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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2011, 2013, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * lustre/mdt/mdt_internal.h
38 * Lustre Metadata Target (mdt) request handler
40 * Author: Peter Braam <braam@clusterfs.com>
41 * Author: Andreas Dilger <adilger@clusterfs.com>
42 * Author: Phil Schwan <phil@clusterfs.com>
43 * Author: Mike Shaver <shaver@clusterfs.com>
44 * Author: Nikita Danilov <nikita@clusterfs.com>
45 * Author: Huang Hua <huanghua@clusterfs.com>
48 #ifndef _MDT_INTERNAL_H
49 #define _MDT_INTERNAL_H
51 #if defined(__KERNEL__)
53 #include <lustre_net.h>
54 #include <lustre/lustre_idl.h>
55 #include <obd_class.h>
56 #include <lustre_disk.h>
57 #include <lu_target.h>
58 #include <md_object.h>
59 #include <lustre_fid.h>
60 #include <lustre_fld.h>
61 #include <lustre_req_layout.h>
62 #include <lustre_sec.h>
63 #include <lustre_idmap.h>
64 #include <lustre_eacl.h>
65 #include <lustre_quota.h>
67 /* check if request's xid is equal to last one or not*/
68 static inline int req_xid_is_last(struct ptlrpc_request *req)
70 struct lsd_client_data *lcd = req->rq_export->exp_target_data.ted_lcd;
71 return (req->rq_xid == lcd->lcd_last_xid ||
72 req->rq_xid == lcd->lcd_last_close_xid);
77 /* file data for open files on MDS */
78 struct mdt_file_data {
79 struct portals_handle mfd_handle; /* must be first */
80 __u64 mfd_mode; /* open mode provided by client */
81 cfs_list_t mfd_list; /* protected by med_open_lock */
82 __u64 mfd_xid; /* xid of the open request */
83 struct lustre_handle mfd_old_handle; /* old handle in replay case */
84 struct mdt_object *mfd_object; /* point to opened object */
87 #define CDT_NONBLOCKING_RESTORE 0x0000000000000001ULL
88 #define CDT_NORETRY_ACTION 0x0000000000000002ULL
89 #define CDT_POLICY_MASK CDT_NONBLOCKING_RESTORE | \
92 /* when adding a new policy, do not forget to update
93 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
95 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
97 enum cdt_states { CDT_STOPPED = 0,
103 /* when multiple lock are needed, the lock order is
111 struct ptlrpc_thread cdt_thread; /**< coordinator thread */
112 struct lu_env cdt_env; /**< coordinator lustre
114 struct lu_context cdt_session; /** session for lu_ucred */
115 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
116 __u64 cdt_policy; /**< flags to defined
118 enum cdt_states cdt_state; /**< state */
119 atomic_t cdt_compound_id; /**< compound id counter */
120 __u64 cdt_last_cookie; /**< last cookie allocated */
121 struct mutex cdt_llog_lock; /**< protect llog access */
122 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
123 struct rw_semaphore cdt_request_lock; /**< protect request list */
124 struct mutex cdt_restore_lock; /**< protect restore list */
125 cfs_time_t cdt_loop_period; /**< llog scan period */
126 cfs_time_t cdt_delay; /**< request grace delay */
127 cfs_time_t cdt_timeout; /**< request timeout */
128 __u64 cdt_max_request; /**< max count of started
130 atomic_t cdt_request_count; /**< current count of
131 * started requests */
132 struct list_head cdt_requests; /**< list of started
134 struct list_head cdt_agents; /**< list of register
136 struct list_head cdt_restore_hdl; /**< list of restore lock
140 /* mdt state flag bits */
141 #define MDT_FL_CFGLOG 0
142 #define MDT_FL_SYNCED 1
146 struct lu_device mdt_lu_dev;
147 struct seq_server_site mdt_seq_site;
148 /* DLM name-space for meta-data locks maintained by this server */
149 struct ldlm_namespace *mdt_namespace;
150 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
151 struct ptlrpc_client *mdt_ldlm_client;
152 /* underlying device */
153 struct obd_export *mdt_child_exp;
154 struct md_device *mdt_child;
155 struct dt_device *mdt_bottom;
156 struct obd_export *mdt_bottom_exp;
158 struct lu_target mdt_lut;
160 * Options bit-fields.
163 unsigned int mo_user_xattr:1,
171 /* mdt state flags */
172 unsigned long mdt_state;
173 /* lock to protect IOepoch */
174 spinlock_t mdt_ioepoch_lock;
177 /* transaction callbacks */
178 struct dt_txn_callback mdt_txn_cb;
180 /* these values should be updated from lov if necessary.
181 * or should be placed somewhere else. */
183 int mdt_max_cookiesize;
187 struct upcall_cache *mdt_identity_cache;
190 rwlock_t mdt_sptlrpc_lock;
191 struct sptlrpc_rule_set mdt_sptlrpc_rset;
193 /* capability keys */
194 unsigned long mdt_capa_timeout;
196 struct dt_object *mdt_ck_obj;
197 unsigned long mdt_ck_timeout;
198 unsigned long mdt_ck_expiry;
199 cfs_timer_t mdt_ck_timer;
200 struct ptlrpc_thread mdt_ck_thread;
201 struct lustre_capa_key mdt_capa_keys[2];
202 unsigned int mdt_capa_conf:1,
204 /* Enable remote dir on non-MDT0 */
205 mdt_enable_remote_dir:1;
207 gid_t mdt_enable_remote_dir_gid;
208 /* statfs optimization: we cache a bit */
209 struct obd_statfs mdt_osfs;
211 spinlock_t mdt_osfs_lock;
214 uid_t mdt_squash_uid;
215 gid_t mdt_squash_gid;
216 cfs_list_t mdt_nosquash_nids;
217 char *mdt_nosquash_str;
218 int mdt_nosquash_strlen;
219 struct rw_semaphore mdt_squash_sem;
222 struct rename_stats mdt_rename_stats;
223 struct lu_fid mdt_md_root_fid;
225 /* connection to quota master */
226 struct obd_export *mdt_qmt_exp;
227 /* quota master device associated with this MDT */
228 struct lu_device *mdt_qmt_dev;
230 struct coordinator mdt_coordinator;
233 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
234 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
235 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
236 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
237 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
238 #define MDT_COS_DEFAULT (0)
241 struct lu_object_header mot_header;
242 struct lu_object mot_obj;
245 int mot_ioepoch_count;
247 /* Lock to protect object's IO epoch. */
248 struct mutex mot_ioepoch_mutex;
249 /* Lock to protect create_data */
250 struct mutex mot_lov_mutex;
251 /* Lock to protect lease open.
252 * Lease open acquires write lock; normal open acquires read lock */
253 struct rw_semaphore mot_open_sem;
254 atomic_t mot_lease_count;
255 atomic_t mot_open_count;
256 /* A lock to protect EA data from racing setxattr and getxattrall */
257 struct rw_semaphore mot_xattr_sem;
260 enum mdt_object_flags {
261 /** SOM attributes are changed. */
262 MOF_SOM_CHANGE = (1 << 0),
264 * The SOM recovery state for mdt object.
265 * This state is an in-memory equivalent of an absent SOM EA, used
266 * instead of invalidating SOM EA while IOEpoch is still opened when
267 * a client eviction occurs or a client fails to obtain SOM attributes.
268 * It indicates that the last IOEpoch holder will need to obtain SOM
269 * attributes under [0;EOF] extent lock to flush all the client's
270 * cached of evicted from MDS clients (but not necessary evicted from
271 * OST) before taking ost attributes.
273 MOF_SOM_RECOV = (1 << 1),
274 /** File has been just created. */
275 MOF_SOM_CREATED = (1 << 2),
276 /** lov object has been created. */
277 MOF_LOV_CREATED = (1 << 3),
280 struct mdt_lock_handle {
281 /* Lock type, reg for cross-ref use or pdo lock. */
285 struct lustre_handle mlh_reg_lh;
286 ldlm_mode_t mlh_reg_mode;
289 struct lustre_handle mlh_pdo_lh;
290 ldlm_mode_t mlh_pdo_mode;
291 unsigned int mlh_pdo_hash;
293 /* Remote regular lock */
294 struct lustre_handle mlh_rreg_lh;
295 ldlm_mode_t mlh_rreg_mode;
299 MDT_LH_PARENT, /* parent lockh */
300 MDT_LH_CHILD, /* child lockh */
301 MDT_LH_OLD, /* old lockh for rename */
302 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
303 MDT_LH_NEW, /* new lockh for rename */
304 MDT_LH_RMT, /* used for return lh to caller */
305 MDT_LH_LOCAL, /* local lock never return to client */
314 struct mdt_reint_record {
315 mdt_reint_t rr_opcode;
316 const struct lustre_handle *rr_handle;
317 const struct lu_fid *rr_fid1;
318 const struct lu_fid *rr_fid2;
323 const void *rr_eadata;
326 const struct llog_cookie *rr_logcookies;
330 enum mdt_reint_flag {
331 MRF_OPEN_TRUNC = 1 << 0,
334 struct mdt_thread_info;
336 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
340 tx_exec_func_t exec_fn;
341 tx_exec_func_t undo_fn;
342 struct dt_object *object;
344 struct update_reply *reply;
349 const struct dt_rec *rec;
350 const struct dt_key *key;
365 struct dt_allocation_hint hint;
366 struct dt_object_format dof;
374 struct ost_body *body;
379 #define TX_MAX_OPS 10
380 struct thandle_exec_args {
381 struct thandle *ta_handle;
382 struct dt_device *ta_dev;
384 struct tx_arg ta_args[TX_MAX_OPS];
385 int ta_argno; /* used args */
389 * Common data shared by mdt-level handlers. This is allocated per-thread to
390 * reduce stack consumption.
392 struct mdt_thread_info {
395 * The following members will be filled explicitly
396 * with specific data in mdt_thread_info_init().
398 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
399 * request handling may migrate from one server thread to another.
401 struct req_capsule *mti_pill;
403 /* although we have export in req, there are cases when it is not
404 * available, e.g. closing files upon export destroy */
405 struct obd_export *mti_exp;
407 * A couple of lock handles.
409 struct mdt_lock_handle mti_lh[MDT_LH_NR];
411 struct mdt_device *mti_mdt;
412 const struct lu_env *mti_env;
413 /* XXX: temporary flag to have healthy mti during OUT calls
414 * to be removed upon moving MDT to the unified target code */
418 * Additional fail id that can be set by handler. Passed to
419 * target_send_reply().
423 /* transaction number of current request */
429 * The following members will be filled expilictly
430 * with zero in mdt_thread_info_init(). These members may be used
437 struct md_attr mti_attr;
439 * Body for "habeo corpus" operations.
441 const struct mdt_body *mti_body;
443 * Host object. This is released at the end of mdt_handler().
445 struct mdt_object *mti_object;
447 * Lock request for "habeo clavis" operations.
449 const struct ldlm_request *mti_dlm_req;
451 __u32 mti_has_trans:1, /* has txn already? */
454 /* opdata for mdt_reint_open(), has the same as
455 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
456 * value onto disk for recovery when mdt_trans_stop_cb() is called.
462 * The following members will be filled explicitly
463 * with zero in mdt_reint_unpack(), because they are only used
464 * by reint requests (including mdt_reint_open()).
468 * reint record. contains information for reint operations.
470 struct mdt_reint_record mti_rr;
472 /** md objects included in operation */
473 struct mdt_object *mti_mos;
474 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
476 * Operation specification (currently create and lookup)
478 struct md_op_spec mti_spec;
482 * The following members will _NOT_ be initialized at all.
483 * DO NOT expect them to contain any valid value.
484 * They should be initialized explicitly by the user themselves.
487 /* XXX: If something is in a union, make sure they do not conflict */
489 struct lu_fid mti_tmp_fid1;
490 struct lu_fid mti_tmp_fid2;
491 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
492 * mdt_rename_lock() */
493 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
496 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
497 char ns_name[48]; /* for mdt_init0() */
498 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
499 struct obd_statfs osfs; /* for mdt_statfs() */
501 /* for mdt_readpage() */
502 struct lu_rdpg mti_rdpg;
503 /* for mdt_sendpage() */
504 struct l_wait_info mti_wait_info;
508 struct md_som_data data;
511 struct dt_object_format mti_update_dof;
512 struct update_reply *mti_update_reply;
513 struct update *mti_update;
514 int mti_update_reply_index;
515 struct obdo mti_obdo;
516 struct dt_object *mti_dt_object;
520 /* IO epoch related stuff. */
521 struct mdt_ioepoch *mti_ioepoch;
522 __u64 mti_replayepoch;
525 struct lu_buf mti_buf;
526 struct lu_buf mti_big_buf;
527 struct lustre_capa_key mti_capa_key;
529 /* Ops object filename */
530 struct lu_name mti_name;
531 /* per-thread values, can be re-used */
534 /* big_lmm buffer was used and must be used in reply */
535 int mti_big_lmm_used;
536 /* should be enough to fit lustre_mdt_attrs */
537 char mti_xattr_buf[128];
538 struct thandle_exec_args mti_handle;
539 struct ldlm_enqueue_info mti_einfo;
542 /* ptlrpc request handler for MDT. All handlers are
543 * grouped into several slices - struct mdt_opc_slice,
544 * and stored in an array - mdt_handlers[].
547 /* The name of this handler. */
549 /* Fail id for this handler, checked at the beginning of this handler*/
551 /* Operation code for this handler */
553 /* flags are listed in enum mdt_handler_flags below. */
555 /* The actual handler function to execute. */
556 int (*mh_act)(struct mdt_thread_info *info);
557 /* Request format for this request. */
558 const struct req_format *mh_fmt;
561 struct mdt_opc_slice {
564 struct mdt_handler *mos_hs;
567 struct cdt_req_progress {
568 struct mutex crp_lock; /**< protect tree */
569 struct interval_node *crp_root; /**< tree to track extent
571 struct interval_node **crp_node; /**< buffer for tree nodes
572 * vector of fixed size
574 int crp_cnt; /**< # of used nodes */
575 int crp_max; /**< # of allocated nodes */
578 struct cdt_agent_req {
579 cfs_list_t car_request_list; /**< to chain all the req. */
580 atomic_t car_refcount; /**< reference counter */
581 __u64 car_compound_id; /**< compound id */
582 __u64 car_flags; /**< request original flags */
583 struct obd_uuid car_uuid; /**< agent doing the req. */
584 __u32 car_archive_id; /**< archive id */
585 int car_canceled; /**< request was canceled */
586 cfs_time_t car_req_start; /**< start time */
587 cfs_time_t car_req_update; /**< last update time */
588 struct hsm_action_item *car_hai; /**< req. to the agent */
589 struct cdt_req_progress car_progress; /**< track data mvt
592 extern struct kmem_cache *mdt_hsm_car_kmem;
595 cfs_list_t ha_list; /**< to chain the agents */
596 struct obd_uuid ha_uuid; /**< agent uuid */
597 __u32 *ha_archive_id; /**< archive id */
598 int ha_archive_cnt; /**< number of archive entries
599 * 0 means any archive */
600 atomic_t ha_requests; /**< current request count */
601 atomic_t ha_success; /**< number of successful
603 atomic_t ha_failure; /**< number of failed actions */
606 struct cdt_restore_handle {
607 cfs_list_t crh_list; /**< to chain the handle */
608 struct lu_fid crh_fid; /**< fid of the object */
609 struct ldlm_extent crh_extent; /**< extent of the restore */
610 struct mdt_lock_handle crh_lh; /**< lock handle */
612 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
614 static inline const struct md_device_operations *
615 mdt_child_ops(struct mdt_device * m)
617 LASSERT(m->mdt_child);
618 return m->mdt_child->md_ops;
621 static inline struct md_object *mdt_object_child(struct mdt_object *o)
624 return lu2md(lu_object_next(&o->mot_obj));
627 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
629 return info->mti_pill ? info->mti_pill->rc_req : NULL;
632 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
634 LASSERT(info->mti_exp);
635 return exp_connect_flags(info->mti_exp);
638 static inline void mdt_object_get(const struct lu_env *env,
639 struct mdt_object *o)
642 lu_object_get(&o->mot_obj);
646 static inline void mdt_object_put(const struct lu_env *env,
647 struct mdt_object *o)
650 lu_object_put(env, &o->mot_obj);
654 static inline int mdt_object_exists(const struct mdt_object *o)
656 return lu_object_exists(&o->mot_obj);
659 static inline int mdt_object_remote(const struct mdt_object *o)
661 return lu_object_remote(&o->mot_obj);
664 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
666 return lu_object_fid(&o->mot_obj);
669 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
671 return mdt->mdt_lu_dev.ld_site;
674 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
676 return &mdt->mdt_seq_site;
679 static inline void mdt_export_evict(struct obd_export *exp)
681 class_fail_export(exp);
682 class_export_put(exp);
685 /* Here we use LVB_TYPE to check dne client, because it is
686 * also landed on 2.4. */
687 static inline int mdt_is_dne_client(struct obd_export *exp)
689 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
692 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
693 void mdt_set_disposition(struct mdt_thread_info *info,
694 struct ldlm_reply *rep, int flag);
695 void mdt_clear_disposition(struct mdt_thread_info *info,
696 struct ldlm_reply *rep, int flag);
698 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
699 ldlm_mode_t lm, const char *name,
702 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
705 int mdt_lock_setup(struct mdt_thread_info *info,
706 struct mdt_object *o,
707 struct mdt_lock_handle *lh);
709 int mdt_object_lock(struct mdt_thread_info *,
711 struct mdt_lock_handle *,
714 int mdt_object_lock_try(struct mdt_thread_info *,
716 struct mdt_lock_handle *,
719 void mdt_object_unlock(struct mdt_thread_info *,
721 struct mdt_lock_handle *,
724 struct mdt_object *mdt_object_new(const struct lu_env *,
726 const struct lu_fid *);
727 struct mdt_object *mdt_object_find(const struct lu_env *,
729 const struct lu_fid *);
730 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
731 const struct lu_fid *,
732 struct mdt_lock_handle *,
734 void mdt_object_unlock_put(struct mdt_thread_info *,
736 struct mdt_lock_handle *,
739 void mdt_client_compatibility(struct mdt_thread_info *info);
741 int mdt_remote_object_lock(struct mdt_thread_info *mti,
742 struct mdt_object *o, struct lustre_handle *lh,
743 ldlm_mode_t mode, __u64 ibits);
744 int mdt_close_unpack(struct mdt_thread_info *info);
745 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
746 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
747 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
748 const struct lu_attr *attr, const struct lu_fid *fid);
750 int mdt_getxattr(struct mdt_thread_info *info);
751 int mdt_reint_setxattr(struct mdt_thread_info *info,
752 struct mdt_lock_handle *lh);
754 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
755 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
757 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
758 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
759 struct mdt_lock_handle *lhc);
761 extern void target_recovery_fini(struct obd_device *obd);
762 extern void target_recovery_init(struct lu_target *lut,
763 svc_handler_t handler);
764 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
765 struct obd_device *, struct lustre_sb_info *lsi);
766 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
768 int mdt_export_stats_init(struct obd_device *obd,
769 struct obd_export *exp,
772 int mdt_pin(struct mdt_thread_info* info);
774 int mdt_lock_new_child(struct mdt_thread_info *info,
775 struct mdt_object *o,
776 struct mdt_lock_handle *child_lockh);
778 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
781 int mdt_reint_open(struct mdt_thread_info *info,
782 struct mdt_lock_handle *lhc);
784 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
785 const struct lustre_handle *handle,
789 MDT_IOEPOCH_CLOSED = 0,
790 MDT_IOEPOCH_OPENED = 1,
791 MDT_IOEPOCH_GETATTR = 2,
799 int mdt_attr_get_complex(struct mdt_thread_info *info,
800 struct mdt_object *o, struct md_attr *ma);
801 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
803 int mdt_object_is_som_enabled(struct mdt_object *mo);
804 int mdt_write_get(struct mdt_object *o);
805 void mdt_write_put(struct mdt_object *o);
806 int mdt_write_read(struct mdt_object *o);
807 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
808 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
809 void mdt_mfd_free(struct mdt_file_data *mfd);
810 int mdt_close(struct mdt_thread_info *info);
811 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
812 struct md_attr *ma, int flags);
813 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
815 int mdt_done_writing(struct mdt_thread_info *info);
816 int mdt_fix_reply(struct mdt_thread_info *info);
817 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
818 const struct md_attr *);
819 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
821 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
822 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
823 const void *area, ssize_t len);
825 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm);
827 int mdt_check_ucred(struct mdt_thread_info *);
828 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
829 int mdt_init_ucred_reint(struct mdt_thread_info *);
830 void mdt_exit_ucred(struct mdt_thread_info *);
831 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
832 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
833 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
835 int mdt_handle_common(struct ptlrpc_request *req,
836 struct mdt_opc_slice *supported);
837 int mdt_connect(struct mdt_thread_info *info);
838 int mdt_disconnect(struct mdt_thread_info *info);
839 int mdt_set_info(struct mdt_thread_info *info);
840 int mdt_get_info(struct mdt_thread_info *info);
841 int mdt_getstatus(struct mdt_thread_info *info);
842 int mdt_getattr(struct mdt_thread_info *info);
843 int mdt_getattr_name(struct mdt_thread_info *info);
844 int mdt_statfs(struct mdt_thread_info *info);
845 int mdt_reint(struct mdt_thread_info *info);
846 int mdt_sync(struct mdt_thread_info *info);
847 int mdt_is_subdir(struct mdt_thread_info *info);
848 int mdt_obd_ping(struct mdt_thread_info *info);
849 int mdt_obd_log_cancel(struct mdt_thread_info *info);
850 int mdt_obd_qc_callback(struct mdt_thread_info *info);
851 int mdt_enqueue(struct mdt_thread_info *info);
852 int mdt_convert(struct mdt_thread_info *info);
853 int mdt_bl_callback(struct mdt_thread_info *info);
854 int mdt_cp_callback(struct mdt_thread_info *info);
855 int mdt_llog_create(struct mdt_thread_info *info);
856 int mdt_llog_destroy(struct mdt_thread_info *info);
857 int mdt_llog_read_header(struct mdt_thread_info *info);
858 int mdt_llog_next_block(struct mdt_thread_info *info);
859 int mdt_llog_prev_block(struct mdt_thread_info *info);
860 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
861 int mdt_readpage(struct mdt_thread_info *info);
862 int mdt_obd_idx_read(struct mdt_thread_info *info);
863 int mdt_tgt_connect(struct tgt_session_info *tsi);
864 void mdt_thread_info_init(struct ptlrpc_request *req,
865 struct mdt_thread_info *mti);
866 void mdt_thread_info_fini(struct mdt_thread_info *mti);
868 extern struct mdt_opc_slice mdt_regular_handlers[];
869 extern struct mdt_opc_slice mdt_seq_handlers[];
870 extern struct mdt_opc_slice mdt_fld_handlers[];
872 int mdt_quotacheck(struct mdt_thread_info *info);
873 int mdt_quotactl(struct mdt_thread_info *info);
874 int mdt_quota_dqacq(struct mdt_thread_info *info);
875 int mdt_swap_layouts(struct mdt_thread_info *info);
877 extern struct lprocfs_vars lprocfs_mds_module_vars[];
878 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
880 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
881 const struct md_hsm *mh);
883 struct mdt_handler *mdt_handler_find(__u32 opc,
884 struct mdt_opc_slice *supported);
886 int mdt_init_sec_level(struct mdt_thread_info *);
887 int mdt_init_idmap(struct mdt_thread_info *);
888 void mdt_cleanup_idmap(struct mdt_export_data *);
889 int mdt_handle_idmap(struct mdt_thread_info *);
890 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
891 struct ptlrpc_user_desc *);
892 void mdt_body_reverse_idmap(struct mdt_thread_info *,
894 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
895 struct mdt_remote_perm *);
896 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
898 static inline struct mdt_device *mdt_dev(struct lu_device *d)
900 return container_of0(d, struct mdt_device, mdt_lu_dev);
903 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
905 struct lu_object *lo;
906 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
908 lo = lu_object_locate(mo->mot_obj.lo_header,
909 mdt->mdt_bottom->dd_lu_dev.ld_type);
914 /* mdt/mdt_identity.c */
915 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
917 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
919 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
921 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
923 void mdt_flush_identity(struct upcall_cache *, int);
925 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
927 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
930 int mdt_hsm_state_get(struct mdt_thread_info *info);
931 int mdt_hsm_state_set(struct mdt_thread_info *info);
932 int mdt_hsm_action(struct mdt_thread_info *info);
933 int mdt_hsm_progress(struct mdt_thread_info *info);
934 int mdt_hsm_ct_register(struct mdt_thread_info *info);
935 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
936 int mdt_hsm_request(struct mdt_thread_info *info);
937 /* mdt/mdt_hsm_cdt_actions.c */
938 extern const struct file_operations mdt_agent_actions_fops;
939 void dump_llog_agent_req_rec(const char *prefix,
940 const struct llog_agent_req_rec *larr);
941 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
942 llog_cb_t cb, void *data);
943 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
944 __u64 compound_id, __u32 archive_id,
945 __u64 flags, struct hsm_action_item *hai);
946 int mdt_agent_record_update(const struct lu_env *env,
947 struct mdt_device *mdt, __u64 *cookies,
948 int cookies_count, enum agent_req_status status);
949 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
950 struct llog_handle *llh,
951 struct llog_agent_req_rec *larr);
953 /* mdt/mdt_hsm_cdt_agent.c */
954 extern const struct file_operations mdt_hsm_agent_fops;
955 int mdt_hsm_agent_register(struct mdt_thread_info *info,
956 const struct obd_uuid *uuid,
957 int nr_archives, __u32 *archive_num);
958 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
959 const struct obd_uuid *uuid,
961 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
962 const struct obd_uuid *uuid);
963 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
964 int succ_rq, int fail_rq, int new_rq,
965 const struct obd_uuid *uuid);
966 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
967 struct obd_uuid *uuid);
968 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
970 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
971 struct hsm_progress_kernel *pgs);
972 /* mdt/mdt_hsm_cdt_client.c */
973 int mdt_hsm_add_actions(struct mdt_thread_info *info,
974 struct hsm_action_list *hal, __u64 *compound_id);
975 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
976 struct hsm_action_list *hal);
977 int mdt_hsm_get_running(struct mdt_thread_info *mti,
978 struct hsm_action_list *hal);
979 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
980 const struct lu_fid *fid);
981 /* mdt/mdt_hsm_cdt_requests.c */
982 extern const struct file_operations mdt_hsm_request_fops;
983 void dump_requests(char *prefix, struct coordinator *cdt);
984 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
985 __u64 flags, struct obd_uuid *uuid,
986 struct hsm_action_item *hai);
987 void mdt_cdt_free_request(struct cdt_agent_req *car);
988 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
989 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
991 const struct lu_fid *fid);
992 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
993 void mdt_cdt_get_request(struct cdt_agent_req *car);
994 void mdt_cdt_put_request(struct cdt_agent_req *car);
995 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
996 const struct hsm_progress_kernel *pgs);
997 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
998 /* mdt/mdt_coordinator.c */
999 void mdt_hsm_dump_hal(int level, const char *prefix,
1000 struct hsm_action_list *hal);
1001 /* coordinator management */
1002 int mdt_hsm_cdt_init(struct mdt_device *mdt);
1003 int mdt_hsm_cdt_start(struct mdt_device *mdt);
1004 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
1005 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
1006 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
1008 /* coordinator control /proc interface */
1009 int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
1010 unsigned long count, void *data);
1011 int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
1012 int count, int *eof, void *data);
1013 /* md_hsm helpers */
1014 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1015 const struct lu_fid *fid,
1016 struct md_hsm *hsm);
1017 /* actions/request helpers */
1018 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1019 struct hsm_action_list *hal, struct obd_uuid *uuid);
1020 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1021 const int hal_an, const __u64 rq_flags,
1022 const struct md_hsm *hsm);
1023 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1024 struct hsm_progress_kernel *pgs,
1025 const int update_record);
1027 extern struct lu_context_key mdt_thread_key;
1028 /* debug issues helper starts here*/
1029 static inline int mdt_fail_write(const struct lu_env *env,
1030 struct dt_device *dd, int id)
1032 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1033 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1035 return dd->dd_ops->dt_ro(env, dd);
1036 /* We set FAIL_ONCE because we never "un-fail" a device */
1042 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1044 return &req->rq_export->exp_mdt_data;
1047 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1048 struct mdt_lock_handle *lhc);
1049 static inline int mdt_check_resent(struct mdt_thread_info *info,
1050 mdt_reconstruct_t reconstruct,
1051 struct mdt_lock_handle *lhc)
1053 struct ptlrpc_request *req = mdt_info_req(info);
1056 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1057 if (req_xid_is_last(req)) {
1058 reconstruct(info, lhc);
1061 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
1062 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
1067 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1068 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1070 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1072 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1075 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1077 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1078 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1079 struct lustre_handle *lh,
1081 ldlm_policy_data_t *policy,
1082 const struct ldlm_res_id *res_id,
1083 __u64 flags, const __u64 *client_cookie)
1087 LASSERT(ns != NULL);
1088 LASSERT(lh != NULL);
1090 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1091 mode, &flags, mdt_blocking_ast,
1092 ldlm_completion_ast, NULL, NULL, 0,
1093 LVB_T_NONE, client_cookie, lh);
1094 return rc == ELDLM_OK ? 0 : -EIO;
1097 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1100 ldlm_lock_decref(lh, mode);
1103 extern mdl_mode_t mdt_mdl_lock_modes[];
1104 extern ldlm_mode_t mdt_dlm_lock_modes[];
1106 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1108 LASSERT(IS_PO2(mode));
1109 return mdt_mdl_lock_modes[mode];
1112 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1114 LASSERT(IS_PO2(mode));
1115 return mdt_dlm_lock_modes[mode];
1119 extern struct ldlm_valblock_ops mdt_lvbo;
1121 static inline struct lu_name *mdt_name(const struct lu_env *env,
1122 char *name, int namelen)
1124 struct lu_name *lname;
1125 struct mdt_thread_info *mti;
1127 LASSERT(namelen > 0);
1128 /* trailing '\0' in buffer */
1129 LASSERT(name[namelen] == '\0');
1131 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
1132 lname = &mti->mti_name;
1133 lname->ln_name = name;
1134 lname->ln_namelen = namelen;
1138 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
1139 struct lu_name *slname)
1144 tlname->ln_name = slname->ln_name;
1145 tlname->ln_namelen = slname->ln_namelen;
1149 void mdt_enable_cos(struct mdt_device *, int);
1150 int mdt_cos_is_enabled(struct mdt_device *);
1168 LPROC_MDT_SAMEDIR_RENAME,
1169 LPROC_MDT_CROSSDIR_RENAME,
1172 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1173 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1174 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1175 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1176 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1177 void mdt_procfs_fini(struct mdt_device *mdt);
1178 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1179 struct mdt_device *mdt,
1180 struct ptlrpc_request *req,
1181 struct mdt_object *src, struct mdt_object *tgt);
1184 int mdt_ck_thread_start(struct mdt_device *mdt);
1185 void mdt_ck_thread_stop(struct mdt_device *mdt);
1186 void mdt_ck_timer_callback(unsigned long castmeharder);
1187 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1188 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1189 const struct lu_fid *fid, struct lustre_capa *capa);
1190 void mdt_dump_capainfo(struct mdt_thread_info *info);
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 int lu_device_is_mdt(struct lu_device *d)
1201 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1204 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1206 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1207 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1208 return container_of0(d, struct mdt_device, mdt_lu_dev);
1211 static inline char *mdt_obd_name(struct mdt_device *mdt)
1213 return mdt->mdt_lu_dev.ld_obd->obd_name;
1216 int mds_mod_init(void);
1217 void mds_mod_exit(void);
1219 #endif /* __KERNEL__ */