] == BYPASS_CAPA) {
1042 CERROR("bypass for index %d "DFID"\n",
1043 i, PFID(&ci->mc_fid[i]));
1046 DEBUG_CAPA(D_ERROR, ci->mc_capa[i], "index %d", i);
1050 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1052 return mdt->mdt_md_dev.md_lu_dev.ld_obd;
1055 extern const struct lu_device_operations mdt_lu_ops;
1057 static inline int lu_device_is_mdt(struct lu_device *d)
1059 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1062 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1064 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1065 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1066 return container_of0(d, struct mdt_device, mdt_md_dev.md_lu_dev);
1069 static inline char *mdt_obd_name(struct mdt_device *mdt)
1071 return mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name;
1074 int mds_mod_init(void);
1075 void mds_mod_exit(void);
1077 /* Update handlers */
1078 int out_handle(struct mdt_thread_info *info);
1080 #define out_tx_create(info, obj, attr, fid, dof, th, reply, idx) \
1081 __out_tx_create(info, obj, attr, fid, dof, th, reply, idx, \
1084 #define out_tx_attr_set(info, obj, attr, th, reply, idx) \
1085 __out_tx_attr_set(info, obj, attr, th, reply, idx, \
1088 #define out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx) \
1089 __out_tx_xattr_set(info, obj, buf, name, fl, th, reply, idx, \
1092 #define out_tx_ref_add(info, obj, th, reply, idx) \
1093 __out_tx_ref_add(info, obj, th, reply, idx, __FILE__, __LINE__)
1095 #define out_tx_ref_del(info, obj, th, reply, idx) \
1096 __out_tx_ref_del(info, obj, th, reply, idx, __FILE__, __LINE__)
1098 #define out_tx_index_insert(info, obj, th, name, fid, reply, idx) \
1099 __out_tx_index_insert(info, obj, th, name, fid, reply, idx, \
1102 #define out_tx_index_delete(info, obj, th, name, reply, idx) \
1103 __out_tx_index_delete(info, obj, th, name, reply, idx, \
1107 #endif /* __KERNEL__ */