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 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 time64_t cdt_loop_period; /**< llog scan period */
151 time64_t cdt_grace_delay; /**< request grace
153 time64_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 /** Current count of active requests */
160 atomic_t cdt_request_count; /** total */
161 atomic_t cdt_archive_count;
162 atomic_t cdt_restore_count;
163 atomic_t cdt_remove_count;
165 /* started requests (struct cdt_agent_req:car_cookie_hash)
166 * indexed by cookie */
167 struct cfs_hash *cdt_request_cookie_hash;
168 /* started requests (struct cdt_agent_req:car_request_list) */
169 struct list_head cdt_request_list;
170 struct list_head cdt_agents; /**< list of register
172 struct list_head cdt_restore_handle_list;
174 /* Hash of cookies to locations of record locations in agent
176 struct cfs_hash *cdt_agent_record_hash;
178 /* Bitmasks indexed by the HSMA_XXX constants. */
179 __u64 cdt_user_request_mask;
180 __u64 cdt_group_request_mask;
181 __u64 cdt_other_request_mask;
183 /* Remove archive on last unlink policy */
184 bool cdt_remove_archive_on_last_unlink;
186 bool cdt_wakeup_coordinator;
189 /* mdt state flag bits */
190 #define MDT_FL_CFGLOG 0
191 #define MDT_FL_SYNCED 1
195 struct lu_device mdt_lu_dev;
196 struct seq_server_site mdt_seq_site;
197 /* DLM name-space for meta-data locks maintained by this server */
198 struct ldlm_namespace *mdt_namespace;
199 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
200 struct ptlrpc_client *mdt_ldlm_client;
201 /* underlying device */
202 struct obd_export *mdt_child_exp;
203 struct md_device *mdt_child;
204 struct dt_device *mdt_bottom;
205 struct obd_export *mdt_bottom_exp;
207 struct lu_target mdt_lut;
209 * Options bit-fields.
212 unsigned int mo_user_xattr:1,
218 /* mdt state flags */
219 unsigned long mdt_state;
221 /* transaction callbacks */
222 struct dt_txn_callback mdt_txn_cb;
224 /* these values should be updated from lov if necessary.
225 * or should be placed somewhere else. */
230 /* preferred BRW size, decided by storage type and capability */
233 struct upcall_cache *mdt_identity_cache;
235 unsigned int mdt_capa_conf:1,
236 /* Enable remote dir on non-MDT0 */
237 mdt_enable_remote_dir:1,
240 gid_t mdt_enable_remote_dir_gid;
242 /* lock for osfs and md_root */
246 struct root_squash_info mdt_squash;
248 struct rename_stats mdt_rename_stats;
249 struct lu_fid mdt_md_root_fid;
251 /* connection to quota master */
252 struct obd_export *mdt_qmt_exp;
253 /* quota master device associated with this MDT */
254 struct lu_device *mdt_qmt_dev;
256 struct coordinator mdt_coordinator;
258 /* inter-MDT connection count */
259 atomic_t mdt_mds_mds_conns;
261 /* MDT device async commit count, used for debug and sanity test */
262 atomic_t mdt_async_commit_count;
264 struct mdt_object *mdt_md_root;
267 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
268 #define MDT_COS_DEFAULT (0)
270 #define ENOENT_VERSION 1 /** 'virtual' version of non-existent object */
273 struct lu_object_header mot_header;
274 struct lu_object mot_obj;
275 unsigned int mot_lov_created:1, /* lov object created */
276 mot_cache_attr:1; /* enable remote object
279 spinlock_t mot_write_lock;
280 /* Lock to protect create_data */
281 struct mutex mot_lov_mutex;
282 /* lock to protect read/write stages for Data-on-MDT files */
283 struct rw_semaphore mot_dom_sem;
284 /* Lock to protect lease open.
285 * Lease open acquires write lock; normal open acquires read lock */
286 struct rw_semaphore mot_open_sem;
287 atomic_t mot_lease_count;
288 atomic_t mot_open_count;
291 struct mdt_lock_handle {
292 /* Lock type, reg for cross-ref use or pdo lock. */
296 struct lustre_handle mlh_reg_lh;
297 enum ldlm_mode mlh_reg_mode;
300 struct lustre_handle mlh_pdo_lh;
301 enum ldlm_mode mlh_pdo_mode;
302 unsigned int mlh_pdo_hash;
304 /* Remote regular lock */
305 struct lustre_handle mlh_rreg_lh;
306 enum ldlm_mode mlh_rreg_mode;
310 MDT_LH_PARENT, /* parent lockh */
311 MDT_LH_CHILD, /* child lockh */
312 MDT_LH_OLD, /* old lockh for rename */
313 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
314 MDT_LH_NEW, /* new lockh for rename */
315 MDT_LH_RMT, /* used for return lh to caller */
316 MDT_LH_LOCAL, /* local lock never return to client */
325 /* Special magical errno for communicaiton between mdt_reint_open()
326 * and mdt_intent_reint() which means return the lock to the client
327 * for subsequent cross ref open. Previously we used plain -EREMOTE
328 * but other functions called in that path might return it too and
329 * confuse us. This is not returned to the client. See LU-5370. */
330 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
332 struct mdt_reint_record {
333 enum mds_reint_op rr_opcode;
334 const struct lustre_handle *rr_handle;
335 const struct lu_fid *rr_fid1;
336 const struct lu_fid *rr_fid2;
337 struct lu_name rr_name;
338 struct lu_name rr_tgt_name;
344 enum mdt_reint_flag {
345 MRF_OPEN_TRUNC = 1 << 0,
349 * Common data shared by mdt-level handlers. This is allocated per-thread to
350 * reduce stack consumption.
352 struct mdt_thread_info {
355 * The following members will be filled explicitly
356 * with specific data in mdt_thread_info_init().
358 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
359 * request handling may migrate from one server thread to another.
361 struct req_capsule *mti_pill;
363 /* although we have export in req, there are cases when it is not
364 * available, e.g. closing files upon export destroy */
365 struct obd_export *mti_exp;
367 * A couple of lock handles.
369 struct mdt_lock_handle mti_lh[MDT_LH_NR];
371 struct mdt_device *mti_mdt;
372 const struct lu_env *mti_env;
374 /* transaction number of current request */
380 * The following members will be filled expilictly
381 * with zero in mdt_thread_info_init(). These members may be used
388 struct md_attr mti_attr;
390 * Body for "habeo corpus" operations.
392 const struct mdt_body *mti_body;
394 * Host object. This is released at the end of mdt_handler().
396 struct mdt_object *mti_object;
398 * Lock request for "habeo clavis" operations.
400 const struct ldlm_request *mti_dlm_req;
402 __u32 mti_has_trans:1, /* has txn already? */
404 /* big_lmm buffer was used and must be used in reply */
409 /* opdata for mdt_reint_open(), has the same as
410 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
411 * value onto disk for recovery when mdt_trans_stop_cb() is called.
417 * The following members will be filled explicitly
418 * with zero in mdt_reint_unpack(), because they are only used
419 * by reint requests (including mdt_reint_open()).
423 * reint record. contains information for reint operations.
425 struct mdt_reint_record mti_rr;
427 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
429 * Operation specification (currently create and lookup)
431 struct md_op_spec mti_spec;
435 * The following members will _NOT_ be initialized at all.
436 * DO NOT expect them to contain any valid value.
437 * They should be initialized explicitly by the user themselves.
440 /* XXX: If something is in a union, make sure they do not conflict */
441 struct lu_fid mti_tmp_fid1;
442 struct lu_fid mti_tmp_fid2;
443 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
444 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
446 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
447 char ns_name[48];/* for mdt_init0() */
448 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
449 struct obd_statfs osfs; /* for mdt_statfs() */
451 /* for mdt_readpage() */
452 struct lu_rdpg mti_rdpg;
453 /* for mdt_sendpage() */
454 struct l_wait_info mti_wait_info;
461 struct lustre_handle mti_close_handle;
463 struct lu_buf mti_buf;
464 struct lu_buf mti_big_buf;
466 /* Ops object filename */
467 struct lu_name mti_name;
468 /* per-thread values, can be re-used, may be vmalloc'd */
473 /* should be enough to fit lustre_mdt_attrs */
474 char mti_xattr_buf[128];
475 struct ldlm_enqueue_info mti_einfo;
476 struct tg_reply_data *mti_reply_data;
478 struct lustre_som_attrs mti_som;
480 /* FLR: layout change API */
481 struct md_layout_change mti_layout;
484 extern struct lu_context_key mdt_thread_key;
486 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
488 struct mdt_thread_info *mti;
490 lu_env_refill((void *)env);
491 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
496 struct cdt_req_progress {
497 struct mutex crp_lock; /**< protect tree */
498 struct interval_node *crp_root; /**< tree to track extent
500 struct interval_node **crp_node; /**< buffer for tree nodes
501 * vector of fixed size
503 int crp_cnt; /**< # of used nodes */
504 int crp_max; /**< # of allocated nodes */
507 struct cdt_agent_req {
508 struct hlist_node car_cookie_hash; /**< find req by cookie */
509 struct list_head car_request_list; /**< to chain all the req. */
510 atomic_t car_refcount; /**< reference counter */
511 __u64 car_compound_id; /**< compound id */
512 __u64 car_flags; /**< request original flags */
513 struct obd_uuid car_uuid; /**< agent doing the req. */
514 __u32 car_archive_id; /**< archive id */
515 int car_canceled; /**< request was canceled */
516 time64_t car_req_start; /**< start time */
517 time64_t car_req_update; /**< last update time */
518 struct hsm_action_item *car_hai; /**< req. to the agent */
519 struct cdt_req_progress car_progress; /**< track data mvt
522 extern struct kmem_cache *mdt_hsm_car_kmem;
525 struct list_head ha_list; /**< to chain the agents */
526 struct obd_uuid ha_uuid; /**< agent uuid */
527 __u32 *ha_archive_id; /**< archive id */
528 int ha_archive_cnt; /**< number of archive entries
529 * 0 means any archive */
530 atomic_t ha_requests; /**< current request count */
531 atomic_t ha_success; /**< number of successful
533 atomic_t ha_failure; /**< number of failed actions */
536 struct cdt_restore_handle {
537 struct list_head crh_list; /**< to chain the handle */
538 struct lu_fid crh_fid; /**< fid of the object */
539 struct ldlm_extent crh_extent; /**< extent of the restore */
540 struct mdt_lock_handle crh_lh; /**< lock handle */
542 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
544 struct hsm_record_update {
546 enum agent_req_status status;
549 static inline const struct md_device_operations *
550 mdt_child_ops(struct mdt_device * m)
552 LASSERT(m->mdt_child);
553 return m->mdt_child->md_ops;
556 static inline struct md_object *mdt_object_child(struct mdt_object *o)
559 return lu2md(lu_object_next(&o->mot_obj));
562 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
564 return info->mti_pill ? info->mti_pill->rc_req : NULL;
567 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
569 LASSERT(info->mti_exp);
570 return exp_connect_flags(info->mti_exp);
573 static inline void mdt_object_get(const struct lu_env *env,
574 struct mdt_object *o)
577 lu_object_get(&o->mot_obj);
581 static inline void mdt_object_put(const struct lu_env *env,
582 struct mdt_object *o)
585 lu_object_put(env, &o->mot_obj);
589 static inline int mdt_object_exists(const struct mdt_object *o)
591 return lu_object_exists(&o->mot_obj);
594 static inline int mdt_object_remote(const struct mdt_object *o)
596 return lu_object_remote(&o->mot_obj);
599 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
601 return lu_object_fid(&o->mot_obj);
604 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
606 return mdt->mdt_lu_dev.ld_site;
609 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
611 return &mdt->mdt_seq_site;
614 static inline void mdt_export_evict(struct obd_export *exp)
616 class_fail_export(exp);
619 /* Here we use LVB_TYPE to check dne client, because it is
620 * also landed on 2.4. */
621 static inline bool mdt_is_dne_client(struct obd_export *exp)
623 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
626 static inline bool mdt_is_striped_client(struct obd_export *exp)
628 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
637 /* XXX Look into layout in MDT layer. This must be done in LOD. */
638 static inline int mdt_lmm_dom_entry(struct lov_mds_md *lmm)
640 struct lov_comp_md_v1 *comp_v1;
641 struct lov_mds_md *v1;
645 if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
648 comp_v1 = (struct lov_comp_md_v1 *)lmm;
649 off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
650 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
652 /* DoM entry is the first entry always */
653 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
656 for (i = 1; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
659 off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
660 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
662 for (j = 0; j < le16_to_cpu(v1->lmm_stripe_count); j++) {
663 /* if there is any object on OST */
664 if (le32_to_cpu(v1->lmm_objects[j].l_ost_idx) !=
672 static inline bool mdt_lmm_is_flr(struct lov_mds_md *lmm)
674 struct lov_comp_md_v1 *lcm = (typeof(lcm))lmm;
676 return le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1 &&
677 le16_to_cpu(lcm->lcm_mirror_count) > 0;
680 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
681 void mdt_set_disposition(struct mdt_thread_info *info,
682 struct ldlm_reply *rep, __u64 op_flag);
683 void mdt_clear_disposition(struct mdt_thread_info *info,
684 struct ldlm_reply *rep, __u64 op_flag);
686 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
687 const struct lu_name *lname);
689 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
691 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
692 struct mdt_lock_handle *lh);
694 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
695 struct mdt_lock_handle *lhc);
697 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *mo,
698 struct mdt_lock_handle *lh, __u64 ibits);
700 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
701 struct mdt_lock_handle *lh, __u64 ibits,
704 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
705 struct mdt_lock_handle *lh, __u64 *ibits,
706 __u64 trybits, bool cos_incompat);
708 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
709 struct mdt_lock_handle *lh, int decref);
710 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
711 enum ldlm_mode mode, int decref);
713 struct mdt_object *mdt_object_new(const struct lu_env *env,
715 const struct lu_fid *);
716 struct mdt_object *mdt_object_find(const struct lu_env *,
718 const struct lu_fid *);
719 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
720 const struct lu_fid *,
721 struct mdt_lock_handle *,
723 void mdt_object_unlock_put(struct mdt_thread_info *,
725 struct mdt_lock_handle *,
728 void mdt_client_compatibility(struct mdt_thread_info *info);
730 int mdt_remote_object_lock(struct mdt_thread_info *mti,
731 struct mdt_object *o, const struct lu_fid *fid,
732 struct lustre_handle *lh,
733 enum ldlm_mode mode, __u64 ibits, bool cache);
735 enum mdt_name_flags {
739 int mdt_name_unpack(struct req_capsule *pill,
740 const struct req_msg_field *field,
742 enum mdt_name_flags flags);
743 int mdt_close_unpack(struct mdt_thread_info *info);
744 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
745 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata);
746 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
747 #ifdef CONFIG_FS_POSIX_ACL
748 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
749 struct mdt_object *o, struct lu_nodemap *nodemap);
751 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
752 const struct lu_attr *attr, const struct lu_fid *fid);
753 int mdt_pack_size2body(struct mdt_thread_info *info,
754 const struct lu_fid *fid, bool dom_lock);
755 int mdt_getxattr(struct mdt_thread_info *info);
756 int mdt_reint_setxattr(struct mdt_thread_info *info,
757 struct mdt_lock_handle *lh);
759 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
760 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
762 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
763 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
764 struct mdt_lock_handle *lhc);
766 extern void target_recovery_fini(struct obd_device *obd);
767 extern void target_recovery_init(struct lu_target *lut,
768 svc_handler_t handler);
769 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
770 struct obd_device *, struct lustre_sb_info *lsi);
771 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
773 int mdt_export_stats_init(struct obd_device *obd,
774 struct obd_export *exp,
777 int mdt_lock_new_child(struct mdt_thread_info *info,
778 struct mdt_object *o,
779 struct mdt_lock_handle *child_lockh);
781 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
784 int mdt_reint_open(struct mdt_thread_info *info,
785 struct mdt_lock_handle *lhc);
787 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
788 const struct lustre_handle *handle,
791 int mdt_get_info(struct tgt_session_info *tsi);
792 int mdt_attr_get_complex(struct mdt_thread_info *info,
793 struct mdt_object *o, struct md_attr *ma);
794 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
796 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
797 struct md_attr *ma, const char *name);
798 int mdt_write_get(struct mdt_object *o);
799 void mdt_write_put(struct mdt_object *o);
800 int mdt_write_read(struct mdt_object *o);
801 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
802 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
803 void mdt_mfd_free(struct mdt_file_data *mfd);
804 int mdt_close(struct tgt_session_info *tsi);
805 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
807 int mdt_fix_reply(struct mdt_thread_info *info);
808 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
810 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
811 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
812 struct md_layout_change *spec);
814 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
815 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
816 const void *area, ssize_t len);
818 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
819 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
821 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
822 int mdt_check_ucred(struct mdt_thread_info *);
823 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
824 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *, struct mdt_body *);
825 int mdt_init_ucred_reint(struct mdt_thread_info *);
826 void mdt_exit_ucred(struct mdt_thread_info *);
827 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
828 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
829 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
831 void mdt_thread_info_init(struct ptlrpc_request *req,
832 struct mdt_thread_info *mti);
833 void mdt_thread_info_fini(struct mdt_thread_info *mti);
834 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
835 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
836 struct ldlm_lock *new_lock,
837 struct mdt_lock_handle *lh, __u64 flags);
838 int mdt_intent_lock_replace(struct mdt_thread_info *info,
839 struct ldlm_lock **lockp,
840 struct mdt_lock_handle *lh,
841 __u64 flags, int result);
843 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
844 const struct md_hsm *mh);
846 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
847 void *data, int flag);
848 int mdt_links_read(struct mdt_thread_info *info,
849 struct mdt_object *mdt_obj,
850 struct linkea_data *ldata);
851 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
852 struct mdt_body *repbody);
854 static inline struct mdt_device *mdt_dev(struct lu_device *d)
856 return container_of0(d, struct mdt_device, mdt_lu_dev);
859 static inline struct mdt_object *mdt_obj(struct lu_object *o)
861 return container_of0(o, struct mdt_object, mot_obj);
864 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
866 struct lu_object *lo;
867 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
869 lo = lu_object_locate(mo->mot_obj.lo_header,
870 mdt->mdt_bottom->dd_lu_dev.ld_type);
875 static inline bool agent_req_in_final_state(enum agent_req_status ars)
877 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
878 (ars == ARS_CANCELED));
881 /* mdt/mdt_identity.c */
882 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
884 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
886 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
888 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
890 void mdt_flush_identity(struct upcall_cache *, int);
892 __u32 mdt_identity_get_perm(struct md_identity *, lnet_nid_t);
894 /* mdt/mdt_recovery.c */
895 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
898 int mdt_hsm_state_get(struct tgt_session_info *tsi);
899 int mdt_hsm_state_set(struct tgt_session_info *tsi);
900 int mdt_hsm_action(struct tgt_session_info *tsi);
901 int mdt_hsm_progress(struct tgt_session_info *tsi);
902 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
903 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
904 int mdt_hsm_request(struct tgt_session_info *tsi);
906 /* mdt/mdt_hsm_cdt_actions.c */
907 extern const struct file_operations mdt_hsm_actions_fops;
908 void dump_llog_agent_req_rec(const char *prefix,
909 const struct llog_agent_req_rec *larr);
910 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
911 llog_cb_t cb, void *data, u32 start_cat_idx,
912 u32 start_rec_idx, int rw);
913 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
914 __u64 compound_id, __u32 archive_id,
915 __u64 flags, struct hsm_action_item *hai);
916 int mdt_agent_record_update(const struct lu_env *env, struct mdt_device *mdt,
917 struct hsm_record_update *updates,
918 unsigned int updates_count);
919 void cdt_agent_record_hash_add(struct coordinator *cdt, u64 cookie, u32 cat_idt,
921 void cdt_agent_record_hash_lookup(struct coordinator *cdt, u64 cookie,
922 u32 *cat_idt, u32 *rec_idx);
923 void cdt_agent_record_hash_del(struct coordinator *cdt, u64 cookie);
925 /* mdt/mdt_hsm_cdt_agent.c */
926 extern const struct file_operations mdt_hsm_agent_fops;
927 int mdt_hsm_agent_register(struct mdt_thread_info *info,
928 const struct obd_uuid *uuid,
929 int nr_archives, __u32 *archive_num);
930 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
931 const struct obd_uuid *uuid,
933 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
934 const struct obd_uuid *uuid);
935 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
936 int succ_rq, int fail_rq, int new_rq,
937 const struct obd_uuid *uuid);
938 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
939 struct obd_uuid *uuid);
940 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
942 /* mdt/mdt_hsm_cdt_client.c */
943 int mdt_hsm_add_actions(struct mdt_thread_info *info,
944 struct hsm_action_list *hal);
945 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
946 struct hsm_action_list *hal);
947 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
948 const struct lu_fid *fid);
949 /* mdt/mdt_hsm_cdt_requests.c */
950 extern struct cfs_hash_ops cdt_request_cookie_hash_ops;
951 extern struct cfs_hash_ops cdt_agent_record_hash_ops;
952 extern const struct file_operations mdt_hsm_active_requests_fops;
953 void dump_requests(char *prefix, struct coordinator *cdt);
954 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
955 __u64 flags, struct obd_uuid *uuid,
956 struct hsm_action_item *hai);
957 void mdt_cdt_free_request(struct cdt_agent_req *car);
958 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
959 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt, u64 cookie);
960 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
961 void mdt_cdt_get_request(struct cdt_agent_req *car);
962 void mdt_cdt_put_request(struct cdt_agent_req *car);
963 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
964 const struct hsm_progress_kernel *pgs);
965 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
966 /* mdt/mdt_coordinator.c */
967 void mdt_hsm_dump_hal(int level, const char *prefix,
968 struct hsm_action_list *hal);
969 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
970 const struct lu_fid *fid,
971 const struct hsm_extent *he);
972 struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
973 const struct lu_fid *fid);
974 void cdt_restore_handle_del(struct mdt_thread_info *mti,
975 struct coordinator *cdt, const struct lu_fid *fid);
976 /* coordinator management */
977 int mdt_hsm_cdt_init(struct mdt_device *mdt);
978 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
979 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
982 * Signal the coordinator has work to do
983 * \param cdt [IN] coordinator
985 static inline void mdt_hsm_cdt_event(struct coordinator *cdt)
987 cdt->cdt_event = true;
990 /* coordinator control /proc interface */
991 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file,
992 const char __user *buffer,
993 size_t count, loff_t *off);
994 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
995 int hsm_cdt_procfs_init(struct mdt_device *mdt);
996 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
997 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
999 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1000 const struct lu_fid *fid,
1001 struct md_hsm *hsm);
1002 /* actions/request helpers */
1003 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1004 struct hsm_action_list *hal, struct obd_uuid *uuid);
1005 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1006 u32 archive_id, u64 rq_flags,
1007 const struct md_hsm *hsm);
1008 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1009 struct hsm_progress_kernel *pgs);
1011 int mdt_close_swap_layouts(struct mdt_thread_info *info,
1012 struct mdt_object *o, struct md_attr *ma);
1014 extern struct lu_context_key mdt_thread_key;
1016 /* debug issues helper starts here*/
1017 static inline int mdt_fail_write(const struct lu_env *env,
1018 struct dt_device *dd, int id)
1020 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1021 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1023 return dt_ro(env, dd);
1024 /* We set FAIL_ONCE because we never "un-fail" a device */
1030 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1032 return &req->rq_export->exp_mdt_data;
1035 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
1037 return mdt_dev(exp->exp_obd->obd_lu_dev);
1040 static inline bool mdt_rdonly(struct obd_export *exp)
1042 if (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
1043 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly)
1048 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1049 struct mdt_lock_handle *lhc);
1050 static inline int mdt_check_resent(struct mdt_thread_info *info,
1051 mdt_reconstruct_t reconstruct,
1052 struct mdt_lock_handle *lhc)
1054 struct ptlrpc_request *req = mdt_info_req(info);
1058 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1059 OBD_ALLOC_PTR(info->mti_reply_data);
1060 if (info->mti_reply_data == NULL)
1063 if (req_can_reconstruct(req, info->mti_reply_data)) {
1064 reconstruct(info, lhc);
1067 DEBUG_REQ(D_HA, req,
1068 "no reply data found for RESENT req");
1071 OBD_FREE_PTR(info->mti_reply_data);
1072 info->mti_reply_data = NULL;
1077 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1078 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1080 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1082 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1085 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1087 static int mdt_dom_glimpse_ast(struct ldlm_lock *lock, void *reqp)
1089 return -ELDLM_NO_LOCK_DATA;
1092 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1093 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1094 struct lustre_handle *lh, enum ldlm_mode mode,
1095 union ldlm_policy_data *policy,
1096 const struct ldlm_res_id *res_id,
1097 __u64 flags, const __u64 *client_cookie)
1100 bool glimpse = policy->l_inodebits.bits & MDS_INODELOCK_DOM;
1102 LASSERT(ns != NULL);
1103 LASSERT(lh != NULL);
1105 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1106 mode, &flags, mdt_blocking_ast,
1107 ldlm_completion_ast,
1108 glimpse ? mdt_dom_glimpse_ast : NULL,
1109 NULL, 0, LVB_T_NONE, client_cookie, lh);
1110 return rc == ELDLM_OK ? 0 : -EIO;
1113 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1115 ldlm_lock_decref(lh, mode);
1118 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
1120 return mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL;
1123 extern mdl_mode_t mdt_mdl_lock_modes[];
1124 extern enum ldlm_mode mdt_dlm_lock_modes[];
1126 /* LCK_MINMODE which is zero returns false for is_power_of_2 */
1128 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
1130 LASSERT(mode == LCK_MINMODE || is_power_of_2(mode));
1131 return mdt_mdl_lock_modes[mode];
1134 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1136 LASSERT(mode == MDL_MINMODE || is_power_of_2(mode));
1137 return mdt_dlm_lock_modes[mode];
1141 int mdt_set_som(struct mdt_thread_info *info, struct mdt_object *obj,
1142 struct lu_attr *attr);
1143 int mdt_get_som(struct mdt_thread_info *info, struct mdt_object *obj,
1144 struct lu_attr *attr);
1147 extern struct ldlm_valblock_ops mdt_lvbo;
1148 int mdt_dom_lvb_is_valid(struct ldlm_resource *res);
1149 int mdt_dom_lvbo_update(struct ldlm_resource *res, struct ldlm_lock *lock,
1150 struct ptlrpc_request *req, bool increase_only);
1152 void mdt_enable_cos(struct mdt_device *dev, bool enable);
1153 int mdt_cos_is_enabled(struct mdt_device *);
1171 LPROC_MDT_SAMEDIR_RENAME,
1172 LPROC_MDT_CROSSDIR_RENAME,
1178 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1179 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1180 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1181 void mdt_procfs_fini(struct mdt_device *mdt);
1183 /* lustre/mdt_mdt_lproc.c */
1184 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1186 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1187 struct mdt_device *mdt,
1188 struct ptlrpc_request *req,
1189 struct mdt_object *src, struct mdt_object *tgt);
1191 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1193 return mdt->mdt_lu_dev.ld_obd;
1196 extern const struct lu_device_operations mdt_lu_ops;
1198 static inline char *mdt_obd_name(struct mdt_device *mdt)
1200 return mdt->mdt_lu_dev.ld_obd->obd_name;
1203 int mds_mod_init(void);
1204 void mds_mod_exit(void);
1206 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1208 struct obd_export *exp = req->rq_export;
1211 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1212 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1219 #define VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
1220 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
1222 int mdt_obd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
1223 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1224 struct niobuf_remote *rnb, int *nr_local,
1225 struct niobuf_local *lnb);
1227 int mdt_obd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1228 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1229 struct niobuf_remote *rnb, int npages,
1230 struct niobuf_local *lnb, int old_rc);
1231 int mdt_punch_hdl(struct tgt_session_info *tsi);
1232 int mdt_glimpse_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
1233 struct ldlm_lock **lockp, __u64 flags);
1234 int mdt_brw_enqueue(struct mdt_thread_info *info, struct ldlm_namespace *ns,
1235 struct ldlm_lock **lockp, __u64 flags);
1236 void mdt_dom_discard_data(struct mdt_thread_info *info,
1237 const struct lu_fid *fid);
1238 int mdt_dom_disk_lvbo_update(const struct lu_env *env, struct mdt_object *mo,
1239 struct ldlm_resource *res, bool increase_only);
1240 void mdt_dom_obj_lvb_update(const struct lu_env *env, struct mdt_object *mo,
1241 bool increase_only);
1242 int mdt_dom_lvb_alloc(struct ldlm_resource *res);
1244 static inline void mdt_dom_check_and_discard(struct mdt_thread_info *mti,
1245 struct mdt_object *mo)
1247 if (lu_object_is_dying(&mo->mot_header) &&
1248 S_ISREG(lu_object_attr(&mo->mot_obj)))
1249 mdt_dom_discard_data(mti, mdt_object_fid(mo));
1252 int mdt_dom_object_size(const struct lu_env *env, struct mdt_device *mdt,
1253 const struct lu_fid *fid, struct mdt_body *mb,
1255 bool mdt_dom_client_has_lock(struct mdt_thread_info *info,
1256 const struct lu_fid *fid);
1257 void mdt_hp_brw(struct tgt_session_info *tsi);
1258 void mdt_hp_punch(struct tgt_session_info *tsi);
1259 int mdt_data_version_get(struct tgt_session_info *tsi);
1262 long mdt_grant_connect(const struct lu_env *env, struct obd_export *exp,
1263 u64 want, bool conservative);
1264 extern struct kmem_cache *ldlm_glimpse_work_kmem;
1266 #endif /* _MDT_INTERNAL_H */