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
193 struct lu_device mdt_lu_dev;
194 struct seq_server_site mdt_seq_site;
195 /* DLM name-space for meta-data locks maintained by this server */
196 struct ldlm_namespace *mdt_namespace;
197 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
198 struct ptlrpc_client *mdt_ldlm_client;
199 /* underlying device */
200 struct obd_export *mdt_child_exp;
201 struct md_device *mdt_child;
202 struct dt_device *mdt_bottom;
203 struct obd_export *mdt_bottom_exp;
204 struct local_oid_storage *mdt_los;
206 struct lu_target mdt_lut;
208 * Options bit-fields.
211 unsigned int mo_user_xattr:1,
217 /* mdt state flags */
218 unsigned long mdt_state;
220 /* transaction callbacks */
221 struct dt_txn_callback mdt_txn_cb;
223 /* these values should be updated from lov if necessary.
224 * or should be placed somewhere else. */
229 /* preferred BRW size, decided by storage type and capability */
232 struct upcall_cache *mdt_identity_cache;
234 unsigned int mdt_capa_conf:1,
235 /* Enable remote dir on non-MDT0 */
236 mdt_enable_remote_dir:1,
239 gid_t mdt_enable_remote_dir_gid;
241 /* lock for osfs and md_root */
245 struct root_squash_info mdt_squash;
247 struct rename_stats mdt_rename_stats;
248 struct lu_fid mdt_md_root_fid;
250 /* connection to quota master */
251 struct obd_export *mdt_qmt_exp;
252 /* quota master device associated with this MDT */
253 struct lu_device *mdt_qmt_dev;
255 struct coordinator mdt_coordinator;
257 /* inter-MDT connection count */
258 atomic_t mdt_mds_mds_conns;
260 /* MDT device async commit count, used for debug and sanity test */
261 atomic_t mdt_async_commit_count;
263 struct mdt_object *mdt_md_root;
266 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
267 #define MDT_COS_DEFAULT (0)
269 #define ENOENT_VERSION 1 /** 'virtual' version of non-existent object */
272 struct lu_object_header mot_header;
273 struct lu_object mot_obj;
274 unsigned int mot_lov_created:1, /* lov object created */
275 mot_cache_attr:1; /* enable remote object
278 spinlock_t mot_write_lock;
279 /* Lock to protect create_data */
280 struct mutex mot_lov_mutex;
281 /* lock to protect read/write stages for Data-on-MDT files */
282 struct rw_semaphore mot_dom_sem;
283 /* Lock to protect lease open.
284 * Lease open acquires write lock; normal open acquires read lock */
285 struct rw_semaphore mot_open_sem;
286 atomic_t mot_lease_count;
287 atomic_t mot_open_count;
290 struct mdt_lock_handle {
291 /* Lock type, reg for cross-ref use or pdo lock. */
295 struct lustre_handle mlh_reg_lh;
296 enum ldlm_mode mlh_reg_mode;
299 struct lustre_handle mlh_pdo_lh;
300 enum ldlm_mode mlh_pdo_mode;
301 unsigned int mlh_pdo_hash;
303 /* Remote regular lock */
304 struct lustre_handle mlh_rreg_lh;
305 enum ldlm_mode mlh_rreg_mode;
309 MDT_LH_PARENT, /* parent lockh */
310 MDT_LH_CHILD, /* child lockh */
311 MDT_LH_OLD, /* old lockh for rename */
312 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
313 MDT_LH_NEW, /* new lockh for rename */
314 MDT_LH_RMT, /* used for return lh to caller */
315 MDT_LH_LOCAL, /* local lock never return to client */
324 /* Special magical errno for communicaiton between mdt_reint_open()
325 * and mdt_intent_reint() which means return the lock to the client
326 * for subsequent cross ref open. Previously we used plain -EREMOTE
327 * but other functions called in that path might return it too and
328 * confuse us. This is not returned to the client. See LU-5370. */
329 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
331 struct mdt_reint_record {
332 enum mds_reint_op rr_opcode;
333 const struct lustre_handle *rr_handle;
334 const struct lu_fid *rr_fid1;
335 const struct lu_fid *rr_fid2;
336 struct lu_name rr_name;
337 struct lu_name rr_tgt_name;
343 enum mdt_reint_flag {
344 MRF_OPEN_TRUNC = 1 << 0,
348 * Common data shared by mdt-level handlers. This is allocated per-thread to
349 * reduce stack consumption.
351 struct mdt_thread_info {
354 * The following members will be filled explicitly
355 * with specific data in mdt_thread_info_init().
357 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
358 * request handling may migrate from one server thread to another.
360 struct req_capsule *mti_pill;
362 /* although we have export in req, there are cases when it is not
363 * available, e.g. closing files upon export destroy */
364 struct obd_export *mti_exp;
366 * A couple of lock handles.
368 struct mdt_lock_handle mti_lh[MDT_LH_NR];
370 struct mdt_device *mti_mdt;
371 const struct lu_env *mti_env;
373 /* transaction number of current request */
379 * The following members will be filled expilictly
380 * with zero in mdt_thread_info_init(). These members may be used
387 struct md_attr mti_attr;
389 * Body for "habeo corpus" operations.
391 const struct mdt_body *mti_body;
393 * Host object. This is released at the end of mdt_handler().
395 struct mdt_object *mti_object;
397 * Lock request for "habeo clavis" operations.
399 const struct ldlm_request *mti_dlm_req;
401 __u32 mti_has_trans:1, /* has txn already? */
403 /* big_lmm buffer was used and must be used in reply */
408 /* opdata for mdt_reint_open(), has the same as
409 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
410 * value onto disk for recovery when mdt_trans_stop_cb() is called.
416 * The following members will be filled explicitly
417 * with zero in mdt_reint_unpack(), because they are only used
418 * by reint requests (including mdt_reint_open()).
422 * reint record. contains information for reint operations.
424 struct mdt_reint_record mti_rr;
426 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
428 * Operation specification (currently create and lookup)
430 struct md_op_spec mti_spec;
434 * The following members will _NOT_ be initialized at all.
435 * DO NOT expect them to contain any valid value.
436 * They should be initialized explicitly by the user themselves.
439 /* XXX: If something is in a union, make sure they do not conflict */
440 struct lu_fid mti_tmp_fid1;
441 struct lu_fid mti_tmp_fid2;
442 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
443 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
445 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
446 char ns_name[48];/* for mdt_init0() */
447 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
448 struct obd_statfs osfs; /* for mdt_statfs() */
450 /* for mdt_readpage() */
451 struct lu_rdpg mti_rdpg;
452 /* for mdt_sendpage() */
453 struct l_wait_info mti_wait_info;
460 struct lustre_handle mti_close_handle;
462 struct lu_buf mti_buf;
463 struct lu_buf mti_big_buf;
465 /* Ops object filename */
466 struct lu_name mti_name;
467 /* per-thread values, can be re-used, may be vmalloc'd */
472 /* should be enough to fit lustre_mdt_attrs */
473 char mti_xattr_buf[128];
474 struct ldlm_enqueue_info mti_einfo;
475 struct tg_reply_data *mti_reply_data;
477 struct lustre_som_attrs mti_som;
479 /* FLR: layout change API */
480 struct md_layout_change mti_layout;
483 extern struct lu_context_key mdt_thread_key;
485 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
487 struct mdt_thread_info *mti;
489 lu_env_refill((void *)env);
490 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
495 struct cdt_req_progress {
496 struct mutex crp_lock; /**< protect tree */
497 struct interval_node *crp_root; /**< tree to track extent
499 struct interval_node **crp_node; /**< buffer for tree nodes
500 * vector of fixed size
502 int crp_cnt; /**< # of used nodes */
503 int crp_max; /**< # of allocated nodes */
506 struct cdt_agent_req {
507 struct hlist_node car_cookie_hash; /**< find req by cookie */
508 struct list_head car_request_list; /**< to chain all the req. */
509 atomic_t car_refcount; /**< reference counter */
510 __u64 car_flags; /**< request original flags */
511 struct obd_uuid car_uuid; /**< agent doing the req. */
512 __u32 car_archive_id; /**< archive id */
513 int car_canceled; /**< request was canceled */
514 time64_t car_req_start; /**< start time */
515 time64_t car_req_update; /**< last update time */
516 struct hsm_action_item *car_hai; /**< req. to the agent */
517 struct cdt_req_progress car_progress; /**< track data mvt
520 extern struct kmem_cache *mdt_hsm_car_kmem;
523 struct list_head ha_list; /**< to chain the agents */
524 struct obd_uuid ha_uuid; /**< agent uuid */
525 __u32 *ha_archive_id; /**< archive id */
526 int ha_archive_cnt; /**< number of archive entries
527 * 0 means any archive */
528 atomic_t ha_requests; /**< current request count */
529 atomic_t ha_success; /**< number of successful
531 atomic_t ha_failure; /**< number of failed actions */
534 struct cdt_restore_handle {
535 struct list_head crh_list; /**< to chain the handle */
536 struct lu_fid crh_fid; /**< fid of the object */
537 struct ldlm_extent crh_extent; /**< extent of the restore */
538 struct mdt_lock_handle crh_lh; /**< lock handle */
540 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
542 struct hsm_record_update {
544 enum agent_req_status status;
547 static inline const struct md_device_operations *
548 mdt_child_ops(struct mdt_device * m)
550 LASSERT(m->mdt_child);
551 return m->mdt_child->md_ops;
554 static inline struct md_object *mdt_object_child(struct mdt_object *o)
557 return lu2md(lu_object_next(&o->mot_obj));
560 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
562 return info->mti_pill ? info->mti_pill->rc_req : NULL;
565 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
567 LASSERT(info->mti_exp);
568 return exp_connect_flags(info->mti_exp);
571 static inline void mdt_object_get(const struct lu_env *env,
572 struct mdt_object *o)
575 lu_object_get(&o->mot_obj);
579 static inline void mdt_object_put(const struct lu_env *env,
580 struct mdt_object *o)
583 lu_object_put(env, &o->mot_obj);
587 static inline int mdt_object_exists(const struct mdt_object *o)
589 return lu_object_exists(&o->mot_obj);
592 static inline int mdt_object_remote(const struct mdt_object *o)
594 return lu_object_remote(&o->mot_obj);
597 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
599 return lu_object_fid(&o->mot_obj);
602 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
604 return mdt->mdt_lu_dev.ld_site;
607 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
609 return &mdt->mdt_seq_site;
612 static inline void mdt_export_evict(struct obd_export *exp)
614 class_fail_export(exp);
617 /* Here we use LVB_TYPE to check dne client, because it is
618 * also landed on 2.4. */
619 static inline bool mdt_is_dne_client(struct obd_export *exp)
621 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
624 static inline bool mdt_is_striped_client(struct obd_export *exp)
626 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
635 /* XXX Look into layout in MDT layer. This must be done in LOD. */
636 static inline int mdt_lmm_dom_entry(struct lov_mds_md *lmm)
638 struct lov_comp_md_v1 *comp_v1;
639 struct lov_mds_md *v1;
643 if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
646 comp_v1 = (struct lov_comp_md_v1 *)lmm;
647 off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
648 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
650 /* DoM entry is the first entry always */
651 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
654 for (i = 1; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
657 off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
658 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
660 for (j = 0; j < le16_to_cpu(v1->lmm_stripe_count); j++) {
661 /* if there is any object on OST */
662 if (le32_to_cpu(v1->lmm_objects[j].l_ost_idx) !=
670 static inline bool mdt_lmm_is_flr(struct lov_mds_md *lmm)
672 struct lov_comp_md_v1 *lcm = (typeof(lcm))lmm;
674 return le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1 &&
675 le16_to_cpu(lcm->lcm_mirror_count) > 0;
678 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
679 void mdt_set_disposition(struct mdt_thread_info *info,
680 struct ldlm_reply *rep, __u64 op_flag);
681 void mdt_clear_disposition(struct mdt_thread_info *info,
682 struct ldlm_reply *rep, __u64 op_flag);
684 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
685 const struct lu_name *lname);
687 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
689 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
690 struct mdt_lock_handle *lh);
692 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
693 struct mdt_lock_handle *lhc);
695 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *mo,
696 struct mdt_lock_handle *lh, __u64 ibits);
698 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
699 struct mdt_lock_handle *lh, __u64 ibits,
702 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
703 struct mdt_lock_handle *lh, __u64 *ibits,
704 __u64 trybits, bool cos_incompat);
706 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
707 struct mdt_lock_handle *lh, int decref);
708 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
709 enum ldlm_mode mode, int decref);
711 struct mdt_object *mdt_object_new(const struct lu_env *env,
713 const struct lu_fid *);
714 struct mdt_object *mdt_object_find(const struct lu_env *,
716 const struct lu_fid *);
717 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
718 const struct lu_fid *,
719 struct mdt_lock_handle *,
721 void mdt_object_unlock_put(struct mdt_thread_info *,
723 struct mdt_lock_handle *,
726 void mdt_client_compatibility(struct mdt_thread_info *info);
728 int mdt_remote_object_lock(struct mdt_thread_info *mti,
729 struct mdt_object *o, const struct lu_fid *fid,
730 struct lustre_handle *lh,
731 enum ldlm_mode mode, __u64 ibits, bool cache);
733 enum mdt_name_flags {
737 int mdt_name_unpack(struct req_capsule *pill,
738 const struct req_msg_field *field,
740 enum mdt_name_flags flags);
741 int mdt_close_unpack(struct mdt_thread_info *info);
742 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
743 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata);
744 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
745 #ifdef CONFIG_FS_POSIX_ACL
746 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
747 struct mdt_object *o, struct lu_nodemap *nodemap);
749 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
750 const struct lu_attr *attr, const struct lu_fid *fid);
751 int mdt_pack_size2body(struct mdt_thread_info *info,
752 const struct lu_fid *fid, bool dom_lock);
753 int mdt_getxattr(struct mdt_thread_info *info);
754 int mdt_reint_setxattr(struct mdt_thread_info *info,
755 struct mdt_lock_handle *lh);
757 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
758 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
760 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
761 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
762 struct mdt_lock_handle *lhc);
764 extern void target_recovery_fini(struct obd_device *obd);
765 extern void target_recovery_init(struct lu_target *lut,
766 svc_handler_t handler);
767 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
768 struct obd_device *, struct lustre_sb_info *lsi);
769 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
771 int mdt_export_stats_init(struct obd_device *obd,
772 struct obd_export *exp,
775 int mdt_lock_new_child(struct mdt_thread_info *info,
776 struct mdt_object *o,
777 struct mdt_lock_handle *child_lockh);
779 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
782 int mdt_reint_open(struct mdt_thread_info *info,
783 struct mdt_lock_handle *lhc);
785 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
786 const struct lustre_handle *handle,
789 int mdt_get_info(struct tgt_session_info *tsi);
790 int mdt_attr_get_complex(struct mdt_thread_info *info,
791 struct mdt_object *o, struct md_attr *ma);
792 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
794 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
795 struct md_attr *ma, const char *name);
796 int mdt_write_get(struct mdt_object *o);
797 void mdt_write_put(struct mdt_object *o);
798 int mdt_write_read(struct mdt_object *o);
799 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
800 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
801 void mdt_mfd_free(struct mdt_file_data *mfd);
802 int mdt_close(struct tgt_session_info *tsi);
803 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
805 int mdt_fix_reply(struct mdt_thread_info *info);
806 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
808 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
809 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
810 struct md_layout_change *spec);
812 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
813 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
814 const void *area, ssize_t len);
816 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
817 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
819 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
820 int mdt_check_ucred(struct mdt_thread_info *);
821 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
822 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *, struct mdt_body *);
823 int mdt_init_ucred_reint(struct mdt_thread_info *);
824 void mdt_exit_ucred(struct mdt_thread_info *);
825 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
826 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
827 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
829 void mdt_thread_info_init(struct ptlrpc_request *req,
830 struct mdt_thread_info *mti);
831 void mdt_thread_info_fini(struct mdt_thread_info *mti);
832 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
833 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
834 struct ldlm_lock *new_lock,
835 struct mdt_lock_handle *lh, __u64 flags);
836 int mdt_intent_lock_replace(struct mdt_thread_info *info,
837 struct ldlm_lock **lockp,
838 struct mdt_lock_handle *lh,
839 __u64 flags, int result);
841 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
842 const struct md_hsm *mh);
844 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
845 void *data, int flag);
846 int mdt_links_read(struct mdt_thread_info *info,
847 struct mdt_object *mdt_obj,
848 struct linkea_data *ldata);
849 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
850 struct mdt_body *repbody);
852 static inline struct mdt_device *mdt_dev(struct lu_device *d)
854 return container_of0(d, struct mdt_device, mdt_lu_dev);
857 static inline struct mdt_object *mdt_obj(struct lu_object *o)
859 return container_of0(o, struct mdt_object, mot_obj);
862 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
864 struct lu_object *lo;
865 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
867 lo = lu_object_locate(mo->mot_obj.lo_header,
868 mdt->mdt_bottom->dd_lu_dev.ld_type);
873 static inline bool agent_req_in_final_state(enum agent_req_status ars)
875 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
876 (ars == ARS_CANCELED));
879 /* mdt/mdt_identity.c */
880 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
882 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
884 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
886 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
888 void mdt_flush_identity(struct upcall_cache *, int);
890 __u32 mdt_identity_get_perm(struct md_identity *, lnet_nid_t);
892 /* mdt/mdt_recovery.c */
893 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
896 int mdt_hsm_state_get(struct tgt_session_info *tsi);
897 int mdt_hsm_state_set(struct tgt_session_info *tsi);
898 int mdt_hsm_action(struct tgt_session_info *tsi);
899 int mdt_hsm_progress(struct tgt_session_info *tsi);
900 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
901 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
902 int mdt_hsm_request(struct tgt_session_info *tsi);
904 /* mdt/mdt_hsm_cdt_actions.c */
905 extern const struct file_operations mdt_hsm_actions_fops;
906 void dump_llog_agent_req_rec(const char *prefix,
907 const struct llog_agent_req_rec *larr);
908 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
909 llog_cb_t cb, void *data, u32 start_cat_idx,
910 u32 start_rec_idx, int rw);
911 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
912 __u32 archive_id, __u64 flags,
913 struct hsm_action_item *hai);
914 int mdt_agent_record_update(const struct lu_env *env, struct mdt_device *mdt,
915 struct hsm_record_update *updates,
916 unsigned int updates_count);
917 void cdt_agent_record_hash_add(struct coordinator *cdt, u64 cookie, u32 cat_idt,
919 void cdt_agent_record_hash_lookup(struct coordinator *cdt, u64 cookie,
920 u32 *cat_idt, u32 *rec_idx);
921 void cdt_agent_record_hash_del(struct coordinator *cdt, u64 cookie);
923 /* mdt/mdt_hsm_cdt_agent.c */
924 extern const struct file_operations mdt_hsm_agent_fops;
925 int mdt_hsm_agent_register(struct mdt_thread_info *info,
926 const struct obd_uuid *uuid,
927 int nr_archives, __u32 *archive_num);
928 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
929 const struct obd_uuid *uuid,
931 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
932 const struct obd_uuid *uuid);
933 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
934 int succ_rq, int fail_rq, int new_rq,
935 const struct obd_uuid *uuid);
936 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
937 struct obd_uuid *uuid);
938 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
940 /* mdt/mdt_hsm_cdt_client.c */
941 int mdt_hsm_add_actions(struct mdt_thread_info *info,
942 struct hsm_action_list *hal);
943 int mdt_hsm_get_action(struct mdt_thread_info *mti,
944 const struct lu_fid *fid,
945 enum hsm_copytool_action *action,
946 enum agent_req_status *status,
947 struct hsm_extent *extent);
948 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
949 const struct lu_fid *fid);
950 /* mdt/mdt_hsm_cdt_requests.c */
951 extern struct cfs_hash_ops cdt_request_cookie_hash_ops;
952 extern struct cfs_hash_ops cdt_agent_record_hash_ops;
953 extern const struct file_operations mdt_hsm_active_requests_fops;
954 void dump_requests(char *prefix, struct coordinator *cdt);
955 struct cdt_agent_req *mdt_cdt_alloc_request(__u32 archive_id, __u64 flags,
956 struct obd_uuid *uuid,
957 struct hsm_action_item *hai);
958 void mdt_cdt_free_request(struct cdt_agent_req *car);
959 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
960 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt, u64 cookie);
961 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
962 void mdt_cdt_get_request(struct cdt_agent_req *car);
963 void mdt_cdt_put_request(struct cdt_agent_req *car);
964 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
965 const struct hsm_progress_kernel *pgs);
966 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
967 /* mdt/mdt_coordinator.c */
968 void mdt_hsm_dump_hal(int level, const char *prefix,
969 struct hsm_action_list *hal);
970 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
971 const struct lu_fid *fid,
972 const struct hsm_extent *he);
973 struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
974 const struct lu_fid *fid);
975 void cdt_restore_handle_del(struct mdt_thread_info *mti,
976 struct coordinator *cdt, const struct lu_fid *fid);
977 /* coordinator management */
978 int mdt_hsm_cdt_init(struct mdt_device *mdt);
979 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
980 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
983 * Signal the coordinator has work to do
984 * \param cdt [IN] coordinator
986 static inline void mdt_hsm_cdt_event(struct coordinator *cdt)
988 cdt->cdt_event = true;
991 /* coordinator control /proc interface */
992 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file,
993 const char __user *buffer,
994 size_t count, loff_t *off);
995 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
996 int hsm_cdt_procfs_init(struct mdt_device *mdt);
997 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
998 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
1000 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1001 const struct lu_fid *fid,
1002 struct md_hsm *hsm);
1003 /* actions/request helpers */
1004 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1005 struct hsm_action_list *hal, struct obd_uuid *uuid);
1006 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1007 u32 archive_id, u64 rq_flags,
1008 const struct md_hsm *hsm);
1009 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1010 struct hsm_progress_kernel *pgs);
1012 int mdt_close_swap_layouts(struct mdt_thread_info *info,
1013 struct mdt_object *o, struct md_attr *ma);
1015 extern struct lu_context_key mdt_thread_key;
1017 /* debug issues helper starts here*/
1018 static inline int mdt_fail_write(const struct lu_env *env,
1019 struct dt_device *dd, int id)
1021 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1022 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1024 return dt_ro(env, dd);
1025 /* We set FAIL_ONCE because we never "un-fail" a device */
1031 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1033 return &req->rq_export->exp_mdt_data;
1036 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
1038 return mdt_dev(exp->exp_obd->obd_lu_dev);
1041 static inline bool mdt_rdonly(struct obd_export *exp)
1043 if (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
1044 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly)
1049 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1050 struct mdt_lock_handle *lhc);
1051 static inline int mdt_check_resent(struct mdt_thread_info *info,
1052 mdt_reconstruct_t reconstruct,
1053 struct mdt_lock_handle *lhc)
1055 struct ptlrpc_request *req = mdt_info_req(info);
1059 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1060 OBD_ALLOC_PTR(info->mti_reply_data);
1061 if (info->mti_reply_data == NULL)
1064 if (req_can_reconstruct(req, info->mti_reply_data)) {
1065 reconstruct(info, lhc);
1068 DEBUG_REQ(D_HA, req,
1069 "no reply data found for RESENT req");
1072 OBD_FREE_PTR(info->mti_reply_data);
1073 info->mti_reply_data = NULL;
1078 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1079 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1081 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1083 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1086 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1088 static int mdt_dom_glimpse_ast(struct ldlm_lock *lock, void *reqp)
1090 return -ELDLM_NO_LOCK_DATA;
1093 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1094 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1095 struct lustre_handle *lh, enum ldlm_mode mode,
1096 union ldlm_policy_data *policy,
1097 const struct ldlm_res_id *res_id,
1098 __u64 flags, const __u64 *client_cookie)
1101 bool glimpse = policy->l_inodebits.bits & MDS_INODELOCK_DOM;
1103 LASSERT(ns != NULL);
1104 LASSERT(lh != NULL);
1106 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1107 mode, &flags, mdt_blocking_ast,
1108 ldlm_completion_ast,
1109 glimpse ? mdt_dom_glimpse_ast : NULL,
1110 NULL, 0, LVB_T_NONE, client_cookie, lh);
1111 return rc == ELDLM_OK ? 0 : -EIO;
1114 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1116 ldlm_lock_decref(lh, mode);
1119 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
1121 return mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL;
1124 extern mdl_mode_t mdt_mdl_lock_modes[];
1125 extern enum ldlm_mode mdt_dlm_lock_modes[];
1127 /* LCK_MINMODE which is zero returns false for is_power_of_2 */
1129 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
1131 LASSERT(mode == LCK_MINMODE || is_power_of_2(mode));
1132 return mdt_mdl_lock_modes[mode];
1135 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1137 LASSERT(mode == MDL_MINMODE || is_power_of_2(mode));
1138 return mdt_dlm_lock_modes[mode];
1142 int mdt_set_som(struct mdt_thread_info *info, struct mdt_object *obj,
1143 struct lu_attr *attr);
1144 int mdt_get_som(struct mdt_thread_info *info, struct mdt_object *obj,
1145 struct lu_attr *attr);
1148 extern struct ldlm_valblock_ops mdt_lvbo;
1149 int mdt_dom_lvb_is_valid(struct ldlm_resource *res);
1150 int mdt_dom_lvbo_update(struct ldlm_resource *res, struct ldlm_lock *lock,
1151 struct ptlrpc_request *req, bool increase_only);
1153 void mdt_enable_cos(struct mdt_device *dev, bool enable);
1154 int mdt_cos_is_enabled(struct mdt_device *);
1172 LPROC_MDT_SAMEDIR_RENAME,
1173 LPROC_MDT_CROSSDIR_RENAME,
1179 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1180 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1181 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1182 void mdt_procfs_fini(struct mdt_device *mdt);
1184 /* lustre/mdt_mdt_lproc.c */
1185 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1187 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1188 struct mdt_device *mdt,
1189 struct ptlrpc_request *req,
1190 struct mdt_object *src, struct mdt_object *tgt);
1192 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1194 return mdt->mdt_lu_dev.ld_obd;
1197 extern const struct lu_device_operations mdt_lu_ops;
1199 static inline char *mdt_obd_name(struct mdt_device *mdt)
1201 return mdt->mdt_lu_dev.ld_obd->obd_name;
1204 int mds_mod_init(void);
1205 void mds_mod_exit(void);
1207 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1209 struct obd_export *exp = req->rq_export;
1212 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1213 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1220 #define VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
1221 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
1223 int mdt_obd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
1224 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1225 struct niobuf_remote *rnb, int *nr_local,
1226 struct niobuf_local *lnb);
1228 int mdt_obd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1229 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1230 struct niobuf_remote *rnb, int npages,
1231 struct niobuf_local *lnb, int old_rc);
1232 int mdt_punch_hdl(struct tgt_session_info *tsi);
1233 int mdt_glimpse_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
1234 struct ldlm_lock **lockp, __u64 flags);
1235 int mdt_brw_enqueue(struct mdt_thread_info *info, struct ldlm_namespace *ns,
1236 struct ldlm_lock **lockp, __u64 flags);
1237 void mdt_dom_discard_data(struct mdt_thread_info *info,
1238 const struct lu_fid *fid);
1239 int mdt_dom_disk_lvbo_update(const struct lu_env *env, struct mdt_object *mo,
1240 struct ldlm_resource *res, bool increase_only);
1241 void mdt_dom_obj_lvb_update(const struct lu_env *env, struct mdt_object *mo,
1242 bool increase_only);
1243 int mdt_dom_lvb_alloc(struct ldlm_resource *res);
1245 static inline void mdt_dom_check_and_discard(struct mdt_thread_info *mti,
1246 struct mdt_object *mo)
1248 if (lu_object_is_dying(&mo->mot_header) &&
1249 S_ISREG(lu_object_attr(&mo->mot_obj)))
1250 mdt_dom_discard_data(mti, mdt_object_fid(mo));
1253 int mdt_dom_object_size(const struct lu_env *env, struct mdt_device *mdt,
1254 const struct lu_fid *fid, struct mdt_body *mb,
1256 bool mdt_dom_client_has_lock(struct mdt_thread_info *info,
1257 const struct lu_fid *fid);
1258 void mdt_hp_brw(struct tgt_session_info *tsi);
1259 void mdt_hp_punch(struct tgt_session_info *tsi);
1260 int mdt_data_version_get(struct tgt_session_info *tsi);
1263 long mdt_grant_connect(const struct lu_env *env, struct obd_export *exp,
1264 u64 want, bool conservative);
1265 extern struct kmem_cache *ldlm_glimpse_work_kmem;
1267 #endif /* _MDT_INTERNAL_H */