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, 2017, 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 struct mutex cdt_state_lock; /**< cdt_state lock */
139 __u64 cdt_last_cookie; /**< last cookie
141 struct rw_semaphore cdt_llog_lock; /**< protect llog
143 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
144 struct rw_semaphore cdt_request_lock; /**< protect request
146 struct mutex cdt_restore_lock; /**< protect restore
148 time64_t cdt_loop_period; /**< llog scan period */
149 time64_t cdt_grace_delay; /**< request grace
151 time64_t cdt_active_req_timeout; /**< request timeout */
152 __u32 cdt_default_archive_id; /**< archive id used
155 __u64 cdt_max_requests; /**< max count of started
157 /** Current count of active requests */
158 atomic_t cdt_request_count; /** total */
159 atomic_t cdt_archive_count;
160 atomic_t cdt_restore_count;
161 atomic_t cdt_remove_count;
163 /* started requests (struct cdt_agent_req:car_cookie_hash)
164 * indexed by cookie */
165 struct cfs_hash *cdt_request_cookie_hash;
166 /* started requests (struct cdt_agent_req:car_request_list) */
167 struct list_head cdt_request_list;
168 struct list_head cdt_agents; /**< list of register
170 struct list_head cdt_restore_handle_list;
172 /* Hash of cookies to locations of record locations in agent
174 struct cfs_hash *cdt_agent_record_hash;
176 /* Bitmasks indexed by the HSMA_XXX constants. */
177 __u64 cdt_user_request_mask;
178 __u64 cdt_group_request_mask;
179 __u64 cdt_other_request_mask;
181 /* Remove archive on last unlink policy */
182 bool cdt_remove_archive_on_last_unlink;
184 bool cdt_wakeup_coordinator;
187 /* mdt state flag bits */
188 #define MDT_FL_CFGLOG 0
189 #define MDT_FL_SYNCED 1
191 /* possible values for mo_dom_lock */
193 NO_DOM_LOCK_ON_OPEN = 0,
194 TRYLOCK_DOM_ON_OPEN = 1,
195 ALWAYS_DOM_LOCK_ON_OPEN = 2,
196 NUM_DOM_LOCK_ON_OPEN_MODES
201 struct lu_device mdt_lu_dev;
202 struct seq_server_site mdt_seq_site;
203 /* DLM name-space for meta-data locks maintained by this server */
204 struct ldlm_namespace *mdt_namespace;
205 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
206 struct ptlrpc_client *mdt_ldlm_client;
207 /* underlying device */
208 struct obd_export *mdt_child_exp;
209 struct md_device *mdt_child;
210 struct dt_device *mdt_bottom;
211 struct obd_export *mdt_bottom_exp;
212 struct local_oid_storage *mdt_los;
214 struct lu_target mdt_lut;
216 * Options bit-fields.
219 unsigned int mo_user_xattr:1,
224 unsigned int mo_dom_lock;
226 /* mdt state flags */
227 unsigned long mdt_state;
229 /* transaction callbacks */
230 struct dt_txn_callback mdt_txn_cb;
232 /* these values should be updated from lov if necessary.
233 * or should be placed somewhere else. */
238 /* preferred BRW size, decided by storage type and capability */
241 struct upcall_cache *mdt_identity_cache;
243 unsigned int mdt_capa_conf:1,
244 /* Enable remote dir on non-MDT0 */
245 mdt_enable_remote_dir:1,
248 gid_t mdt_enable_remote_dir_gid;
250 /* lock for osfs and md_root */
254 struct root_squash_info mdt_squash;
256 struct rename_stats mdt_rename_stats;
257 struct lu_fid mdt_md_root_fid;
259 /* connection to quota master */
260 struct obd_export *mdt_qmt_exp;
261 /* quota master device associated with this MDT */
262 struct lu_device *mdt_qmt_dev;
264 struct coordinator mdt_coordinator;
266 /* inter-MDT connection count */
267 atomic_t mdt_mds_mds_conns;
269 /* MDT device async commit count, used for debug and sanity test */
270 atomic_t mdt_async_commit_count;
272 struct mdt_object *mdt_md_root;
275 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
276 #define MDT_COS_DEFAULT (0)
278 #define ENOENT_VERSION 1 /** 'virtual' version of non-existent object */
281 struct lu_object_header mot_header;
282 struct lu_object mot_obj;
283 unsigned int mot_lov_created:1, /* lov object created */
284 mot_cache_attr:1; /* enable remote object
287 spinlock_t mot_write_lock;
288 /* Lock to protect object's SOM update. */
289 struct mutex mot_som_mutex;
290 /* Lock to protect create_data */
291 struct mutex mot_lov_mutex;
292 /* lock to protect read/write stages for Data-on-MDT files */
293 struct rw_semaphore mot_dom_sem;
294 /* Lock to protect lease open.
295 * Lease open acquires write lock; normal open acquires read lock */
296 struct rw_semaphore mot_open_sem;
297 atomic_t mot_lease_count;
298 atomic_t mot_open_count;
301 struct mdt_lock_handle {
302 /* Lock type, reg for cross-ref use or pdo lock. */
306 struct lustre_handle mlh_reg_lh;
307 enum ldlm_mode mlh_reg_mode;
310 struct lustre_handle mlh_pdo_lh;
311 enum ldlm_mode mlh_pdo_mode;
312 unsigned int mlh_pdo_hash;
314 /* Remote regular lock */
315 struct lustre_handle mlh_rreg_lh;
316 enum ldlm_mode mlh_rreg_mode;
320 MDT_LH_PARENT, /* parent lockh */
321 MDT_LH_CHILD, /* child lockh */
322 MDT_LH_OLD, /* old lockh for rename */
323 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
324 MDT_LH_NEW, /* new lockh for rename */
325 MDT_LH_RMT, /* used for return lh to caller */
326 MDT_LH_LOCAL, /* local lock never return to client */
335 /* Special magical errno for communicaiton between mdt_reint_open()
336 * and mdt_intent_reint() which means return the lock to the client
337 * for subsequent cross ref open. Previously we used plain -EREMOTE
338 * but other functions called in that path might return it too and
339 * confuse us. This is not returned to the client. See LU-5370. */
340 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
342 struct mdt_reint_record {
343 enum mds_reint_op rr_opcode;
344 const struct lustre_handle *rr_handle;
345 const struct lu_fid *rr_fid1;
346 const struct lu_fid *rr_fid2;
347 struct lu_name rr_name;
348 struct lu_name rr_tgt_name;
354 enum mdt_reint_flag {
355 MRF_OPEN_TRUNC = 1 << 0,
359 * Common data shared by mdt-level handlers. This is allocated per-thread to
360 * reduce stack consumption.
362 struct mdt_thread_info {
365 * The following members will be filled explicitly
366 * with specific data in mdt_thread_info_init().
368 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
369 * request handling may migrate from one server thread to another.
371 struct req_capsule *mti_pill;
373 /* although we have export in req, there are cases when it is not
374 * available, e.g. closing files upon export destroy */
375 struct obd_export *mti_exp;
377 * A couple of lock handles.
379 struct mdt_lock_handle mti_lh[MDT_LH_NR];
381 struct mdt_device *mti_mdt;
382 const struct lu_env *mti_env;
384 /* transaction number of current request */
390 * The following members will be filled expilictly
391 * with zero in mdt_thread_info_init(). These members may be used
398 struct md_attr mti_attr;
400 * Body for "habeo corpus" operations.
402 const struct mdt_body *mti_body;
404 * Host object. This is released at the end of mdt_handler().
406 struct mdt_object *mti_object;
408 * Lock request for "habeo clavis" operations.
410 const struct ldlm_request *mti_dlm_req;
412 __u32 mti_has_trans:1, /* has txn already? */
414 /* big_lmm buffer was used and must be used in reply */
419 /* opdata for mdt_reint_open(), has the same as
420 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
421 * value onto disk for recovery when mdt_trans_stop_cb() is called.
427 * The following members will be filled explicitly
428 * with zero in mdt_reint_unpack(), because they are only used
429 * by reint requests (including mdt_reint_open()).
433 * reint record. contains information for reint operations.
435 struct mdt_reint_record mti_rr;
437 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
439 * Operation specification (currently create and lookup)
441 struct md_op_spec mti_spec;
445 * The following members will _NOT_ be initialized at all.
446 * DO NOT expect them to contain any valid value.
447 * They should be initialized explicitly by the user themselves.
450 /* XXX: If something is in a union, make sure they do not conflict */
451 struct lu_fid mti_tmp_fid1;
452 struct lu_fid mti_tmp_fid2;
453 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
454 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
456 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
457 char ns_name[48];/* for mdt_init0() */
458 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
459 struct obd_statfs osfs; /* for mdt_statfs() */
461 /* for mdt_readpage() */
462 struct lu_rdpg mti_rdpg;
463 /* for mdt_sendpage() */
464 struct l_wait_info mti_wait_info;
474 struct lustre_handle mti_close_handle;
476 struct lu_buf mti_buf;
477 struct lu_buf mti_big_buf;
479 /* Ops object filename */
480 struct lu_name mti_name;
481 /* per-thread values, can be re-used, may be vmalloc'd */
486 /* should be enough to fit lustre_mdt_attrs */
487 char mti_xattr_buf[128];
488 struct ldlm_enqueue_info mti_einfo;
489 /* einfo used by mdt_remote_object_lock_try() */
490 struct ldlm_enqueue_info mti_remote_einfo;
491 struct tg_reply_data *mti_reply_data;
493 /* FLR: layout change API */
494 struct md_layout_change mti_layout;
497 extern struct lu_context_key mdt_thread_key;
499 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
501 struct mdt_thread_info *mti;
503 lu_env_refill((void *)env);
504 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
509 struct cdt_req_progress {
510 struct mutex crp_lock; /**< protect tree */
511 struct interval_node *crp_root; /**< tree to track extent
513 struct interval_node **crp_node; /**< buffer for tree nodes
514 * vector of fixed size
516 int crp_cnt; /**< # of used nodes */
517 int crp_max; /**< # of allocated nodes */
520 struct cdt_agent_req {
521 struct hlist_node car_cookie_hash; /**< find req by cookie */
522 struct list_head car_request_list; /**< to chain all the req. */
523 atomic_t car_refcount; /**< reference counter */
524 __u64 car_flags; /**< request original flags */
525 struct obd_uuid car_uuid; /**< agent doing the req. */
526 __u32 car_archive_id; /**< archive id */
527 int car_canceled; /**< request was canceled */
528 time64_t car_req_start; /**< start time */
529 time64_t car_req_update; /**< last update time */
530 struct hsm_action_item *car_hai; /**< req. to the agent */
531 struct cdt_req_progress car_progress; /**< track data mvt
534 extern struct kmem_cache *mdt_hsm_car_kmem;
537 struct list_head ha_list; /**< to chain the agents */
538 struct obd_uuid ha_uuid; /**< agent uuid */
539 __u32 *ha_archive_id; /**< archive id */
540 int ha_archive_cnt; /**< number of archive entries
541 * 0 means any archive */
542 atomic_t ha_requests; /**< current request count */
543 atomic_t ha_success; /**< number of successful
545 atomic_t ha_failure; /**< number of failed actions */
548 struct cdt_restore_handle {
549 struct list_head crh_list; /**< to chain the handle */
550 struct lu_fid crh_fid; /**< fid of the object */
551 struct ldlm_extent crh_extent; /**< extent of the restore */
552 struct mdt_lock_handle crh_lh; /**< lock handle */
554 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
556 struct hsm_record_update {
558 enum agent_req_status status;
561 static inline const struct md_device_operations *
562 mdt_child_ops(struct mdt_device * m)
564 LASSERT(m->mdt_child);
565 return m->mdt_child->md_ops;
568 static inline struct md_object *mdt_object_child(struct mdt_object *o)
571 return lu2md(lu_object_next(&o->mot_obj));
574 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
576 return info->mti_pill ? info->mti_pill->rc_req : NULL;
579 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
581 LASSERT(info->mti_exp);
582 return exp_connect_flags(info->mti_exp);
585 static inline void mdt_object_get(const struct lu_env *env,
586 struct mdt_object *o)
589 lu_object_get(&o->mot_obj);
593 static inline void mdt_object_put(const struct lu_env *env,
594 struct mdt_object *o)
597 lu_object_put(env, &o->mot_obj);
601 static inline int mdt_object_exists(const struct mdt_object *o)
603 return lu_object_exists(&o->mot_obj);
606 static inline int mdt_object_remote(const struct mdt_object *o)
608 return lu_object_remote(&o->mot_obj);
611 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
613 return lu_object_fid(&o->mot_obj);
616 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
618 return mdt->mdt_lu_dev.ld_site;
621 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
623 return &mdt->mdt_seq_site;
626 static inline void mdt_export_evict(struct obd_export *exp)
628 class_fail_export(exp);
631 /* Here we use LVB_TYPE to check dne client, because it is
632 * also landed on 2.4. */
633 static inline bool mdt_is_dne_client(struct obd_export *exp)
635 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
638 static inline bool mdt_is_striped_client(struct obd_export *exp)
640 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
649 /* XXX Look into layout in MDT layer. This must be done in LOD. */
650 static inline int mdt_lmm_dom_entry(struct lov_mds_md *lmm)
652 struct lov_comp_md_v1 *comp_v1;
653 struct lov_mds_md *v1;
657 if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
660 comp_v1 = (struct lov_comp_md_v1 *)lmm;
661 off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
662 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
664 /* DoM entry is the first entry always */
665 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
668 for (i = 1; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
671 off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
672 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
674 for (j = 0; j < le16_to_cpu(v1->lmm_stripe_count); j++) {
675 /* if there is any object on OST */
676 if (le32_to_cpu(v1->lmm_objects[j].l_ost_idx) !=
684 static inline bool mdt_lmm_is_flr(struct lov_mds_md *lmm)
686 struct lov_comp_md_v1 *lcm = (typeof(lcm))lmm;
688 return le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1 &&
689 le16_to_cpu(lcm->lcm_mirror_count) > 0;
692 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
693 void mdt_set_disposition(struct mdt_thread_info *info,
694 struct ldlm_reply *rep, __u64 op_flag);
695 void mdt_clear_disposition(struct mdt_thread_info *info,
696 struct ldlm_reply *rep, __u64 op_flag);
698 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
699 const struct lu_name *lname);
701 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
703 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
704 struct mdt_lock_handle *lh);
706 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
707 struct mdt_lock_handle *lhc);
709 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *mo,
710 struct mdt_lock_handle *lh, __u64 ibits);
712 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
713 struct mdt_lock_handle *lh, __u64 ibits,
716 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
717 struct mdt_lock_handle *lh, __u64 *ibits,
718 __u64 trybits, bool cos_incompat);
720 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
721 struct mdt_lock_handle *lh, int decref);
722 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
723 enum ldlm_mode mode, int decref);
725 struct mdt_object *mdt_object_new(const struct lu_env *env,
727 const struct lu_fid *);
728 struct mdt_object *mdt_object_find(const struct lu_env *,
730 const struct lu_fid *);
731 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
732 const struct lu_fid *,
733 struct mdt_lock_handle *,
735 void mdt_object_unlock_put(struct mdt_thread_info *,
737 struct mdt_lock_handle *,
740 void mdt_client_compatibility(struct mdt_thread_info *info);
742 int mdt_remote_object_lock(struct mdt_thread_info *mti,
743 struct mdt_object *o, const struct lu_fid *fid,
744 struct lustre_handle *lh,
745 enum ldlm_mode mode, __u64 ibits, bool cache);
747 enum mdt_name_flags {
751 int mdt_name_unpack(struct req_capsule *pill,
752 const struct req_msg_field *field,
754 enum mdt_name_flags flags);
755 int mdt_close_unpack(struct mdt_thread_info *info);
756 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
757 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata);
758 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
759 #ifdef CONFIG_FS_POSIX_ACL
760 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
761 struct mdt_object *o, struct lu_nodemap *nodemap);
763 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
764 const struct lu_attr *attr, const struct lu_fid *fid);
765 int mdt_pack_size2body(struct mdt_thread_info *info,
766 const struct lu_fid *fid, struct lustre_handle *lh);
767 int mdt_getxattr(struct mdt_thread_info *info);
768 int mdt_reint_setxattr(struct mdt_thread_info *info,
769 struct mdt_lock_handle *lh);
771 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
772 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
774 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
775 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
776 struct mdt_lock_handle *lhc);
778 extern void target_recovery_fini(struct obd_device *obd);
779 extern void target_recovery_init(struct lu_target *lut,
780 svc_handler_t handler);
781 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
782 struct obd_device *, struct lustre_sb_info *lsi);
783 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
785 int mdt_export_stats_init(struct obd_device *obd,
786 struct obd_export *exp,
789 int mdt_lock_new_child(struct mdt_thread_info *info,
790 struct mdt_object *o,
791 struct mdt_lock_handle *child_lockh);
793 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
796 int mdt_reint_open(struct mdt_thread_info *info,
797 struct mdt_lock_handle *lhc);
799 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
800 const struct lustre_handle *handle,
803 int mdt_get_info(struct tgt_session_info *tsi);
804 int mdt_attr_get_complex(struct mdt_thread_info *info,
805 struct mdt_object *o, struct md_attr *ma);
806 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
808 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
809 struct md_attr *ma, const char *name);
810 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
811 struct lu_fid *pfid);
812 int mdt_write_get(struct mdt_object *o);
813 void mdt_write_put(struct mdt_object *o);
814 int mdt_write_read(struct mdt_object *o);
815 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
816 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
817 void mdt_mfd_free(struct mdt_file_data *mfd);
818 int mdt_close(struct tgt_session_info *tsi);
819 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
821 int mdt_fix_reply(struct mdt_thread_info *info);
822 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
824 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
825 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
826 struct md_layout_change *spec);
828 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
829 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
830 const void *area, ssize_t len);
832 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
833 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
835 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
836 int mdt_check_ucred(struct mdt_thread_info *);
837 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
838 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *, struct mdt_body *);
839 int mdt_init_ucred_reint(struct mdt_thread_info *);
840 void mdt_exit_ucred(struct mdt_thread_info *);
841 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
842 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
843 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
845 void mdt_thread_info_init(struct ptlrpc_request *req,
846 struct mdt_thread_info *mti);
847 void mdt_thread_info_fini(struct mdt_thread_info *mti);
848 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
849 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
850 struct ldlm_lock *new_lock,
851 struct mdt_lock_handle *lh, __u64 flags);
852 int mdt_intent_lock_replace(struct mdt_thread_info *info,
853 struct ldlm_lock **lockp,
854 struct mdt_lock_handle *lh,
855 __u64 flags, int result);
857 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
858 const struct md_hsm *mh);
860 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
861 void *data, int flag);
862 int mdt_links_read(struct mdt_thread_info *info,
863 struct mdt_object *mdt_obj,
864 struct linkea_data *ldata);
865 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
866 struct mdt_body *repbody);
868 static inline struct mdt_device *mdt_dev(struct lu_device *d)
870 return container_of0(d, struct mdt_device, mdt_lu_dev);
873 static inline struct mdt_object *mdt_obj(struct lu_object *o)
875 return container_of0(o, struct mdt_object, mot_obj);
878 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
880 struct lu_object *lo;
881 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
883 lo = lu_object_locate(mo->mot_obj.lo_header,
884 mdt->mdt_bottom->dd_lu_dev.ld_type);
889 static inline bool agent_req_in_final_state(enum agent_req_status ars)
891 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
892 (ars == ARS_CANCELED));
895 /* mdt/mdt_identity.c */
896 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
898 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
900 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
902 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
904 void mdt_flush_identity(struct upcall_cache *, int);
906 __u32 mdt_identity_get_perm(struct md_identity *, lnet_nid_t);
908 /* mdt/mdt_recovery.c */
909 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
912 int mdt_hsm_state_get(struct tgt_session_info *tsi);
913 int mdt_hsm_state_set(struct tgt_session_info *tsi);
914 int mdt_hsm_action(struct tgt_session_info *tsi);
915 int mdt_hsm_progress(struct tgt_session_info *tsi);
916 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
917 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
918 int mdt_hsm_request(struct tgt_session_info *tsi);
920 /* mdt/mdt_hsm_cdt_actions.c */
921 extern const struct file_operations mdt_hsm_actions_fops;
922 void dump_llog_agent_req_rec(const char *prefix,
923 const struct llog_agent_req_rec *larr);
924 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
925 llog_cb_t cb, void *data, u32 start_cat_idx,
926 u32 start_rec_idx, int rw);
927 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
928 __u32 archive_id, __u64 flags,
929 struct hsm_action_item *hai);
930 int mdt_agent_record_update(const struct lu_env *env, struct mdt_device *mdt,
931 struct hsm_record_update *updates,
932 unsigned int updates_count);
933 void cdt_agent_record_hash_add(struct coordinator *cdt, u64 cookie, u32 cat_idt,
935 void cdt_agent_record_hash_lookup(struct coordinator *cdt, u64 cookie,
936 u32 *cat_idt, u32 *rec_idx);
937 void cdt_agent_record_hash_del(struct coordinator *cdt, u64 cookie);
939 /* mdt/mdt_hsm_cdt_agent.c */
940 extern const struct file_operations mdt_hsm_agent_fops;
941 int mdt_hsm_agent_register(struct mdt_thread_info *info,
942 const struct obd_uuid *uuid,
943 int nr_archives, __u32 *archive_num);
944 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
945 const struct obd_uuid *uuid,
947 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
948 const struct obd_uuid *uuid);
949 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
950 int succ_rq, int fail_rq, int new_rq,
951 const struct obd_uuid *uuid);
952 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
953 struct obd_uuid *uuid);
954 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
956 /* mdt/mdt_hsm_cdt_client.c */
957 int mdt_hsm_add_actions(struct mdt_thread_info *info,
958 struct hsm_action_list *hal);
959 int mdt_hsm_get_action(struct mdt_thread_info *mti,
960 const struct lu_fid *fid,
961 enum hsm_copytool_action *action,
962 enum agent_req_status *status,
963 struct hsm_extent *extent);
964 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
965 const struct lu_fid *fid);
966 /* mdt/mdt_hsm_cdt_requests.c */
967 extern struct cfs_hash_ops cdt_request_cookie_hash_ops;
968 extern struct cfs_hash_ops cdt_agent_record_hash_ops;
969 extern const struct file_operations mdt_hsm_active_requests_fops;
970 void dump_requests(char *prefix, struct coordinator *cdt);
971 struct cdt_agent_req *mdt_cdt_alloc_request(__u32 archive_id, __u64 flags,
972 struct obd_uuid *uuid,
973 struct hsm_action_item *hai);
974 void mdt_cdt_free_request(struct cdt_agent_req *car);
975 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
976 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt, u64 cookie);
977 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
978 void mdt_cdt_get_request(struct cdt_agent_req *car);
979 void mdt_cdt_put_request(struct cdt_agent_req *car);
980 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
981 const struct hsm_progress_kernel *pgs);
982 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
983 /* mdt/mdt_coordinator.c */
984 void mdt_hsm_dump_hal(int level, const char *prefix,
985 struct hsm_action_list *hal);
986 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
987 const struct lu_fid *fid,
988 const struct hsm_extent *he);
989 struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
990 const struct lu_fid *fid);
991 void cdt_restore_handle_del(struct mdt_thread_info *mti,
992 struct coordinator *cdt, const struct lu_fid *fid);
993 /* coordinator management */
994 int mdt_hsm_cdt_init(struct mdt_device *mdt);
995 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
996 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
999 * Signal the coordinator has work to do
1000 * \param cdt [IN] coordinator
1002 static inline void mdt_hsm_cdt_event(struct coordinator *cdt)
1004 cdt->cdt_event = true;
1007 /* coordinator control /proc interface */
1008 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file,
1009 const char __user *buffer,
1010 size_t count, loff_t *off);
1011 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
1012 int hsm_cdt_procfs_init(struct mdt_device *mdt);
1013 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
1014 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
1015 /* md_hsm helpers */
1016 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1017 const struct lu_fid *fid,
1018 struct md_hsm *hsm);
1019 /* actions/request helpers */
1020 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1021 struct hsm_action_list *hal, struct obd_uuid *uuid);
1022 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1023 u32 archive_id, u64 rq_flags,
1024 const struct md_hsm *hsm);
1025 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1026 struct hsm_progress_kernel *pgs);
1028 int mdt_close_swap_layouts(struct mdt_thread_info *info,
1029 struct mdt_object *o, struct md_attr *ma);
1031 extern struct lu_context_key mdt_thread_key;
1033 /* debug issues helper starts here*/
1034 static inline int mdt_fail_write(const struct lu_env *env,
1035 struct dt_device *dd, int id)
1037 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1038 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1040 return dt_ro(env, dd);
1041 /* We set FAIL_ONCE because we never "un-fail" a device */
1047 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1049 return &req->rq_export->exp_mdt_data;
1052 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
1054 return mdt_dev(exp->exp_obd->obd_lu_dev);
1057 static inline bool mdt_rdonly(struct obd_export *exp)
1059 if (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
1060 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly)
1065 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1066 struct mdt_lock_handle *lhc);
1067 static inline int mdt_check_resent(struct mdt_thread_info *info,
1068 mdt_reconstruct_t reconstruct,
1069 struct mdt_lock_handle *lhc)
1071 struct ptlrpc_request *req = mdt_info_req(info);
1075 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1076 OBD_ALLOC_PTR(info->mti_reply_data);
1077 if (info->mti_reply_data == NULL)
1080 if (req_can_reconstruct(req, info->mti_reply_data)) {
1081 reconstruct(info, lhc);
1084 DEBUG_REQ(D_HA, req,
1085 "no reply data found for RESENT req");
1088 OBD_FREE_PTR(info->mti_reply_data);
1089 info->mti_reply_data = NULL;
1094 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1095 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1097 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1099 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1102 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1104 static int mdt_dom_glimpse_ast(struct ldlm_lock *lock, void *reqp)
1106 return -ELDLM_NO_LOCK_DATA;
1109 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1110 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1111 struct lustre_handle *lh, enum ldlm_mode mode,
1112 union ldlm_policy_data *policy,
1113 const struct ldlm_res_id *res_id,
1114 __u64 flags, const __u64 *client_cookie)
1117 bool glimpse = policy->l_inodebits.bits & MDS_INODELOCK_DOM;
1119 LASSERT(ns != NULL);
1120 LASSERT(lh != NULL);
1122 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1123 mode, &flags, mdt_blocking_ast,
1124 ldlm_completion_ast,
1125 glimpse ? mdt_dom_glimpse_ast : NULL,
1126 NULL, 0, LVB_T_NONE, client_cookie, lh);
1127 return rc == ELDLM_OK ? 0 : -EIO;
1130 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1132 ldlm_lock_decref(lh, mode);
1135 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
1137 return mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL;
1140 extern mdl_mode_t mdt_mdl_lock_modes[];
1141 extern enum ldlm_mode mdt_dlm_lock_modes[];
1143 /* LCK_MINMODE which is zero returns false for is_power_of_2 */
1145 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
1147 LASSERT(mode == LCK_MINMODE || is_power_of_2(mode));
1148 return mdt_mdl_lock_modes[mode];
1151 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1153 LASSERT(mode == MDL_MINMODE || is_power_of_2(mode));
1154 return mdt_dlm_lock_modes[mode];
1158 int mdt_set_som(struct mdt_thread_info *info, struct mdt_object *obj,
1159 enum lustre_som_flags flag, __u64 size, __u64 blocks);
1160 int mdt_get_som(struct mdt_thread_info *info, struct mdt_object *obj,
1161 struct md_attr *ma);
1162 int mdt_lsom_downgrade(struct mdt_thread_info *info, struct mdt_object *obj);
1163 int mdt_lsom_update(struct mdt_thread_info *info, struct mdt_object *obj,
1167 extern struct ldlm_valblock_ops mdt_lvbo;
1168 int mdt_dom_lvb_is_valid(struct ldlm_resource *res);
1169 int mdt_dom_lvbo_update(struct ldlm_resource *res, struct ldlm_lock *lock,
1170 struct ptlrpc_request *req, bool increase_only);
1172 void mdt_enable_cos(struct mdt_device *dev, bool enable);
1173 int mdt_cos_is_enabled(struct mdt_device *);
1191 LPROC_MDT_SAMEDIR_RENAME,
1192 LPROC_MDT_CROSSDIR_RENAME,
1198 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1199 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1200 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1201 void mdt_procfs_fini(struct mdt_device *mdt);
1203 /* lustre/mdt_mdt_lproc.c */
1204 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1206 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1207 struct mdt_device *mdt,
1208 struct ptlrpc_request *req,
1209 struct mdt_object *src, struct mdt_object *tgt);
1211 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1213 return mdt->mdt_lu_dev.ld_obd;
1216 extern const struct lu_device_operations mdt_lu_ops;
1218 static inline char *mdt_obd_name(struct mdt_device *mdt)
1220 return mdt->mdt_lu_dev.ld_obd->obd_name;
1223 int mds_mod_init(void);
1224 void mds_mod_exit(void);
1226 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1228 struct obd_export *exp = req->rq_export;
1231 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1232 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1239 #define VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
1240 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
1242 int mdt_obd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
1243 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1244 struct niobuf_remote *rnb, int *nr_local,
1245 struct niobuf_local *lnb);
1247 int mdt_obd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1248 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1249 struct niobuf_remote *rnb, int npages,
1250 struct niobuf_local *lnb, int old_rc);
1251 int mdt_punch_hdl(struct tgt_session_info *tsi);
1252 int mdt_glimpse_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
1253 struct ldlm_lock **lockp, __u64 flags);
1254 int mdt_brw_enqueue(struct mdt_thread_info *info, struct ldlm_namespace *ns,
1255 struct ldlm_lock **lockp, __u64 flags);
1256 int mdt_dom_read_on_open(struct mdt_thread_info *mti, struct mdt_device *mdt,
1257 struct lustre_handle *lh);
1258 void mdt_dom_discard_data(struct mdt_thread_info *info,
1259 const struct lu_fid *fid);
1260 int mdt_dom_disk_lvbo_update(const struct lu_env *env, struct mdt_object *mo,
1261 struct ldlm_resource *res, bool increase_only);
1262 void mdt_dom_obj_lvb_update(const struct lu_env *env, struct mdt_object *mo,
1263 bool increase_only);
1264 int mdt_dom_lvb_alloc(struct ldlm_resource *res);
1266 static inline void mdt_dom_check_and_discard(struct mdt_thread_info *mti,
1267 struct mdt_object *mo)
1269 if (lu_object_is_dying(&mo->mot_header) &&
1270 S_ISREG(lu_object_attr(&mo->mot_obj)))
1271 mdt_dom_discard_data(mti, mdt_object_fid(mo));
1274 int mdt_dom_object_size(const struct lu_env *env, struct mdt_device *mdt,
1275 const struct lu_fid *fid, struct mdt_body *mb,
1277 bool mdt_dom_client_has_lock(struct mdt_thread_info *info,
1278 const struct lu_fid *fid);
1279 void mdt_hp_brw(struct tgt_session_info *tsi);
1280 void mdt_hp_punch(struct tgt_session_info *tsi);
1281 int mdt_data_version_get(struct tgt_session_info *tsi);
1284 long mdt_grant_connect(const struct lu_env *env, struct obd_export *exp,
1285 u64 want, bool conservative);
1286 extern struct kmem_cache *ldlm_glimpse_work_kmem;
1288 #endif /* _MDT_INTERNAL_H */