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 (1ULL << 0)
88 #define CDT_NORETRY_ACTION (1ULL << 1)
89 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
90 #define CDT_POLICY_SHIFT_COUNT 2
91 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
94 /* when adding a new policy, do not forget to update
95 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
97 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
99 enum cdt_states { CDT_STOPPED = 0,
105 /* when multiple lock are needed, the lock order is
113 struct ptlrpc_thread cdt_thread; /**< coordinator thread */
114 struct lu_env cdt_env; /**< coordinator lustre
116 struct lu_context cdt_session; /** session for lu_ucred */
117 struct proc_dir_entry *cdt_proc_dir; /**< cdt /proc directory */
118 __u64 cdt_policy; /**< policy flags */
119 enum cdt_states cdt_state; /**< state */
120 atomic_t cdt_compound_id; /**< compound id
122 __u64 cdt_last_cookie; /**< last cookie
124 struct mutex cdt_llog_lock; /**< protect llog
126 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
127 struct rw_semaphore cdt_request_lock; /**< protect request
129 struct mutex cdt_restore_lock; /**< protect restore
131 cfs_time_t cdt_loop_period; /**< llog scan period */
132 cfs_time_t cdt_grace_delay; /**< request grace
134 cfs_time_t cdt_active_req_timeout; /**< request timeout */
135 __u32 cdt_default_archive_id; /**< archive id used
138 __u64 cdt_max_requests; /**< max count of started
140 atomic_t cdt_request_count; /**< current count of
141 * started requests */
142 struct list_head cdt_requests; /**< list of started
144 struct list_head cdt_agents; /**< list of register
146 struct list_head cdt_restore_hdl; /**< list of restore lock
148 /* Bitmasks indexed by the HSMA_XXX constants. */
149 __u64 cdt_user_request_mask;
150 __u64 cdt_group_request_mask;
151 __u64 cdt_other_request_mask;
154 /* mdt state flag bits */
155 #define MDT_FL_CFGLOG 0
156 #define MDT_FL_SYNCED 1
160 struct lu_device mdt_lu_dev;
161 struct seq_server_site mdt_seq_site;
162 /* DLM name-space for meta-data locks maintained by this server */
163 struct ldlm_namespace *mdt_namespace;
164 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
165 struct ptlrpc_client *mdt_ldlm_client;
166 /* underlying device */
167 struct obd_export *mdt_child_exp;
168 struct md_device *mdt_child;
169 struct dt_device *mdt_bottom;
170 struct obd_export *mdt_bottom_exp;
172 struct lu_target mdt_lut;
174 * Options bit-fields.
177 unsigned int mo_user_xattr:1,
185 /* mdt state flags */
186 unsigned long mdt_state;
187 /* lock to protect IOepoch */
188 spinlock_t mdt_ioepoch_lock;
191 /* transaction callbacks */
192 struct dt_txn_callback mdt_txn_cb;
194 /* these values should be updated from lov if necessary.
195 * or should be placed somewhere else. */
200 struct upcall_cache *mdt_identity_cache;
203 rwlock_t mdt_sptlrpc_lock;
204 struct sptlrpc_rule_set mdt_sptlrpc_rset;
206 /* capability keys */
207 unsigned long mdt_capa_timeout;
209 struct dt_object *mdt_ck_obj;
210 unsigned long mdt_ck_timeout;
211 unsigned long mdt_ck_expiry;
212 struct timer_list mdt_ck_timer;
213 struct ptlrpc_thread mdt_ck_thread;
214 struct lustre_capa_key mdt_capa_keys[2];
215 unsigned int mdt_capa_conf:1,
217 /* Enable remote dir on non-MDT0 */
218 mdt_enable_remote_dir:1;
220 gid_t mdt_enable_remote_dir_gid;
221 /* statfs optimization: we cache a bit */
222 struct obd_statfs mdt_osfs;
224 spinlock_t mdt_osfs_lock;
227 struct root_squash_info mdt_squash;
230 struct rename_stats mdt_rename_stats;
231 struct lu_fid mdt_md_root_fid;
233 /* connection to quota master */
234 struct obd_export *mdt_qmt_exp;
235 /* quota master device associated with this MDT */
236 struct lu_device *mdt_qmt_dev;
238 struct coordinator mdt_coordinator;
241 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
242 #define MDT_ROCOMPAT_SUPP (OBD_ROCOMPAT_LOVOBJID)
243 #define MDT_INCOMPAT_SUPP (OBD_INCOMPAT_MDT | OBD_INCOMPAT_COMMON_LR | \
244 OBD_INCOMPAT_FID | OBD_INCOMPAT_IAM_DIR | \
245 OBD_INCOMPAT_LMM_VER | OBD_INCOMPAT_MULTI_OI)
246 #define MDT_COS_DEFAULT (0)
249 struct lu_object_header mot_header;
250 struct lu_object mot_obj;
253 int mot_ioepoch_count;
255 /* Lock to protect object's IO epoch. */
256 struct mutex mot_ioepoch_mutex;
257 /* Lock to protect create_data */
258 struct mutex mot_lov_mutex;
259 /* Lock to protect lease open.
260 * Lease open acquires write lock; normal open acquires read lock */
261 struct rw_semaphore mot_open_sem;
262 atomic_t mot_lease_count;
263 atomic_t mot_open_count;
266 enum mdt_object_flags {
267 /** SOM attributes are changed. */
268 MOF_SOM_CHANGE = (1 << 0),
270 * The SOM recovery state for mdt object.
271 * This state is an in-memory equivalent of an absent SOM EA, used
272 * instead of invalidating SOM EA while IOEpoch is still opened when
273 * a client eviction occurs or a client fails to obtain SOM attributes.
274 * It indicates that the last IOEpoch holder will need to obtain SOM
275 * attributes under [0;EOF] extent lock to flush all the client's
276 * cached of evicted from MDS clients (but not necessary evicted from
277 * OST) before taking ost attributes.
279 MOF_SOM_RECOV = (1 << 1),
280 /** File has been just created. */
281 MOF_SOM_CREATED = (1 << 2),
282 /** lov object has been created. */
283 MOF_LOV_CREATED = (1 << 3),
286 struct mdt_lock_handle {
287 /* Lock type, reg for cross-ref use or pdo lock. */
291 struct lustre_handle mlh_reg_lh;
292 ldlm_mode_t mlh_reg_mode;
295 struct lustre_handle mlh_pdo_lh;
296 ldlm_mode_t mlh_pdo_mode;
297 unsigned int mlh_pdo_hash;
299 /* Remote regular lock */
300 struct lustre_handle mlh_rreg_lh;
301 ldlm_mode_t mlh_rreg_mode;
305 MDT_LH_PARENT, /* parent lockh */
306 MDT_LH_CHILD, /* child lockh */
307 MDT_LH_OLD, /* old lockh for rename */
308 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
309 MDT_LH_NEW, /* new lockh for rename */
310 MDT_LH_RMT, /* used for return lh to caller */
311 MDT_LH_LOCAL, /* local lock never return to client */
320 struct mdt_reint_record {
321 mdt_reint_t rr_opcode;
322 const struct lustre_handle *rr_handle;
323 const struct lu_fid *rr_fid1;
324 const struct lu_fid *rr_fid2;
329 const void *rr_eadata;
332 const struct llog_cookie *rr_logcookies;
336 enum mdt_reint_flag {
337 MRF_OPEN_TRUNC = 1 << 0,
340 struct mdt_thread_info;
342 typedef int (*tx_exec_func_t)(const struct lu_env *, struct thandle *,
346 tx_exec_func_t exec_fn;
347 tx_exec_func_t undo_fn;
348 struct dt_object *object;
350 struct update_reply *reply;
355 const struct dt_rec *rec;
356 const struct dt_key *key;
371 struct dt_allocation_hint hint;
372 struct dt_object_format dof;
380 struct ost_body *body;
385 #define TX_MAX_OPS 10
386 struct thandle_exec_args {
387 struct thandle *ta_handle;
388 struct dt_device *ta_dev;
390 struct tx_arg ta_args[TX_MAX_OPS];
391 int ta_argno; /* used args */
395 * Common data shared by mdt-level handlers. This is allocated per-thread to
396 * reduce stack consumption.
398 struct mdt_thread_info {
401 * The following members will be filled explicitly
402 * with specific data in mdt_thread_info_init().
404 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
405 * request handling may migrate from one server thread to another.
407 struct req_capsule *mti_pill;
409 /* although we have export in req, there are cases when it is not
410 * available, e.g. closing files upon export destroy */
411 struct obd_export *mti_exp;
413 * A couple of lock handles.
415 struct mdt_lock_handle mti_lh[MDT_LH_NR];
417 struct mdt_device *mti_mdt;
418 const struct lu_env *mti_env;
419 /* XXX: temporary flag to have healthy mti during OUT calls
420 * to be removed upon moving MDT to the unified target code */
424 * Additional fail id that can be set by handler. Passed to
425 * target_send_reply().
429 /* transaction number of current request */
435 * The following members will be filled expilictly
436 * with zero in mdt_thread_info_init(). These members may be used
443 struct md_attr mti_attr;
445 * Body for "habeo corpus" operations.
447 const struct mdt_body *mti_body;
449 * Host object. This is released at the end of mdt_handler().
451 struct mdt_object *mti_object;
453 * Lock request for "habeo clavis" operations.
455 const struct ldlm_request *mti_dlm_req;
457 __u32 mti_has_trans:1, /* has txn already? */
460 /* opdata for mdt_reint_open(), has the same as
461 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
462 * value onto disk for recovery when mdt_trans_stop_cb() is called.
468 * The following members will be filled explicitly
469 * with zero in mdt_reint_unpack(), because they are only used
470 * by reint requests (including mdt_reint_open()).
474 * reint record. contains information for reint operations.
476 struct mdt_reint_record mti_rr;
478 /** md objects included in operation */
479 struct mdt_object *mti_mos;
480 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
482 * Operation specification (currently create and lookup)
484 struct md_op_spec mti_spec;
488 * The following members will _NOT_ be initialized at all.
489 * DO NOT expect them to contain any valid value.
490 * They should be initialized explicitly by the user themselves.
493 /* XXX: If something is in a union, make sure they do not conflict */
495 struct lu_fid mti_tmp_fid1;
496 struct lu_fid mti_tmp_fid2;
497 ldlm_policy_data_t mti_policy; /* for mdt_object_lock() and
498 * mdt_rename_lock() */
499 struct ldlm_res_id mti_res_id; /* for mdt_object_lock() and
502 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
503 char ns_name[48]; /* for mdt_init0() */
504 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
505 struct obd_statfs osfs; /* for mdt_statfs() */
507 /* for mdt_readpage() */
508 struct lu_rdpg mti_rdpg;
509 /* for mdt_sendpage() */
510 struct l_wait_info mti_wait_info;
514 struct md_som_data data;
517 struct dt_object_format mti_update_dof;
518 struct update_reply *mti_update_reply;
519 struct update *mti_update;
520 int mti_update_reply_index;
521 struct obdo mti_obdo;
522 struct dt_object *mti_dt_object;
526 /* IO epoch related stuff. */
527 struct mdt_ioepoch *mti_ioepoch;
528 __u64 mti_replayepoch;
531 struct lu_buf mti_buf;
532 struct lu_buf mti_big_buf;
533 struct lustre_capa_key mti_capa_key;
535 /* Ops object filename */
536 struct lu_name mti_name;
537 /* per-thread values, can be re-used */
540 /* big_lmm buffer was used and must be used in reply */
541 int mti_big_lmm_used;
542 /* should be enough to fit lustre_mdt_attrs */
543 char mti_xattr_buf[128];
544 struct thandle_exec_args mti_handle;
545 struct ldlm_enqueue_info mti_einfo;
548 /* ptlrpc request handler for MDT. All handlers are
549 * grouped into several slices - struct mdt_opc_slice,
550 * and stored in an array - mdt_handlers[].
553 /* The name of this handler. */
555 /* Fail id for this handler, checked at the beginning of this handler*/
557 /* Operation code for this handler */
559 /* flags are listed in enum mdt_handler_flags below. */
561 /* The actual handler function to execute. */
562 int (*mh_act)(struct mdt_thread_info *info);
563 /* Request format for this request. */
564 const struct req_format *mh_fmt;
567 struct mdt_opc_slice {
570 struct mdt_handler *mos_hs;
573 struct cdt_req_progress {
574 struct mutex crp_lock; /**< protect tree */
575 struct interval_node *crp_root; /**< tree to track extent
577 struct interval_node **crp_node; /**< buffer for tree nodes
578 * vector of fixed size
580 int crp_cnt; /**< # of used nodes */
581 int crp_max; /**< # of allocated nodes */
584 struct cdt_agent_req {
585 cfs_list_t car_request_list; /**< to chain all the req. */
586 atomic_t car_refcount; /**< reference counter */
587 __u64 car_compound_id; /**< compound id */
588 __u64 car_flags; /**< request original flags */
589 struct obd_uuid car_uuid; /**< agent doing the req. */
590 __u32 car_archive_id; /**< archive id */
591 int car_canceled; /**< request was canceled */
592 cfs_time_t car_req_start; /**< start time */
593 cfs_time_t car_req_update; /**< last update time */
594 struct hsm_action_item *car_hai; /**< req. to the agent */
595 struct cdt_req_progress car_progress; /**< track data mvt
598 extern struct kmem_cache *mdt_hsm_car_kmem;
601 cfs_list_t ha_list; /**< to chain the agents */
602 struct obd_uuid ha_uuid; /**< agent uuid */
603 __u32 *ha_archive_id; /**< archive id */
604 int ha_archive_cnt; /**< number of archive entries
605 * 0 means any archive */
606 atomic_t ha_requests; /**< current request count */
607 atomic_t ha_success; /**< number of successful
609 atomic_t ha_failure; /**< number of failed actions */
612 struct cdt_restore_handle {
613 cfs_list_t crh_list; /**< to chain the handle */
614 struct lu_fid crh_fid; /**< fid of the object */
615 struct ldlm_extent crh_extent; /**< extent of the restore */
616 struct mdt_lock_handle crh_lh; /**< lock handle */
618 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
620 static inline const struct md_device_operations *
621 mdt_child_ops(struct mdt_device * m)
623 LASSERT(m->mdt_child);
624 return m->mdt_child->md_ops;
627 static inline struct md_object *mdt_object_child(struct mdt_object *o)
630 return lu2md(lu_object_next(&o->mot_obj));
633 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
635 return info->mti_pill ? info->mti_pill->rc_req : NULL;
638 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
640 LASSERT(info->mti_exp);
641 return exp_connect_flags(info->mti_exp);
644 static inline void mdt_object_get(const struct lu_env *env,
645 struct mdt_object *o)
648 lu_object_get(&o->mot_obj);
652 static inline void mdt_object_put(const struct lu_env *env,
653 struct mdt_object *o)
656 lu_object_put(env, &o->mot_obj);
660 static inline int mdt_object_exists(const struct mdt_object *o)
662 return lu_object_exists(&o->mot_obj);
665 static inline int mdt_object_remote(const struct mdt_object *o)
667 return lu_object_remote(&o->mot_obj);
670 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
672 return lu_object_fid(&o->mot_obj);
675 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
677 return mdt->mdt_lu_dev.ld_site;
680 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
682 return &mdt->mdt_seq_site;
685 static inline void mdt_export_evict(struct obd_export *exp)
687 class_fail_export(exp);
688 class_export_put(exp);
691 /* Here we use LVB_TYPE to check dne client, because it is
692 * also landed on 2.4. */
693 static inline int mdt_is_dne_client(struct obd_export *exp)
695 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
698 int mdt_get_disposition(struct ldlm_reply *rep, int flag);
699 void mdt_set_disposition(struct mdt_thread_info *info,
700 struct ldlm_reply *rep, int flag);
701 void mdt_clear_disposition(struct mdt_thread_info *info,
702 struct ldlm_reply *rep, int flag);
704 void mdt_lock_pdo_init(struct mdt_lock_handle *lh,
705 ldlm_mode_t lm, const char *name,
708 void mdt_lock_reg_init(struct mdt_lock_handle *lh,
711 int mdt_lock_setup(struct mdt_thread_info *info,
712 struct mdt_object *o,
713 struct mdt_lock_handle *lh);
715 int mdt_check_resent_lock(struct mdt_thread_info *info,
716 struct mdt_object *mo,
717 struct mdt_lock_handle *lhc);
719 int mdt_object_lock(struct mdt_thread_info *,
721 struct mdt_lock_handle *,
724 int mdt_object_lock_try(struct mdt_thread_info *,
726 struct mdt_lock_handle *,
729 void mdt_object_unlock(struct mdt_thread_info *,
731 struct mdt_lock_handle *,
734 struct mdt_object *mdt_object_new(const struct lu_env *,
736 const struct lu_fid *);
737 struct mdt_object *mdt_object_find(const struct lu_env *,
739 const struct lu_fid *);
740 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *,
741 const struct lu_fid *,
742 struct mdt_lock_handle *,
744 void mdt_object_unlock_put(struct mdt_thread_info *,
746 struct mdt_lock_handle *,
749 void mdt_client_compatibility(struct mdt_thread_info *info);
751 int mdt_remote_object_lock(struct mdt_thread_info *mti,
752 struct mdt_object *o, struct lustre_handle *lh,
753 ldlm_mode_t mode, __u64 ibits);
754 int mdt_close_unpack(struct mdt_thread_info *info);
755 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
756 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
757 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
758 const struct lu_attr *attr, const struct lu_fid *fid);
760 int mdt_getxattr(struct mdt_thread_info *info);
761 int mdt_reint_setxattr(struct mdt_thread_info *info,
762 struct mdt_lock_handle *lh);
764 void mdt_lock_handle_init(struct mdt_lock_handle *lh);
765 void mdt_lock_handle_fini(struct mdt_lock_handle *lh);
767 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
768 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
769 struct mdt_lock_handle *lhc);
771 extern void target_recovery_fini(struct obd_device *obd);
772 extern void target_recovery_init(struct lu_target *lut,
773 svc_handler_t handler);
774 int mdt_fs_setup(const struct lu_env *, struct mdt_device *,
775 struct obd_device *, struct lustre_sb_info *lsi);
776 void mdt_fs_cleanup(const struct lu_env *, struct mdt_device *);
778 int mdt_export_stats_init(struct obd_device *obd,
779 struct obd_export *exp,
782 int mdt_pin(struct mdt_thread_info* info);
784 int mdt_lock_new_child(struct mdt_thread_info *info,
785 struct mdt_object *o,
786 struct mdt_lock_handle *child_lockh);
788 void mdt_mfd_set_mode(struct mdt_file_data *mfd,
791 int mdt_reint_open(struct mdt_thread_info *info,
792 struct mdt_lock_handle *lhc);
794 struct mdt_file_data *mdt_handle2mfd(struct mdt_export_data *med,
795 const struct lustre_handle *handle,
799 MDT_IOEPOCH_CLOSED = 0,
800 MDT_IOEPOCH_OPENED = 1,
801 MDT_IOEPOCH_GETATTR = 2,
809 int mdt_attr_get_complex(struct mdt_thread_info *info,
810 struct mdt_object *o, struct md_attr *ma);
811 int mdt_ioepoch_open(struct mdt_thread_info *info, struct mdt_object *o,
813 int mdt_object_is_som_enabled(struct mdt_object *mo);
814 int mdt_write_get(struct mdt_object *o);
815 void mdt_write_put(struct mdt_object *o);
816 int mdt_write_read(struct mdt_object *o);
817 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
818 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
819 void mdt_mfd_free(struct mdt_file_data *mfd);
820 int mdt_close(struct mdt_thread_info *info);
821 int mdt_attr_set(struct mdt_thread_info *info, struct mdt_object *mo,
822 struct md_attr *ma, int flags);
823 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
825 int mdt_done_writing(struct mdt_thread_info *info);
826 int mdt_fix_reply(struct mdt_thread_info *info);
827 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
828 const struct md_attr *);
829 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
831 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
832 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
833 const void *area, ssize_t len);
835 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
837 int mdt_check_ucred(struct mdt_thread_info *);
838 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
839 int mdt_init_ucred_reint(struct mdt_thread_info *);
840 void mdt_exit_ucred(struct mdt_thread_info *);
841 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
842 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
843 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
845 int mdt_handle_common(struct ptlrpc_request *req,
846 struct mdt_opc_slice *supported);
847 int mdt_connect(struct mdt_thread_info *info);
848 int mdt_disconnect(struct mdt_thread_info *info);
849 int mdt_set_info(struct mdt_thread_info *info);
850 int mdt_get_info(struct mdt_thread_info *info);
851 int mdt_getstatus(struct mdt_thread_info *info);
852 int mdt_getattr(struct mdt_thread_info *info);
853 int mdt_getattr_name(struct mdt_thread_info *info);
854 int mdt_statfs(struct mdt_thread_info *info);
855 int mdt_reint(struct mdt_thread_info *info);
856 int mdt_sync(struct mdt_thread_info *info);
857 int mdt_is_subdir(struct mdt_thread_info *info);
858 int mdt_obd_ping(struct mdt_thread_info *info);
859 int mdt_obd_log_cancel(struct mdt_thread_info *info);
860 int mdt_obd_qc_callback(struct mdt_thread_info *info);
861 int mdt_enqueue(struct mdt_thread_info *info);
862 int mdt_convert(struct mdt_thread_info *info);
863 int mdt_bl_callback(struct mdt_thread_info *info);
864 int mdt_cp_callback(struct mdt_thread_info *info);
865 int mdt_llog_create(struct mdt_thread_info *info);
866 int mdt_llog_destroy(struct mdt_thread_info *info);
867 int mdt_llog_read_header(struct mdt_thread_info *info);
868 int mdt_llog_next_block(struct mdt_thread_info *info);
869 int mdt_llog_prev_block(struct mdt_thread_info *info);
870 int mdt_sec_ctx_handle(struct mdt_thread_info *info);
871 int mdt_readpage(struct mdt_thread_info *info);
872 int mdt_obd_idx_read(struct mdt_thread_info *info);
873 int mdt_tgt_connect(struct tgt_session_info *tsi);
874 void mdt_thread_info_init(struct ptlrpc_request *req,
875 struct mdt_thread_info *mti);
876 void mdt_thread_info_fini(struct mdt_thread_info *mti);
878 extern struct mdt_opc_slice mdt_regular_handlers[];
879 extern struct mdt_opc_slice mdt_seq_handlers[];
880 extern struct mdt_opc_slice mdt_fld_handlers[];
882 int mdt_quotacheck(struct mdt_thread_info *info);
883 int mdt_quotactl(struct mdt_thread_info *info);
884 int mdt_quota_dqacq(struct mdt_thread_info *info);
885 int mdt_swap_layouts(struct mdt_thread_info *info);
887 extern struct lprocfs_vars lprocfs_mds_module_vars[];
888 extern struct lprocfs_vars lprocfs_mds_obd_vars[];
890 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
891 const struct md_hsm *mh);
893 struct mdt_handler *mdt_handler_find(__u32 opc,
894 struct mdt_opc_slice *supported);
896 int mdt_init_sec_level(struct mdt_thread_info *);
897 int mdt_init_idmap(struct mdt_thread_info *);
898 void mdt_cleanup_idmap(struct mdt_export_data *);
899 int mdt_handle_idmap(struct mdt_thread_info *);
900 int ptlrpc_user_desc_do_idmap(struct ptlrpc_request *,
901 struct ptlrpc_user_desc *);
902 void mdt_body_reverse_idmap(struct mdt_thread_info *,
904 int mdt_remote_perm_reverse_idmap(struct ptlrpc_request *,
905 struct mdt_remote_perm *);
906 int mdt_fix_attr_ucred(struct mdt_thread_info *, __u32);
908 static inline struct mdt_device *mdt_dev(struct lu_device *d)
910 return container_of0(d, struct mdt_device, mdt_lu_dev);
913 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
915 struct lu_object *lo;
916 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
918 lo = lu_object_locate(mo->mot_obj.lo_header,
919 mdt->mdt_bottom->dd_lu_dev.ld_type);
924 /* mdt/mdt_identity.c */
925 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
927 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
929 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
931 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
933 void mdt_flush_identity(struct upcall_cache *, int);
935 __u32 mdt_identity_get_perm(struct md_identity *, __u32, lnet_nid_t);
937 int mdt_pack_remote_perm(struct mdt_thread_info *, struct mdt_object *, void *);
940 int mdt_hsm_state_get(struct mdt_thread_info *info);
941 int mdt_hsm_state_set(struct mdt_thread_info *info);
942 int mdt_hsm_action(struct mdt_thread_info *info);
943 int mdt_hsm_progress(struct mdt_thread_info *info);
944 int mdt_hsm_ct_register(struct mdt_thread_info *info);
945 int mdt_hsm_ct_unregister(struct mdt_thread_info *info);
946 int mdt_hsm_request(struct mdt_thread_info *info);
947 /* mdt/mdt_hsm_cdt_actions.c */
948 extern const struct file_operations mdt_hsm_actions_fops;
949 void dump_llog_agent_req_rec(const char *prefix,
950 const struct llog_agent_req_rec *larr);
951 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
952 llog_cb_t cb, void *data);
953 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
954 __u64 compound_id, __u32 archive_id,
955 __u64 flags, struct hsm_action_item *hai);
956 int mdt_agent_record_update(const struct lu_env *env,
957 struct mdt_device *mdt, __u64 *cookies,
958 int cookies_count, enum agent_req_status status);
959 int mdt_agent_llog_update_rec(const struct lu_env *env, struct mdt_device *mdt,
960 struct llog_handle *llh,
961 struct llog_agent_req_rec *larr);
963 /* mdt/mdt_hsm_cdt_agent.c */
964 extern const struct file_operations mdt_hsm_agent_fops;
965 int mdt_hsm_agent_register(struct mdt_thread_info *info,
966 const struct obd_uuid *uuid,
967 int nr_archives, __u32 *archive_num);
968 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
969 const struct obd_uuid *uuid,
971 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
972 const struct obd_uuid *uuid);
973 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
974 int succ_rq, int fail_rq, int new_rq,
975 const struct obd_uuid *uuid);
976 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
977 struct obd_uuid *uuid);
978 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
980 int mdt_hsm_coordinator_update(struct mdt_thread_info *mti,
981 struct hsm_progress_kernel *pgs);
982 /* mdt/mdt_hsm_cdt_client.c */
983 int mdt_hsm_add_actions(struct mdt_thread_info *info,
984 struct hsm_action_list *hal, __u64 *compound_id);
985 int mdt_hsm_get_actions(struct mdt_thread_info *mti,
986 struct hsm_action_list *hal);
987 int mdt_hsm_get_running(struct mdt_thread_info *mti,
988 struct hsm_action_list *hal);
989 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
990 const struct lu_fid *fid);
991 /* mdt/mdt_hsm_cdt_requests.c */
992 extern const struct file_operations mdt_hsm_active_requests_fops;
993 void dump_requests(char *prefix, struct coordinator *cdt);
994 struct cdt_agent_req *mdt_cdt_alloc_request(__u64 compound_id, __u32 archive_id,
995 __u64 flags, struct obd_uuid *uuid,
996 struct hsm_action_item *hai);
997 void mdt_cdt_free_request(struct cdt_agent_req *car);
998 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
999 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt,
1001 const struct lu_fid *fid);
1002 void mdt_cdt_get_work_done(struct cdt_agent_req *car, __u64 *done_sz);
1003 void mdt_cdt_get_request(struct cdt_agent_req *car);
1004 void mdt_cdt_put_request(struct cdt_agent_req *car);
1005 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
1006 const struct hsm_progress_kernel *pgs);
1007 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
1008 /* mdt/mdt_coordinator.c */
1009 void mdt_hsm_dump_hal(int level, const char *prefix,
1010 struct hsm_action_list *hal);
1011 /* coordinator management */
1012 int mdt_hsm_cdt_init(struct mdt_device *mdt);
1013 int mdt_hsm_cdt_start(struct mdt_device *mdt);
1014 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
1015 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
1016 int mdt_hsm_cdt_wakeup(struct mdt_device *mdt);
1018 /* coordinator control /proc interface */
1019 int lprocfs_wr_hsm_cdt_control(struct file *file, const char *buffer,
1020 unsigned long count, void *data);
1021 int lprocfs_rd_hsm_cdt_control(char *page, char **start, off_t off,
1022 int count, int *eof, void *data);
1023 int hsm_cdt_procfs_init(struct mdt_device *mdt);
1024 void hsm_cdt_procfs_fini(struct mdt_device *mdt);
1025 struct lprocfs_vars *hsm_cdt_get_proc_vars(void);
1026 /* md_hsm helpers */
1027 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1028 const struct lu_fid *fid,
1029 struct md_hsm *hsm);
1030 /* actions/request helpers */
1031 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1032 struct hsm_action_list *hal, struct obd_uuid *uuid);
1033 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1034 const int hal_an, const __u64 rq_flags,
1035 const struct md_hsm *hsm);
1036 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1037 struct hsm_progress_kernel *pgs,
1038 const int update_record);
1040 extern struct lu_context_key mdt_thread_key;
1041 /* debug issues helper starts here*/
1042 static inline int mdt_fail_write(const struct lu_env *env,
1043 struct dt_device *dd, int id)
1045 if (OBD_FAIL_CHECK_ORSET(id, OBD_FAIL_ONCE)) {
1046 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1048 return dd->dd_ops->dt_ro(env, dd);
1049 /* We set FAIL_ONCE because we never "un-fail" a device */
1055 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1057 return &req->rq_export->exp_mdt_data;
1060 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1061 struct mdt_lock_handle *lhc);
1062 static inline int mdt_check_resent(struct mdt_thread_info *info,
1063 mdt_reconstruct_t reconstruct,
1064 struct mdt_lock_handle *lhc)
1066 struct ptlrpc_request *req = mdt_info_req(info);
1069 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1070 if (req_xid_is_last(req)) {
1071 reconstruct(info, lhc);
1074 DEBUG_REQ(D_HA, req, "no reply for RESENT req (have "LPD64")",
1075 req->rq_export->exp_target_data.ted_lcd->lcd_last_xid);
1080 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1081 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1083 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1085 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1088 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1090 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1091 static inline int mdt_fid_lock(struct ldlm_namespace *ns,
1092 struct lustre_handle *lh,
1094 ldlm_policy_data_t *policy,
1095 const struct ldlm_res_id *res_id,
1096 __u64 flags, const __u64 *client_cookie)
1100 LASSERT(ns != NULL);
1101 LASSERT(lh != NULL);
1103 rc = ldlm_cli_enqueue_local(ns, res_id, LDLM_IBITS, policy,
1104 mode, &flags, mdt_blocking_ast,
1105 ldlm_completion_ast, NULL, NULL, 0,
1106 LVB_T_NONE, client_cookie, lh);
1107 return rc == ELDLM_OK ? 0 : -EIO;
1110 static inline void mdt_fid_unlock(struct lustre_handle *lh,
1113 ldlm_lock_decref(lh, mode);
1116 extern mdl_mode_t mdt_mdl_lock_modes[];
1117 extern ldlm_mode_t mdt_dlm_lock_modes[];
1119 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(ldlm_mode_t mode)
1121 LASSERT(IS_PO2(mode));
1122 return mdt_mdl_lock_modes[mode];
1125 static inline ldlm_mode_t mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1127 LASSERT(IS_PO2(mode));
1128 return mdt_dlm_lock_modes[mode];
1132 extern struct ldlm_valblock_ops mdt_lvbo;
1134 static inline struct lu_name *mdt_name(const struct lu_env *env,
1135 char *name, int namelen)
1137 struct lu_name *lname;
1138 struct mdt_thread_info *mti;
1140 LASSERT(namelen > 0);
1141 /* trailing '\0' in buffer */
1142 LASSERT(name[namelen] == '\0');
1144 mti = lu_context_key_get(&env->le_ctx, &mdt_thread_key);
1145 lname = &mti->mti_name;
1146 lname->ln_name = name;
1147 lname->ln_namelen = namelen;
1151 static inline struct lu_name *mdt_name_copy(struct lu_name *tlname,
1152 struct lu_name *slname)
1157 tlname->ln_name = slname->ln_name;
1158 tlname->ln_namelen = slname->ln_namelen;
1162 void mdt_enable_cos(struct mdt_device *, int);
1163 int mdt_cos_is_enabled(struct mdt_device *);
1181 LPROC_MDT_SAMEDIR_RENAME,
1182 LPROC_MDT_CROSSDIR_RENAME,
1185 void mdt_counter_incr(struct ptlrpc_request *req, int opcode);
1186 void mdt_stats_counter_init(struct lprocfs_stats *stats);
1187 void lprocfs_mdt_init_vars(struct lprocfs_static_vars *lvars);
1188 void lprocfs_mds_init_vars(struct lprocfs_static_vars *lvars);
1189 int mdt_procfs_init(struct mdt_device *mdt, const char *name);
1190 void mdt_procfs_fini(struct mdt_device *mdt);
1192 /* lustre/mdt_mdt_lproc.c */
1193 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1195 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1196 struct mdt_device *mdt,
1197 struct ptlrpc_request *req,
1198 struct mdt_object *src, struct mdt_object *tgt);
1201 int mdt_ck_thread_start(struct mdt_device *mdt);
1202 void mdt_ck_thread_stop(struct mdt_device *mdt);
1203 void mdt_ck_timer_callback(unsigned long castmeharder);
1204 int mdt_capa_keys_init(const struct lu_env *env, struct mdt_device *mdt);
1205 void mdt_set_capainfo(struct mdt_thread_info *info, int offset,
1206 const struct lu_fid *fid, struct lustre_capa *capa);
1207 void mdt_dump_capainfo(struct mdt_thread_info *info);
1209 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1211 return mdt->mdt_lu_dev.ld_obd;
1214 extern const struct lu_device_operations mdt_lu_ops;
1216 static inline int lu_device_is_mdt(struct lu_device *d)
1218 return ergo(d != NULL && d->ld_ops != NULL, d->ld_ops == &mdt_lu_ops);
1221 static inline struct mdt_device *lu2mdt_dev(struct lu_device *d)
1223 LASSERTF(lu_device_is_mdt(d), "It is %s instead of MDT %p %p\n",
1224 d->ld_type->ldt_name, d->ld_ops, &mdt_lu_ops);
1225 return container_of0(d, struct mdt_device, mdt_lu_dev);
1228 static inline char *mdt_obd_name(struct mdt_device *mdt)
1230 return mdt->mdt_lu_dev.ld_obd->obd_name;
1233 int mds_mod_init(void);
1234 void mds_mod_exit(void);
1236 #endif /* __KERNEL__ */