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/
31 * lustre/mdt/mdt_internal.h
33 * Lustre Metadata Target (mdt) request handler
35 * Author: Peter Braam <braam@clusterfs.com>
36 * Author: Andreas Dilger <adilger@clusterfs.com>
37 * Author: Phil Schwan <phil@clusterfs.com>
38 * Author: Mike Shaver <shaver@clusterfs.com>
39 * Author: Nikita Danilov <nikita@clusterfs.com>
40 * Author: Huang Hua <huanghua@clusterfs.com>
43 #ifndef _MDT_INTERNAL_H
44 #define _MDT_INTERNAL_H
46 #include <libcfs/libcfs.h>
47 #include <libcfs/libcfs_hash.h>
48 #include <upcall_cache.h>
49 #include <obd_class.h>
50 #include <lustre_disk.h>
51 #include <lu_target.h>
52 #include <md_object.h>
53 #include <lustre_fid.h>
54 #include <lustre_fld.h>
55 #include <lustre_req_layout.h>
56 #include <lustre_sec.h>
57 #include <lustre_idmap.h>
58 #include <lustre_quota.h>
59 #include <lustre_linkea.h>
60 #include <lustre_lmv.h>
64 /* file data for open files on MDS */
65 struct mdt_file_data {
66 /** portals handle must be first */
67 struct portals_handle mfd_open_handle;
68 /* export data of portals_handle */
69 const struct mdt_export_data *mfd_owner;
70 /** open mode provided by client */
72 /** protected by med_open_lock */
73 struct list_head mfd_list;
74 /** xid of the open request */
76 /** old handle in replay case */
77 struct lustre_handle mfd_open_handle_old;
78 /** point to opened object */
79 struct mdt_object *mfd_object;
82 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
83 #define CDT_NORETRY_ACTION (1ULL << 1)
84 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
85 #define CDT_POLICY_SHIFT_COUNT 2
86 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
89 /* when adding a new policy, do not forget to update
90 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
92 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
94 /* Coordinator states. Keep the cdt_transition table in sync. */
95 enum cdt_states { CDT_STOPPED = 0,
104 static inline char *cdt_mdt_state2str(int state)
122 /* when multiple lock are needed, the lock order is
129 refcount_t cdt_ref; /**< cdt refcount */
130 wait_queue_head_t cdt_waitq; /**< cdt wait queue */
131 bool cdt_event; /**< coordinator event */
132 struct task_struct *cdt_task; /**< cdt thread handle */
133 struct lu_env cdt_env; /**< coordinator lustre
135 struct lu_context cdt_session; /** session for lu_ucred */
136 struct dentry *cdt_debugfs_dir; /**< cdt debugfs directory */
137 struct completion cdt_kobj_unregister;
138 struct kobject cdt_hsm_kobj; /* hsm sysfs object */
139 __u64 cdt_policy; /**< policy flags */
140 enum cdt_states cdt_state; /**< state */
141 struct mutex cdt_state_lock; /**< cdt_state lock */
142 __u64 cdt_last_cookie; /**< last cookie
144 struct rw_semaphore cdt_llog_lock; /**< protect llog
146 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
147 struct rw_semaphore cdt_request_lock; /**< protect request
149 timeout_t cdt_loop_period; /**< llog scan period */
150 timeout_t cdt_grace_delay; /**< request grace
152 timeout_t cdt_active_req_timeout; /**< request timeout */
153 __u32 cdt_default_archive_id; /**< archive id used
156 u64 cdt_max_requests; /**< max count of started
158 /** Current count of active requests */
159 atomic_t cdt_request_count; /** total */
160 atomic_t cdt_archive_count;
161 atomic_t cdt_restore_count;
162 atomic_t cdt_remove_count;
164 /* started requests (struct cdt_agent_req:car_cookie_hash)
165 * indexed by cookie */
166 struct cfs_hash *cdt_request_cookie_hash;
167 /* started requests (struct cdt_agent_req:car_request_list) */
168 struct list_head cdt_request_list;
169 struct list_head cdt_agents; /**< list of register
171 struct rhashtable cdt_restore_hash; /* rhashtable for
175 /* Hash of cookies to locations of record locations in agent
177 struct cfs_hash *cdt_agent_record_hash;
179 /* Bitmasks indexed by the HSMA_XXX constants. */
180 __u64 cdt_user_request_mask;
181 __u64 cdt_group_request_mask;
182 __u64 cdt_other_request_mask;
184 /* Remove archive on last unlink policy */
185 bool cdt_remove_archive_on_last_unlink;
187 bool cdt_wakeup_coordinator;
190 /* mdt state flag bits */
191 #define MDT_FL_CFGLOG 0
192 #define MDT_FL_SYNCED 1
194 /* possible values for mo_dom_lock */
196 NO_DOM_LOCK_ON_OPEN = 0,
197 TRYLOCK_DOM_ON_OPEN = 1,
198 /* not used anymore, left here for compatibility */
199 ALWAYS_DOM_LOCK_ON_OPEN = 2,
200 NUM_DOM_LOCK_ON_OPEN_MODES
203 struct mdt_statfs_cache {
204 struct obd_statfs msf_osfs;
208 enum mdt_rename_type {
209 RENAME_SAMEDIR_SIZE = 0,
210 RENAME_CROSSDIR_SRC_SIZE,
211 RENAME_CROSSDIR_TGT_SIZE,
215 struct rename_stats {
217 struct obd_histogram rs_hist[RENAME_LAST];
220 /* split directory automatically when sub file count exceeds 50k */
221 #define DIR_SPLIT_COUNT_DEFAULT 50000
223 /* directory auto-split allocate delta new stripes each time */
224 #define DIR_SPLIT_DELTA_DEFAULT 4
226 struct mdt_dir_restriper {
227 struct lu_env mdr_env;
228 struct lu_context mdr_session;
229 struct task_struct *mdr_task;
230 /* lock for below fields */
232 /* auto split when plain dir/shard sub files exceed threshold */
233 u64 mdr_dir_split_count;
234 /* auto split growth delta */
235 u32 mdr_dir_split_delta;
236 /* directories to split */
237 struct list_head mdr_auto_splitting;
238 /* directories under which sub files are migrating */
239 struct list_head mdr_migrating;
240 /* directories waiting to update layout after migration */
241 struct list_head mdr_updating;
242 /* time to update directory layout after migration */
243 time64_t mdr_update_time;
244 /* lum used in split/migrate/layout_change */
245 union lmv_mds_md mdr_lmv;
246 /* page used in readdir */
247 struct page *mdr_page;
252 struct lu_device mdt_lu_dev;
253 struct seq_server_site mdt_seq_site;
254 /* DLM name-space for meta-data locks maintained by this server */
255 struct ldlm_namespace *mdt_namespace;
256 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
257 struct ptlrpc_client *mdt_ldlm_client;
258 /* underlying device */
259 struct obd_export *mdt_child_exp;
260 struct md_device *mdt_child;
261 struct dt_device *mdt_bottom;
262 struct obd_export *mdt_bottom_exp;
263 struct local_oid_storage *mdt_los;
265 struct lu_target mdt_lut;
267 * Options bit-fields.
270 unsigned int mo_user_xattr:1,
273 unsigned int mo_dom_lock;
275 /* mdt state flags */
276 unsigned long mdt_state;
278 /* transaction callbacks */
279 struct dt_txn_callback mdt_txn_cb;
281 /* these values should be updated from lov if necessary.
282 * or should be placed somewhere else. */
287 /* Max modify RPCs, replaces max_mod_rpcs_per_client module param */
288 unsigned int mdt_max_mod_rpcs_in_flight;
290 /* preferred BRW size, decided by storage type and capability */
293 struct upcall_cache *mdt_identity_cache;
295 unsigned int mdt_evict_tgt_nids:1,
297 mdt_migrate_hsm_allowed:1,
298 mdt_enable_strict_som:1,
299 /* Enable remote dir on non-MDT0 */
300 mdt_enable_dir_migration:1,
301 mdt_enable_dir_restripe:1,
302 mdt_enable_dir_auto_split:1,
303 mdt_enable_parallel_rename_dir:1,
304 mdt_enable_parallel_rename_file:1,
305 mdt_enable_parallel_rename_crossdir:1,
306 mdt_enable_remote_dir:1,
307 mdt_enable_remote_rename:1,
308 mdt_enable_striped_dir:1,
311 /* dir restripe migrate dirent only */
312 mdt_dir_restripe_nsonly:1,
313 /* this is enabled by default, but once an
314 * old client joins, disable this to handle
315 * inherited default LMV on server.
317 mdt_enable_dmv_implicit_inherit:1,
318 /* set dmv by setxattr, off by default */
319 mdt_enable_dmv_xattr:1;
321 /* user with gid can create remote/striped
322 * dir, and set default dir stripe */
323 gid_t mdt_enable_remote_dir_gid;
324 /* user with this gid can change projid */
325 gid_t mdt_enable_chprojid_gid;
326 kernel_cap_t mdt_enable_cap_mask;
328 /* lock for osfs and md_root */
331 /* statfs optimization: we cache a bit */
332 struct mdt_statfs_cache mdt_sum_osfs;
333 struct mdt_statfs_cache mdt_osfs;
336 struct root_squash_info mdt_squash;
338 struct rename_stats mdt_rename_stats;
339 struct lu_fid mdt_md_root_fid;
341 /* connection to quota master */
342 struct obd_export *mdt_qmt_exp;
343 /* quota master device associated with this MDT */
344 struct lu_device *mdt_qmt_dev;
346 struct coordinator mdt_coordinator;
348 /* inter-MDT connection count */
349 atomic_t mdt_mds_mds_conns;
351 /* MDT device async commit count, used for debug and sanity test */
352 atomic_t mdt_async_commit_count;
354 struct mdt_object *mdt_md_root;
356 struct mdt_dir_restriper mdt_restriper;
358 /* count of old clients that doesn't support DMV implicite inherit */
359 atomic_t mdt_dmv_old_client_count;
361 /* name of xattr used to store jobid in mdt inode */
362 char mdt_job_xattr[XATTR_JOB_MAX_LEN];
365 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
366 #define MDT_COS_DEFAULT (0)
368 #define ENOENT_VERSION 1 /** 'virtual' version of non-existent object */
371 struct lu_object_header mot_header;
372 struct lu_object mot_obj;
373 unsigned int mot_lov_created:1, /* lov object created */
374 mot_cache_attr:1, /* enable remote object
376 mot_restriping:1, /* dir restriping */
377 /* dir auto-split disabled */
378 mot_auto_split_disabled:1,
379 mot_lsom_inited:1; /* lsom was inited */
381 spinlock_t mot_write_lock;
382 /* Lock to protect create_data */
383 struct mutex mot_lov_mutex;
384 /* Lock to protect object's SOM update. */
385 struct mutex mot_som_mutex;
387 __u64 mot_lsom_blocks;
388 /* lock to protect read/write stages for Data-on-MDT files */
389 struct rw_semaphore mot_dom_sem;
390 /* Lock to protect lease open.
391 * Lease open acquires write lock; normal open acquires read lock */
392 struct rw_semaphore mot_open_sem;
393 atomic_t mot_lease_count;
394 atomic_t mot_open_count;
395 /* directory offset, used in sub file migration in dir restripe */
396 loff_t mot_restripe_offset;
397 /* link to mdt_restriper auto_splitting/migrating/updating */
398 struct list_head mot_restripe_linkage;
401 struct mdt_lock_handle {
402 /* Lock type, reg for cross-ref use or pdo lock. */
406 struct lustre_handle mlh_reg_lh;
407 enum ldlm_mode mlh_reg_mode;
411 struct lustre_handle mlh_pdo_lh;
412 enum ldlm_mode mlh_pdo_mode;
413 unsigned int mlh_pdo_hash;
414 unsigned int mlh_pdo_remote:1;
416 /* Remote regular lock */
417 struct lustre_handle mlh_rreg_lh;
418 enum ldlm_mode mlh_rreg_mode;
422 MDT_LH_PARENT, /* parent lockh */
423 MDT_LH_CHILD, /* child lockh */
424 MDT_LH_OLD, /* old lockh for rename */
425 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
426 MDT_LH_NEW, /* new lockh for rename */
427 MDT_LH_RMT, /* used for return lh to caller */
428 MDT_LH_LOCAL, /* local lock never return to client */
429 MDT_LH_LOOKUP, /* lookup lock for source object in rename/migrate if
430 * it's remote object */
439 /* Special magical errno for communicaiton between mdt_reint_open()
440 * and mdt_intent_reint() which means return the lock to the client
441 * for subsequent cross ref open. Previously we used plain -EREMOTE
442 * but other functions called in that path might return it too and
443 * confuse us. This is not returned to the client. See LU-5370. */
444 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
446 struct mdt_reint_record {
447 enum mds_reint_op rr_opcode;
448 const struct lustre_handle *rr_open_handle;
449 const struct lustre_handle *rr_lease_handle;
450 const struct lu_fid *rr_fid1;
451 const struct lu_fid *rr_fid2;
452 struct lu_name rr_name;
453 struct lu_name rr_tgt_name;
460 enum mdt_reint_flag {
461 MRF_OPEN_TRUNC = BIT(0),
462 MRF_OPEN_RESEND = BIT(1),
466 * Common data shared by mdt-level handlers. This is allocated per-thread to
467 * reduce stack consumption.
469 struct mdt_thread_info {
472 * The following members will be filled explicitly
473 * with specific data in mdt_thread_info_init().
475 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
476 * request handling may migrate from one server thread to another.
478 struct req_capsule *mti_pill;
481 * SUB request pill in a batch request.
483 struct req_capsule mti_sub_pill;
485 /* although we have export in req, there are cases when it is not
486 * available, e.g. closing files upon export destroy */
487 struct obd_export *mti_exp;
489 * A couple of lock handles.
491 struct mdt_lock_handle mti_lh[MDT_LH_NR];
493 struct mdt_device *mti_mdt;
494 const struct lu_env *mti_env;
496 /* transaction number of current request */
501 * The following members will be filled expilictly
502 * with zero in mdt_thread_info_init(). These members may be used
509 struct md_attr mti_attr;
510 struct md_attr mti_attr2; /* mdt_lvb.c */
512 * Body for "habeo corpus" operations.
514 const struct mdt_body *mti_body;
516 * Host object. This is released at the end of mdt_handler().
518 struct mdt_object *mti_object;
520 * Lock request for "habeo clavis" operations.
522 const struct ldlm_request *mti_dlm_req;
524 __u32 mti_cross_ref:1,
525 /* big_lmm buffer was used and must be used in reply */
529 /* Batch processing environment */
533 /* opdata for mdt_reint_open(), has the same as
534 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
535 * value onto disk for recovery when mdt_trans_stop_cb() is called.
541 * The following members will be filled explicitly
542 * with zero in mdt_reint_unpack(), because they are only used
543 * by reint requests (including mdt_reint_open()).
547 * reint record. contains information for reint operations.
549 struct mdt_reint_record mti_rr;
551 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
553 * Operation specification (currently create and lookup)
555 struct md_op_spec mti_spec;
559 * The following members will _NOT_ be initialized at all.
560 * DO NOT expect them to contain any valid value.
561 * They should be initialized explicitly by the user themselves.
564 /* XXX: If something is in a union, make sure they do not conflict */
565 struct lu_fid mti_tmp_fid1;
566 struct lu_fid mti_tmp_fid2;
567 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
568 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
570 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
571 char ns_name[48];/* for mdt_init0() */
572 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
573 struct obd_statfs osfs; /* for mdt_statfs() */
575 /* for mdt_readpage() */
576 struct lu_rdpg mti_rdpg;
586 struct lustre_handle mti_open_handle;
588 struct lu_buf mti_buf;
589 struct lu_buf mti_big_buf;
591 /* Ops object filename */
592 struct lu_name mti_name;
593 char mti_filename[NAME_MAX + 1];
594 /* per-thread values, can be re-used, may be vmalloc'd */
599 /* should be enough to fit lustre_mdt_attrs */
600 char mti_xattr_buf[128];
601 struct ldlm_enqueue_info mti_einfo;
602 /* einfo used by mdt_remote_object_lock_try() */
603 struct ldlm_enqueue_info mti_remote_einfo;
604 struct tg_reply_data *mti_reply_data;
606 /* FLR: layout change API */
607 struct md_layout_change mti_mlc;
609 struct lu_seq_range mti_range;
612 extern struct lu_context_key mdt_thread_key;
614 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
616 return lu_env_info(env, &mdt_thread_key);
619 struct cdt_req_progress {
620 spinlock_t crp_lock; /**< protect tree */
621 struct interval_tree_root crp_root; /**< tree to track extent
626 struct cdt_agent_req {
627 struct hlist_node car_cookie_hash; /**< find req by cookie */
628 struct list_head car_request_list; /**< to chain all the req. */
629 struct kref car_refcount; /**< reference counter */
630 __u64 car_flags; /**< request original flags */
631 struct obd_uuid car_uuid; /**< agent doing the req. */
632 __u32 car_archive_id; /**< archive id */
633 int car_canceled; /**< request was canceled */
634 time64_t car_req_start; /**< start time */
635 time64_t car_req_update; /**< last update time */
636 struct hsm_action_item *car_hai; /**< req. to the agent */
637 struct cdt_req_progress car_progress; /**< track data mvt
640 extern struct kmem_cache *mdt_hsm_car_kmem;
643 struct list_head ha_list; /**< to chain the agents */
644 struct obd_uuid ha_uuid; /**< agent uuid */
645 __u32 *ha_archive_id; /**< archive id */
646 int ha_archive_cnt; /**< number of archive entries
647 * 0 means any archive */
648 atomic_t ha_requests; /**< current request count */
649 atomic_t ha_success; /**< number of successful
651 atomic_t ha_failure; /**< number of failed actions */
654 struct cdt_restore_handle {
655 struct rhash_head crh_hash; /**< link to cdt_restore_hash */
656 struct lu_fid crh_fid; /**< fid of the object */
657 struct ldlm_extent crh_extent; /**< extent of the restore */
658 struct mdt_lock_handle crh_lh; /**< lock handle */
659 atomic_t crh_refc; /**< crh ref counter */
660 struct rcu_head crh_rcu; /**< crh rcu head */
662 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
664 struct hsm_record_update {
666 enum agent_req_status status;
669 static inline const struct md_device_operations *
670 mdt_child_ops(struct mdt_device * m)
672 LASSERT(m->mdt_child);
673 return m->mdt_child->md_ops;
676 static inline struct md_object *mdt_object_child(struct mdt_object *o)
679 return lu2md(lu_object_next(&o->mot_obj));
682 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
684 return info->mti_pill ? info->mti_pill->rc_req : NULL;
687 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
689 LASSERT(info->mti_exp);
690 return exp_connect_flags(info->mti_exp);
693 static inline void mdt_object_get(const struct lu_env *env,
694 struct mdt_object *o)
697 lu_object_get(&o->mot_obj);
701 static inline void mdt_object_put(const struct lu_env *env,
702 struct mdt_object *o)
705 lu_object_put(env, &o->mot_obj);
709 static inline int mdt_object_exists(const struct mdt_object *o)
711 return lu_object_exists(&o->mot_obj);
714 static inline int mdt_object_remote(const struct mdt_object *o)
716 return lu_object_remote(&o->mot_obj);
719 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
721 return lu_object_fid(&o->mot_obj);
724 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
726 return mdt->mdt_lu_dev.ld_site;
729 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
731 return &mdt->mdt_seq_site;
734 static inline u32 mdt_node_id(const struct mdt_device *mdt)
736 return mdt->mdt_seq_site.ss_node_id;
739 static inline void mdt_export_evict(struct obd_export *exp)
741 class_fail_export(exp);
744 /* Here we use LVB_TYPE to check dne client, because it is
745 * also landed on 2.4. */
746 static inline bool mdt_is_dne_client(struct obd_export *exp)
748 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
751 static inline bool mdt_is_striped_client(struct obd_export *exp)
753 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
756 __u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *dom_only);
758 static inline bool mdt_lmm_dom_only(struct lov_mds_md *lmm)
762 mdt_lmm_dom_entry_check(lmm, &dom_only);
766 static inline __u32 mdt_lmm_dom_stripesize(struct lov_mds_md *lmm)
768 return mdt_lmm_dom_entry_check(lmm, NULL);
771 static inline bool mdt_lmm_is_flr(struct lov_mds_md *lmm)
773 struct lov_comp_md_v1 *lcm = (typeof(lcm))lmm;
775 return le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1 &&
776 le16_to_cpu(lcm->lcm_mirror_count) > 0;
779 static inline bool lmm_is_overstriping(struct lov_mds_md *lmm)
781 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1 ||
782 le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V3)
783 return le16_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_OVERSTRIPING;
788 static inline bool mdt_lmm_comp_overstriping(struct lov_mds_md *lmm)
790 struct lov_comp_md_v1 *comp_v1;
791 struct lov_mds_md *v1;
795 comp_v1 = (struct lov_comp_md_v1 *)lmm;
797 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
798 off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
799 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
801 if (lmm_is_overstriping(v1))
808 static inline bool mdt_lmm_is_overstriping(struct lov_mds_md *lmm)
810 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1)
811 return mdt_lmm_comp_overstriping(lmm);
813 return lmm_is_overstriping(lmm);
816 static inline bool mdt_is_sum_statfs_client(struct obd_export *exp)
818 return exp_connect_flags(exp) & OBD_CONNECT_FLAGS2 &&
819 exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS;
822 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
823 void mdt_set_disposition(struct mdt_thread_info *info,
824 struct ldlm_reply *rep, __u64 op_flag);
825 void mdt_clear_disposition(struct mdt_thread_info *info,
826 struct ldlm_reply *rep, __u64 op_flag);
828 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
829 const struct lu_name *lname);
831 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
832 void mdt_lh_reg_init(struct mdt_lock_handle *lh, struct ldlm_lock *lock);
834 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
835 struct mdt_lock_handle *lh);
837 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
838 struct mdt_lock_handle *lhc);
840 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *obj,
841 struct mdt_lock_handle *lh, __u64 ibits,
842 enum ldlm_mode mode);
843 int mdt_parent_lock(struct mdt_thread_info *info, struct mdt_object *o,
844 struct mdt_lock_handle *lh, const struct lu_name *lname,
845 enum ldlm_mode mode);
846 int mdt_object_stripes_lock(struct mdt_thread_info *info,
847 struct mdt_object *pobj, struct mdt_object *o,
848 struct mdt_lock_handle *lh,
849 struct ldlm_enqueue_info *einfo, __u64 ibits,
850 enum ldlm_mode mode);
851 int mdt_object_check_lock(struct mdt_thread_info *info,
852 struct mdt_object *parent, struct mdt_object *child,
853 struct mdt_lock_handle *lh, __u64 ibits,
854 enum ldlm_mode mode);
855 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
856 struct mdt_lock_handle *lh, __u64 *ibits,
857 __u64 trybits, enum ldlm_mode mode);
859 /* below three lock functions are used internally */
860 int mdt_object_lock_internal(struct mdt_thread_info *info,
861 struct mdt_object *obj, const struct lu_fid *fid,
862 struct mdt_lock_handle *lh, __u64 *ibits,
863 __u64 trybits, bool cache);
864 int mdt_object_pdo_lock(struct mdt_thread_info *info, struct mdt_object *obj,
865 struct mdt_lock_handle *lh, const struct lu_name *name,
866 enum ldlm_mode mode, bool pdo_lock);
867 int mdt_object_lookup_lock(struct mdt_thread_info *info,
868 struct mdt_object *pobj, struct mdt_object *obj,
869 struct mdt_lock_handle *lh, enum ldlm_mode mode);
871 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
872 struct mdt_lock_handle *lh, int decref);
873 void mdt_object_stripes_unlock(struct mdt_thread_info *info,
874 struct mdt_object *o, struct mdt_lock_handle *lh,
875 struct ldlm_enqueue_info *einfo, int decref);
877 struct mdt_object *mdt_object_new(const struct lu_env *env,
879 const struct lu_fid *);
880 struct mdt_object *mdt_object_find(const struct lu_env *,
882 const struct lu_fid *);
883 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
884 const struct lu_fid *f,
885 struct mdt_lock_handle *lh,
886 __u64 ibits, enum ldlm_mode mode);
887 void mdt_object_unlock_put(struct mdt_thread_info *info,
888 struct mdt_object *o,
889 struct mdt_lock_handle *lh,
892 void mdt_client_compatibility(struct mdt_thread_info *info);
894 enum mdt_name_flags {
898 int mdt_name_unpack(struct req_capsule *pill,
899 const struct req_msg_field *field,
901 enum mdt_name_flags flags);
902 int mdt_close_unpack(struct mdt_thread_info *info);
903 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
904 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata);
905 int mdt_reint_rec(struct mdt_thread_info *, struct mdt_lock_handle *);
906 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
907 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
908 struct mdt_object *o, struct lu_nodemap *nodemap);
910 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
911 const struct lu_attr *attr, const struct lu_fid *fid);
912 int mdt_pack_size2body(struct mdt_thread_info *info,
913 const struct lu_fid *fid, struct lustre_handle *lh);
914 int mdt_getxattr(struct mdt_thread_info *info);
915 int mdt_reint_setxattr(struct mdt_thread_info *info,
916 struct mdt_lock_handle *lh);
918 void mdt_reconstruct(struct mdt_thread_info *, struct mdt_lock_handle *);
919 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
920 struct mdt_lock_handle *lhc);
922 int mdt_export_stats_init(struct obd_device *obd,
923 struct obd_export *exp,
926 int mdt_lock_new_child(struct mdt_thread_info *info,
927 struct mdt_object *o,
928 struct mdt_lock_handle *child_lockh);
929 void mdt_mfd_set_mode(struct mdt_file_data *mfd, u64 open_flags);
930 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc);
931 struct mdt_file_data *mdt_open_handle2mfd(struct mdt_export_data *med,
932 const struct lustre_handle *open_handle,
934 int mdt_revoke_remote_lookup_lock(struct mdt_thread_info *info,
935 struct mdt_object *pobj,
936 struct mdt_object *obj);
938 int mdt_get_info(struct tgt_session_info *tsi);
939 int mdt_attr_get_complex(struct mdt_thread_info *info,
940 struct mdt_object *o, struct md_attr *ma);
941 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
943 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
944 struct md_attr *ma, const char *name);
945 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
946 struct md_attr *ma, const char *name);
947 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
948 struct lu_fid *pfid);
949 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
950 struct lu_fid *pfid, struct lu_name *lname);
951 int mdt_write_get(struct mdt_object *o);
952 void mdt_write_put(struct mdt_object *o);
953 int mdt_write_read(struct mdt_object *o);
954 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
955 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
956 void mdt_mfd_free(struct mdt_file_data *mfd);
957 int mdt_close(struct tgt_session_info *tsi);
958 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
960 int mdt_fix_reply(struct mdt_thread_info *info);
961 int mdt_handle_last_unlink(struct mdt_thread_info *, struct mdt_object *,
963 void mdt_reconstruct_open(struct mdt_thread_info *, struct mdt_lock_handle *);
964 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
965 struct mdt_lock_handle *lhc,
966 struct md_layout_change *spec);
967 int find_name_matching_hash(struct mdt_thread_info *info, struct lu_name *lname,
968 struct mdt_object *parent, struct mdt_object *child);
969 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt);
971 struct lu_buf *mdt_buf(const struct lu_env *env, void *area, ssize_t len);
972 const struct lu_buf *mdt_buf_const(const struct lu_env *env,
973 const void *area, ssize_t len);
975 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
976 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
978 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
979 int mdt_check_ucred(struct mdt_thread_info *);
980 int mdt_init_ucred(struct mdt_thread_info *, struct mdt_body *);
981 int mdt_init_ucred_reint(struct mdt_thread_info *);
982 void mdt_exit_ucred(struct mdt_thread_info *);
983 int mdt_version_get_check(struct mdt_thread_info *, struct mdt_object *, int);
984 void mdt_version_get_save(struct mdt_thread_info *, struct mdt_object *, int);
985 int mdt_version_get_check_save(struct mdt_thread_info *, struct mdt_object *,
987 int mdt_lookup_version_check(struct mdt_thread_info *info,
988 struct mdt_object *p,
989 const struct lu_name *lname,
990 struct lu_fid *fid, int idx);
991 void mdt_thread_info_reset(struct mdt_thread_info *info);
992 void mdt_thread_info_init(struct ptlrpc_request *req,
993 struct mdt_thread_info *mti);
994 void mdt_thread_info_fini(struct mdt_thread_info *mti);
995 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
996 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
997 struct ldlm_lock *new_lock,
998 struct mdt_lock_handle *lh, __u64 flags);
999 int mdt_intent_lock_replace(struct mdt_thread_info *info,
1000 struct ldlm_lock **lockp,
1001 struct mdt_lock_handle *lh,
1002 __u64 flags, int result);
1004 int hsm_init_ucred(struct lu_ucred *uc);
1005 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
1006 const struct md_hsm *mh);
1008 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1009 void *data, int flag);
1010 int mdt_links_read(struct mdt_thread_info *info,
1011 struct mdt_object *mdt_obj,
1012 struct linkea_data *ldata);
1013 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
1014 struct mdt_body *repbody);
1015 int mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
1016 struct mdt_object *child);
1017 int mdt_pack_encctx_in_reply(struct mdt_thread_info *info,
1018 struct mdt_object *child);
1019 void mdt_prep_ma_buf_from_rep(struct mdt_thread_info *info,
1020 struct mdt_object *obj, struct md_attr *ma,
1023 static inline struct mdt_device *mdt_dev(struct lu_device *d)
1025 return container_of_safe(d, struct mdt_device, mdt_lu_dev);
1028 static inline struct mdt_object *mdt_obj(struct lu_object *o)
1030 return container_of_safe(o, struct mdt_object, mot_obj);
1033 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
1035 struct lu_object *lo;
1036 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
1038 lo = lu_object_locate(mo->mot_obj.lo_header,
1039 mdt->mdt_bottom->dd_lu_dev.ld_type);
1044 static inline bool agent_req_in_final_state(enum agent_req_status ars)
1046 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
1047 (ars == ARS_CANCELED));
1050 /* mdt/mdt_identity.c */
1051 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
1052 #define UC_IDCACHE_HASH_SIZE 128
1053 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
1055 struct md_identity *mdt_identity_get(struct upcall_cache *, __u32);
1057 void mdt_identity_put(struct upcall_cache *, struct md_identity *);
1059 void mdt_flush_identity(struct upcall_cache *, int);
1061 __u32 mdt_identity_get_perm(struct md_identity *identity, struct lnet_nid *nid);
1063 /* mdt/mdt_recovery.c */
1064 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
1066 /* mdt/mdt_batch.c */
1067 int mdt_batch(struct tgt_session_info *tsi);
1070 int mdt_hsm_state_get(struct tgt_session_info *tsi);
1071 int mdt_hsm_state_set(struct tgt_session_info *tsi);
1072 int mdt_hsm_action(struct tgt_session_info *tsi);
1073 int mdt_hsm_progress(struct tgt_session_info *tsi);
1074 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
1075 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
1076 int mdt_hsm_request(struct tgt_session_info *tsi);
1078 /* mdt/mdt_hsm_cdt_actions.c */
1079 extern const struct file_operations mdt_hsm_actions_fops;
1080 void dump_llog_agent_req_rec(const char *prefix,
1081 const struct llog_agent_req_rec *larr);
1082 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
1083 llog_cb_t cb, void *data, u32 start_cat_idx,
1084 u32 start_rec_idx, int rw);
1085 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
1086 __u32 archive_id, __u64 flags,
1087 struct hsm_action_item *hai);
1088 int mdt_agent_record_update(struct mdt_thread_info *mti,
1089 struct hsm_record_update *updates,
1090 unsigned int updates_count);
1091 void cdt_agent_record_hash_add(struct coordinator *cdt, u64 cookie, u32 cat_idt,
1093 void cdt_agent_record_hash_lookup(struct coordinator *cdt, u64 cookie,
1094 u32 *cat_idt, u32 *rec_idx);
1095 void cdt_agent_record_hash_del(struct coordinator *cdt, u64 cookie);
1097 /* mdt/mdt_hsm_cdt_agent.c */
1098 extern const struct file_operations mdt_hsm_agent_fops;
1099 int mdt_hsm_agent_register(struct mdt_thread_info *info,
1100 const struct obd_uuid *uuid,
1101 int nr_archives, __u32 *archive_num);
1102 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
1103 const struct obd_uuid *uuid,
1104 __u32 archive_mask);
1105 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
1106 const struct obd_uuid *uuid);
1107 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
1108 int succ_rq, int fail_rq, int new_rq,
1109 const struct obd_uuid *uuid);
1110 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
1111 struct obd_uuid *uuid);
1112 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_action_list *hal,
1114 /* mdt/mdt_hsm_cdt_client.c */
1115 int mdt_hsm_add_actions(struct mdt_thread_info *info,
1116 struct hsm_action_list *hal);
1117 int mdt_hsm_get_action(struct mdt_thread_info *mti,
1118 const struct lu_fid *fid,
1119 enum hsm_copytool_action *action,
1120 enum agent_req_status *status,
1121 struct hsm_extent *extent);
1122 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
1123 const struct lu_fid *fid);
1124 /* mdt/mdt_hsm_cdt_requests.c */
1125 extern struct cfs_hash_ops cdt_request_cookie_hash_ops;
1126 extern struct cfs_hash_ops cdt_agent_record_hash_ops;
1127 extern const struct file_operations mdt_hsm_active_requests_fops;
1128 void dump_requests(char *prefix, struct coordinator *cdt);
1129 struct cdt_agent_req *mdt_cdt_alloc_request(__u32 archive_id, __u64 flags,
1130 struct obd_uuid *uuid,
1131 struct hsm_action_item *hai);
1132 void mdt_cdt_free_request(struct cdt_agent_req *car);
1133 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
1134 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt, u64 cookie);
1135 void mdt_cdt_get_request(struct cdt_agent_req *car);
1136 void mdt_cdt_put_request(struct cdt_agent_req *car);
1137 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
1138 const struct hsm_progress_kernel *pgs);
1139 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
1140 /* mdt/mdt_coordinator.c */
1141 void mdt_hsm_dump_hal(int level, const char *prefix,
1142 struct hsm_action_list *hal);
1143 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
1144 const struct lu_fid *fid,
1145 const struct hsm_extent *he);
1146 bool cdt_restore_handle_exists(struct coordinator *cdt,
1147 const struct lu_fid *fid);
1148 void cdt_restore_handle_del(struct mdt_thread_info *mti,
1149 struct coordinator *cdt, const struct lu_fid *fid);
1150 int cdt_getref_try(struct coordinator *cdt);
1151 void cdt_putref(struct coordinator *cdt);
1152 /* coordinator management */
1153 int mdt_hsm_cdt_init(struct mdt_device *mdt);
1154 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
1155 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
1158 * Signal the coordinator has work to do
1159 * \param cdt [IN] coordinator
1161 static inline void mdt_hsm_cdt_event(struct coordinator *cdt)
1163 cdt->cdt_event = true;
1166 /* coordinator control sysfs interface */
1167 ssize_t hsm_control_show(struct kobject *kobj, struct attribute *attr,
1169 ssize_t hsm_control_store(struct kobject *kobj, struct attribute *attr,
1170 const char *buffer, size_t count);
1171 int hsm_cdt_tunables_init(struct mdt_device *mdt);
1172 void hsm_cdt_tunables_fini(struct mdt_device *mdt);
1173 /* md_hsm helpers */
1174 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1175 const struct lu_fid *fid,
1176 struct md_hsm *hsm);
1177 /* actions/request helpers */
1178 int mdt_hsm_add_hal(struct mdt_thread_info *mti,
1179 struct hsm_action_list *hal, struct obd_uuid *uuid);
1180 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1181 u32 archive_id, u64 rq_flags,
1182 const struct md_hsm *hsm);
1183 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1184 struct hsm_progress_kernel *pgs);
1186 extern struct lu_context_key mdt_thread_key;
1188 /* debug issues helper starts here*/
1189 static inline int mdt_fail_write(const struct lu_env *env,
1190 struct dt_device *dd, int id)
1192 if (CFS_FAIL_CHECK_ORSET(id, CFS_FAIL_ONCE)) {
1193 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1195 return dt_ro(env, dd);
1196 /* We set FAIL_ONCE because we never "un-fail" a device */
1202 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1204 return &req->rq_export->exp_mdt_data;
1207 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
1209 return mdt_dev(exp->exp_obd->obd_lu_dev);
1212 static inline bool mdt_rdonly(struct obd_export *exp)
1214 return (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
1215 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly ||
1216 mdt_exp2dev(exp)->mdt_readonly);
1219 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1220 struct mdt_lock_handle *lhc);
1221 static inline int mdt_check_resent(struct mdt_thread_info *info,
1222 mdt_reconstruct_t reconstruct,
1223 struct mdt_lock_handle *lhc)
1225 struct ptlrpc_request *req = mdt_info_req(info);
1229 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1230 OBD_ALLOC_PTR(info->mti_reply_data);
1231 if (info->mti_reply_data == NULL)
1234 rc = req_can_reconstruct(req, info->mti_reply_data);
1236 reconstruct(info, lhc);
1238 DEBUG_REQ(D_HA, req,
1239 "no reply data found for RESENT req, rc = %d",
1242 OBD_FREE_PTR(info->mti_reply_data);
1243 info->mti_reply_data = NULL;
1248 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1249 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1251 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1253 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1256 int mdt_blocking_ast(struct ldlm_lock*, struct ldlm_lock_desc*, void*, int);
1258 static int mdt_dom_glimpse_ast(struct ldlm_lock *lock, void *reqp)
1260 return -ELDLM_NO_LOCK_DATA;
1263 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1264 static inline int mdt_fid_lock(const struct lu_env *env,
1265 struct ldlm_namespace *ns,
1266 struct lustre_handle *lh, enum ldlm_mode mode,
1267 union ldlm_policy_data *policy,
1268 const struct ldlm_res_id *res_id,
1269 __u64 flags, const __u64 *client_cookie)
1272 bool glimpse = policy->l_inodebits.bits & MDS_INODELOCK_DOM;
1274 LASSERT(ns != NULL);
1275 LASSERT(lh != NULL);
1277 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS, policy,
1278 mode, &flags, mdt_blocking_ast,
1279 ldlm_completion_ast,
1280 glimpse ? mdt_dom_glimpse_ast : NULL,
1281 NULL, 0, LVB_T_NONE, client_cookie, lh);
1282 return rc == ELDLM_OK ? 0 : -EIO;
1285 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1287 ldlm_lock_decref(lh, mode);
1290 /* this is identical to say this is a DNE system */
1291 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
1293 return mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING;
1296 extern mdl_mode_t mdt_mdl_lock_modes[];
1297 extern enum ldlm_mode mdt_dlm_lock_modes[];
1299 /* LCK_MINMODE which is zero returns false for is_power_of_2 */
1301 static inline mdl_mode_t mdt_dlm_mode2mdl_mode(enum ldlm_mode mode)
1303 LASSERT(mode == LCK_MINMODE || is_power_of_2(mode));
1304 return mdt_mdl_lock_modes[mode];
1307 static inline enum ldlm_mode mdt_mdl_mode2dlm_mode(mdl_mode_t mode)
1309 LASSERT(mode == MDL_MINMODE || is_power_of_2(mode));
1310 return mdt_dlm_lock_modes[mode];
1314 int mdt_set_som(struct mdt_thread_info *info, struct mdt_object *obj,
1315 enum lustre_som_flags flag, __u64 size, __u64 blocks);
1316 int mdt_get_som(struct mdt_thread_info *info, struct mdt_object *obj,
1317 struct md_attr *ma);
1318 int mdt_lsom_downgrade(struct mdt_thread_info *info, struct mdt_object *obj);
1319 int mdt_lsom_update(struct mdt_thread_info *info, struct mdt_object *obj,
1323 extern struct ldlm_valblock_ops mdt_lvbo;
1324 int mdt_dom_lvb_is_valid(struct ldlm_resource *res);
1325 int mdt_dom_lvbo_update(struct ldlm_resource *res, struct ldlm_lock *lock,
1326 struct ptlrpc_request *req, bool increase_only);
1328 void mdt_enable_cos(struct mdt_device *dev, bool enable);
1329 int mdt_cos_is_enabled(struct mdt_device *);
1347 LPROC_MDT_RENAME_SAMEDIR,
1348 LPROC_MDT_RENAME_PAR_FILE,
1349 LPROC_MDT_RENAME_PAR_DIR,
1350 LPROC_MDT_RENAME_CROSSDIR,
1353 LPROC_MDT_IO_READ_BYTES,
1354 LPROC_MDT_IO_WRITE_BYTES,
1357 LPROC_MDT_FALLOCATE,
1361 void mdt_counter_incr(struct ptlrpc_request *req, int opcode, long amount);
1362 void mdt_stats_counter_init(struct lprocfs_stats *stats, unsigned int offset,
1363 enum lprocfs_counter_config cntr_umask);
1364 int mdt_tunables_init(struct mdt_device *mdt, const char *name);
1365 void mdt_tunables_fini(struct mdt_device *mdt);
1367 /* lustre/mdt_mdt_lproc.c */
1368 int lprocfs_mdt_open_files_seq_open(struct inode *inode,
1370 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1371 struct mdt_device *mdt,
1372 struct ptlrpc_request *req,
1373 struct mdt_object *src, struct mdt_object *tgt,
1374 enum mdt_stat_idx msi, s64 count);
1376 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1378 return mdt->mdt_lu_dev.ld_obd;
1381 extern const struct lu_device_operations mdt_lu_ops;
1383 static inline char *mdt_obd_name(struct mdt_device *mdt)
1385 return mdt->mdt_lu_dev.ld_obd->obd_name;
1388 int mds_mod_init(void);
1389 void mds_mod_exit(void);
1391 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1393 struct obd_export *exp = req->rq_export;
1396 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1397 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1404 #define VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
1405 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
1407 int mdt_obd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
1408 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1409 struct niobuf_remote *rnb, int *nr_local,
1410 struct niobuf_local *lnb);
1412 int mdt_obd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1413 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1414 struct niobuf_remote *rnb, int npages,
1415 struct niobuf_local *lnb, int old_rc, int nob,
1417 int mdt_punch_hdl(struct tgt_session_info *tsi);
1418 int mdt_fallocate_hdl(struct tgt_session_info *tsi);
1419 int mdt_glimpse_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
1420 struct ldlm_lock **lockp, __u64 flags);
1421 int mdt_brw_enqueue(struct mdt_thread_info *info, struct ldlm_namespace *ns,
1422 struct ldlm_lock **lockp, __u64 flags);
1423 int mdt_dom_read_on_open(struct mdt_thread_info *mti, struct mdt_device *mdt,
1424 struct lustre_handle *lh);
1425 void mdt_dom_discard_data(struct mdt_thread_info *info, struct mdt_object *mo);
1426 int mdt_dom_disk_lvbo_update(const struct lu_env *env, struct mdt_object *mo,
1427 struct ldlm_resource *res, bool increase_only);
1428 void mdt_dom_obj_lvb_update(const struct lu_env *env, struct mdt_object *mo,
1429 struct obdo *oa, bool increase_only);
1430 int mdt_dom_lvb_alloc(struct ldlm_resource *res);
1432 static inline bool mdt_dom_check_for_discard(struct mdt_thread_info *mti,
1433 struct mdt_object *mo)
1435 return lu_object_is_dying(&mo->mot_header) &&
1436 S_ISREG(lu_object_attr(&mo->mot_obj));
1439 int mdt_dom_object_size(const struct lu_env *env, struct mdt_device *mdt,
1440 const struct lu_fid *fid, struct mdt_body *mb,
1442 bool mdt_dom_client_has_lock(struct mdt_thread_info *info,
1443 const struct lu_fid *fid);
1444 void mdt_hp_brw(struct tgt_session_info *tsi);
1445 void mdt_hp_punch(struct tgt_session_info *tsi);
1446 int mdt_data_version_get(struct tgt_session_info *tsi);
1447 int mdt_io_set_info(struct tgt_session_info *tsi);
1450 long mdt_grant_connect(const struct lu_env *env, struct obd_export *exp,
1451 u64 want, bool conservative);
1452 extern struct kmem_cache *ldlm_glimpse_work_kmem;
1454 static inline bool mdt_changelog_allow(struct mdt_thread_info *info)
1456 struct lu_ucred *uc = NULL;
1460 if (info == NULL || info->mti_body == NULL)
1461 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 17, 3, 0)
1462 /* return true in case old client did not send mdt body */
1468 rc = mdt_init_ucred(info, (struct mdt_body *)info->mti_body);
1472 uc = mdt_ucred(info);
1473 is_admin = (uc->uc_uid == 0 && uc->uc_gid == 0 &&
1474 cap_raised(uc->uc_cap, CAP_SYS_ADMIN) &&
1475 uc->uc_rbac_chlg_ops);
1477 mdt_exit_ucred(info);
1482 /* convert a capability into an integer to print or manage more easily */
1483 static inline u64 mdt_cap2num(kernel_cap_t cap)
1485 #ifdef CAP_FOR_EACH_U32
1486 /* kernels before v6.2-13111-gf122a08b197d had a more complex
1487 * kernel_cap_t structure with an array of __u32 values, but this
1488 * was then fixed to have a single __u64 value. There are accessor
1489 * functions for the old kernel_cap_t but since that is now dead code
1490 * it isn't worthwhile to jump through hoops for compatibility for it.
1492 return ((u64)cap.cap[1] << 32) | cap.cap[0];
1498 /* convert an integer into a capabilityt */
1499 static inline kernel_cap_t mdt_num2cap(u64 num)
1503 #ifdef CAP_FOR_EACH_U32
1505 cap.cap[1] = (num >> 32);
1513 /* We forbid operations from encryption-unaware clients if they try to
1514 * manipulate encrypted files/directories.
1516 static inline int mdt_check_enc(struct mdt_thread_info *info,
1517 struct mdt_object *obj)
1519 struct obd_export *exp;
1520 struct md_attr ma = { 0 };
1523 if (!mdt_info_req(info))
1524 /* no req, so cannot tell about client connection flags:
1528 exp = mdt_info_req(info)->rq_export;
1529 if (exp_connect_encrypt(exp))
1530 /* client is encryption aware: fine */
1533 ma.ma_need = MA_INODE;
1534 mdt_attr_get_complex(info, obj, &ma);
1535 if (ma.ma_attr.la_valid & LA_FLAGS &&
1536 ma.ma_attr.la_flags & LUSTRE_ENCRYPT_FL)
1540 CDEBUG(D_SEC, "%s: operation on encrypted "DFID
1541 " from encryption-unaware client %s: %d\n",
1542 mdt_obd_name(info->mti_mdt),
1543 PFID(lu_object_fid(&obj->mot_obj)),
1544 libcfs_nidstr(&exp->exp_connection->c_peer.nid), rc);
1549 int mdt_fids_different_target(struct mdt_thread_info *info,
1550 const struct lu_fid *fid1,
1551 const struct lu_fid *fid2);
1553 int mdt_reint_migrate(struct mdt_thread_info *info,
1554 struct mdt_lock_handle *unused);
1555 int mdt_dir_layout_update(struct mdt_thread_info *info);
1557 /* directory restripe */
1558 int mdt_restripe_internal(struct mdt_thread_info *info,
1559 struct mdt_object *parent,
1560 struct mdt_object *child,
1561 const struct lu_name *lname,
1562 struct lu_fid *tfid,
1563 struct md_op_spec *spec,
1564 struct md_attr *ma);
1565 int mdt_restriper_start(struct mdt_device *mdt);
1566 void mdt_restriper_stop(struct mdt_device *mdt);
1567 void mdt_auto_split_add(struct mdt_thread_info *info, struct mdt_object *o);
1568 void mdt_restripe_migrate_add(struct mdt_thread_info *info,
1569 struct mdt_object *o);
1570 void mdt_restripe_update_add(struct mdt_thread_info *info,
1571 struct mdt_object *o);
1572 int mdt_is_remote_object(struct mdt_thread_info *info,
1573 struct mdt_object *parent,
1574 struct mdt_object *child);
1576 static unsigned int max_mod_rpcs_per_client = OBD_MAX_RIF_DEFAULT;
1577 #if OBD_OCD_VERSION(3, 0, 53, 0) > LUSTRE_VERSION_CODE
1578 static inline bool mdt_max_mod_rpcs_changed(struct mdt_device *mdt)
1580 return max_mod_rpcs_per_client != OBD_MAX_RIF_DEFAULT &&
1581 mdt->mdt_max_mod_rpcs_in_flight == OBD_MAX_RIF_DEFAULT;
1584 #define mdt_max_mod_rpcs_changed(mdt) false
1587 #endif /* _MDT_INTERNAL_H */