1 /* SPDX-License-Identifier: GPL-2.0 */
4 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
5 * Use is subject to license terms.
7 * Copyright (c) 2011, 2017, Intel Corporation.
11 * This file is part of Lustre, http://www.lustre.org/
13 * Lustre Metadata Target (mdt) request handler
15 * Author: Peter Braam <braam@clusterfs.com>
16 * Author: Andreas Dilger <adilger@clusterfs.com>
17 * Author: Phil Schwan <phil@clusterfs.com>
18 * Author: Mike Shaver <shaver@clusterfs.com>
19 * Author: Nikita Danilov <nikita@clusterfs.com>
20 * Author: Huang Hua <huanghua@clusterfs.com>
23 #ifndef _MDT_INTERNAL_H
24 #define _MDT_INTERNAL_H
26 #include <libcfs/libcfs.h>
28 #include <upcall_cache.h>
29 #include <obd_class.h>
30 #include <lustre_disk.h>
31 #include <lu_target.h>
32 #include <md_object.h>
33 #include <lustre_fid.h>
34 #include <lustre_fld.h>
35 #include <lustre_req_layout.h>
36 #include <lustre_sec.h>
37 #include <lustre_idmap.h>
38 #include <lustre_quota.h>
39 #include <lustre_linkea.h>
40 #include <lustre_lmv.h>
44 /* file data for open files on MDS */
45 struct mdt_file_data {
46 /** portals handle must be first */
47 struct portals_handle mfd_open_handle;
48 /* export data of portals_handle */
49 const struct mdt_export_data *mfd_owner;
50 /** open mode provided by client */
51 enum mds_open_flags mfd_open_flags;
52 /** protected by med_open_lock */
53 struct list_head mfd_list;
54 /** xid of the open request */
56 /** old handle in replay case */
57 struct lustre_handle mfd_open_handle_old;
58 /** point to opened object */
59 struct mdt_object *mfd_object;
62 #define CDT_NONBLOCKING_RESTORE (1ULL << 0)
63 #define CDT_NORETRY_ACTION (1ULL << 1)
64 #define CDT_POLICY_LAST CDT_NORETRY_ACTION
65 #define CDT_POLICY_SHIFT_COUNT 2
66 #define CDT_POLICY_ALL (CDT_NONBLOCKING_RESTORE | \
69 /* when adding a new policy, do not forget to update
70 * lustre/mdt/mdt_coordinator.c::hsm_policy_names[]
72 #define CDT_DEFAULT_POLICY CDT_NORETRY_ACTION
74 /* Coordinator states. Keep the cdt_transition table in sync. */
75 enum cdt_states { CDT_STOPPED = 0,
84 static inline char *cdt_mdt_state2str(int state)
102 /* when multiple lock are needed, the lock order is
108 refcount_t cdt_ref; /**< cdt refcount */
109 wait_queue_head_t cdt_waitq; /**< cdt wait queue */
110 wait_queue_head_t cdt_cancel_all; /**< cancel_all wait */
111 bool cdt_event; /**< coordinator event */
112 struct task_struct *cdt_task; /**< cdt thread handle */
113 struct lu_env cdt_env; /**< coordinator lustre
115 struct lu_context cdt_session; /** session for lu_ucred */
116 struct dentry *cdt_debugfs_dir; /**< cdt debugfs directory */
117 struct completion cdt_kobj_unregister;
118 struct kobject cdt_hsm_kobj; /* hsm sysfs object */
119 __u64 cdt_policy; /**< policy flags */
120 enum cdt_states cdt_state; /**< state */
121 struct mutex cdt_state_lock; /**< cdt_state lock */
122 atomic64_t cdt_last_cookie; /**< last cookie
124 struct rw_semaphore cdt_agent_lock; /**< protect agent list */
125 struct rw_semaphore cdt_request_lock; /**< protect request
127 timeout_t cdt_loop_period; /**< llog scan period */
128 timeout_t cdt_grace_delay; /**< request grace
130 timeout_t cdt_active_req_timeout; /**< request timeout */
131 __u32 cdt_default_archive_id; /**< archive id used
134 u64 cdt_max_requests; /**< max count of started
136 /** Current count of active requests */
137 atomic_t cdt_request_count; /** total */
138 atomic_t cdt_archive_count;
139 atomic_t cdt_restore_count;
140 atomic_t cdt_remove_count;
142 /* started requests (struct cdt_agent_req:car_cookie_hash)
145 struct cfs_hash *cdt_request_cookie_hash;
146 /* started requests (struct cdt_agent_req:car_request_list) */
147 struct list_head cdt_request_list;
148 struct list_head cdt_agents; /**< list of register
150 struct rhashtable cdt_restore_hash; /* rhashtable for
154 /* Bitmasks indexed by the HSMA_XXX constants. */
155 __u64 cdt_user_request_mask;
156 __u64 cdt_group_request_mask;
157 __u64 cdt_other_request_mask;
159 /* Remove archive on last unlink policy */
160 bool cdt_remove_archive_on_last_unlink;
162 bool cdt_wakeup_coordinator;
166 /* mdt state flag bits */
167 #define MDT_FL_CFGLOG 0
168 #define MDT_FL_SYNCED 1
170 /* possible values for mo_dom_lock */
172 NO_DOM_LOCK_ON_OPEN = 0,
173 TRYLOCK_DOM_ON_OPEN = 1,
174 /* not used anymore, left here for compatibility */
175 ALWAYS_DOM_LOCK_ON_OPEN = 2,
176 NUM_DOM_LOCK_ON_OPEN_MODES
179 struct mdt_statfs_cache {
180 struct obd_statfs msf_osfs;
184 enum mdt_rename_type {
185 RENAME_SAMEDIR_SIZE = 0,
186 RENAME_CROSSDIR_SRC_SIZE,
187 RENAME_CROSSDIR_TGT_SIZE,
191 struct rename_stats {
193 struct obd_histogram rs_hist[RENAME_LAST];
196 /* split directory automatically when sub file count exceeds 50k */
197 #define DIR_SPLIT_COUNT_DEFAULT 50000
199 /* directory auto-split allocate delta new stripes each time */
200 #define DIR_SPLIT_DELTA_DEFAULT 4
202 struct mdt_dir_restriper {
203 struct lu_env mdr_env;
204 struct lu_context mdr_session;
205 struct task_struct *mdr_task;
206 /* lock for below fields */
208 /* auto split when plain dir/shard sub files exceed threshold */
209 u64 mdr_dir_split_count;
210 /* auto split growth delta */
211 u32 mdr_dir_split_delta;
212 /* directories to split */
213 struct list_head mdr_auto_splitting;
214 /* directories under which sub files are migrating */
215 struct list_head mdr_migrating;
216 /* directories waiting to update layout after migration */
217 struct list_head mdr_updating;
218 /* time to update directory layout after migration */
219 time64_t mdr_update_time;
220 /* lum used in split/migrate/layout_change */
221 union lmv_mds_md mdr_lmv;
222 /* page used in readdir */
223 struct page *mdr_page;
228 struct lu_device mdt_lu_dev;
229 struct seq_server_site mdt_seq_site;
230 /* DLM name-space for meta-data locks maintained by this server */
231 struct ldlm_namespace *mdt_namespace;
232 /* ptlrpc handle for MDS->client connections (for lock ASTs). */
233 struct ptlrpc_client *mdt_ldlm_client;
234 /* underlying device */
235 struct obd_export *mdt_child_exp;
236 struct md_device *mdt_child;
237 struct dt_device *mdt_bottom;
238 struct obd_export *mdt_bottom_exp;
239 struct local_oid_storage *mdt_los;
241 struct lu_target mdt_lut;
242 /* Options bit-fields. */
244 unsigned int mo_user_xattr:1,
247 unsigned int mo_dom_lock;
249 /* mdt state flags */
250 unsigned long mdt_state;
252 /* transaction callbacks */
253 struct dt_txn_callback mdt_txn_cb;
255 /* these values should be updated from lov if necessary.
256 * or should be placed somewhere else.
262 /* Max modify RPCs, replaces max_mod_rpcs_per_client module param */
263 unsigned int mdt_max_mod_rpcs_in_flight;
265 /* preferred BRW size, decided by storage type and capability */
268 struct upcall_cache *mdt_identity_cache;
269 struct upcall_cache *mdt_identity_cache_int;
271 unsigned int mdt_evict_tgt_nids:1,
273 mdt_migrate_hsm_allowed:1,
274 mdt_enable_strict_som:1,
275 /* Enable remote dir on non-MDT0 */
276 mdt_enable_dir_migration:1,
277 mdt_enable_dir_restripe:1,
278 mdt_enable_dir_auto_split:1,
279 mdt_enable_parallel_rename_dir:1,
280 mdt_enable_parallel_rename_file:1,
281 mdt_enable_parallel_rename_crossdir:1,
282 mdt_enable_remote_dir:1,
283 mdt_enable_remote_rename:1,
284 mdt_enable_rename_trylock:1,
285 mdt_enable_striped_dir:1,
288 /* dir restripe migrate dirent only */
289 mdt_dir_restripe_nsonly:1,
290 /* this is enabled by default, but once an
291 * old client joins, disable this to handle
292 * inherited default LMV on server.
294 mdt_enable_dmv_implicit_inherit:1,
295 /* set dmv by setxattr, off by default */
296 mdt_enable_dmv_xattr:1;
298 /* user with gid can create remote/striped
299 * dir, and set default dir stripe
301 gid_t mdt_enable_remote_dir_gid;
302 /* user with this gid can change projid */
303 gid_t mdt_enable_chprojid_gid;
304 kernel_cap_t mdt_enable_cap_mask;
305 /* user with this gid can pin/unpin file */
306 gid_t mdt_enable_pin_gid;
308 /* lock for osfs and md_root */
311 /* statfs optimization: we cache a bit */
312 struct mdt_statfs_cache mdt_sum_osfs;
313 struct mdt_statfs_cache mdt_osfs;
316 struct root_squash_info mdt_squash;
318 struct rename_stats mdt_rename_stats;
319 struct lu_fid mdt_md_root_fid;
321 /* connection to quota master */
322 struct obd_export *mdt_qmt_exp;
323 /* quota master device associated with this MDT */
324 struct lu_device *mdt_qmt_dev;
326 struct coordinator mdt_coordinator;
328 /* inter-MDT connection count */
329 atomic_t mdt_mds_mds_conns;
331 /* MDT device async commit count, used for debug and sanity test */
332 atomic_t mdt_async_commit_count;
334 struct mdt_object *mdt_md_root;
336 struct mdt_dir_restriper mdt_restriper;
338 /* count of old clients that doesn't support DMV implicite inherit */
339 atomic_t mdt_dmv_old_client_count;
341 /* name of xattr used to store jobid in mdt inode */
342 char mdt_job_xattr[XATTR_JOB_MAX_LEN];
345 #define MDT_SERVICE_WATCHDOG_FACTOR (2)
346 #define MDT_COS_DEFAULT (0)
348 #define ENOENT_VERSION 1 /** 'virtual' version of non-existent object */
351 struct lu_object_header mot_header;
352 struct lu_object mot_obj;
353 unsigned int mot_lov_created:1, /* lov object created */
354 mot_cache_attr:1, /* enable remote object
356 mot_restriping:1, /* dir restriping */
357 /* dir auto-split disabled */
358 mot_auto_split_disabled:1,
359 mot_lsom_inited:1, /* lsom was inited */
360 mot_discard_done:1; /* discard lock was sent */
362 spinlock_t mot_write_lock;
363 /* Lock to protect create_data */
364 struct mutex mot_lov_mutex;
365 /* Lock to protect object's SOM update. */
366 struct mutex mot_som_mutex;
368 __u64 mot_lsom_blocks;
369 /* lock to protect read/write stages for Data-on-MDT files */
370 struct rw_semaphore mot_dom_sem;
371 /* Lock to protect lease open.
372 * Lease open acquires write lock; normal open acquires read lock
374 struct rw_semaphore mot_open_sem;
375 atomic_t mot_lease_count;
376 atomic_t mot_open_count;
377 /* directory offset, used in sub file migration in dir restripe */
378 loff_t mot_restripe_offset;
379 /* link to mdt_restriper auto_splitting/migrating/updating */
380 struct list_head mot_restripe_linkage;
383 struct mdt_lock_handle {
384 /* Lock type, reg for cross-ref use or pdo lock. */
388 struct lustre_handle mlh_reg_lh;
389 enum ldlm_mode mlh_reg_mode;
393 struct lustre_handle mlh_pdo_lh;
394 enum ldlm_mode mlh_pdo_mode;
395 unsigned int mlh_pdo_hash;
396 unsigned int mlh_pdo_remote:1;
398 /* Remote regular lock */
399 struct lustre_handle mlh_rreg_lh;
400 enum ldlm_mode mlh_rreg_mode;
404 MDT_LH_PARENT, /* parent lockh */
405 MDT_LH_CHILD, /* child lockh */
406 MDT_LH_OLD, /* old lockh for rename */
407 MDT_LH_LAYOUT = MDT_LH_OLD, /* layout lock */
408 MDT_LH_NEW, /* new lockh for rename */
409 MDT_LH_RMT, /* used for return lh to caller */
410 MDT_LH_LOCAL, /* local lock never return to client */
411 MDT_LH_LOOKUP, /* lookup lock for source object in rename/migrate if
412 * it's remote object */
421 /* Special magical errno for communicaiton between mdt_reint_open()
422 * and mdt_intent_reint() which means return the lock to the client
423 * for subsequent cross ref open. Previously we used plain -EREMOTE
424 * but other functions called in that path might return it too and
425 * confuse us. This is not returned to the client. See LU-5370.
427 #define MDT_EREMOTE_OPEN (EREMOTE + 1024)
429 struct mdt_reint_record {
430 enum mds_reint_op rr_opcode;
431 const struct lustre_handle *rr_open_handle;
432 const struct lustre_handle *rr_lease_handle;
433 const struct lu_fid *rr_fid1;
434 const struct lu_fid *rr_fid2;
435 struct lu_name rr_name;
436 struct lu_name rr_tgt_name;
443 enum mdt_reint_flag {
444 MRF_OPEN_TRUNC = BIT(0),
445 MRF_OPEN_RESEND = BIT(1),
449 * Common data shared by mdt-level handlers. This is allocated per-thread to
450 * reduce stack consumption.
452 struct mdt_thread_info {
455 * The following members will be filled explicitly
456 * with specific data in mdt_thread_info_init().
458 /* TODO: move this into mdt_session_key(with LCT_SESSION), because
459 * request handling may migrate from one server thread to another.
461 struct req_capsule *mti_pill;
463 /* SUB request pill in a batch request. */
464 struct req_capsule mti_sub_pill;
466 /* although we have export in req, there are cases when it is not
467 * available, e.g. closing files upon export destroy
469 struct obd_export *mti_exp;
470 /* A couple of lock handles. */
471 struct mdt_lock_handle mti_lh[MDT_LH_NR];
473 struct mdt_device *mti_mdt;
474 const struct lu_env *mti_env;
476 /* transaction number of current request */
481 * The following members will be filled expilictly
482 * with zero in mdt_thread_info_init(). These members may be used
486 /* Object attributes. */
487 struct md_attr mti_attr;
488 struct md_attr mti_attr2; /* mdt_lvb.c */
489 /* Body for "habeo corpus" operations. */
490 const struct mdt_body *mti_body;
491 /* Host object. This is released at the end of mdt_handler(). */
492 struct mdt_object *mti_object;
494 * Lock request for "habeo clavis" operations.
496 const struct ldlm_request *mti_dlm_req;
498 __u32 mti_cross_ref:1,
499 /* big_lmm buffer was used and must be used in reply */
504 /* Batch processing environment */
508 /* opdata for mdt_reint_open(), has the same as
509 * ldlm_reply:lock_policy_res1. mdt_update_last_rcvd() stores this
510 * value onto disk for recovery when mdt_trans_stop_cb() is called.
516 * The following members will be filled explicitly
517 * with zero in mdt_reint_unpack(), because they are only used
518 * by reint requests (including mdt_reint_open()).
521 /* reint record. contains information for reint operations. */
522 struct mdt_reint_record mti_rr;
524 __u64 mti_ver[PTLRPC_NUM_VERSIONS];
525 /* Operation specification (currently create and lookup) */
526 struct md_op_spec mti_spec;
530 * The following members will _NOT_ be initialized at all.
531 * DO NOT expect them to contain any valid value.
532 * They should be initialized explicitly by the user themselves.
535 /* XXX: If something is in a union, make sure they do not conflict */
536 struct lu_fid mti_tmp_fid1;
537 struct lu_fid mti_tmp_fid2;
538 union ldlm_policy_data mti_policy; /* for mdt_object_lock() */
539 struct ldlm_res_id mti_res_id; /* and mdt_rename_lock() */
541 struct obd_uuid uuid[2]; /* for mdt_seq_init_cli() */
542 char ns_name[48];/* for mdt_init0() */
543 struct lustre_cfg_bufs bufs; /* for mdt_stack_fini() */
544 struct obd_statfs osfs; /* for mdt_statfs() */
546 /* for mdt_readpage() */
547 struct lu_rdpg mti_rdpg;
557 struct lustre_handle mti_open_handle;
559 struct lu_buf mti_buf;
560 struct lu_buf mti_big_buf;
562 /* Ops object filename */
563 struct lu_name mti_name;
564 char mti_filename[NAME_MAX + 1];
565 /* per-thread values, can be re-used, may be vmalloc'd */
566 void *mti_big_lov; /* was _lmm */
572 /* should be enough to fit lustre_mdt_attrs */
573 char mti_xattr_buf[128];
574 struct ldlm_enqueue_info mti_einfo;
575 /* einfo used by mdt_remote_object_lock_try() */
576 struct ldlm_enqueue_info mti_remote_einfo;
577 struct tg_reply_data *mti_reply_data;
579 /* FLR: layout change API */
580 struct md_layout_change mti_mlc;
582 struct lu_seq_range mti_range;
585 extern struct lu_context_key mdt_thread_key;
587 static inline struct mdt_thread_info *mdt_th_info(const struct lu_env *env)
589 return lu_env_info_get(env, &mdt_thread_key);
592 struct cdt_req_progress {
593 spinlock_t crp_lock; /**< protect tree */
594 struct interval_tree_root crp_root; /**< tree to track extent
599 struct cdt_agent_req {
600 struct hlist_node car_cookie_hash; /**< find req by cookie */
601 struct list_head car_request_list; /**< to chain all the req. */
602 struct list_head car_scan_list; /**< list for scan process */
603 struct kref car_refcount; /**< reference counter */
604 struct obd_uuid car_uuid; /**< agent doing the req. */
605 struct hsm_mem_req_rec *car_hmm; /**< llog rec with cookies */
606 struct cdt_req_progress car_progress; /**< track data mvt
608 struct cdt_agent_req *car_cancel; /**< corresponding cancel */
610 #define car_flags car_hmm->mr_rec.arr_flags /**< request original flags */
611 #define car_archive_id car_hmm->mr_rec.arr_archive_id /**< archive id */
612 #define car_req_update car_hmm->mr_rec.arr_req_change /**< last update time */
613 #define car_hai car_hmm->mr_rec.arr_hai /**< req. to the agent */
615 extern struct kmem_cache *mdt_hsm_car_kmem;
618 struct list_head ha_list; /* to chain the agents */
619 struct obd_uuid ha_uuid; /* agent uuid */
620 __u32 *ha_archive_id; /* archive id */
621 int ha_archive_cnt; /* number of archive entries
622 * 0 means any archive */
623 atomic_t ha_requests; /* current request count */
624 atomic_t ha_success; /* num of successful actions */
625 atomic_t ha_failure; /* number of failed actions */
628 struct cdt_restore_handle {
629 struct rhash_head crh_hash; /**< link to cdt_restore_hash */
630 struct lu_fid crh_fid; /**< fid of the object */
631 struct ldlm_extent crh_extent; /**< extent of the restore */
632 struct mdt_lock_handle crh_lh; /**< lock handle */
633 atomic_t crh_refc; /**< crh ref counter */
634 struct rcu_head crh_rcu; /**< crh rcu head */
636 extern struct kmem_cache *mdt_hsm_cdt_kmem; /** restore handle slab cache */
638 struct hsm_record_update {
640 enum agent_req_status status;
643 struct hsm_mem_req_rec {
644 struct llog_logid mr_lid;
646 struct llog_agent_req_rec mr_rec;
650 * data passed to llog_cat_process() callback
651 * to scan requests and take actions
653 struct hsm_scan_request {
654 struct list_head hsr_cars;
661 static inline u32 hsr_get_archive_id(struct hsm_scan_request *rq)
663 struct cdt_agent_req *car;
665 if (rq->hsr_count > 0) {
666 car = list_first_entry(&rq->hsr_cars, struct cdt_agent_req,
669 return car->car_archive_id;
675 const struct md_device_operations *mdt_child_ops(struct mdt_device *m)
677 LASSERT(m->mdt_child);
678 return m->mdt_child->md_ops;
681 static inline struct md_object *mdt_object_child(struct mdt_object *o)
684 return lu2md(lu_object_next(&o->mot_obj));
687 static inline struct ptlrpc_request *mdt_info_req(struct mdt_thread_info *info)
689 return info->mti_pill ? info->mti_pill->rc_req : NULL;
692 static inline __u64 mdt_conn_flags(struct mdt_thread_info *info)
694 LASSERT(info->mti_exp);
695 return exp_connect_flags(info->mti_exp);
698 static inline void mdt_object_get(const struct lu_env *env,
699 struct mdt_object *o)
702 lu_object_get(&o->mot_obj);
706 static inline void mdt_object_put(const struct lu_env *env,
707 struct mdt_object *o)
710 lu_object_put(env, &o->mot_obj);
714 static inline int mdt_object_exists(const struct mdt_object *o)
716 return lu_object_exists(&o->mot_obj);
719 static inline int mdt_object_remote(const struct mdt_object *o)
721 return lu_object_remote(&o->mot_obj);
724 int mdt_object_striped(struct mdt_thread_info *mti, struct mdt_object *obj);
726 static inline const struct lu_fid *mdt_object_fid(const struct mdt_object *o)
728 return lu_object_fid(&o->mot_obj);
731 static inline struct lu_site *mdt_lu_site(const struct mdt_device *mdt)
733 return mdt->mdt_lu_dev.ld_site;
736 static inline struct seq_server_site *mdt_seq_site(struct mdt_device *mdt)
738 return &mdt->mdt_seq_site;
741 static inline u32 mdt_node_id(const struct mdt_device *mdt)
743 return mdt->mdt_seq_site.ss_node_id;
746 static inline void mdt_export_evict(struct obd_export *exp)
748 class_fail_export(exp);
751 /* Here we use LVB_TYPE to check dne client, because it is
752 * also landed on 2.4.
754 static inline bool mdt_is_dne_client(struct obd_export *exp)
756 return !!(exp_connect_flags(exp) & OBD_CONNECT_LVB_TYPE);
759 static inline bool mdt_is_striped_client(struct obd_export *exp)
761 return exp_connect_flags(exp) & OBD_CONNECT_DIR_STRIPE;
764 __u32 mdt_lmm_dom_entry_check(struct lov_mds_md *lmm, int *dom_only);
766 static inline bool mdt_lmm_dom_only(struct lov_mds_md *lmm)
770 mdt_lmm_dom_entry_check(lmm, &dom_only);
774 static inline __u32 mdt_lmm_dom_stripesize(struct lov_mds_md *lmm)
776 return mdt_lmm_dom_entry_check(lmm, NULL);
779 static inline bool mdt_lmm_is_flr(struct lov_mds_md *lmm)
781 struct lov_comp_md_v1 *lcm = (typeof(lcm))lmm;
783 return le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1 &&
784 le16_to_cpu(lcm->lcm_mirror_count) > 0;
787 static inline bool lmm_is_overstriping(struct lov_mds_md *lmm)
789 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V1 ||
790 le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_V3)
791 return le16_to_cpu(lmm->lmm_pattern) & LOV_PATTERN_OVERSTRIPING;
796 static inline bool mdt_lmm_comp_overstriping(struct lov_mds_md *lmm)
798 struct lov_comp_md_v1 *comp_v1;
799 struct lov_mds_md *v1;
803 comp_v1 = (struct lov_comp_md_v1 *)lmm;
805 for (i = 0; i < le16_to_cpu(comp_v1->lcm_entry_count); i++) {
806 off = le32_to_cpu(comp_v1->lcm_entries[i].lcme_offset);
807 v1 = (struct lov_mds_md *)((char *)comp_v1 + off);
809 if (lmm_is_overstriping(v1))
816 static inline bool mdt_lmm_is_overstriping(struct lov_mds_md *lmm)
818 if (le32_to_cpu(lmm->lmm_magic) == LOV_MAGIC_COMP_V1)
819 return mdt_lmm_comp_overstriping(lmm);
821 return lmm_is_overstriping(lmm);
824 static inline bool mdt_is_sum_statfs_client(struct obd_export *exp)
826 return exp_connect_flags(exp) & OBD_CONNECT_FLAGS2 &&
827 exp_connect_flags2(exp) & OBD_CONNECT2_SUM_STATFS;
830 __u64 mdt_get_disposition(struct ldlm_reply *rep, __u64 op_flag);
831 void mdt_set_disposition(struct mdt_thread_info *info,
832 struct ldlm_reply *rep, __u64 op_flag);
833 void mdt_clear_disposition(struct mdt_thread_info *info,
834 struct ldlm_reply *rep, __u64 op_flag);
836 void mdt_lock_pdo_init(struct mdt_lock_handle *lh, enum ldlm_mode lock_mode,
837 const struct lu_name *lname);
839 void mdt_lock_reg_init(struct mdt_lock_handle *lh, enum ldlm_mode lm);
840 void mdt_lh_reg_init(struct mdt_lock_handle *lh, struct ldlm_lock *lock);
842 int mdt_lock_setup(struct mdt_thread_info *info, struct mdt_object *mo,
843 struct mdt_lock_handle *lh);
845 int mdt_check_resent_lock(struct mdt_thread_info *info, struct mdt_object *mo,
846 struct mdt_lock_handle *lhc);
848 int mdt_object_lock(struct mdt_thread_info *info, struct mdt_object *obj,
849 struct mdt_lock_handle *lh, enum mds_ibits_locks ibits,
850 enum ldlm_mode mode);
851 int mdt_parent_lock(struct mdt_thread_info *info, struct mdt_object *o,
852 struct mdt_lock_handle *lh, const struct lu_name *lname,
853 enum ldlm_mode mode);
854 int mdt_object_stripes_lock(struct mdt_thread_info *info,
855 struct mdt_object *pobj, struct mdt_object *o,
856 struct mdt_lock_handle *lh,
857 struct ldlm_enqueue_info *einfo,
858 enum mds_ibits_locks ibits,
859 enum ldlm_mode mode);
860 int mdt_object_check_lock(struct mdt_thread_info *info,
861 struct mdt_object *parent, struct mdt_object *child,
862 struct mdt_lock_handle *lh,
863 enum mds_ibits_locks ibits, enum ldlm_mode mode);
864 int mdt_object_lock_try(struct mdt_thread_info *info, struct mdt_object *mo,
865 struct mdt_lock_handle *lh,
866 enum mds_ibits_locks *ibits,
867 enum mds_ibits_locks trybits, enum ldlm_mode mode);
869 /* below three lock functions are used internally */
870 int mdt_object_lock_internal(struct mdt_thread_info *info,
871 struct mdt_object *obj, const struct lu_fid *fid,
872 struct mdt_lock_handle *lh,
873 enum mds_ibits_locks *ibits,
874 enum mds_ibits_locks trybits,
876 int mdt_object_pdo_lock(struct mdt_thread_info *info, struct mdt_object *obj,
877 struct mdt_lock_handle *lh, const struct lu_name *name,
878 enum ldlm_mode mode, bool pdo_lock);
879 int mdt_object_lookup_lock(struct mdt_thread_info *info,
880 struct mdt_object *pobj, struct mdt_object *obj,
881 struct mdt_lock_handle *lh, enum ldlm_mode mode);
883 void mdt_object_unlock(struct mdt_thread_info *info, struct mdt_object *mo,
884 struct mdt_lock_handle *lh, int decref);
885 void mdt_object_stripes_unlock(struct mdt_thread_info *info,
886 struct mdt_object *o, struct mdt_lock_handle *lh,
887 struct ldlm_enqueue_info *einfo, int decref);
889 struct mdt_object *mdt_object_new(const struct lu_env *env,
890 struct mdt_device *mdt,
891 const struct lu_fid *fid);
892 struct mdt_object *mdt_object_find(const struct lu_env *env,
893 struct mdt_device *mdt,
894 const struct lu_fid *fid);
895 struct mdt_object *mdt_object_find_lock(struct mdt_thread_info *info,
896 const struct lu_fid *f,
897 struct mdt_lock_handle *lh,
898 enum mds_ibits_locks ibits,
899 enum ldlm_mode mode);
900 void mdt_object_unlock_put(struct mdt_thread_info *info,
901 struct mdt_object *o,
902 struct mdt_lock_handle *lh,
905 void mdt_client_compatibility(struct mdt_thread_info *info);
907 enum mdt_name_flags {
911 int mdt_name_unpack(struct req_capsule *pill,
912 const struct req_msg_field *field,
914 enum mdt_name_flags flags);
915 int mdt_close_unpack(struct mdt_thread_info *info);
916 int mdt_reint_unpack(struct mdt_thread_info *info, __u32 op);
917 void mdt_fix_lov_magic(struct mdt_thread_info *info, void *eadata);
918 int mdt_reint_rec(struct mdt_thread_info *info, struct mdt_lock_handle *lh);
919 #ifdef CONFIG_LUSTRE_FS_POSIX_ACL
920 int mdt_pack_acl2body(struct mdt_thread_info *info, struct mdt_body *repbody,
921 struct mdt_object *o, struct lu_nodemap *nodemap);
923 void mdt_pack_attr2body(struct mdt_thread_info *info, struct mdt_body *b,
924 const struct lu_attr *attr, const struct lu_fid *fid);
925 int mdt_pack_size2body(struct mdt_thread_info *info,
926 const struct lu_fid *fid, struct lustre_handle *lh);
927 int mdt_getxattr(struct mdt_thread_info *info);
928 int mdt_reint_setxattr(struct mdt_thread_info *info,
929 struct mdt_lock_handle *lh);
931 void mdt_reconstruct(struct mdt_thread_info *info, struct mdt_lock_handle *lh);
932 void mdt_reconstruct_generic(struct mdt_thread_info *mti,
933 struct mdt_lock_handle *lhc);
935 int mdt_export_stats_init(struct obd_device *obd, struct obd_export *exp,
938 int mdt_lock_new_child(struct mdt_thread_info *info,
939 struct mdt_object *o,
940 struct mdt_lock_handle *child_lockh);
941 void mdt_mfd_set_mode(struct mdt_file_data *mfd, enum mds_open_flags o_flags);
942 int mdt_reint_open(struct mdt_thread_info *info, struct mdt_lock_handle *lhc);
943 struct mdt_file_data *mdt_open_handle2mfd(struct mdt_export_data *med,
944 const struct lustre_handle *open_handle,
946 int mdt_revoke_remote_lookup_lock(struct mdt_thread_info *info,
947 struct mdt_object *pobj,
948 struct mdt_object *obj);
950 int mdt_get_info(struct tgt_session_info *tsi);
951 int mdt_attr_get_complex(struct mdt_thread_info *info,
952 struct mdt_object *o, struct md_attr *ma);
953 int mdt_big_xattr_get(struct mdt_thread_info *info, struct mdt_object *o,
955 int __mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
956 struct md_attr *ma, const char *name);
957 int mdt_stripe_get(struct mdt_thread_info *info, struct mdt_object *o,
958 struct md_attr *ma, const char *name);
959 int mdt_attr_get_pfid(struct mdt_thread_info *info, struct mdt_object *o,
960 struct lu_fid *pfid);
961 int mdt_attr_get_pfid_name(struct mdt_thread_info *info, struct mdt_object *o,
962 struct lu_fid *pfid, struct lu_name *lname);
963 int mdt_write_get(struct mdt_object *o);
964 void mdt_write_put(struct mdt_object *o);
965 int mdt_write_read(struct mdt_object *o);
966 struct mdt_file_data *mdt_mfd_new(const struct mdt_export_data *med);
967 int mdt_mfd_close(struct mdt_thread_info *info, struct mdt_file_data *mfd);
968 void mdt_mfd_free(struct mdt_file_data *mfd);
969 int mdt_close(struct tgt_session_info *tsi);
970 int mdt_add_dirty_flag(struct mdt_thread_info *info, struct mdt_object *mo,
972 int mdt_fix_reply(struct mdt_thread_info *info);
973 int mdt_handle_last_unlink(struct mdt_thread_info *info, struct mdt_object *dir,
974 struct md_attr *attr);
975 void mdt_reconstruct_open(struct mdt_thread_info *info,
976 struct mdt_lock_handle *lh);
977 int mdt_layout_change(struct mdt_thread_info *info, struct mdt_object *obj,
978 struct mdt_lock_handle *lhc,
979 struct md_layout_change *spec);
980 int find_name_matching_hash(struct mdt_thread_info *info, struct lu_name *lname,
981 struct mdt_object *parent, struct mdt_object *child);
982 int mdt_device_sync(const struct lu_env *env, struct mdt_device *mdt);
984 void mdt_dump_lmm(int level, const struct lov_mds_md *lmm, __u64 valid);
985 void mdt_dump_lmv(unsigned int level, const union lmv_mds_md *lmv);
987 bool allow_client_chgrp(struct mdt_thread_info *info, struct lu_ucred *uc);
988 int mdt_check_ucred(struct mdt_thread_info *info);
989 int mdt_init_ucred(struct mdt_thread_info *info, struct mdt_body *body);
990 int mdt_init_ucred_reint(struct mdt_thread_info *info);
991 void mdt_exit_ucred(struct mdt_thread_info *info);
992 int mdt_check_resource_ids(struct mdt_thread_info *info,
993 struct mdt_object *obj);
994 int mdt_version_get_check(struct mdt_thread_info *info, struct mdt_object *mto,
996 void mdt_version_get_save(struct mdt_thread_info *info, struct mdt_object *mto,
998 int mdt_version_get_check_save(struct mdt_thread_info *info,
999 struct mdt_object *mto, int index);
1000 int mdt_lookup_version_check(struct mdt_thread_info *info,
1001 struct mdt_object *p,
1002 const struct lu_name *lname,
1003 struct lu_fid *fid, int idx);
1004 void mdt_thread_info_reset(struct mdt_thread_info *info);
1005 void mdt_thread_info_init(struct ptlrpc_request *req,
1006 struct mdt_thread_info *mti);
1007 void mdt_thread_info_fini(struct mdt_thread_info *mti);
1008 struct mdt_thread_info *tsi2mdt_info(struct tgt_session_info *tsi);
1009 void mdt_intent_fixup_resent(struct mdt_thread_info *info,
1010 struct ldlm_lock *new_lock,
1011 struct mdt_lock_handle *lh, __u64 flags);
1012 int mdt_intent_lock_replace(struct mdt_thread_info *info,
1013 struct ldlm_lock **lockp,
1014 struct mdt_lock_handle *lh,
1015 __u64 flags, int result);
1017 int hsm_init_ucred(struct lu_ucred *uc);
1018 int mdt_hsm_attr_set(struct mdt_thread_info *info, struct mdt_object *obj,
1019 const struct md_hsm *mh);
1021 int mdt_remote_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1022 void *data, int flag);
1023 int mdt_links_read(struct mdt_thread_info *info,
1024 struct mdt_object *mdt_obj,
1025 struct linkea_data *ldata);
1026 int mdt_close_internal(struct mdt_thread_info *info, struct ptlrpc_request *req,
1027 struct mdt_body *repbody);
1028 int mdt_pack_secctx_in_reply(struct mdt_thread_info *info,
1029 struct mdt_object *child);
1030 int mdt_pack_encctx_in_reply(struct mdt_thread_info *info,
1031 struct mdt_object *child);
1032 void mdt_prep_ma_buf_from_rep(struct mdt_thread_info *info,
1033 struct mdt_object *obj, struct md_attr *ma,
1034 enum mds_open_flags open_flags);
1037 static inline struct mdt_device *mdt_dev(struct lu_device *d)
1039 return container_of_safe(d, struct mdt_device, mdt_lu_dev);
1042 static inline struct mdt_object *mdt_obj(struct lu_object *o)
1044 return container_of_safe(o, struct mdt_object, mot_obj);
1047 static inline struct dt_object *mdt_obj2dt(struct mdt_object *mo)
1049 struct lu_object *lo;
1050 struct mdt_device *mdt = mdt_dev(mo->mot_obj.lo_dev);
1052 lo = lu_object_locate(mo->mot_obj.lo_header,
1053 mdt->mdt_bottom->dd_lu_dev.ld_type);
1058 static inline bool agent_req_in_final_state(enum agent_req_status ars)
1060 return ((ars == ARS_SUCCEED) || (ars == ARS_FAILED) ||
1061 (ars == ARS_CANCELED));
1064 /* mdt/mdt_identity.c */
1065 #define MDT_IDENTITY_UPCALL_PATH "/usr/sbin/l_getidentity"
1066 #define UC_IDCACHE_HASH_SIZE 128
1067 extern struct upcall_cache_ops mdt_identity_upcall_cache_ops;
1069 struct md_identity *mdt_identity_get(struct upcall_cache *cache, __u32 uid,
1070 struct mdt_thread_info *info);
1072 void mdt_identity_put(struct upcall_cache *cache, struct md_identity *identity);
1074 void mdt_flush_identity(struct upcall_cache *cache, int uid);
1076 __u32 mdt_identity_get_perm(struct md_identity *identity, struct lnet_nid *nid);
1078 /* mdt/mdt_recovery.c */
1079 __u64 mdt_req_from_lrd(struct ptlrpc_request *req, struct tg_reply_data *trd);
1081 /* mdt/mdt_batch.c */
1082 int mdt_batch(struct tgt_session_info *tsi);
1085 int mdt_hsm_state_get(struct tgt_session_info *tsi);
1086 int mdt_hsm_state_set(struct tgt_session_info *tsi);
1087 int mdt_hsm_action(struct tgt_session_info *tsi);
1088 int mdt_hsm_progress(struct tgt_session_info *tsi);
1089 int mdt_hsm_ct_register(struct tgt_session_info *tsi);
1090 int mdt_hsm_ct_unregister(struct tgt_session_info *tsi);
1091 int mdt_hsm_request(struct tgt_session_info *tsi);
1092 int mdt_hsm_data_version(struct tgt_session_info *tsi);
1094 /* mdt/mdt_hsm_cdt_actions.c */
1095 extern const struct file_operations mdt_hsm_actions_fops;
1096 void dump_llog_agent_req_rec(const char *prefix,
1097 const struct llog_agent_req_rec *larr);
1098 int cdt_llog_process(const struct lu_env *env, struct mdt_device *mdt,
1099 llog_cb_t cb, void *data, u32 start_cat_idx,
1101 int mdt_agent_record_add(const struct lu_env *env, struct mdt_device *mdt,
1102 __u32 archive_id, __u64 flags,
1103 struct hsm_action_item *hai);
1104 int mdt_hsm_agent_modify_record(const struct lu_env *env,
1105 struct mdt_device *mdt,
1106 struct hsm_mem_req_rec *hmm);
1107 /* mdt/mdt_hsm_cdt_agent.c */
1108 extern const struct file_operations mdt_hsm_agent_fops;
1109 int mdt_hsm_agent_register(struct mdt_thread_info *info,
1110 const struct obd_uuid *uuid,
1111 int nr_archives, __u32 *archive_num);
1112 int mdt_hsm_agent_register_mask(struct mdt_thread_info *info,
1113 const struct obd_uuid *uuid,
1114 __u32 archive_mask);
1115 int mdt_hsm_agent_unregister(struct mdt_thread_info *info,
1116 const struct obd_uuid *uuid);
1117 int mdt_hsm_agent_update_statistics(struct coordinator *cdt,
1118 int succ_rq, int fail_rq, int new_rq,
1119 const struct obd_uuid *uuid);
1120 int mdt_hsm_find_best_agent(struct coordinator *cdt, __u32 archive,
1121 struct obd_uuid *uuid);
1122 int mdt_hsm_agent_send(struct mdt_thread_info *mti, struct hsm_scan_request *rq,
1124 /* mdt/mdt_hsm_cdt_client.c */
1125 int mdt_hsm_add_actions(struct mdt_thread_info *info,
1126 struct hsm_action_list *hal);
1127 int mdt_hsm_get_action(struct mdt_thread_info *mti,
1128 const struct lu_fid *fid,
1129 enum hsm_copytool_action *action,
1130 enum agent_req_status *status,
1131 struct hsm_extent *extent);
1132 bool mdt_hsm_restore_is_running(struct mdt_thread_info *mti,
1133 const struct lu_fid *fid);
1134 /* mdt/mdt_hsm_cdt_requests.c */
1135 extern struct cfs_hash_ops cdt_request_cookie_hash_ops;
1136 extern const struct file_operations mdt_hsm_active_requests_fops;
1137 void dump_requests(char *prefix, struct coordinator *cdt);
1138 struct cdt_agent_req *mdt_cdt_alloc_request(struct obd_uuid *uuid,
1139 struct llog_agent_req_rec *rec);
1140 void mdt_cdt_free_request(struct cdt_agent_req *car);
1141 int mdt_cdt_add_request(struct coordinator *cdt, struct cdt_agent_req *new_car);
1142 struct cdt_agent_req *mdt_cdt_find_request(struct coordinator *cdt, u64 cookie);
1143 void mdt_cdt_get_request(struct cdt_agent_req *car);
1144 void mdt_cdt_put_request(struct cdt_agent_req *car);
1145 struct cdt_agent_req *mdt_cdt_update_request(struct coordinator *cdt,
1146 const struct hsm_progress_kernel *pgs);
1147 int mdt_cdt_remove_request(struct coordinator *cdt, __u64 cookie);
1148 /* mdt/mdt_coordinator.c */
1149 void mdt_hsm_dump_hal(int level, const char *prefix,
1150 struct hsm_action_list *hal);
1151 int cdt_restore_handle_add(struct mdt_thread_info *mti, struct coordinator *cdt,
1152 const struct lu_fid *fid,
1153 const struct hsm_extent *he);
1154 bool cdt_restore_handle_exists(struct coordinator *cdt,
1155 const struct lu_fid *fid);
1156 void cdt_restore_handle_del(struct mdt_thread_info *mti,
1157 struct coordinator *cdt, const struct lu_fid *fid);
1158 int cdt_getref_try(struct coordinator *cdt);
1159 void cdt_putref(struct coordinator *cdt);
1160 /* coordinator management */
1161 int mdt_hsm_cdt_init(struct mdt_device *mdt);
1162 int mdt_hsm_cdt_stop(struct mdt_device *mdt);
1163 int mdt_hsm_cdt_fini(struct mdt_device *mdt);
1166 * Signal the coordinator has work to do
1167 * \param cdt [IN] coordinator
1169 static inline void mdt_hsm_cdt_event(struct coordinator *cdt)
1171 cdt->cdt_event = true;
1174 /* coordinator control sysfs interface */
1175 ssize_t hsm_control_show(struct kobject *kobj, struct attribute *attr,
1177 ssize_t hsm_control_store(struct kobject *kobj, struct attribute *attr,
1178 const char *buffer, size_t count);
1179 int hsm_cdt_tunables_init(struct mdt_device *mdt);
1180 void hsm_cdt_tunables_fini(struct mdt_device *mdt);
1181 /* md_hsm helpers */
1182 struct mdt_object *mdt_hsm_get_md_hsm(struct mdt_thread_info *mti,
1183 const struct lu_fid *fid,
1184 struct md_hsm *hsm);
1185 /* actions/request helpers */
1186 int mdt_hsm_add_hsr(struct mdt_thread_info *mti, struct hsm_scan_request *hsr,
1187 struct obd_uuid *uuid);
1188 bool mdt_hsm_is_action_compat(const struct hsm_action_item *hai,
1189 u32 archive_id, u64 rq_flags,
1190 const struct md_hsm *hsm);
1191 int mdt_hsm_update_request_state(struct mdt_thread_info *mti,
1192 struct hsm_progress_kernel *pgs);
1194 extern struct lu_context_key mdt_thread_key;
1196 /* debug issues helper starts here*/
1197 static inline int mdt_fail_write(const struct lu_env *env,
1198 struct dt_device *dd, int id)
1200 if (CFS_FAIL_CHECK_ORSET(id, CFS_FAIL_ONCE)) {
1201 CERROR(LUSTRE_MDT_NAME": cfs_fail_loc=%x, fail write ops\n",
1203 return dt_ro(env, dd);
1204 /* We set FAIL_ONCE because we never "un-fail" a device */
1210 static inline struct mdt_export_data *mdt_req2med(struct ptlrpc_request *req)
1212 return &req->rq_export->exp_mdt_data;
1215 static inline struct mdt_device *mdt_exp2dev(struct obd_export *exp)
1217 return mdt_dev(exp->exp_obd->obd_lu_dev);
1220 static inline bool mdt_rdonly(struct obd_export *exp)
1222 return (exp_connect_flags(exp) & OBD_CONNECT_RDONLY ||
1223 mdt_exp2dev(exp)->mdt_bottom->dd_rdonly ||
1224 mdt_exp2dev(exp)->mdt_readonly);
1227 typedef void (*mdt_reconstruct_t)(struct mdt_thread_info *mti,
1228 struct mdt_lock_handle *lhc);
1229 static inline int mdt_check_resent(struct mdt_thread_info *info,
1230 mdt_reconstruct_t reconstruct,
1231 struct mdt_lock_handle *lhc)
1233 struct ptlrpc_request *req = mdt_info_req(info);
1237 if (lustre_msg_get_flags(req->rq_reqmsg) & MSG_RESENT) {
1238 OBD_ALLOC_PTR(info->mti_reply_data);
1239 if (info->mti_reply_data == NULL)
1242 rc = req_can_reconstruct(req, info->mti_reply_data);
1244 reconstruct(info, lhc);
1246 DEBUG_REQ(D_HA, req,
1247 "no reply data found for RESENT req, rc = %d",
1250 OBD_FREE_PTR(info->mti_reply_data);
1251 info->mti_reply_data = NULL;
1256 struct lu_ucred *mdt_ucred(const struct mdt_thread_info *info);
1257 struct lu_ucred *mdt_ucred_check(const struct mdt_thread_info *info);
1259 static inline int is_identity_get_disabled(struct upcall_cache *cache)
1261 return cache ? (strcmp(cache->uc_upcall, "NONE") == 0) : 1;
1264 int mdt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1265 void *data, int flag);
1267 static int mdt_dom_glimpse_ast(struct ldlm_lock *lock, void *reqp)
1269 return -ELDLM_NO_LOCK_DATA;
1272 /* Issues dlm lock on passed @ns, @f stores it lock handle into @lh. */
1273 static inline int mdt_fid_lock(const struct lu_env *env,
1274 struct ldlm_namespace *ns,
1275 struct lustre_handle *lh, enum ldlm_mode mode,
1276 union ldlm_policy_data *policy,
1277 const struct ldlm_res_id *res_id,
1278 __u64 flags, const __u64 *client_cookie)
1281 bool glimpse = policy->l_inodebits.bits & MDS_INODELOCK_DOM;
1283 LASSERT(ns != NULL);
1284 LASSERT(lh != NULL);
1286 rc = ldlm_cli_enqueue_local(env, ns, res_id, LDLM_IBITS, policy,
1287 mode, &flags, mdt_blocking_ast,
1288 ldlm_completion_ast,
1289 glimpse ? mdt_dom_glimpse_ast : NULL,
1290 NULL, 0, LVB_T_NONE, client_cookie, lh);
1291 return rc == ELDLM_OK ? 0 : -EIO;
1294 static inline void mdt_fid_unlock(struct lustre_handle *lh, enum ldlm_mode mode)
1296 ldlm_lock_decref(lh, mode);
1299 /* this is identical to say this is a DNE system */
1300 static inline bool mdt_slc_is_enabled(struct mdt_device *mdt)
1302 return mdt->mdt_lut.lut_sync_lock_cancel == SYNC_LOCK_CANCEL_BLOCKING;
1306 int mdt_set_som(struct mdt_thread_info *info, struct mdt_object *obj,
1307 enum lustre_som_flags flag, __u64 size, __u64 blocks);
1308 int mdt_get_som(struct mdt_thread_info *info, struct mdt_object *obj,
1309 struct md_attr *ma);
1310 int mdt_lsom_downgrade(struct mdt_thread_info *info, struct mdt_object *obj);
1311 int mdt_lsom_update(struct mdt_thread_info *info, struct mdt_object *obj,
1315 extern struct ldlm_valblock_ops mdt_lvbo;
1316 int mdt_dom_lvb_is_valid(struct ldlm_resource *res);
1317 int mdt_dom_lvbo_update(struct ldlm_resource *res, struct ldlm_lock *lock,
1318 struct ptlrpc_request *req, bool increase_only);
1320 void mdt_enable_cos(struct mdt_device *dev, bool enable);
1321 int mdt_cos_is_enabled(struct mdt_device *mdt);
1339 LPROC_MDT_RENAME_SAMEDIR,
1340 LPROC_MDT_RENAME_PAR_FILE,
1341 LPROC_MDT_RENAME_PAR_DIR,
1342 LPROC_MDT_RENAME_CROSSDIR,
1343 LPROC_MDT_RENAME_TRYLOCK,
1346 LPROC_MDT_IO_READ_BYTES,
1347 LPROC_MDT_IO_WRITE_BYTES,
1350 LPROC_MDT_FALLOCATE,
1354 void mdt_counter_incr(struct ptlrpc_request *req, int opcode, long amount);
1355 void mdt_stats_counter_init(struct lprocfs_stats *stats, unsigned int offset,
1356 enum lprocfs_counter_config cntr_umask);
1357 int mdt_tunables_init(struct mdt_device *mdt, const char *name);
1358 void mdt_tunables_fini(struct mdt_device *mdt);
1360 /* lustre/mdt_mdt_lproc.c */
1361 int ldebugfs_mdt_open_files_seq_open(struct inode *inode,
1363 void mdt_rename_counter_tally(struct mdt_thread_info *info,
1364 struct mdt_device *mdt,
1365 struct ptlrpc_request *req,
1366 struct mdt_object *src, struct mdt_object *tgt,
1367 enum mdt_stat_idx msi, s64 count);
1369 static inline struct obd_device *mdt2obd_dev(const struct mdt_device *mdt)
1371 return mdt->mdt_lu_dev.ld_obd;
1374 extern const struct lu_device_operations mdt_lu_ops;
1376 static inline char *mdt_obd_name(struct mdt_device *mdt)
1378 return mdt->mdt_lu_dev.ld_obd->obd_name;
1381 int mds_mod_init(void);
1382 void mds_mod_exit(void);
1384 static inline char *mdt_req_get_jobid(struct ptlrpc_request *req)
1386 struct obd_export *exp = req->rq_export;
1389 if (exp_connect_flags(exp) & OBD_CONNECT_JOBSTATS)
1390 jobid = lustre_msg_get_jobid(req->rq_reqmsg);
1396 #define VALID_FLAGS (LA_TYPE | LA_MODE | LA_SIZE | LA_BLOCKS | \
1397 LA_BLKSIZE | LA_ATIME | LA_MTIME | LA_CTIME)
1399 int mdt_obd_preprw(const struct lu_env *env, int cmd, struct obd_export *exp,
1400 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1401 struct niobuf_remote *rnb, int *nr_local,
1402 struct niobuf_local *lnb);
1404 int mdt_obd_commitrw(const struct lu_env *env, int cmd, struct obd_export *exp,
1405 struct obdo *oa, int objcount, struct obd_ioobj *obj,
1406 struct niobuf_remote *rnb, int npages,
1407 struct niobuf_local *lnb, int old_rc, int nob,
1409 int mdt_punch_hdl(struct tgt_session_info *tsi);
1410 int mdt_fallocate_hdl(struct tgt_session_info *tsi);
1411 int mdt_fiemap_get(struct tgt_session_info *tsi);
1412 int mdt_glimpse_enqueue(struct mdt_thread_info *mti, struct ldlm_namespace *ns,
1413 struct ldlm_lock **lockp, __u64 flags);
1414 int mdt_brw_enqueue(struct mdt_thread_info *info, struct ldlm_namespace *ns,
1415 struct ldlm_lock **lockp, __u64 flags);
1416 int mdt_dom_read_on_open(struct mdt_thread_info *mti, struct mdt_device *mdt,
1417 struct lustre_handle *lh);
1418 void mdt_dom_discard_data(struct mdt_thread_info *info, struct mdt_object *mo);
1419 int mdt_dom_disk_lvbo_update(const struct lu_env *env, struct mdt_object *mo,
1420 struct ldlm_resource *res, bool increase_only);
1421 void mdt_dom_obj_lvb_update(const struct lu_env *env, struct mdt_object *mo,
1422 struct obdo *oa, bool increase_only);
1423 int mdt_dom_lvb_alloc(struct ldlm_resource *res);
1425 static inline bool mdt_dom_check_for_discard(struct mdt_thread_info *mti,
1426 struct mdt_object *mo)
1428 return lu_object_is_dying(&mo->mot_header) &&
1429 S_ISREG(lu_object_attr(&mo->mot_obj)) &&
1430 !mo->mot_discard_done;
1433 int mdt_dom_object_size(const struct lu_env *env, struct mdt_device *mdt,
1434 const struct lu_fid *fid, struct mdt_body *mb,
1436 bool mdt_dom_client_has_lock(struct mdt_thread_info *info,
1437 const struct lu_fid *fid);
1438 void mdt_hp_brw(struct tgt_session_info *tsi);
1439 void mdt_hp_punch(struct tgt_session_info *tsi);
1440 int mdt_data_version_get(struct tgt_session_info *tsi);
1441 int mdt_io_set_info(struct tgt_session_info *tsi);
1444 long mdt_grant_connect(const struct lu_env *env, struct obd_export *exp,
1445 u64 want, bool conservative);
1446 extern struct kmem_cache *ldlm_glimpse_work_kmem;
1448 static inline bool mdt_changelog_allow(struct mdt_thread_info *info)
1450 struct lu_ucred *uc = NULL;
1454 if (info == NULL || info->mti_body == NULL)
1455 #if LUSTRE_VERSION_CODE < OBD_OCD_VERSION(2, 17, 3, 0)
1456 /* return true in case old client did not send mdt body */
1462 rc = mdt_init_ucred(info, (struct mdt_body *)info->mti_body);
1466 uc = mdt_ucred(info);
1467 is_admin = (uc->uc_uid == 0 && uc->uc_gid == 0 &&
1468 cap_raised(uc->uc_cap, CAP_SYS_ADMIN) &&
1469 uc->uc_rbac_chlg_ops);
1471 mdt_exit_ucred(info);
1476 /* We forbid operations from encryption-unaware clients if they try to
1477 * manipulate encrypted files/directories.
1479 static inline int mdt_check_enc(struct mdt_thread_info *info,
1480 struct mdt_object *obj)
1482 struct obd_export *exp;
1483 struct md_attr ma = { 0 };
1486 if (!mdt_info_req(info))
1487 /* no req, so cannot tell about client connection flags:
1491 exp = mdt_info_req(info)->rq_export;
1492 if (exp_connect_encrypt(exp))
1493 /* client is encryption aware: fine */
1496 ma.ma_need = MA_INODE;
1497 mdt_attr_get_complex(info, obj, &ma);
1498 if (ma.ma_attr.la_valid & LA_FLAGS &&
1499 ma.ma_attr.la_flags & LUSTRE_ENCRYPT_FL)
1503 CDEBUG(D_SEC, "%s: operation on encrypted "DFID
1504 " from encryption-unaware client %s: %d\n",
1505 mdt_obd_name(info->mti_mdt),
1506 PFID(lu_object_fid(&obj->mot_obj)),
1507 libcfs_nidstr(&exp->exp_connection->c_peer.nid), rc);
1512 int mdt_fids_different_target(struct mdt_thread_info *info,
1513 const struct lu_fid *fid1,
1514 const struct lu_fid *fid2);
1516 int mdt_reint_migrate(struct mdt_thread_info *info,
1517 struct mdt_lock_handle *unused);
1518 int mdt_dir_layout_update(struct mdt_thread_info *info);
1520 /* directory restripe */
1521 int mdt_restripe_internal(struct mdt_thread_info *info,
1522 struct mdt_object *parent,
1523 struct mdt_object *child,
1524 const struct lu_name *lname,
1525 struct lu_fid *tfid,
1526 struct md_op_spec *spec,
1527 struct md_attr *ma);
1528 int mdt_restriper_start(struct mdt_device *mdt);
1529 void mdt_restriper_stop(struct mdt_device *mdt);
1530 void mdt_auto_split_add(struct mdt_thread_info *info, struct mdt_object *o);
1531 void mdt_restripe_migrate_add(struct mdt_thread_info *info,
1532 struct mdt_object *o);
1533 void mdt_restripe_update_add(struct mdt_thread_info *info,
1534 struct mdt_object *o);
1535 int mdt_is_remote_object(struct mdt_thread_info *info,
1536 struct mdt_object *parent,
1537 struct mdt_object *child);
1539 static unsigned int max_mod_rpcs_per_client = OBD_MAX_RIF_DEFAULT;
1540 #if OBD_OCD_VERSION(3, 0, 53, 0) > LUSTRE_VERSION_CODE
1541 static inline bool mdt_max_mod_rpcs_changed(struct mdt_device *mdt)
1543 return max_mod_rpcs_per_client != OBD_MAX_RIF_DEFAULT &&
1544 mdt->mdt_max_mod_rpcs_in_flight == OBD_MAX_RIF_DEFAULT;
1547 #define mdt_max_mod_rpcs_changed(mdt) false
1550 #endif /* _MDT_INTERNAL_H */