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_open_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_open_handle_old;
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
199 struct mdt_statfs_cache {
200 struct obd_statfs msf_osfs;
206 struct lu_device mdt_lu_dev;
207 struct seq_server_site mdt_seq_site;
208 /* DLM name-space for meta-data locks maintained by this server */
209 struct ldlm_namespace *mdt_namespace;
210 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
211 struct ptlrpc_client *mdt_ldlm_client;
212 /* underlying device */
213 struct obd_export *mdt_child_exp;
214 struct md_device *mdt_child;
215 struct dt_device *mdt_bottom;
216 struct obd_export *mdt_bottom_exp;
217 struct local_oid_storage *mdt_los;
219 struct lu_target mdt_lut;
221 * Options bit-fields.
224 unsigned int mo_user_xattr:1,
229 mo_migrate_hsm_allowed:1;
230 unsigned int mo_dom_lock;
232 /* mdt state flags */
233 unsigned long mdt_state;
235 /* transaction callbacks */
236 struct dt_txn_callback mdt_txn_cb;
238 /* these values should be updated from lov if necessary.
239 * or should be placed somewhere else. */
244 /* preferred BRW size, decided by storage type and capability */
247 struct upcall_cache *mdt_identity_cache;
249 unsigned int mdt_capa_conf:1,
250 /* Enable remote dir on non-MDT0 */
251 mdt_enable_remote_dir:1,
252 mdt_enable_striped_dir:1,
253 mdt_enable_dir_migration:1,
254 mdt_enable_remote_rename:1,
258 /* user with gid can create remote/striped
259 * dir, and set default dir stripe */
260 gid_t mdt_enable_remote_dir_gid;
262 /* lock for osfs and md_root */
265 /* statfs optimization: we cache a bit */
266 struct mdt_statfs_cache mdt_sum_osfs;
267 struct mdt_statfs_cache mdt_osfs;
270 struct root_squash_info mdt_squash;
272 struct rename_stats mdt_rename_stats;
273 struct lu_fid mdt_md_root_fid;
275 /* connection to quota master */
276 struct obd_export *mdt_qmt_exp;
277 /* quota master device associated with this MDT */
278 struct lu_device *mdt_qmt_dev;
280 struct coordinator mdt_coordinator;
282 /* inter-MDT connection count */
283 atomic_t mdt_mds_mds_conns;
285 /* MDT device async commit count, used for debug and sanity test */
286 atomic_t mdt_async_commit_count;
288 struct mdt_object *mdt_md_root;
291 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
292 #define MDT_COS_DEFAULT (0)
294 #define ENOENT_VERSION 1 /** 'virtual' version of non-existent object */
297 struct lu_object_header mot_header;
298 struct lu_object mot_obj;
299 unsigned int mot_lov_created:1, /* lov object created */
300 mot_cache_attr:1; /* enable remote object
303 spinlock_t mot_write_lock;
304 /* Lock to protect object's SOM update. */
305 struct mutex mot_som_mutex;
306 /* Lock to protect create_data */
307 struct mutex mot_lov_mutex;
308 /* lock to protect read/write stages for Data-on-MDT files */
309 struct rw_semaphore mot_dom_sem;
310 /* Lock to protect lease open.
311 * Lease open acquires write lock; normal open acquires read lock */
312 struct rw_semaphore mot_open_sem;
313 atomic_t mot_lease_count;
314 atomic_t mot_open_count;
317 struct mdt_lock_handle {
318 /* Lock type, reg for cross-ref use or pdo lock. */
322 struct lustre_handle mlh_reg_lh;
323 enum ldlm_mode mlh_reg_mode;
326 struct lustre_handle mlh_pdo_lh;
327 enum ldlm_mode mlh_pdo_mode;
328 unsigned int mlh_pdo_hash;
330 /* Remote regular lock */
331 struct lustre_handle mlh_rreg_lh;
332 enum ldlm_mode mlh_rreg_mode;
336 MDT_LH_PARENT, /* parent lockh */
337 MDT_LH_CHILD, /* child lockh */
338 MDT_LH_OLD, /* old lockh for rename */
339 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
340 MDT_LH_NEW, /* new lockh for rename */
341 MDT_LH_RMT, /* used for return lh to caller */
342 MDT_LH_LOCAL, /* local lock never return to client */
351 /* Special magical errno for communicaiton between mdt_reint_open()
352 * and mdt_intent_reint() which means return the lock to the client
353 * for subsequent cross ref open. Previously we used plain -EREMOTE
354 * but other functions called in that path might return it too and
355 * confuse us. This is not returned to the client. See LU-5370. */
356 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
358 struct mdt_reint_record {
359 enum mds_reint_op rr_opcode;
360 const struct lustre_handle *rr_open_handle;
361 const struct lustre_handle *rr_lease_handle;
362 const struct lu_fid *rr_fid1;
363 const struct lu_fid *rr_fid2;
364 struct lu_name rr_name;
365 struct lu_name rr_tgt_name;
372 enum mdt_reint_flag {
373 MRF_OPEN_TRUNC = 1 << 0,
377 * Common data shared by mdt-level handlers. This is allocated per-thread to
378 * reduce stack consumption.
380 struct mdt_thread_info {
383 * The following members will be filled explicitly
384 * with specific data in mdt_thread_info_init().
386 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
387 * request handling may migrate from one server thread to another.
389 struct req_capsule *mti_pill;
391 /* although we have export in req, there are cases when it is not
392 * available, e.g. closing files upon export destroy */
393 struct obd_export *mti_exp;
395 * A couple of lock handles.
397 struct mdt_lock_handle mti_lh[MDT_LH_NR];
399 struct mdt_device *mti_mdt;
400 const struct lu_env *mti_env;
402 /* transaction number of current request */
408 * The following members will be filled expilictly
409 * with zero in mdt_thread_info_init(). These members may be used
416 struct md_attr mti_attr;
417 struct md_attr mti_attr2; /* mdt_lvb.c */
419 * Body for "habeo corpus" operations.
421 const struct mdt_body *mti_body;
423 * Host object. This is released at the end of mdt_handler().
425 struct mdt_object *mti_object;
427 * Lock request for "habeo clavis" operations.
429 const struct ldlm_request *mti_dlm_req;
431 __u32 mti_has_trans:1, /* has txn already? */
433 /* big_lmm buffer was used and must be used in reply */
438 /* opdata for mdt_reint_open(), has the same as
439 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
440 * value onto disk for recovery when mdt_trans_stop_cb() is called.
446 * The following members will be filled explicitly
447 * with zero in mdt_reint_unpack(), because they are only used
448 * by reint requests (including mdt_reint_open()).
452 * reint record. contains information for reint operations.
454 struct mdt_reint_record mti_rr;
456 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
458 * Operation specification (currently create and lookup)
460 struct md_op_spec mti_spec;
464 * The following members will _NOT_ be initialized at all.
465 * DO NOT expect them to contain any valid value.
466 * They should be initialized explicitly by the user themselves.
469 /* XXX: If something is in a union, make sure they do not conflict */
470 struct lu_fid mti_tmp_fid1;
471 struct lu_fid mti_tmp_fid2;
472 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
473 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
475 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
476 char ns_name[48];/* for mdt_init0() */
477 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
478 struct obd_statfs osfs; /* for mdt_statfs() */
480 /* for mdt_readpage() */
481 struct lu_rdpg mti_rdpg;
482 /* for mdt_sendpage() */
483 struct l_wait_info mti_wait_info;
493 struct lustre_handle mti_open_handle;
495 struct lu_buf mti_buf;
496 struct lu_buf mti_big_buf;
498 /* Ops object filename */
499 struct lu_name mti_name;
500 char mti_filename[NAME_MAX + 1];
501 /* per-thread values, can be re-used, may be vmalloc'd */
506 /* should be enough to fit lustre_mdt_attrs */
507 char mti_xattr_buf[128];
508 struct ldlm_enqueue_info mti_einfo[2];
509 /* einfo used by mdt_remote_object_lock_try() */
510 struct ldlm_enqueue_info mti_remote_einfo;
511 struct tg_reply_data *mti_reply_data;
513 /* FLR: layout change API */
514 struct md_layout_change mti_layout;
517 extern struct lu_context_key mdt_thread_key;
519 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
521 struct mdt_thread_info *mti;
523 lu_env_refill((void *)env);
524 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
529 struct cdt_req_progress {
530 struct mutex crp_lock; /**< protect tree */
531 struct interval_node *crp_root; /**< tree to track extent
533 struct interval_node **crp_node; /**< buffer for tree nodes
534 * vector of fixed size
536 int crp_cnt; /**< # of used nodes */
537 int crp_max; /**< # of allocated nodes */
540 struct cdt_agent_req {
541 struct hlist_node car_cookie_hash; /**< find req by cookie */
542 struct list_head car_request_list; /**< to chain all the req. */
543 atomic_t car_refcount; /**< reference counter */
544 __u64 car_flags; /**< request original flags */
545 struct obd_uuid car_uuid; /**< agent doing the req. */
546 __u32 car_archive_id; /**< archive id */
547 int car_canceled; /**< request was canceled */
548 time64_t car_req_start; /**< start time */
549 time64_t car_req_update; /**< last update time */
550 struct hsm_action_item *car_hai; /**< req. to the agent */
551 struct cdt_req_progress car_progress; /**< track data mvt
554 extern struct kmem_cache *mdt_hsm_car_kmem;
557 struct list_head ha_list; /**< to chain the agents */
558 struct obd_uuid ha_uuid; /**< agent uuid */
559 __u32 *ha_archive_id; /**< archive id */
560 int ha_archive_cnt; /**< number of archive entries
561 * 0 means any archive */
562 atomic_t ha_requests; /**< current request count */
563 atomic_t ha_success; /**< number of successful
565 atomic_t ha_failure; /**< number of failed actions */
568 struct cdt_restore_handle {
569 struct list_head crh_list; /**< to chain the handle */
570 struct lu_fid crh_fid; /**< fid of the object */
571 struct ldlm_extent crh_extent; /**< extent of the restore */
572 struct mdt_lock_handle crh_lh; /**< lock handle */
574 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
576 struct hsm_record_update {
578 enum agent_req_status status;
581 static inline const struct md_device_operations *
582 mdt_child_ops(struct mdt_device * m)
584 LASSERT(m->mdt_child);
585 return m->mdt_child->md_ops;
588 static inline struct md_object *mdt_object_child(struct mdt_object *o)
591 return lu2md(lu_object_next(&o->mot_obj));
594 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
596 return info->mti_pill ? info->mti_pill->rc_req : NULL;
599 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
601 LASSERT(info->mti_exp);
602 return exp_connect_flags(info->mti_exp);
605 static inline void mdt_object_get(const struct lu_env *env,
606 struct mdt_object *o)
609 lu_object_get(&o->mot_obj);
613 static inline void mdt_object_put(const struct lu_env *env,
614 struct mdt_object *o)
617 lu_object_put(env, &o->mot_obj);
621 static inline int mdt_object_exists(const struct mdt_object *o)
623 return lu_object_exists(&o->mot_obj);
626 static inline int mdt_object_remote(const struct mdt_object *o)
628 return lu_object_remote(&o->mot_obj);
631 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
633 return lu_object_fid(&o->mot_obj);
636 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
638 return mdt->mdt_lu_dev.ld_site;
641 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
643 return &mdt->mdt_seq_site;
646 static inline void mdt_export_evict(struct obd_export *exp)
648 class_fail_export(exp);
651 /* Here we use LVB_TYPE to check dne client, because it is
652 * also landed on 2.4. */
653 static inline bool mdt_is_dne_client(struct obd_export *exp)
655 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
658 static inline bool mdt_is_striped_client(struct obd_export *exp)
660 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
669 /* XXX Look into layout in MDT layer. This must be done in LOD. */
670 static inline int mdt_lmm_dom_entry(struct lov_mds_md *lmm)
672 struct lov_comp_md_v1 *comp_v1;
673 struct lov_mds_md *v1;
677 if (le32_to_cpu(lmm->lmm_magic) != LOV_MAGIC_COMP_V1)
680 comp_v1 = (struct lov_comp_md_v1 *)lmm;
681 off = le32_to_cpu(comp_v1->lcm_entries[0].lcme_offset);
682 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
684 /* DoM entry is the first entry always */
685 if (lov_pattern(le32_to_cpu(v1->lmm_pattern)) != LOV_PATTERN_MDT)
688 for (i = 1; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
691 off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
692 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
694 for (j = 0; j < le16_to_cpu(v1->lmm_stripe_count); j++) {
695 /* if there is any object on OST */
696 if (le32_to_cpu(v1->lmm_objects[j].l_ost_idx) !=
704 static inline bool mdt_lmm_is_flr(struct lov_mds_md *lmm)
706 struct lov_comp_md_v1 *lcm = (typeof(lcm))lmm;
708 return le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1 &&
709 le16_to_cpu(lcm->lcm_mirror_count) > 0;
712 static inline bool mdt_is_sum_statfs_client(struct obd_export *exp)
714 return exp_connect_flags(exp) & OBD_CONNECT_FLAGS2 &&
715 exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS;
718 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
719 void mdt_set_disposition(struct mdt_thread_info *info,
720 struct ldlm_reply *rep, __u64 op_flag);
721 void mdt_clear_disposition(struct mdt_thread_info *info,
722 struct ldlm_reply *rep, __u64 op_flag);
724 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
725 const struct lu_name *lname);
727 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
729 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
730 struct mdt_lock_handle *lh);
732 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
733 struct mdt_lock_handle *lhc);
735 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *mo,
736 struct mdt_lock_handle *lh, __u64 ibits);
738 int mdt_reint_object_lock(struct mdt_thread_info *info, struct mdt_object *o,
739 struct mdt_lock_handle *lh, __u64 ibits,
742 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
743 struct mdt_lock_handle *lh, __u64 *ibits,
744 __u64 trybits, bool cos_incompat);
746 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
747 struct mdt_lock_handle *lh, int decref);
748 void mdt_save_lock(struct mdt_thread_info *info, struct lustre_handle *h,
749 enum ldlm_mode mode, int decref);
751 struct mdt_object *mdt_object_new(const struct lu_env *env,
753 const struct lu_fid *);
754 struct mdt_object *mdt_object_find(const struct lu_env *,
756 const struct lu_fid *);
757 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
758 const struct lu_fid *,
759 struct mdt_lock_handle *,
761 void mdt_object_unlock_put(struct mdt_thread_info *,
763 struct mdt_lock_handle *,
766 void mdt_client_compatibility(struct mdt_thread_info *info);
768 int mdt_remote_object_lock(struct mdt_thread_info *mti,
769 struct mdt_object *o, const struct lu_fid *fid,
770 struct lustre_handle *lh,
771 enum ldlm_mode mode, __u64 ibits, bool cache);
772 int mdt_reint_striped_lock(struct mdt_thread_info *info,
773 struct mdt_object *o,
774 struct mdt_lock_handle *lh,
776 struct ldlm_enqueue_info *einfo,
778 void mdt_reint_striped_unlock(struct mdt_thread_info *info,
779 struct mdt_object *o,
780 struct mdt_lock_handle *lh,
781 struct ldlm_enqueue_info *einfo, int decref);
783 enum mdt_name_flags {
787 int mdt_name_unpack(struct req_capsule *pill,
788 const struct req_msg_field *field,
790 enum mdt_name_flags flags);
791 int mdt_close_unpack(struct mdt_thread_info *info);
792 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
793 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata);
794 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
795 #ifdef CONFIG_FS_POSIX_ACL
796 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
797 struct mdt_object *o, struct lu_nodemap *nodemap);
799 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
800 const struct lu_attr *attr, const struct lu_fid *fid);
801 int mdt_pack_size2body(struct mdt_thread_info *info,
802 const struct lu_fid *fid, struct lustre_handle *lh);
803 int mdt_getxattr(struct mdt_thread_info *info);
804 int mdt_reint_setxattr(struct mdt_thread_info *info,
805 struct mdt_lock_handle *lh);
807 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
808 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
810 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
811 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
812 struct mdt_lock_handle *lhc);
814 extern void target_recovery_fini(struct obd_device *obd);
815 extern void target_recovery_init(struct lu_target *lut,
816 svc_handler_t handler);
817 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
818 struct obd_device *, struct lustre_sb_info *lsi);
819 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
821 int mdt_export_stats_init(struct obd_device *obd,
822 struct obd_export *exp,
825 int mdt_lock_new_child(struct mdt_thread_info *info,
826 struct mdt_object *o,
827 struct mdt_lock_handle *child_lockh);
828 void mdt_mfd_set_mode(struct mdt_file_data *mfd, u64 open_flags);
829 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc);
830 struct mdt_file_data *mdt_open_handle2mfd(struct mdt_export_data *med,
831 const struct lustre_handle *open_handle,
833 int mdt_revoke_remote_lookup_lock(struct mdt_thread_info *info,
834 struct mdt_object *pobj,
835 struct mdt_object *obj);
837 int mdt_get_info(struct tgt_session_info *tsi);
838 int mdt_attr_get_complex(struct mdt_thread_info *info,
839 struct mdt_object *o, struct md_attr *ma);
840 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
842 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
843 struct md_attr *ma, const char *name);
844 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
845 struct lu_fid *pfid);
846 int mdt_write_get(struct mdt_object *o);
847 void mdt_write_put(struct mdt_object *o);
848 int mdt_write_read(struct mdt_object *o);
849 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
850 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
851 void mdt_mfd_free(struct mdt_file_data *mfd);
852 int mdt_close(struct tgt_session_info *tsi);
853 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
855 int mdt_fix_reply(struct mdt_thread_info *info);
856 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
858 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
859 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
860 struct md_layout_change *spec);
861 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt);
863 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
864 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
865 const void *area, ssize_t len);
867 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
868 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
870 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
871 int mdt_check_ucred(struct mdt_thread_info *);
872 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
873 int mdt_init_ucred_intent_getattr(struct mdt_thread_info *, struct mdt_body *);
874 int mdt_init_ucred_reint(struct mdt_thread_info *);
875 void mdt_exit_ucred(struct mdt_thread_info *);
876 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
877 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
878 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
880 void mdt_thread_info_init(struct ptlrpc_request *req,
881 struct mdt_thread_info *mti);
882 void mdt_thread_info_fini(struct mdt_thread_info *mti);
883 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
884 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
885 struct ldlm_lock *new_lock,
886 struct mdt_lock_handle *lh, __u64 flags);
887 int mdt_intent_lock_replace(struct mdt_thread_info *info,
888 struct ldlm_lock **lockp,
889 struct mdt_lock_handle *lh,
890 __u64 flags, int result);
892 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
893 const struct md_hsm *mh);
895 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
896 void *data, int flag);
897 int mdt_links_read(struct mdt_thread_info *info,
898 struct mdt_object *mdt_obj,
899 struct linkea_data *ldata);
900 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
901 struct mdt_body *repbody);
902 void mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
903 struct mdt_object *child);
905 static inline struct mdt_device *mdt_dev(struct lu_device *d)
907 return container_of0(d, struct mdt_device, mdt_lu_dev);
910 static inline struct mdt_object *mdt_obj(struct lu_object *o)
912 return container_of0(o, struct mdt_object, mot_obj);
915 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
917 struct lu_object *lo;
918 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
920 lo = lu_object_locate(mo->mot_obj.lo_header,
921 mdt->mdt_bottom->dd_lu_dev.ld_type);
926 static inline bool agent_req_in_final_state(enum agent_req_status ars)
928 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
929 (ars == ARS_CANCELED));
932 /* mdt/mdt_identity.c */
933 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
935 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
937 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
939 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
941 void mdt_flush_identity(struct upcall_cache *, int);
943 __u32 mdt_identity_get_perm(struct md_identity *, lnet_nid_t);
945 /* mdt/mdt_recovery.c */
946 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
949 int mdt_hsm_state_get(struct tgt_session_info *tsi);
950 int mdt_hsm_state_set(struct tgt_session_info *tsi);
951 int mdt_hsm_action(struct tgt_session_info *tsi);
952 int mdt_hsm_progress(struct tgt_session_info *tsi);
953 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
954 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
955 int mdt_hsm_request(struct tgt_session_info *tsi);
957 /* mdt/mdt_hsm_cdt_actions.c */
958 extern const struct file_operations mdt_hsm_actions_fops;
959 void dump_llog_agent_req_rec(const char *prefix,
960 const struct llog_agent_req_rec *larr);
961 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
962 llog_cb_t cb, void *data, u32 start_cat_idx,
963 u32 start_rec_idx, int rw);
964 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
965 __u32 archive_id, __u64 flags,
966 struct hsm_action_item *hai);
967 int mdt_agent_record_update(const struct lu_env *env, struct mdt_device *mdt,
968 struct hsm_record_update *updates,
969 unsigned int updates_count);
970 void cdt_agent_record_hash_add(struct coordinator *cdt, u64 cookie, u32 cat_idt,
972 void cdt_agent_record_hash_lookup(struct coordinator *cdt, u64 cookie,
973 u32 *cat_idt, u32 *rec_idx);
974 void cdt_agent_record_hash_del(struct coordinator *cdt, u64 cookie);
976 /* mdt/mdt_hsm_cdt_agent.c */
977 extern const struct file_operations mdt_hsm_agent_fops;
978 int mdt_hsm_agent_register(struct mdt_thread_info *info,
979 const struct obd_uuid *uuid,
980 int nr_archives, __u32 *archive_num);
981 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
982 const struct obd_uuid *uuid,
984 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
985 const struct obd_uuid *uuid);
986 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
987 int succ_rq, int fail_rq, int new_rq,
988 const struct obd_uuid *uuid);
989 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
990 struct obd_uuid *uuid);
991 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
993 /* mdt/mdt_hsm_cdt_client.c */
994 int mdt_hsm_add_actions(struct mdt_thread_info *info,
995 struct hsm_action_list *hal);
996 int mdt_hsm_get_action(struct mdt_thread_info *mti,
997 const struct lu_fid *fid,
998 enum hsm_copytool_action *action,
999 enum agent_req_status *status,
1000 struct hsm_extent *extent);
1001 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
1002 const struct lu_fid *fid);
1003 /* mdt/mdt_hsm_cdt_requests.c */
1004 extern struct cfs_hash_ops cdt_request_cookie_hash_ops;
1005 extern struct cfs_hash_ops cdt_agent_record_hash_ops;
1006 extern const struct file_operations mdt_hsm_active_requests_fops;
1007 void dump_requests(char *prefix, struct coordinator *cdt);
1008 struct cdt_agent_req *mdt_cdt_alloc_request(__u32 archive_id, __u64 flags,
1009 struct obd_uuid *uuid,
1010 struct hsm_action_item *hai);
1011 void mdt_cdt_free_request(struct cdt_agent_req *car);
1012 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
1013 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt, u64 cookie);
1014 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
1015 void mdt_cdt_get_request(struct cdt_agent_req *car);
1016 void mdt_cdt_put_request(struct cdt_agent_req *car);
1017 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
1018 const struct hsm_progress_kernel *pgs);
1019 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
1020 /* mdt/mdt_coordinator.c */
1021 void mdt_hsm_dump_hal(int level, const char *prefix,
1022 struct hsm_action_list *hal);
1023 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
1024 const struct lu_fid *fid,
1025 const struct hsm_extent *he);
1026 struct cdt_restore_handle *cdt_restore_handle_find(struct coordinator *cdt,
1027 const struct lu_fid *fid);
1028 void cdt_restore_handle_del(struct mdt_thread_info *mti,
1029 struct coordinator *cdt, const struct lu_fid *fid);
1030 /* coordinator management */
1031 int mdt_hsm_cdt_init(struct mdt_device *mdt);
1032 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
1033 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
1036 * Signal the coordinator has work to do
1037 * \param cdt [IN] coordinator
1039 static inline void mdt_hsm_cdt_event(struct coordinator *cdt)
1041 cdt->cdt_event = true;
1044 /* coordinator control /proc interface */
1045 ssize_t mdt_hsm_cdt_control_seq_write(struct file *file,
1046 const char __user *buffer,
1047 size_t count, loff_t *off);
1048 int mdt_hsm_cdt_control_seq_show(struct seq_file *m, void *data);
1049 int hsm_cdt_procfs_init(struct mdt_device *mdt);
1050 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
1051 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
1052 /* md_hsm helpers */
1053 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1054 const struct lu_fid *fid,
1055 struct md_hsm *hsm);
1056 /* actions/request helpers */
1057 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1058 struct hsm_action_list *hal, struct obd_uuid *uuid);
1059 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1060 u32 archive_id, u64 rq_flags,
1061 const struct md_hsm *hsm);
1062 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1063 struct hsm_progress_kernel *pgs);
1065 int mdt_close_swap_layouts(struct mdt_thread_info *info,
1066 struct mdt_object *o, struct md_attr *ma);
1068 extern struct lu_context_key mdt_thread_key;
1070 /* debug issues helper starts here*/
1071 static inline int mdt_fail_write(const struct lu_env *env,
1072 struct dt_device *dd, int id)
1074 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1075 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1077 return dt_ro(env, dd);
1078 /* We set FAIL_ONCE because we never "un-fail" a device */
1084 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1086 return &req->rq_export->exp_mdt_data;
1089 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
1091 return mdt_dev(exp->exp_obd->obd_lu_dev);
1094 static inline bool mdt_rdonly(struct obd_export *exp)
1096 return (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
1097 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly ||
1098 mdt_exp2dev(exp)->mdt_readonly);
1101 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1102 struct mdt_lock_handle *lhc);
1103 static inline int mdt_check_resent(struct mdt_thread_info *info,
1104 mdt_reconstruct_t reconstruct,
1105 struct mdt_lock_handle *lhc)
1107 struct ptlrpc_request *req = mdt_info_req(info);
1111 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1112 OBD_ALLOC_PTR(info->mti_reply_data);
1113 if (info->mti_reply_data == NULL)
1116 if (req_can_reconstruct(req, info->mti_reply_data)) {
1117 reconstruct(info, lhc);
1120 DEBUG_REQ(D_HA, req,
1121 "no reply data found for RESENT req");
1124 OBD_FREE_PTR(info->mti_reply_data);
1125 info->mti_reply_data = NULL;
1130 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1131 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1133 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1135 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1138 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1140 static int mdt_dom_glimpse_ast(struct ldlm_lock *lock, void *reqp)
1142 return -ELDLM_NO_LOCK_DATA;
1145 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1146 static inline int mdt_fid_lock(const struct lu_env *env,
1147 struct ldlm_namespace *ns,
1148 struct lustre_handle *lh, enum ldlm_mode mode,
1149 union ldlm_policy_data *policy,
1150 const struct ldlm_res_id *res_id,
1151 __u64 flags, const __u64 *client_cookie)
1154 bool glimpse = policy->l_inodebits.bits & MDS_INODELOCK_DOM;
1156 LASSERT(ns != NULL);
1157 LASSERT(lh != NULL);
1159 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS, policy,
1160 mode, &flags, mdt_blocking_ast,
1161 ldlm_completion_ast,
1162 glimpse ? mdt_dom_glimpse_ast : NULL,
1163 NULL, 0, LVB_T_NONE, client_cookie, lh);
1164 return rc == ELDLM_OK ? 0 : -EIO;
1167 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1169 ldlm_lock_decref(lh, mode);
1172 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
1174 return mdt->mdt_lut.lut_sync_lock_cancel == BLOCKING_SYNC_ON_CANCEL;
1177 extern mdl_mode_t mdt_mdl_lock_modes[];
1178 extern enum ldlm_mode mdt_dlm_lock_modes[];
1180 /* LCK_MINMODE which is zero returns false for is_power_of_2 */
1182 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
1184 LASSERT(mode == LCK_MINMODE || is_power_of_2(mode));
1185 return mdt_mdl_lock_modes[mode];
1188 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1190 LASSERT(mode == MDL_MINMODE || is_power_of_2(mode));
1191 return mdt_dlm_lock_modes[mode];
1195 int mdt_set_som(struct mdt_thread_info *info, struct mdt_object *obj,
1196 enum lustre_som_flags flag, __u64 size, __u64 blocks);
1197 int mdt_get_som(struct mdt_thread_info *info, struct mdt_object *obj,
1198 struct md_attr *ma);
1199 int mdt_lsom_downgrade(struct mdt_thread_info *info, struct mdt_object *obj);
1200 int mdt_lsom_update(struct mdt_thread_info *info, struct mdt_object *obj,
1204 extern struct ldlm_valblock_ops mdt_lvbo;
1205 int mdt_dom_lvb_is_valid(struct ldlm_resource *res);
1206 int mdt_dom_lvbo_update(const struct lu_env *env, struct ldlm_resource *res,
1207 struct ldlm_lock *lock, struct ptlrpc_request *req,
1208 bool increase_only);
1210 void mdt_enable_cos(struct mdt_device *dev, bool enable);
1211 int mdt_cos_is_enabled(struct mdt_device *);
1229 LPROC_MDT_SAMEDIR_RENAME,
1230 LPROC_MDT_CROSSDIR_RENAME,
1237 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1238 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1239 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1240 void mdt_procfs_fini(struct mdt_device *mdt);
1242 /* lustre/mdt_mdt_lproc.c */
1243 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1245 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1246 struct mdt_device *mdt,
1247 struct ptlrpc_request *req,
1248 struct mdt_object *src, struct mdt_object *tgt);
1250 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1252 return mdt->mdt_lu_dev.ld_obd;
1255 extern const struct lu_device_operations mdt_lu_ops;
1257 static inline char *mdt_obd_name(struct mdt_device *mdt)
1259 return mdt->mdt_lu_dev.ld_obd->obd_name;
1262 int mds_mod_init(void);
1263 void mds_mod_exit(void);
1265 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1267 struct obd_export *exp = req->rq_export;
1270 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1271 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1278 #define VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
1279 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
1281 int mdt_obd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
1282 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1283 struct niobuf_remote *rnb, int *nr_local,
1284 struct niobuf_local *lnb);
1286 int mdt_obd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1287 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1288 struct niobuf_remote *rnb, int npages,
1289 struct niobuf_local *lnb, int old_rc);
1290 int mdt_punch_hdl(struct tgt_session_info *tsi);
1291 int mdt_glimpse_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
1292 struct ldlm_lock **lockp, __u64 flags);
1293 int mdt_brw_enqueue(struct mdt_thread_info *info, struct ldlm_namespace *ns,
1294 struct ldlm_lock **lockp, __u64 flags);
1295 int mdt_dom_read_on_open(struct mdt_thread_info *mti, struct mdt_device *mdt,
1296 struct lustre_handle *lh);
1297 void mdt_dom_discard_data(struct mdt_thread_info *info,
1298 const struct lu_fid *fid);
1299 int mdt_dom_disk_lvbo_update(const struct lu_env *env, struct mdt_object *mo,
1300 struct ldlm_resource *res, bool increase_only);
1301 void mdt_dom_obj_lvb_update(const struct lu_env *env, struct mdt_object *mo,
1302 bool increase_only);
1303 int mdt_dom_lvb_alloc(struct ldlm_resource *res);
1305 static inline bool mdt_dom_check_for_discard(struct mdt_thread_info *mti,
1306 struct mdt_object *mo)
1308 return lu_object_is_dying(&mo->mot_header) &&
1309 S_ISREG(lu_object_attr(&mo->mot_obj));
1312 int mdt_dom_object_size(const struct lu_env *env, struct mdt_device *mdt,
1313 const struct lu_fid *fid, struct mdt_body *mb,
1315 bool mdt_dom_client_has_lock(struct mdt_thread_info *info,
1316 const struct lu_fid *fid);
1317 void mdt_hp_brw(struct tgt_session_info *tsi);
1318 void mdt_hp_punch(struct tgt_session_info *tsi);
1319 int mdt_data_version_get(struct tgt_session_info *tsi);
1322 long mdt_grant_connect(const struct lu_env *env, struct obd_export *exp,
1323 u64 want, bool conservative);
1324 extern struct kmem_cache *ldlm_glimpse_work_kmem;
1326 #endif /* _MDT_INTERNAL_H */