4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2010, 2015, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 /** \defgroup LDLM Lustre Distributed Lock Manager
39 * Lustre DLM is based on VAX DLM.
40 * Its two main roles are:
41 * - To provide locking assuring consistency of data on all Lustre nodes.
42 * - To allow clients to cache state protected by a lock by holding the
43 * lock until a conflicting lock is requested or it is expired by the LRU.
48 #ifndef _LUSTRE_DLM_H__
49 #define _LUSTRE_DLM_H__
51 #include "lustre_lib.h"
52 #include "lustre_net.h"
53 #include "lustre_import.h"
54 #include "lustre_handles.h"
55 #include "interval_tree.h" /* for interval_node{}, ldlm_extent */
58 #include "lustre_dlm_flags.h"
63 #define OBD_LDLM_DEVICENAME "ldlm"
65 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
66 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
67 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
70 * LDLM non-error return states
75 ELDLM_LOCK_CHANGED = 300,
76 ELDLM_LOCK_ABORTED = 301,
77 ELDLM_LOCK_REPLACED = 302,
78 ELDLM_NO_LOCK_DATA = 303,
79 ELDLM_LOCK_WOULDBLOCK = 304,
81 ELDLM_NAMESPACE_EXISTS = 400,
82 ELDLM_BAD_NAMESPACE = 401
86 * LDLM namespace type.
87 * The "client" type is actually an indication that this is a narrow local view
88 * into complete namespace on the server. Such namespaces cannot make any
89 * decisions about lack of conflicts or do any autonomous lock granting without
90 * first speaking to a server.
93 LDLM_NAMESPACE_SERVER = 1 << 0,
94 LDLM_NAMESPACE_CLIENT = 1 << 1
98 * The blocking callback is overloaded to perform two functions. These flags
99 * indicate which operation should be performed.
101 #define LDLM_CB_BLOCKING 1
102 #define LDLM_CB_CANCELING 2
105 * \name Lock Compatibility Matrix.
107 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
108 * Lock types are described in their respective implementation files:
109 * ldlm_{extent,flock,inodebits,plain}.c.
111 * There are six lock modes along with a compatibility matrix to indicate if
112 * two locks are compatible.
114 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
116 * - PW: Protective Write (normal write) mode. When a client requests a write
117 * lock from an OST, a lock with PW mode will be issued.
118 * - PR: Protective Read (normal read) mode. When a client requests a read from
119 * an OST, a lock with PR mode will be issued. Also, if the client opens a
120 * file for execution, it is granted a lock with PR mode.
121 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
122 * requests a write lock during a file open operation.
123 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
124 * an inodebit lock with the CR mode on the intermediate path component.
138 #define LCK_COMPAT_EX LCK_NL
139 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
140 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
141 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
142 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
143 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
144 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
145 #define LCK_COMPAT_COS (LCK_COS)
146 /** @} Lock Compatibility Matrix */
148 extern ldlm_mode_t lck_compat_array[];
150 static inline void lockmode_verify(ldlm_mode_t mode)
152 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
155 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
157 return (lck_compat_array[exist_mode] & new_mode);
162 * cluster name spaces
166 #define DLM_OST_NAMESPACE 1
167 #define DLM_MDS_NAMESPACE 2
170 - do we just separate this by security domains and use a prefix for
171 multiple namespaces in the same domain?
176 * Locking rules for LDLM:
181 * waiting_locks_spinlock
196 struct ldlm_resource;
197 struct ldlm_namespace;
200 * Operations on LDLM pools.
201 * LDLM pool is a pool of locks in the namespace without any implicitly
203 * Locks in the pool are organized in LRU.
204 * Local memory pressure or server instructions (e.g. mempressure on server)
205 * can trigger freeing of locks from the pool
207 struct ldlm_pool_ops {
208 /** Recalculate pool \a pl usage */
209 int (*po_recalc)(struct ldlm_pool *pl);
210 /** Cancel at least \a nr locks from pool \a pl */
211 int (*po_shrink)(struct ldlm_pool *pl, int nr,
215 /** One second for pools thread check interval. Each pool has own period. */
216 #define LDLM_POOLS_THREAD_PERIOD (1)
218 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
219 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
221 /** Default recalc period for server side pools in sec. */
222 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
224 /** Default recalc period for client side pools in sec. */
225 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
228 * LDLM pool structure to track granted locks.
229 * For purposes of determining when to release locks on e.g. memory pressure.
230 * This feature is commonly referred to as lru_resize.
233 /** Pool debugfs directory. */
234 struct dentry *pl_debugfs_entry;
235 /** Pool name, must be long enough to hold compound proc entry name. */
237 /** Lock for protecting SLV/CLV updates. */
239 /** Number of allowed locks in in pool, both, client and server side. */
241 /** Number of granted locks in */
243 /** Grant rate per T. */
244 atomic_t pl_grant_rate;
245 /** Cancel rate per T. */
246 atomic_t pl_cancel_rate;
247 /** Server lock volume (SLV). Protected by pl_lock. */
248 __u64 pl_server_lock_volume;
249 /** Current biggest client lock volume. Protected by pl_lock. */
250 __u64 pl_client_lock_volume;
251 /** Lock volume factor. SLV on client is calculated as following:
252 * server_slv * lock_volume_factor. */
253 atomic_t pl_lock_volume_factor;
254 /** Time when last SLV from server was obtained. */
255 time64_t pl_recalc_time;
256 /** Recalculation period for pool. */
257 time64_t pl_recalc_period;
258 /** Recalculation and shrink operations. */
259 const struct ldlm_pool_ops *pl_ops;
260 /** Number of planned locks for next period. */
262 /** Pool statistics. */
263 struct lprocfs_stats *pl_stats;
266 struct kobject pl_kobj;
267 struct completion pl_kobj_unregister;
270 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
274 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
275 * be associated with an LDLM lock and transferred from client to server and
278 * Currently LVBs are used by:
279 * - OSC-OST code to maintain current object size/times
280 * - layout lock code to return the layout when the layout lock is granted
282 struct ldlm_valblock_ops {
283 int (*lvbo_init)(struct ldlm_resource *res);
284 int (*lvbo_update)(struct ldlm_resource *res,
285 struct ptlrpc_request *r,
287 int (*lvbo_free)(struct ldlm_resource *res);
288 /* Return size of lvb data appropriate RPC size can be reserved */
289 int (*lvbo_size)(struct ldlm_lock *lock);
290 /* Called to fill in lvb data to RPC buffer @buf */
291 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
295 * LDLM pools related, type of lock pool in the namespace.
296 * Greedy means release cached locks aggressively
299 LDLM_NAMESPACE_GREEDY = 1 << 0,
300 LDLM_NAMESPACE_MODEST = 1 << 1
303 struct ldlm_ns_bucket {
304 /** back pointer to namespace */
305 struct ldlm_namespace *nsb_namespace;
307 * Estimated lock callback time. Used by adaptive timeout code to
308 * avoid spurious client evictions due to unresponsiveness when in
309 * fact the network or overall system load is at fault
311 struct adaptive_timeout nsb_at_estimate;
315 /** LDLM namespace lock stats */
322 LDLM_NS_TYPE_UNKNOWN = 0,
340 * Namespace serves to contain locks related to a particular service.
341 * There are two kinds of namespaces:
342 * - Server namespace has knowledge of all locks and is therefore authoritative
343 * to make decisions like what locks could be granted and what conflicts
344 * exist during new lock enqueue.
345 * - Client namespace only has limited knowledge about locks in the namespace,
346 * only seeing locks held by the client.
348 * Every Lustre service has one server namespace present on the server serving
349 * that service. Every client connected to the service has a client namespace
351 * Every lock obtained by client in that namespace is actually represented by
352 * two in-memory locks. One on the server and one on the client. The locks are
353 * linked by a special cookie by which one node can tell to the other which lock
354 * it actually means during communications. Such locks are called remote locks.
355 * The locks held by server only without any reference to a client are called
358 struct ldlm_namespace {
359 /** Backward link to OBD, required for LDLM pool to store new SLV. */
360 struct obd_device *ns_obd;
362 /** Flag indicating if namespace is on client instead of server */
363 ldlm_side_t ns_client;
365 /** Resource hash table for namespace. */
366 struct cfs_hash *ns_rs_hash;
371 /** big refcount (by bucket) */
375 * Namespace connect flags supported by server (may be changed via
376 * sysfs, LRU resize may be disabled/enabled).
378 __u64 ns_connect_flags;
380 /** Client side original connect flags supported by server. */
381 __u64 ns_orig_connect_flags;
383 /* namespace debugfs dir entry */
384 struct dentry *ns_debugfs_entry;
387 * Position in global namespace list linking all namespaces on
390 struct list_head ns_list_chain;
393 * List of unused locks for this namespace. This list is also called
395 * Unused locks are locks with zero reader/writer reference counts.
396 * This list is only used on clients for lock caching purposes.
397 * When we want to release some locks voluntarily or if server wants
398 * us to release some locks due to e.g. memory pressure, we take locks
399 * to release from the head of this list.
400 * Locks are linked via l_lru field in \see struct ldlm_lock.
402 struct list_head ns_unused_list;
403 /** Number of locks in the LRU list above */
407 * Maximum number of locks permitted in the LRU. If 0, means locks
408 * are managed by pools and there is no preset limit, rather it is all
409 * controlled by available memory on this client and on server.
411 unsigned int ns_max_unused;
412 /** Maximum allowed age (last used time) for locks in the LRU */
413 unsigned int ns_max_age;
416 * Used to rate-limit ldlm_namespace_dump calls.
417 * \see ldlm_namespace_dump. Increased by 10 seconds every time
420 unsigned long ns_next_dump;
423 * LVB operations for this namespace.
424 * \see struct ldlm_valblock_ops
426 struct ldlm_valblock_ops *ns_lvbo;
429 * Used by filter code to store pointer to OBD of the service.
430 * Should be dropped in favor of \a ns_obd
435 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
436 * a resource is removed.
438 wait_queue_head_t ns_waitq;
439 /** LDLM pool structure for this namespace */
440 struct ldlm_pool ns_pool;
441 /** Definition of how eagerly unused locks will be released from LRU */
442 ldlm_appetite_t ns_appetite;
444 /** Limit of parallel AST RPC count. */
445 unsigned ns_max_parallel_ast;
447 /** Callback to cancel locks before replaying it during recovery. */
448 ldlm_cancel_for_recovery ns_cancel_for_recovery;
450 /** LDLM lock stats */
451 struct lprocfs_stats *ns_stats;
454 * Flag to indicate namespace is being freed. Used to determine if
455 * recalculation of LDLM pool statistics should be skipped.
457 unsigned ns_stopping:1;
459 struct kobject ns_kobj; /* sysfs object */
460 struct completion ns_kobj_unregister;
464 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
466 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
468 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
472 * Returns 1 if this namespace supports lru_resize.
474 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
476 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
479 static inline void ns_register_cancel(struct ldlm_namespace *ns,
480 ldlm_cancel_for_recovery arg)
482 ns->ns_cancel_for_recovery = arg;
487 /** Type for blocking callback function of a lock. */
488 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
489 struct ldlm_lock_desc *new, void *data,
491 /** Type for completion callback function of a lock. */
492 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
494 /** Type for glimpse callback function of a lock. */
495 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
497 /** Work list for sending GL ASTs to multiple locks. */
498 struct ldlm_glimpse_work {
499 struct ldlm_lock *gl_lock; /* lock to glimpse */
500 struct list_head gl_list; /* linkage to other gl work structs */
501 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
502 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
503 * glimpse callback request */
506 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
507 #define LDLM_GL_WORK_NOFREE 0x1
509 /** Interval node data for each LDLM_EXTENT lock. */
510 struct ldlm_interval {
511 struct interval_node li_node; /* node for tree management */
512 struct list_head li_group; /* the locks which have the same
513 * policy - group of the policy */
516 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
519 * Interval tree for extent locks.
520 * The interval tree must be accessed under the resource lock.
521 * Interval trees are used for granted extent locks to speed up conflicts
522 * lookup. See ldlm/interval_tree.c for more details.
524 struct ldlm_interval_tree {
527 ldlm_mode_t lit_mode; /* lock mode */
528 struct interval_node *lit_root; /* actual ldlm_interval */
531 /** Whether to track references to exports by LDLM locks. */
532 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
536 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
537 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
538 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
540 } ldlm_cancel_flags_t;
546 __u64 blocking_owner;
547 struct obd_export *blocking_export;
548 /* Protected by the hash lock */
554 struct ldlm_extent l_extent;
555 struct ldlm_flock l_flock;
556 struct ldlm_inodebits l_inodebits;
557 } ldlm_policy_data_t;
559 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
560 const ldlm_wire_policy_data_t *wpolicy,
561 ldlm_policy_data_t *lpolicy);
571 * LDLM lock structure
573 * Represents a single LDLM lock and its state in memory. Each lock is
574 * associated with a single ldlm_resource, the object which is being
575 * locked. There may be multiple ldlm_locks on a single resource,
576 * depending on the lock type and whether the locks are conflicting or
582 * When remote side wants to tell us about a lock, they address
583 * it by this opaque handle. The handle does not hold a
584 * reference on the ldlm_lock, so it can be safely passed to
585 * other threads or nodes. When the lock needs to be accessed
586 * from the handle, it is looked up again in the lock table, and
587 * may no longer exist.
589 * Must be first in the structure.
591 struct portals_handle l_handle;
593 * Lock reference count.
594 * This is how many users have pointers to actual structure, so that
595 * we do not accidentally free lock structure that is in use.
599 * Internal spinlock protects l_resource. We should hold this lock
600 * first before taking res_lock.
604 * Pointer to actual resource this lock is in.
605 * ldlm_lock_change_resource() can change this.
607 struct ldlm_resource *l_resource;
609 * List item for client side LRU list.
610 * Protected by ns_lock in struct ldlm_namespace.
612 struct list_head l_lru;
614 * Linkage to resource's lock queues according to current lock state.
615 * (could be granted, waiting or converting)
616 * Protected by lr_lock in struct ldlm_resource.
618 struct list_head l_res_link;
620 * Tree node for ldlm_extent.
622 struct ldlm_interval *l_tree_node;
624 * Per export hash of locks.
625 * Protected by per-bucket exp->exp_lock_hash locks.
627 struct hlist_node l_exp_hash;
629 * Per export hash of flock locks.
630 * Protected by per-bucket exp->exp_flock_hash locks.
632 struct hlist_node l_exp_flock_hash;
635 * Protected by lr_lock.
637 ldlm_mode_t l_req_mode;
639 * Granted mode, also protected by lr_lock.
641 ldlm_mode_t l_granted_mode;
642 /** Lock completion handler pointer. Called when lock is granted. */
643 ldlm_completion_callback l_completion_ast;
645 * Lock blocking AST handler pointer.
646 * It plays two roles:
647 * - as a notification of an attempt to queue a conflicting lock (once)
648 * - as a notification when the lock is being cancelled.
650 * As such it's typically called twice: once for the initial conflict
651 * and then once more when the last user went away and the lock is
652 * cancelled (could happen recursively).
654 ldlm_blocking_callback l_blocking_ast;
656 * Lock glimpse handler.
657 * Glimpse handler is used to obtain LVB updates from a client by
660 ldlm_glimpse_callback l_glimpse_ast;
664 * This is a pointer to actual client export for locks that were granted
665 * to clients. Used server-side.
667 struct obd_export *l_export;
669 * Lock connection export.
670 * Pointer to server export on a client.
672 struct obd_export *l_conn_export;
675 * Remote lock handle.
676 * If the lock is remote, this is the handle of the other side lock
679 struct lustre_handle l_remote_handle;
682 * Representation of private data specific for a lock type.
683 * Examples are: extent range for extent lock or bitmask for ibits locks
685 ldlm_policy_data_t l_policy_data;
688 * Lock state flags. Protected by lr_lock.
689 * \see lustre_dlm_flags.h where the bits are defined.
694 * Lock r/w usage counters.
695 * Protected by lr_lock.
700 * If the lock is granted, a process sleeps on this waitq to learn when
701 * it's no longer in use. If the lock is not granted, a process sleeps
702 * on this waitq to learn when it becomes granted.
704 wait_queue_head_t l_waitq;
707 * Seconds. It will be updated if there is any activity related to
708 * the lock, e.g. enqueue the lock or send blocking AST.
710 time64_t l_last_activity;
713 * Time last used by e.g. being matched by lock match.
714 * Jiffies. Should be converted to time if needed.
716 unsigned long l_last_used;
718 /** Originally requested extent for the extent lock. */
719 struct ldlm_extent l_req_extent;
722 * Client-side-only members.
725 enum lvb_type l_lvb_type;
728 * Temporary storage for a LVB received during an enqueue operation.
733 /** Private storage for lock user. Opaque to LDLM. */
737 * Server-side-only members.
741 * Connection cookie for the client originating the operation.
742 * Used by Commit on Share (COS) code. Currently only used for
743 * inodebits locks on MDS.
745 __u64 l_client_cookie;
748 * List item for locks waiting for cancellation from clients.
749 * The lists this could be linked into are:
750 * waiting_locks_list (protected by waiting_locks_spinlock),
751 * then if the lock timed out, it is moved to
752 * expired_lock_thread.elt_expired_locks for further processing.
753 * Protected by elt_lock.
755 struct list_head l_pending_chain;
758 * Set when lock is sent a blocking AST. Time in seconds when timeout
759 * is reached and client holding this lock could be evicted.
760 * This timeout could be further extended by e.g. certain IO activity
762 * \see ost_rw_prolong_locks
764 unsigned long l_callback_timeout;
766 /** Local PID of process which created this lock. */
770 * Number of times blocking AST was sent for this lock.
771 * This is for debugging. Valid values are 0 and 1, if there is an
772 * attempt to send blocking AST more than once, an assertion would be
773 * hit. \see ldlm_work_bl_ast_lock
776 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
777 struct list_head l_bl_ast;
778 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
779 struct list_head l_cp_ast;
780 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
781 struct list_head l_rk_ast;
784 * Pointer to a conflicting lock that caused blocking AST to be sent
787 struct ldlm_lock *l_blocking_lock;
790 * Protected by lr_lock, linkages to "skip lists".
791 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
793 struct list_head l_sl_mode;
794 struct list_head l_sl_policy;
796 /** Reference tracking structure to debug leaked locks. */
797 struct lu_ref l_reference;
798 #if LUSTRE_TRACKS_LOCK_EXP_REFS
799 /* Debugging stuff for bug 20498, for tracking export references. */
800 /** number of export references taken */
802 /** link all locks referencing one export */
803 struct list_head l_exp_refs_link;
804 /** referenced export object */
805 struct obd_export *l_exp_refs_target;
808 * export blocking dlm lock list, protected by
809 * l_export->exp_bl_list_lock.
810 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
811 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
813 struct list_head l_exp_list;
817 * LDLM resource description.
818 * Basically, resource is a representation for a single object.
819 * Object has a name which is currently 4 64-bit integers. LDLM user is
820 * responsible for creation of a mapping between objects it wants to be
821 * protected and resource names.
823 * A resource can only hold locks of a single lock type, though there may be
824 * multiple ldlm_locks on a single resource, depending on the lock type and
825 * whether the locks are conflicting or not.
827 struct ldlm_resource {
828 struct ldlm_ns_bucket *lr_ns_bucket;
831 * List item for list in namespace hash.
832 * protected by ns_lock
834 struct hlist_node lr_hash;
836 /** Spinlock to protect locks under this resource. */
840 * protected by lr_lock
842 /** List of locks in granted state */
843 struct list_head lr_granted;
845 * List of locks that could not be granted due to conflicts and
846 * that are waiting for conflicts to go away */
847 struct list_head lr_waiting;
850 /* XXX No longer needed? Remove ASAP */
851 ldlm_mode_t lr_most_restr;
853 /** Type of locks this resource can hold. Only one type per resource. */
854 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
857 struct ldlm_res_id lr_name;
858 /** Reference count for this resource */
859 atomic_t lr_refcount;
862 * Interval trees (only for extent locks) for all modes of this resource
864 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM];
867 * Server-side-only lock value block elements.
868 * To serialize lvbo_init.
870 struct mutex lr_lvb_mutex;
873 /** When the resource was considered as contended. */
874 unsigned long lr_contention_time;
875 /** List of references to this resource. For debugging. */
876 struct lu_ref lr_reference;
878 struct inode *lr_lvb_inode;
881 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
883 return lock->l_resource->lr_type == LDLM_IBITS &&
884 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
888 ldlm_ns_name(struct ldlm_namespace *ns)
890 return ns->ns_rs_hash->hs_name;
893 static inline struct ldlm_namespace *
894 ldlm_res_to_ns(struct ldlm_resource *res)
896 return res->lr_ns_bucket->nsb_namespace;
899 static inline struct ldlm_namespace *
900 ldlm_lock_to_ns(struct ldlm_lock *lock)
902 return ldlm_res_to_ns(lock->l_resource);
906 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
908 return ldlm_ns_name(ldlm_lock_to_ns(lock));
911 static inline struct adaptive_timeout *
912 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
914 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
917 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
919 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
921 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init)
922 return ns->ns_lvbo->lvbo_init(res);
927 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
929 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
931 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_size)
932 return ns->ns_lvbo->lvbo_size(lock);
937 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
939 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
942 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
947 struct ldlm_ast_work {
948 struct ldlm_lock *w_lock;
950 struct ldlm_lock_desc w_desc;
951 struct list_head w_list;
958 * Common ldlm_enqueue parameters
960 struct ldlm_enqueue_info {
961 __u32 ei_type; /** Type of the lock being enqueued. */
962 __u32 ei_mode; /** Mode of the lock being enqueued. */
963 void *ei_cb_bl; /** blocking lock callback */
964 void *ei_cb_cp; /** lock completion callback */
965 void *ei_cb_gl; /** lock glimpse callback */
966 void *ei_cbdata; /** Data to be passed into callbacks. */
969 extern struct obd_ops ldlm_obd_ops;
971 extern char *ldlm_lockname[];
972 char *ldlm_it2str(int it);
975 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
976 * For the cases where we do not have actual lock to print along
977 * with a debugging message that is ldlm-related
979 #define LDLM_DEBUG_NOLOCK(format, a...) \
980 CDEBUG(D_DLMTRACE, "### " format "\n", ##a)
983 * Support function for lock information printing into debug logs.
986 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
987 CFS_CHECK_STACK(msgdata, mask, cdls); \
989 if (((mask) & D_CANTMASK) != 0 || \
990 ((libcfs_debug & (mask)) != 0 && \
991 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
992 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
995 void _ldlm_lock_debug(struct ldlm_lock *lock,
996 struct libcfs_debug_msg_data *data,
997 const char *fmt, ...)
1001 * Rate-limited version of lock printing function.
1003 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1004 static struct cfs_debug_limit_state _ldlm_cdls; \
1005 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1006 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt, ##a);\
1009 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1010 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1012 /** Non-rate-limited lock printing function for debugging purposes. */
1013 #define LDLM_DEBUG(lock, fmt, a...) do { \
1014 if (likely(lock)) { \
1015 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1016 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1019 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1023 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1024 int first_enq, ldlm_error_t *err,
1025 struct list_head *work_list);
1028 * Return values for lock iterators.
1029 * Also used during deciding of lock grants and cancellations.
1031 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1032 #define LDLM_ITER_STOP 2 /* stop iterating */
1034 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1035 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1037 /** \defgroup ldlm_iterator Lock iterators
1039 * LDLM provides for a way to iterate through every lock on a resource or
1040 * namespace or every resource in a namespace.
1042 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1043 ldlm_iterator_t iter, void *data);
1044 /** @} ldlm_iterator */
1046 int ldlm_replay_locks(struct obd_import *imp);
1049 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1052 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1054 struct ldlm_callback_suite {
1055 ldlm_completion_callback lcs_completion;
1056 ldlm_blocking_callback lcs_blocking;
1057 ldlm_glimpse_callback lcs_glimpse;
1061 int ldlm_get_ref(void);
1062 void ldlm_put_ref(void);
1063 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1066 void ldlm_lock2handle(const struct ldlm_lock *lock,
1067 struct lustre_handle *lockh);
1068 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1069 void ldlm_cancel_callback(struct ldlm_lock *);
1070 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1071 int ldlm_lock_set_data(struct lustre_handle *, void *);
1074 * Obtain a lock reference by its handle.
1076 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1078 return __ldlm_handle2lock(h, 0);
1081 #define LDLM_LOCK_REF_DEL(lock) \
1082 lu_ref_del(&lock->l_reference, "handle", current)
1084 static inline struct ldlm_lock *
1085 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1087 struct ldlm_lock *lock;
1089 lock = __ldlm_handle2lock(h, flags);
1091 LDLM_LOCK_REF_DEL(lock);
1096 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1097 * data from request \a r
1099 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1100 struct ptlrpc_request *r, int increase)
1102 if (ldlm_res_to_ns(res)->ns_lvbo &&
1103 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1104 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1110 int ldlm_error2errno(ldlm_error_t error);
1112 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1113 void ldlm_dump_export_locks(struct obd_export *exp);
1117 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1118 * __ldlm_handle2lock().
1120 #define LDLM_LOCK_PUT(lock) \
1122 LDLM_LOCK_REF_DEL(lock); \
1123 /*LDLM_DEBUG((lock), "put");*/ \
1124 ldlm_lock_put(lock); \
1128 * Release a lock reference obtained by some other means (see
1131 #define LDLM_LOCK_RELEASE(lock) \
1133 /*LDLM_DEBUG((lock), "put");*/ \
1134 ldlm_lock_put(lock); \
1137 #define LDLM_LOCK_GET(lock) \
1139 ldlm_lock_get(lock); \
1140 /*LDLM_DEBUG((lock), "get");*/ \
1144 #define ldlm_lock_list_put(head, member, count) \
1146 struct ldlm_lock *_lock, *_next; \
1148 list_for_each_entry_safe(_lock, _next, head, member) { \
1151 list_del_init(&_lock->member); \
1152 LDLM_LOCK_RELEASE(_lock); \
1157 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1158 void ldlm_lock_put(struct ldlm_lock *lock);
1159 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1160 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1161 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1162 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1163 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1164 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1165 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1166 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1167 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1168 const struct ldlm_res_id *, ldlm_type_t type,
1169 ldlm_policy_data_t *, ldlm_mode_t mode,
1170 struct lustre_handle *, int unref);
1171 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1173 void ldlm_lock_cancel(struct ldlm_lock *lock);
1174 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1175 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1178 struct ldlm_namespace *
1179 ldlm_namespace_new(struct obd_device *obd, char *name,
1180 ldlm_side_t client, ldlm_appetite_t apt,
1181 ldlm_ns_type_t ns_type);
1182 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1183 void ldlm_namespace_get(struct ldlm_namespace *ns);
1184 void ldlm_namespace_put(struct ldlm_namespace *ns);
1185 int ldlm_debugfs_setup(void);
1186 void ldlm_debugfs_cleanup(void);
1188 /* resource.c - internal */
1189 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1190 struct ldlm_resource *parent,
1191 const struct ldlm_res_id *,
1192 ldlm_type_t type, int create);
1193 int ldlm_resource_putref(struct ldlm_resource *res);
1194 void ldlm_resource_add_lock(struct ldlm_resource *res,
1195 struct list_head *head,
1196 struct ldlm_lock *lock);
1197 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1198 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1199 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1200 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1201 void ldlm_resource_dump(int level, struct ldlm_resource *);
1202 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1203 const struct ldlm_res_id *);
1205 #define LDLM_RESOURCE_ADDREF(res) do { \
1206 lu_ref_add_atomic(&(res)->lr_reference, __func__, current); \
1209 #define LDLM_RESOURCE_DELREF(res) do { \
1210 lu_ref_del(&(res)->lr_reference, __func__, current); \
1213 /* ldlm_request.c */
1214 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1215 * These AST handlers are typically used for server-side local locks and are
1216 * also used by client-side lock handlers to perform minimum level base
1219 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1220 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1221 /** @} ldlm_local_ast */
1223 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1224 * These are typically used by client and server (*_local versions)
1225 * to obtain and release locks.
1227 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1228 struct ldlm_enqueue_info *einfo,
1229 const struct ldlm_res_id *res_id,
1230 ldlm_policy_data_t const *policy, __u64 *flags,
1231 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1232 struct lustre_handle *lockh, int async);
1233 int ldlm_prep_enqueue_req(struct obd_export *exp,
1234 struct ptlrpc_request *req,
1235 struct list_head *cancels,
1237 int ldlm_prep_elc_req(struct obd_export *exp,
1238 struct ptlrpc_request *req,
1239 int version, int opc, int canceloff,
1240 struct list_head *cancels, int count);
1242 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1243 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1244 __u64 *flags, void *lvb, __u32 lvb_len,
1245 struct lustre_handle *lockh, int rc);
1246 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1247 int ldlm_cli_cancel(struct lustre_handle *lockh,
1248 ldlm_cancel_flags_t cancel_flags);
1249 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1250 ldlm_cancel_flags_t flags, void *opaque);
1251 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1252 const struct ldlm_res_id *res_id,
1253 ldlm_policy_data_t *policy,
1255 ldlm_cancel_flags_t flags,
1257 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1258 struct list_head *cancels,
1259 ldlm_policy_data_t *policy,
1260 ldlm_mode_t mode, __u64 lock_flags,
1261 ldlm_cancel_flags_t cancel_flags, void *opaque);
1262 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1263 ldlm_cancel_flags_t flags);
1264 int ldlm_cli_cancel_list(struct list_head *head, int count,
1265 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1266 /** @} ldlm_cli_api */
1269 /* This has to be here because recursive inclusion sucks. */
1270 int intent_disposition(struct ldlm_reply *rep, int flag);
1271 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1273 /* ioctls for trying requests */
1274 #define IOC_LDLM_TYPE 'f'
1275 #define IOC_LDLM_MIN_NR 40
1277 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1278 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1279 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1280 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1281 #define IOC_LDLM_MAX_NR 43
1284 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1285 * than one lock_res is dead-lock safe.
1287 enum lock_res_type {
1292 /** Lock resource. */
1293 static inline void lock_res(struct ldlm_resource *res)
1295 spin_lock(&res->lr_lock);
1298 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1299 static inline void lock_res_nested(struct ldlm_resource *res,
1300 enum lock_res_type mode)
1302 spin_lock_nested(&res->lr_lock, mode);
1305 /** Unlock resource. */
1306 static inline void unlock_res(struct ldlm_resource *res)
1308 spin_unlock(&res->lr_lock);
1311 /** Check if resource is already locked, assert if not. */
1312 static inline void check_res_locked(struct ldlm_resource *res)
1314 assert_spin_locked(&res->lr_lock);
1317 struct ldlm_resource *lock_res_and_lock(struct ldlm_lock *lock);
1318 void unlock_res_and_lock(struct ldlm_lock *lock);
1321 /** \defgroup ldlm_pools Various LDLM pool related functions
1322 * There are not used outside of ldlm.
1325 int ldlm_pools_init(void);
1326 void ldlm_pools_fini(void);
1328 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1329 int idx, ldlm_side_t client);
1330 void ldlm_pool_fini(struct ldlm_pool *pl);
1331 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1332 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);