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_xmds_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,
165 /* statfs optimization: we cache a bit */
166 struct obd_statfs mdt_osfs;
168 spinlock_t mdt_osfs_lock;
171 uid_t mdt_squash_uid;
172 gid_t mdt_squash_gid;
173 cfs_list_t mdt_nosquash_nids;
174 char *mdt_nosquash_str;
175 int mdt_nosquash_strlen;
176 struct rw_semaphore mdt_squash_sem;
178 cfs_proc_dir_entry_t *mdt_proc_entry;
179 struct lprocfs_stats *mdt_stats;
181 struct rename_stats mdt_rename_stats;
182 struct lu_fid mdt_md_root_fid;
184 /* connection to quota master */
185 struct obd_export *mdt_qmt_exp;
186 /* quota master device associated with this MDT */
187 struct lu_device *mdt_qmt_dev;
190 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
191 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
192 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
193 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
194 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
195 #define MDT_COS_DEFAULT (0)
198 struct lu_object_header mot_header;
199 struct md_object mot_obj;
202 int mot_ioepoch_count;
204 /* Lock to protect object's IO epoch. */
205 struct mutex mot_ioepoch_mutex;
206 /* Lock to protect create_data */
207 struct mutex mot_lov_mutex;
210 enum mdt_object_flags {
211 /** SOM attributes are changed. */
212 MOF_SOM_CHANGE = (1 << 0),
214 * The SOM recovery state for mdt object.
215 * This state is an in-memory equivalent of an absent SOM EA, used
216 * instead of invalidating SOM EA while IOEpoch is still opened when
217 * a client eviction occurs or a client fails to obtain SOM attributes.
218 * It indicates that the last IOEpoch holder will need to obtain SOM
219 * attributes under [0;EOF] extent lock to flush all the client's
220 * cached of evicted from MDS clients (but not necessary evicted from
221 * OST) before taking ost attributes.
223 MOF_SOM_RECOV = (1 << 1),
224 /** File has been just created. */
225 MOF_SOM_CREATED = (1 << 2),
226 /** lov object has been created. */
227 MOF_LOV_CREATED = (1 << 3),
230 struct mdt_lock_handle {
231 /* Lock type, reg for cross-ref use or pdo lock. */
235 struct lustre_handle mlh_reg_lh;
236 ldlm_mode_t mlh_reg_mode;
239 struct lustre_handle mlh_pdo_lh;
240 ldlm_mode_t mlh_pdo_mode;
241 unsigned int mlh_pdo_hash;
245 MDT_LH_PARENT, /* parent lockh */
246 MDT_LH_CHILD, /* child lockh */
247 MDT_LH_OLD, /* old lockh for rename */
248 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
249 MDT_LH_NEW, /* new lockh for rename */
250 MDT_LH_RMT, /* used for return lh to caller */
259 struct mdt_reint_record {
260 mdt_reint_t rr_opcode;
261 const struct lustre_handle *rr_handle;
262 const struct lu_fid *rr_fid1;
263 const struct lu_fid *rr_fid2;
268 const void *rr_eadata;
271 const struct llog_cookie *rr_logcookies;
275 enum mdt_reint_flag {
276 MRF_OPEN_TRUNC = 1 << 0,
280 * Common data shared by mdt-level handlers. This is allocated per-thread to
281 * reduce stack consumption.
283 struct mdt_thread_info {
286 * The following members will be filled explicitly
287 * with specific data in mdt_thread_info_init().
289 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
290 * request handling may migrate from one server thread to another.
292 struct req_capsule *mti_pill;
294 /* although we have export in req, there are cases when it is not
295 * available, e.g. closing files upon export destroy */
296 struct obd_export *mti_exp;
298 * A couple of lock handles.
300 struct mdt_lock_handle mti_lh[MDT_LH_NR];
302 struct mdt_device *mti_mdt;
303 const struct lu_env *mti_env;
306 * Additional fail id that can be set by handler. Passed to
307 * target_send_reply().
311 /* transaction number of current request */
317 * The following members will be filled expilictly
318 * with zero in mdt_thread_info_init(). These members may be used
325 struct md_attr mti_attr;
327 * Body for "habeo corpus" operations.
329 const struct mdt_body *mti_body;
331 * Host object. This is released at the end of mdt_handler().
333 struct mdt_object *mti_object;
335 * Lock request for "habeo clavis" operations.
337 const struct ldlm_request *mti_dlm_req;
339 __u32 mti_has_trans:1, /* has txn already? */
342 /* opdata for mdt_reint_open(), has the same as
343 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
344 * value onto disk for recovery when mdt_trans_stop_cb() is called.
350 * The following members will be filled explicitly
351 * with zero in mdt_reint_unpack(), because they are only used
352 * by reint requests (including mdt_reint_open()).
356 * reint record. contains information for reint operations.
358 struct mdt_reint_record mti_rr;
360 /** md objects included in operation */
361 struct mdt_object *mti_mos;
362 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
364 * Operation specification (currently create and lookup)
366 struct md_op_spec mti_spec;
370 * The following members will _NOT_ be initialized at all.
371 * DO NOT expect them to contain any valid value.
372 * They should be initialized explicitly by the user themselves.
375 /* XXX: If something is in a union, make sure they do not conflict */
377 struct lu_fid mti_tmp_fid1;
378 struct lu_fid mti_tmp_fid2;
379 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
380 * mdt_rename_lock() */
381 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
384 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
385 char ns_name[48]; /* for mdt_init0() */
386 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
387 struct obd_statfs osfs; /* for mdt_statfs() */
389 /* for mdt_readpage() */
390 struct lu_rdpg mti_rdpg;
391 /* for mdt_sendpage() */
392 struct l_wait_info mti_wait_info;
396 struct md_som_data data;
400 /* IO epoch related stuff. */
401 struct mdt_ioepoch *mti_ioepoch;
402 __u64 mti_replayepoch;
405 struct lu_buf mti_buf;
406 struct lustre_capa_key mti_capa_key;
408 /* Ops object filename */
409 struct lu_name mti_name;
410 /* per-thread values, can be re-used */
413 /* big_lmm buffer was used and must be used in reply */
414 int mti_big_lmm_used;
415 /* should be enough to fit lustre_mdt_attrs */
416 char mti_xattr_buf[128];
419 /* ptlrpc request handler for MDT. All handlers are
420 * grouped into several slices - struct mdt_opc_slice,
421 * and stored in an array - mdt_handlers[].
424 /* The name of this handler. */
426 /* Fail id for this handler, checked at the beginning of this handler*/
428 /* Operation code for this handler */
430 /* flags are listed in enum mdt_handler_flags below. */
432 /* The actual handler function to execute. */
433 int (*mh_act)(struct mdt_thread_info *info);
434 /* Request format for this request. */
435 const struct req_format *mh_fmt;
438 enum mdt_handler_flags {
440 * struct mdt_body is passed in the incoming message, and object
441 * identified by this fid exists on disk.
443 * "habeo corpus" == "I have a body"
445 HABEO_CORPUS = (1 << 0),
447 * struct ldlm_request is passed in the incoming message.
449 * "habeo clavis" == "I have a key"
451 HABEO_CLAVIS = (1 << 1),
453 * this request has fixed reply format, so that reply message can be
454 * packed by generic code.
456 * "habeo refero" == "I have a reply"
458 HABEO_REFERO = (1 << 2),
460 * this request will modify something, so check whether the filesystem
461 * is readonly or not, then return -EROFS to client asap if necessary.
463 * "mutabor" == "I shall modify"
468 struct mdt_opc_slice {
471 struct mdt_handler *mos_hs;
474 static inline const struct md_device_operations *
475 mdt_child_ops(struct mdt_device * m)
477 LASSERT(m->mdt_child);
478 return m->mdt_child->md_ops;
481 static inline struct md_object *mdt_object_child(struct mdt_object *o)
484 return lu2md(lu_object_next(&o->mot_obj.mo_lu));
487 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
489 return info->mti_pill ? info->mti_pill->rc_req : NULL;
492 static inline int req_is_replay(struct ptlrpc_request *req)
494 LASSERT(req->rq_reqmsg);
495 return !!(lustre_msg_get_flags(req->rq_reqmsg) & MSG_REPLAY);
498 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
500 LASSERT(info->mti_exp);
501 return info->mti_exp->exp_connect_flags;
504 static inline void mdt_object_get(const struct lu_env *env,
505 struct mdt_object *o)
508 lu_object_get(&o->mot_obj.mo_lu);
512 static inline void mdt_object_put(const struct lu_env *env,
513 struct mdt_object *o)
516 lu_object_put(env, &o->mot_obj.mo_lu);
520 static inline int mdt_object_exists(const struct mdt_object *o)
522 return lu_object_exists(&o->mot_obj.mo_lu);
525 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
527 return lu_object_fid(&o->mot_obj.mo_lu);
530 static inline int mdt_object_obf(const struct mdt_object *o)
532 return lu_fid_eq(mdt_object_fid(o), &LU_OBF_FID);
535 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
537 return mdt->mdt_md_dev.md_lu_dev.ld_site;
540 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
542 return &mdt->mdt_seq_site;
545 static inline void mdt_export_evict(struct obd_export *exp)
547 class_fail_export(exp);
548 class_export_put(exp);
551 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
552 void mdt_set_disposition(struct mdt_thread_info *info,
553 struct ldlm_reply *rep, int flag);
554 void mdt_clear_disposition(struct mdt_thread_info *info,
555 struct ldlm_reply *rep, int flag);
557 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
558 ldlm_mode_t lm, const char *name,
561 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
564 int mdt_lock_setup(struct mdt_thread_info *info,
565 struct mdt_object *o,
566 struct mdt_lock_handle *lh);
568 int mdt_object_lock(struct mdt_thread_info *,
570 struct mdt_lock_handle *,
573 int mdt_object_lock_try(struct mdt_thread_info *,
575 struct mdt_lock_handle *,
578 void mdt_object_unlock(struct mdt_thread_info *,
580 struct mdt_lock_handle *,
583 struct mdt_object *mdt_object_new(const struct lu_env *,
585 const struct lu_fid *);
586 struct mdt_object *mdt_object_find(const struct lu_env *,
588 const struct lu_fid *);
589 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
590 const struct lu_fid *,
591 struct mdt_lock_handle *,
593 void mdt_object_unlock_put(struct mdt_thread_info *,
595 struct mdt_lock_handle *,
598 void mdt_client_compatibility(struct mdt_thread_info *info);
600 int mdt_close_unpack(struct mdt_thread_info *info);
601 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
602 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
603 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
604 const struct lu_attr *attr, const struct lu_fid *fid);
606 int mdt_getxattr(struct mdt_thread_info *info);
607 int mdt_reint_setxattr(struct mdt_thread_info *info,
608 struct mdt_lock_handle *lh);
610 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
611 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
613 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
614 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
615 struct mdt_lock_handle *lhc);
617 extern void target_recovery_fini(struct obd_device *obd);
618 extern void target_recovery_init(struct lu_target *lut,
619 svc_handler_t handler);
620 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
621 struct obd_device *, struct lustre_sb_info *lsi);
622 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
624 int mdt_export_stats_init(struct obd_device *obd,
625 struct obd_export *exp,
628 int mdt_pin(struct mdt_thread_info* info);
630 int mdt_lock_new_child(struct mdt_thread_info *info,
631 struct mdt_object *o,
632 struct mdt_lock_handle *child_lockh);
634 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
637 int mdt_reint_open(struct mdt_thread_info *info,
638 struct mdt_lock_handle *lhc);
640 struct mdt_file_data *mdt_handle2mfd(struct mdt_thread_info *,
641 const struct lustre_handle *);
644 MDT_IOEPOCH_CLOSED = 0,
645 MDT_IOEPOCH_OPENED = 1,
646 MDT_IOEPOCH_GETATTR = 2,
654 int mdt_attr_get_complex(struct mdt_thread_info *info,
655 struct mdt_object *o, struct md_attr *ma);
656 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
658 int mdt_object_is_som_enabled(struct mdt_object *mo);
659 int mdt_write_get(struct mdt_object *o);
660 void mdt_write_put(struct mdt_object *o);
661 int mdt_write_read(struct mdt_object *o);
662 struct mdt_file_data *mdt_mfd_new(void);
663 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
664 void mdt_mfd_free(struct mdt_file_data *mfd);
665 int mdt_close(struct mdt_thread_info *info);
666 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
667 struct md_attr *ma, int flags);
668 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
670 int mdt_done_writing(struct mdt_thread_info *info);
671 int mdt_fix_reply(struct mdt_thread_info *info);
672 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
673 const struct md_attr *);
674 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
676 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
677 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
678 const void *area, ssize_t len);
680 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
682 int mdt_check_ucred(struct mdt_thread_info *);
683 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
684 int mdt_init_ucred_reint(struct mdt_thread_info *);
685 void mdt_exit_ucred(struct mdt_thread_info *);
686 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
687 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
688 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
690 int mdt_handle_common(struct ptlrpc_request *req,
691 struct mdt_opc_slice *supported);
692 int mdt_connect(struct mdt_thread_info *info);
693 int mdt_disconnect(struct mdt_thread_info *info);
694 int mdt_set_info(struct mdt_thread_info *info);
695 int mdt_get_info(struct mdt_thread_info *info);
696 int mdt_getstatus(struct mdt_thread_info *info);
697 int mdt_getattr(struct mdt_thread_info *info);
698 int mdt_getattr_name(struct mdt_thread_info *info);
699 int mdt_statfs(struct mdt_thread_info *info);
700 int mdt_reint(struct mdt_thread_info *info);
701 int mdt_sync(struct mdt_thread_info *info);
702 int mdt_is_subdir(struct mdt_thread_info *info);
703 int mdt_obd_ping(struct mdt_thread_info *info);
704 int mdt_obd_log_cancel(struct mdt_thread_info *info);
705 int mdt_obd_qc_callback(struct mdt_thread_info *info);
706 int mdt_enqueue(struct mdt_thread_info *info);
707 int mdt_convert(struct mdt_thread_info *info);
708 int mdt_bl_callback(struct mdt_thread_info *info);
709 int mdt_cp_callback(struct mdt_thread_info *info);
710 int mdt_llog_create(struct mdt_thread_info *info);
711 int mdt_llog_destroy(struct mdt_thread_info *info);
712 int mdt_llog_read_header(struct mdt_thread_info *info);
713 int mdt_llog_next_block(struct mdt_thread_info *info);
714 int mdt_llog_prev_block(struct mdt_thread_info *info);
715 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
716 int mdt_readpage(struct mdt_thread_info *info);
717 int mdt_obd_idx_read(struct mdt_thread_info *info);
719 extern struct mdt_opc_slice mdt_regular_handlers[];
720 extern struct mdt_opc_slice mdt_seq_handlers[];
721 extern struct mdt_opc_slice mdt_fld_handlers[];
723 int mdt_quotacheck(struct mdt_thread_info *info);
724 int mdt_quotactl(struct mdt_thread_info *info);
725 int mdt_quota_dqacq(struct mdt_thread_info *info);
727 extern struct lprocfs_vars lprocfs_mds_module_vars[];
728 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
730 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
734 int mdt_init_sec_level(struct mdt_thread_info *);
735 int mdt_init_idmap(struct mdt_thread_info *);
736 void mdt_cleanup_idmap(struct mdt_export_data *);
737 int mdt_handle_idmap(struct mdt_thread_info *);
738 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
739 struct ptlrpc_user_desc *);
740 void mdt_body_reverse_idmap(struct mdt_thread_info *,
742 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
743 struct mdt_remote_perm *);
744 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
746 static inline struct mdt_device *mdt_dev(struct lu_device *d)
748 // LASSERT(lu_device_is_mdt(d));
749 return container_of0(d, struct mdt_device, mdt_md_dev.md_lu_dev);
752 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
754 struct lu_object *lo;
755 struct mdt_device *mdt = mdt_dev(mo->mot_obj.mo_lu.lo_dev);
757 lo = lu_object_locate(mo->mot_obj.mo_lu.lo_header,
758 mdt->mdt_bottom->dd_lu_dev.ld_type);
762 /* mdt/mdt_identity.c */
763 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
765 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
767 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
769 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
771 void mdt_flush_identity(struct upcall_cache *, int);
773 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
775 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
777 extern struct lu_context_key mdt_thread_key;
778 /* debug issues helper starts here*/
779 static inline int mdt_fail_write(const struct lu_env *env,
780 struct dt_device *dd, int id)
782 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
783 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
785 return dd->dd_ops->dt_ro(env, dd);
786 /* We set FAIL_ONCE because we never "un-fail" a device */
792 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
794 return &req->rq_export->exp_mdt_data;
797 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
798 struct mdt_lock_handle *lhc);
799 static inline int mdt_check_resent(struct mdt_thread_info *info,
800 mdt_reconstruct_t reconstruct,
801 struct mdt_lock_handle *lhc)
803 struct ptlrpc_request *req = mdt_info_req(info);
806 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
807 if (req_xid_is_last(req)) {
808 reconstruct(info, lhc);
811 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
812 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
817 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
818 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
820 static inline int is_identity_get_disabled(struct upcall_cache *cache)
822 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
825 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
827 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
828 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
829 struct lustre_handle *lh,
831 ldlm_policy_data_t *policy,
832 const struct ldlm_res_id *res_id,
833 __u64 flags, const __u64 *client_cookie)
840 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
841 mode, &flags, mdt_blocking_ast,
842 ldlm_completion_ast, NULL, NULL, 0,
843 LVB_T_NONE, client_cookie, lh);
844 return rc == ELDLM_OK ? 0 : -EIO;
847 static inline void mdt_fid_unlock(struct lustre_handle *lh,
850 ldlm_lock_decref(lh, mode);
853 extern mdl_mode_t mdt_mdl_lock_modes[];
854 extern ldlm_mode_t mdt_dlm_lock_modes[];
856 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
858 LASSERT(IS_PO2(mode));
859 return mdt_mdl_lock_modes[mode];
862 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
864 LASSERT(IS_PO2(mode));
865 return mdt_dlm_lock_modes[mode];
869 extern struct ldlm_valblock_ops mdt_lvbo;
871 static inline struct lu_name *mdt_name(const struct lu_env *env,
872 char *name, int namelen)
874 struct lu_name *lname;
875 struct mdt_thread_info *mti;
877 LASSERT(namelen > 0);
878 /* trailing '\0' in buffer */
879 LASSERT(name[namelen] == '\0');
881 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
882 lname = &mti->mti_name;
883 lname->ln_name = name;
884 lname->ln_namelen = namelen;
888 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
889 struct lu_name *slname)
894 tlname->ln_name = slname->ln_name;
895 tlname->ln_namelen = slname->ln_namelen;
899 void mdt_enable_cos(struct mdt_device *, int);
900 int mdt_cos_is_enabled(struct mdt_device *);
901 int mdt_hsm_copytool_send(struct obd_export *exp);
919 LPROC_MDT_SAMEDIR_RENAME,
920 LPROC_MDT_CROSSDIR_RENAME,
923 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
924 void mdt_stats_counter_init(struct lprocfs_stats *stats);
925 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
926 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
927 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
928 int mdt_procfs_fini(struct mdt_device *mdt);
929 void mdt_rename_counter_tally(struct mdt_thread_info *info,
930 struct mdt_device *mdt,
931 struct ptlrpc_request *req,
932 struct mdt_object *src, struct mdt_object *tgt);
934 void mdt_time_start(const struct mdt_thread_info *info);
935 void mdt_time_end(const struct mdt_thread_info *info, int idx);
938 int mdt_ck_thread_start(struct mdt_device *mdt);
939 void mdt_ck_thread_stop(struct mdt_device *mdt);
940 void mdt_ck_timer_callback(unsigned long castmeharder);
941 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
943 static inline void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
944 const struct lu_fid *fid,
945 struct lustre_capa *capa)
947 struct md_capainfo *ci;
949 LASSERT(offset >= 0 && offset < MD_CAPAINFO_MAX);
950 if (!info->mti_mdt->mdt_opts.mo_mds_capa ||
951 !(info->mti_exp->exp_connect_flags & OBD_CONNECT_MDS_CAPA))
954 ci = md_capainfo(info->mti_env);
956 ci->mc_fid[offset] = *fid;
957 ci->mc_capa[offset] = capa;
960 static inline void mdt_dump_capainfo(struct mdt_thread_info *info)
962 struct md_capainfo *ci = md_capainfo(info->mti_env);
967 for (i = 0; i < MD_CAPAINFO_MAX; i++) {
968 if (!ci->mc_capa[i]) {
969 CERROR("no capa for index %d "DFID"\n",
970 i, PFID(&ci->mc_fid[i]));
973 if (ci->mc_capa[i] == BYPASS_CAPA) {
974 CERROR("bypass for index %d "DFID"\n",
975 i, PFID(&ci->mc_fid[i]));
978 DEBUG_CAPA(D_ERROR, ci->mc_capa[i], "index %d", i);
982 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
984 return mdt->mdt_md_dev.md_lu_dev.ld_obd;
987 extern const struct lu_device_operations mdt_lu_ops;
989 static inline int lu_device_is_mdt(struct lu_device *d)
991 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
994 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
996 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
997 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
998 return container_of0(d, struct mdt_device, mdt_md_dev.md_lu_dev);
1001 static inline char *mdt_obd_name(struct mdt_device *mdt)
1003 return mdt->mdt_md_dev.md_lu_dev.ld_obd->obd_name;
1006 int mds_mod_init(void);
1007 void mds_mod_exit(void);
1009 #endif /* __KERNEL__ */