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.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2011, 2016, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
32 * lustre/mdt/mdt_internal.h
34 * Lustre Metadata Target (mdt) request handler
36 * Author: Peter Braam <braam@clusterfs.com>
37 * Author: Andreas Dilger <adilger@clusterfs.com>
38 * Author: Phil Schwan <phil@clusterfs.com>
39 * Author: Mike Shaver <shaver@clusterfs.com>
40 * Author: Nikita Danilov <nikita@clusterfs.com>
41 * Author: Huang Hua <huanghua@clusterfs.com>
44 #ifndef _MDT_INTERNAL_H
45 #define _MDT_INTERNAL_H
47 #include <libcfs/libcfs.h>
48 #include <libcfs/libcfs_hash.h>
49 #include <upcall_cache.h>
50 #include <obd_class.h>
51 #include <lustre_disk.h>
52 #include <lu_target.h>
53 #include <md_object.h>
54 #include <lustre_fid.h>
55 #include <lustre_fld.h>
56 #include <lustre_req_layout.h>
57 #include <lustre_sec.h>
58 #include <lustre_idmap.h>
59 #include <lustre_eacl.h>
60 #include <lustre_quota.h>
61 #include <lustre_linkea.h>
65 /* file data for open files on MDS */
66 struct mdt_file_data {
67 /** portals handle must be first */
68 struct portals_handle mfd_handle;
69 /** open mode provided by client */
71 /** protected by med_open_lock */
72 struct list_head mfd_list;
73 /** xid of the open request */
75 /** old handle in replay case */
76 struct lustre_handle mfd_old_handle;
77 /** point to opened object */
78 struct mdt_object *mfd_object;
81 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
82 #define CDT_NORETRY_ACTION (1ULL << 1)
83 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
84 #define CDT_POLICY_SHIFT_COUNT 2
85 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
88 /* when adding a new policy, do not forget to update
89 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
91 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
93 /* Coordinator states. Keep the cdt_transition table in sync. */
94 enum cdt_states { CDT_STOPPED = 0,
103 static inline char *cdt_mdt_state2str(int state)
121 /* when multiple lock are needed, the lock order is
129 wait_queue_head_t cdt_waitq; /**< cdt wait queue */
130 bool cdt_event; /**< coordinator event */
131 struct task_struct *cdt_task; /**< cdt thread handle */
132 struct lu_env cdt_env; /**< coordinator lustre
134 struct lu_context cdt_session; /** session for lu_ucred */
135 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
136 __u64 cdt_policy; /**< policy flags */
137 enum cdt_states cdt_state; /**< state */
138 spinlock_t cdt_state_lock; /**< cdt_state lock */
139 atomic_t cdt_compound_id; /**< compound id
141 __u64 cdt_last_cookie; /**< last cookie
143 struct rw_semaphore cdt_llog_lock; /**< protect llog
145 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
146 struct rw_semaphore cdt_request_lock; /**< protect request
148 struct mutex cdt_restore_lock; /**< protect restore
150 cfs_time_t cdt_loop_period; /**< llog scan period */
151 cfs_time_t cdt_grace_delay; /**< request grace
153 cfs_time_t cdt_active_req_timeout; /**< request timeout */
154 __u32 cdt_default_archive_id; /**< archive id used
157 __u64 cdt_max_requests; /**< max count of started
159 atomic_t cdt_request_count; /**< current count of
160 * started requests */
161 /* started requests (struct cdt_agent_req:car_cookie_hash)
162 * indexed by cookie */
163 struct cfs_hash *cdt_request_cookie_hash;
164 /* started requests (struct cdt_agent_req:car_request_list) */
165 struct list_head cdt_request_list;
166 struct list_head cdt_agents; /**< list of register
168 struct list_head cdt_restore_hdl; /**< list of restore lock
171 /* Hash of cookies to locations of record locations in agent
173 struct cfs_hash *cdt_agent_record_hash;
175 /* Bitmasks indexed by the HSMA_XXX constants. */
176 __u64 cdt_user_request_mask;
177 __u64 cdt_group_request_mask;
178 __u64 cdt_other_request_mask;
180 /* Remove archive on last unlink policy */
181 bool cdt_remove_archive_on_last_unlink;
184 /* mdt state flag bits */
185 #define MDT_FL_CFGLOG 0
186 #define MDT_FL_SYNCED 1
190 struct lu_device mdt_lu_dev;
191 struct seq_server_site mdt_seq_site;
192 /* DLM name-space for meta-data locks maintained by this server */
193 struct ldlm_namespace *mdt_namespace;
194 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
195 struct ptlrpc_client *mdt_ldlm_client;
196 /* underlying device */
197 struct obd_export *mdt_child_exp;
198 struct md_device *mdt_child;
199 struct dt_device *mdt_bottom;
200 struct obd_export *mdt_bottom_exp;
202 struct lu_target mdt_lut;
204 * Options bit-fields.
207 unsigned int mo_user_xattr:1,
213 /* mdt state flags */
214 unsigned long mdt_state;
216 /* transaction callbacks */
217 struct dt_txn_callback mdt_txn_cb;
219 /* these values should be updated from lov if necessary.
220 * or should be placed somewhere else. */
225 /* preferred BRW size, decided by storage type and capability */
228 struct upcall_cache *mdt_identity_cache;
230 unsigned int mdt_capa_conf:1,
231 /* Enable remote dir on non-MDT0 */
232 mdt_enable_remote_dir:1,
235 gid_t mdt_enable_remote_dir_gid;
237 /* lock for osfs and md_root */
241 struct root_squash_info mdt_squash;
243 struct rename_stats mdt_rename_stats;
244 struct lu_fid mdt_md_root_fid;
246 /* connection to quota master */
247 struct obd_export *mdt_qmt_exp;
248 /* quota master device associated with this MDT */
249 struct lu_device *mdt_qmt_dev;
251 struct coordinator mdt_coordinator;
253 /* inter-MDT connection count */
254 atomic_t mdt_mds_mds_conns;
256 /* MDT device async commit count, used for debug and sanity test */
257 atomic_t mdt_async_commit_count;
259 struct mdt_object *mdt_md_root;
262 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
263 #define MDT_COS_DEFAULT (0)
266 struct lu_object_header mot_header;
267 struct lu_object mot_obj;
268 unsigned int mot_lov_created:1, /* lov object created */
269 mot_cache_attr:1; /* enable remote object
272 spinlock_t mot_write_lock;
273 /* Lock to protect create_data */
274 struct mutex mot_lov_mutex;
275 /* lock to protect read/write stages for Data-on-MDT files */
276 struct rw_semaphore mot_dom_sem;
277 /* Lock to protect lease open.
278 * Lease open acquires write lock; normal open acquires read lock */
279 struct rw_semaphore mot_open_sem;
280 atomic_t mot_lease_count;
281 atomic_t mot_open_count;
284 struct mdt_lock_handle {
285 /* Lock type, reg for cross-ref use or pdo lock. */
289 struct lustre_handle mlh_reg_lh;
290 enum ldlm_mode mlh_reg_mode;
293 struct lustre_handle mlh_pdo_lh;
294 enum ldlm_mode mlh_pdo_mode;
295 unsigned int mlh_pdo_hash;
297 /* Remote regular lock */
298 struct lustre_handle mlh_rreg_lh;
299 enum ldlm_mode mlh_rreg_mode;
303 MDT_LH_PARENT, /* parent lockh */
304 MDT_LH_CHILD, /* child lockh */
305 MDT_LH_OLD, /* old lockh for rename */
306 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
307 MDT_LH_NEW, /* new lockh for rename */
308 MDT_LH_RMT, /* used for return lh to caller */
309 MDT_LH_LOCAL, /* local lock never return to client */
318 /* Special magical errno for communicaiton between mdt_reint_open()
319 * and mdt_intent_reint() which means return the lock to the client
320 * for subsequent cross ref open. Previously we used plain -EREMOTE
321 * but other functions called in that path might return it too and
322 * confuse us. This is not returned to the client. See LU-5370. */
323 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
325 struct mdt_reint_record {
326 mdt_reint_t rr_opcode;
327 const struct lustre_handle *rr_handle;
328 const struct lu_fid *rr_fid1;
329 const struct lu_fid *rr_fid2;
330 struct lu_name rr_name;
331 struct lu_name rr_tgt_name;
337 enum mdt_reint_flag {
338 MRF_OPEN_TRUNC = 1 << 0,
342 * Common data shared by mdt-level handlers. This is allocated per-thread to
343 * reduce stack consumption.
345 struct mdt_thread_info {
348 * The following members will be filled explicitly
349 * with specific data in mdt_thread_info_init().
351 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
352 * request handling may migrate from one server thread to another.
354 struct req_capsule *mti_pill;
356 /* although we have export in req, there are cases when it is not
357 * available, e.g. closing files upon export destroy */
358 struct obd_export *mti_exp;
360 * A couple of lock handles.
362 struct mdt_lock_handle mti_lh[MDT_LH_NR];
364 struct mdt_device *mti_mdt;
365 const struct lu_env *mti_env;
367 /* transaction number of current request */
373 * The following members will be filled expilictly
374 * with zero in mdt_thread_info_init(). These members may be used
381 struct md_attr mti_attr;
383 * Body for "habeo corpus" operations.
385 const struct mdt_body *mti_body;
387 * Host object. This is released at the end of mdt_handler().
389 struct mdt_object *mti_object;
391 * Lock request for "habeo clavis" operations.
393 const struct ldlm_request *mti_dlm_req;
395 __u32 mti_has_trans:1, /* has txn already? */
397 /* big_lmm buffer was used and must be used in reply */
401 /* opdata for mdt_reint_open(), has the same as
402 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
403 * value onto disk for recovery when mdt_trans_stop_cb() is called.
409 * The following members will be filled explicitly
410 * with zero in mdt_reint_unpack(), because they are only used
411 * by reint requests (including mdt_reint_open()).
415 * reint record. contains information for reint operations.
417 struct mdt_reint_record mti_rr;
419 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
421 * Operation specification (currently create and lookup)
423 struct md_op_spec mti_spec;
427 * The following members will _NOT_ be initialized at all.
428 * DO NOT expect them to contain any valid value.
429 * They should be initialized explicitly by the user themselves.
432 /* XXX: If something is in a union, make sure they do not conflict */
433 struct lu_fid mti_tmp_fid1;
434 struct lu_fid mti_tmp_fid2;
435 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
436 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
438 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
439 char ns_name[48];/* for mdt_init0() */
440 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
441 struct obd_statfs osfs; /* for mdt_statfs() */
443 /* for mdt_readpage() */
444 struct lu_rdpg mti_rdpg;
445 /* for mdt_sendpage() */
446 struct l_wait_info mti_wait_info;
453 struct lustre_handle mti_close_handle;
455 struct lu_buf mti_buf;
456 struct lu_buf mti_big_buf;
458 /* Ops object filename */
459 struct lu_name mti_name;
460 /* per-thread values, can be re-used, may be vmalloc'd */
465 /* should be enough to fit lustre_mdt_attrs */
466 char mti_xattr_buf[128];
467 struct ldlm_enqueue_info mti_einfo;
468 struct tg_reply_data *mti_reply_data;
471 extern struct lu_context_key mdt_thread_key;
473 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
475 struct mdt_thread_info *mti;
477 lu_env_refill((void *)env);
478 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
483 struct cdt_req_progress {
484 struct mutex crp_lock; /**< protect tree */
485 struct interval_node *crp_root; /**< tree to track extent
487 struct interval_node **crp_node; /**< buffer for tree nodes
488 * vector of fixed size
490 int crp_cnt; /**< # of used nodes */
491 int crp_max; /**< # of allocated nodes */
494 struct cdt_agent_req {
495 struct hlist_node car_cookie_hash; /**< find req by cookie */
496 struct list_head car_request_list; /**< to chain all the req. */
497 atomic_t car_refcount; /**< reference counter */
498 __u64 car_compound_id; /**< compound id */
499 __u64 car_flags; /**< request original flags */
500 struct obd_uuid car_uuid; /**< agent doing the req. */
501 __u32 car_archive_id; /**< archive id */
502 int car_canceled; /**< request was canceled */
503 cfs_time_t car_req_start; /**< start time */
504 cfs_time_t car_req_update; /**< last update time */
505 struct hsm_action_item *car_hai; /**< req. to the agent */
506 struct cdt_req_progress car_progress; /**< track data mvt
509 extern struct kmem_cache *mdt_hsm_car_kmem;
512 struct list_head ha_list; /**< to chain the agents */
513 struct obd_uuid ha_uuid; /**< agent uuid */
514 __u32 *ha_archive_id; /**< archive id */
515 int ha_archive_cnt; /**< number of archive entries
516 * 0 means any archive */
517 atomic_t ha_requests; /**< current request count */
518 atomic_t ha_success; /**< number of successful
520 atomic_t ha_failure; /**< number of failed actions */
523 struct cdt_restore_handle {
524 struct list_head crh_list; /**< to chain the handle */
525 struct lu_fid crh_fid; /**< fid of the object */
526 struct ldlm_extent crh_extent; /**< extent of the restore */
527 struct mdt_lock_handle crh_lh; /**< lock handle */
529 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
531 struct hsm_record_update {
533 enum agent_req_status status;
536 static inline const struct md_device_operations *
537 mdt_child_ops(struct mdt_device * m)
539 LASSERT(m->mdt_child);
540 return m->mdt_child->md_ops;
543 static inline struct md_object *mdt_object_child(struct mdt_object *o)
546 return lu2md(lu_object_next(&o->mot_obj));
549 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
551 return info->mti_pill ? info->mti_pill->rc_req : NULL;
554 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
556 LASSERT(info->mti_exp);
557 return exp_connect_flags(info->mti_exp);
560 static inline void mdt_object_get(const struct lu_env *env,
561 struct mdt_object *o)
564 lu_object_get(&o->mot_obj);
568 static inline void mdt_object_put(const struct lu_env *env,
569 struct mdt_object *o)
572 lu_object_put(env, &o->mot_obj);
576 static inline int mdt_object_exists(const struct mdt_object *o)
578 return lu_object_exists(&o->mot_obj);
581 static inline int mdt_object_remote(const struct mdt_object *o)
583 return lu_object_remote(&o->mot_obj);
586 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
588 return lu_object_fid(&o->mot_obj);
591 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
593 return mdt->mdt_lu_dev.ld_site;
596 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
598 return &mdt->mdt_seq_site;
601 static inline void mdt_export_evict(struct obd_export *exp)
603 class_fail_export(exp);
606 /* Here we use LVB_TYPE to check dne client, because it is
607 * also landed on 2.4. */
608 static inline bool mdt_is_dne_client(struct obd_export *exp)
610 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
613 static inline bool mdt_is_striped_client(struct obd_export *exp)
615 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
624 /* XXX Look into layout in MDT layer. This must be done in LOD. */
625 static inline int mdt_lmm_dom_entry(struct lov_mds_md *lmm)
627 struct lov_comp_md_v1 *comp_v1;
628 struct lov_mds_md *v1;
631 if (lmm->lmm_magic == LOV_MAGIC_COMP_V1) {
632 comp_v1 = (struct lov_comp_md_v1 *)lmm;
633 v1 = (struct lov_mds_md *)((char *)comp_v1 +
634 comp_v1->lcm_entries[0].lcme_offset);
635 /* DoM entry is the first entry always */
636 if (lov_pattern(v1->lmm_pattern) != LOV_PATTERN_MDT)
639 for (i = 1; i < comp_v1->lcm_entry_count; i++) {
642 v1 = (struct lov_mds_md *)((char *)comp_v1 +
643 comp_v1->lcm_entries[i].lcme_offset);
644 for (j = 0; j < v1->lmm_stripe_count; j++) {
645 /* if there is any object on OST */
646 if (v1->lmm_objects[j].l_ost_idx !=
656 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
657 void mdt_set_disposition(struct mdt_thread_info *info,
658 struct ldlm_reply *rep, __u64 op_flag);
659 void mdt_clear_disposition(struct mdt_thread_info *info,
660 struct ldlm_reply *rep, __u64 op_flag);
662 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
663 const struct lu_name *lname);
665 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
667 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
668 struct mdt_lock_handle *lh);
670 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
671 struct mdt_lock_handle *lhc);
673 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *mo,
674 struct mdt_lock_handle *lh, __u64 ibits);
676 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
677 struct mdt_lock_handle *lh, __u64 ibits,
680 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
681 struct mdt_lock_handle *lh, __u64 *ibits,
682 __u64 trybits, bool cos_incompat);
684 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
685 struct mdt_lock_handle *lh, int decref);
686 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
687 enum ldlm_mode mode, int decref);
689 struct mdt_object *mdt_object_new(const struct lu_env *env,
691 const struct lu_fid *);
692 struct mdt_object *mdt_object_find(const struct lu_env *,
694 const struct lu_fid *);
695 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
696 const struct lu_fid *,
697 struct mdt_lock_handle *,
699 void mdt_object_unlock_put(struct mdt_thread_info *,
701 struct mdt_lock_handle *,
704 void mdt_client_compatibility(struct mdt_thread_info *info);
706 int mdt_remote_object_lock(struct mdt_thread_info *mti,
707 struct mdt_object *o, const struct lu_fid *fid,
708 struct lustre_handle *lh,
709 enum ldlm_mode mode, __u64 ibits, bool cache);
711 enum mdt_name_flags {
715 int mdt_name_unpack(struct req_capsule *pill,
716 const struct req_msg_field *field,
718 enum mdt_name_flags flags);
719 int mdt_close_unpack(struct mdt_thread_info *info);
720 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
721 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata);
722 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
723 #ifdef CONFIG_FS_POSIX_ACL
724 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
725 struct mdt_object *o, struct lu_nodemap *nodemap);
727 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
728 const struct lu_attr *attr, const struct lu_fid *fid);
729 int mdt_pack_size2body(struct mdt_thread_info *info,
730 const struct lu_fid *fid, bool dom_lock);
731 int mdt_getxattr(struct mdt_thread_info *info);
732 int mdt_reint_setxattr(struct mdt_thread_info *info,
733 struct mdt_lock_handle *lh);
735 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
736 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
738 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
739 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
740 struct mdt_lock_handle *lhc);
742 extern void target_recovery_fini(struct obd_device *obd);
743 extern void target_recovery_init(struct lu_target *lut,
744 svc_handler_t handler);
745 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
746 struct obd_device *, struct lustre_sb_info *lsi);
747 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
749 int mdt_export_stats_init(struct obd_device *obd,
750 struct obd_export *exp,
753 int mdt_lock_new_child(struct mdt_thread_info *info,
754 struct mdt_object *o,
755 struct mdt_lock_handle *child_lockh);
757 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
760 int mdt_reint_open(struct mdt_thread_info *info,
761 struct mdt_lock_handle *lhc);
763 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
764 const struct lustre_handle *handle,
767 int mdt_get_info(struct tgt_session_info *tsi);
768 int mdt_attr_get_complex(struct mdt_thread_info *info,
769 struct mdt_object *o, struct md_attr *ma);
770 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
772 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
773 struct md_attr *ma, const char *name);
774 int mdt_write_get(struct mdt_object *o);
775 void mdt_write_put(struct mdt_object *o);
776 int mdt_write_read(struct mdt_object *o);
777 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
778 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
779 void mdt_mfd_free(struct mdt_file_data *mfd);
780 int mdt_close(struct tgt_session_info *tsi);
781 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
783 int mdt_fix_reply(struct mdt_thread_info *info);
784 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
786 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
788 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
789 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
790 const void *area, ssize_t len);
792 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
793 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
795 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
796 int mdt_check_ucred(struct mdt_thread_info *);
797 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
798 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *, struct mdt_body *);
799 int mdt_init_ucred_reint(struct mdt_thread_info *);
800 void mdt_exit_ucred(struct mdt_thread_info *);
801 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
802 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
803 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
805 void mdt_thread_info_init(struct ptlrpc_request *req,
806 struct mdt_thread_info *mti);
807 void mdt_thread_info_fini(struct mdt_thread_info *mti);
808 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
809 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
810 struct ldlm_lock *new_lock,
811 struct mdt_lock_handle *lh, __u64 flags);
812 int mdt_intent_lock_replace(struct mdt_thread_info *info,
813 struct ldlm_lock **lockp,
814 struct mdt_lock_handle *lh,
815 __u64 flags, int result);
817 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
818 const struct md_hsm *mh);
820 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
821 void *data, int flag);
822 int mdt_links_read(struct mdt_thread_info *info,
823 struct mdt_object *mdt_obj,
824 struct linkea_data *ldata);
825 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
826 struct mdt_body *repbody);
828 static inline struct mdt_device *mdt_dev(struct lu_device *d)
830 return container_of0(d, struct mdt_device, mdt_lu_dev);
833 static inline struct mdt_object *mdt_obj(struct lu_object *o)
835 return container_of0(o, struct mdt_object, mot_obj);
838 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
840 struct lu_object *lo;
841 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
843 lo = lu_object_locate(mo->mot_obj.lo_header,
844 mdt->mdt_bottom->dd_lu_dev.ld_type);
849 static inline bool agent_req_in_final_state(enum agent_req_status ars)
851 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
852 (ars == ARS_CANCELED));
855 /* mdt/mdt_identity.c */
856 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
858 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
860 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
862 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
864 void mdt_flush_identity(struct upcall_cache *, int);
866 __u32 mdt_identity_get_perm(struct md_identity *, lnet_nid_t);
868 /* mdt/mdt_recovery.c */
869 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
872 int mdt_hsm_state_get(struct tgt_session_info *tsi);
873 int mdt_hsm_state_set(struct tgt_session_info *tsi);
874 int mdt_hsm_action(struct tgt_session_info *tsi);
875 int mdt_hsm_progress(struct tgt_session_info *tsi);
876 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
877 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
878 int mdt_hsm_request(struct tgt_session_info *tsi);
880 /* mdt/mdt_hsm_cdt_actions.c */
881 extern const struct file_operations mdt_hsm_actions_fops;
882 void dump_llog_agent_req_rec(const char *prefix,
883 const struct llog_agent_req_rec *larr);
884 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
885 llog_cb_t cb, void *data, u32 start_cat_idx,
886 u32 start_rec_idx, int rw);
887 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
888 __u64 compound_id, __u32 archive_id,
889 __u64 flags, struct hsm_action_item *hai);
890 int mdt_agent_record_update(const struct lu_env *env, struct mdt_device *mdt,
891 struct hsm_record_update *updates,
892 unsigned int updates_count);
893 void cdt_agent_record_hash_add(struct coordinator *cdt, u64 cookie, u32 cat_idt,
895 void cdt_agent_record_hash_lookup(struct coordinator *cdt, u64 cookie,
896 u32 *cat_idt, u32 *rec_idx);
897 void cdt_agent_record_hash_del(struct coordinator *cdt, u64 cookie);
899 /* mdt/mdt_hsm_cdt_agent.c */
900 extern const struct file_operations mdt_hsm_agent_fops;
901 int mdt_hsm_agent_register(struct mdt_thread_info *info,
902 const struct obd_uuid *uuid,
903 int nr_archives, __u32 *archive_num);
904 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
905 const struct obd_uuid *uuid,
907 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
908 const struct obd_uuid *uuid);
909 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
910 int succ_rq, int fail_rq, int new_rq,
911 const struct obd_uuid *uuid);
912 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
913 struct obd_uuid *uuid);
914 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
916 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
917 struct hsm_progress_kernel *pgs);
918 /* mdt/mdt_hsm_cdt_client.c */
919 int mdt_hsm_add_actions(struct mdt_thread_info *info,
920 struct hsm_action_list *hal);
921 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
922 struct hsm_action_list *hal);
923 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
924 const struct lu_fid *fid);
925 /* mdt/mdt_hsm_cdt_requests.c */
926 extern struct cfs_hash_ops cdt_request_cookie_hash_ops;
927 extern struct cfs_hash_ops cdt_agent_record_hash_ops;
928 extern const struct file_operations mdt_hsm_active_requests_fops;
929 void dump_requests(char *prefix, struct coordinator *cdt);
930 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
931 __u64 flags, struct obd_uuid *uuid,
932 struct hsm_action_item *hai);
933 void mdt_cdt_free_request(struct cdt_agent_req *car);
934 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
935 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt, u64 cookie);
936 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
937 void mdt_cdt_get_request(struct cdt_agent_req *car);
938 void mdt_cdt_put_request(struct cdt_agent_req *car);
939 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
940 const struct hsm_progress_kernel *pgs);
941 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
942 /* mdt/mdt_coordinator.c */
943 void mdt_hsm_dump_hal(int level, const char *prefix,
944 struct hsm_action_list *hal);
945 struct cdt_restore_handle *mdt_hsm_restore_hdl_find(struct coordinator *cdt,
946 const struct lu_fid *fid);
947 /* coordinator management */
948 int mdt_hsm_cdt_init(struct mdt_device *mdt);
949 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
950 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
953 * Signal the coordinator has work to do
954 * \param cdt [IN] coordinator
956 static inline void mdt_hsm_cdt_event(struct coordinator *cdt)
958 cdt->cdt_event = true;
961 /* coordinator control /proc interface */
962 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file,
963 const char __user *buffer,
964 size_t count, loff_t *off);
965 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
966 int hsm_cdt_procfs_init(struct mdt_device *mdt);
967 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
968 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
970 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
971 const struct lu_fid *fid,
973 /* actions/request helpers */
974 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
975 struct hsm_action_list *hal, struct obd_uuid *uuid);
976 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
977 u32 archive_id, u64 rq_flags,
978 const struct md_hsm *hsm);
979 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
980 struct hsm_progress_kernel *pgs,
981 const int update_record);
983 int mdt_close_swap_layouts(struct mdt_thread_info *info,
984 struct mdt_object *o, struct md_attr *ma);
986 extern struct lu_context_key mdt_thread_key;
988 /* debug issues helper starts here*/
989 static inline int mdt_fail_write(const struct lu_env *env,
990 struct dt_device *dd, int id)
992 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
993 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
995 return dt_ro(env, dd);
996 /* We set FAIL_ONCE because we never "un-fail" a device */
1002 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1004 return &req->rq_export->exp_mdt_data;
1007 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
1009 return mdt_dev(exp->exp_obd->obd_lu_dev);
1012 static inline bool mdt_rdonly(struct obd_export *exp)
1014 if (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
1015 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly)
1020 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1021 struct mdt_lock_handle *lhc);
1022 static inline int mdt_check_resent(struct mdt_thread_info *info,
1023 mdt_reconstruct_t reconstruct,
1024 struct mdt_lock_handle *lhc)
1026 struct ptlrpc_request *req = mdt_info_req(info);
1030 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1031 OBD_ALLOC_PTR(info->mti_reply_data);
1032 if (info->mti_reply_data == NULL)
1035 if (req_can_reconstruct(req, info->mti_reply_data)) {
1036 reconstruct(info, lhc);
1039 DEBUG_REQ(D_HA, req,
1040 "no reply data found for RESENT req");
1043 OBD_FREE_PTR(info->mti_reply_data);
1044 info->mti_reply_data = NULL;
1049 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1050 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1052 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1054 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1057 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1059 static int mdt_dom_glimpse_ast(struct ldlm_lock *lock, void *reqp)
1061 return -ELDLM_NO_LOCK_DATA;
1064 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1065 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1066 struct lustre_handle *lh, enum ldlm_mode mode,
1067 union ldlm_policy_data *policy,
1068 const struct ldlm_res_id *res_id,
1069 __u64 flags, const __u64 *client_cookie)
1072 bool glimpse = policy->l_inodebits.bits & MDS_INODELOCK_DOM;
1074 LASSERT(ns != NULL);
1075 LASSERT(lh != NULL);
1077 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1078 mode, &flags, mdt_blocking_ast,
1079 ldlm_completion_ast,
1080 glimpse ? mdt_dom_glimpse_ast : NULL,
1081 NULL, 0, LVB_T_NONE, client_cookie, lh);
1082 return rc == ELDLM_OK ? 0 : -EIO;
1085 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1087 ldlm_lock_decref(lh, mode);
1090 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
1092 return mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL;
1095 extern mdl_mode_t mdt_mdl_lock_modes[];
1096 extern enum ldlm_mode mdt_dlm_lock_modes[];
1098 /* LCK_MINMODE which is zero returns false for is_power_of_2 */
1100 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
1102 LASSERT(mode == LCK_MINMODE || is_power_of_2(mode));
1103 return mdt_mdl_lock_modes[mode];
1106 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1108 LASSERT(mode == MDL_MINMODE || is_power_of_2(mode));
1109 return mdt_dlm_lock_modes[mode];
1113 extern struct ldlm_valblock_ops mdt_lvbo;
1114 int mdt_dom_lvb_is_valid(struct ldlm_resource *res);
1115 int mdt_dom_lvbo_update(struct ldlm_resource *res, struct ldlm_lock *lock,
1116 struct ptlrpc_request *req, bool increase_only);
1118 void mdt_enable_cos(struct mdt_device *, int);
1119 int mdt_cos_is_enabled(struct mdt_device *);
1137 LPROC_MDT_SAMEDIR_RENAME,
1138 LPROC_MDT_CROSSDIR_RENAME,
1144 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1145 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1146 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1147 void mdt_procfs_fini(struct mdt_device *mdt);
1149 /* lustre/mdt_mdt_lproc.c */
1150 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1152 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1153 struct mdt_device *mdt,
1154 struct ptlrpc_request *req,
1155 struct mdt_object *src, struct mdt_object *tgt);
1157 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1159 return mdt->mdt_lu_dev.ld_obd;
1162 extern const struct lu_device_operations mdt_lu_ops;
1164 static inline char *mdt_obd_name(struct mdt_device *mdt)
1166 return mdt->mdt_lu_dev.ld_obd->obd_name;
1169 int mds_mod_init(void);
1170 void mds_mod_exit(void);
1172 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1174 struct obd_export *exp = req->rq_export;
1177 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1178 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1185 #define VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
1186 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
1188 int mdt_obd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
1189 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1190 struct niobuf_remote *rnb, int *nr_local,
1191 struct niobuf_local *lnb);
1193 int mdt_obd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1194 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1195 struct niobuf_remote *rnb, int npages,
1196 struct niobuf_local *lnb, int old_rc);
1197 int mdt_punch_hdl(struct tgt_session_info *tsi);
1198 int mdt_glimpse_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
1199 struct ldlm_lock **lockp, __u64 flags);
1200 int mdt_brw_enqueue(struct mdt_thread_info *info, struct ldlm_namespace *ns,
1201 struct ldlm_lock **lockp, __u64 flags);
1202 void mdt_dom_discard_data(struct mdt_thread_info *info,
1203 const struct lu_fid *fid);
1204 int mdt_dom_disk_lvbo_update(const struct lu_env *env, struct mdt_object *mo,
1205 struct ldlm_resource *res, bool increase_only);
1206 void mdt_dom_obj_lvb_update(const struct lu_env *env, struct mdt_object *mo,
1207 bool increase_only);
1208 int mdt_dom_lvb_alloc(struct ldlm_resource *res);
1210 static inline void mdt_dom_check_and_discard(struct mdt_thread_info *mti,
1211 struct mdt_object *mo)
1213 if (lu_object_is_dying(&mo->mot_header) &&
1214 S_ISREG(lu_object_attr(&mo->mot_obj)))
1215 mdt_dom_discard_data(mti, mdt_object_fid(mo));
1218 int mdt_dom_object_size(const struct lu_env *env, struct mdt_device *mdt,
1219 const struct lu_fid *fid, struct mdt_body *mb,
1221 bool mdt_dom_client_has_lock(struct mdt_thread_info *info,
1222 const struct lu_fid *fid);
1224 long mdt_grant_connect(const struct lu_env *env, struct obd_export *exp,
1225 u64 want, bool conservative);
1226 extern struct kmem_cache *ldlm_glimpse_work_kmem;
1228 #endif /* _MDT_INTERNAL_H */