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
74 ELDLM_LOCK_MATCHED = 1,
76 ELDLM_LOCK_CHANGED = 300,
77 ELDLM_LOCK_ABORTED = 301,
78 ELDLM_LOCK_REPLACED = 302,
79 ELDLM_NO_LOCK_DATA = 303,
80 ELDLM_LOCK_WOULDBLOCK = 304,
82 ELDLM_NAMESPACE_EXISTS = 400,
83 ELDLM_BAD_NAMESPACE = 401
87 * LDLM namespace type.
88 * The "client" type is actually an indication that this is a narrow local view
89 * into complete namespace on the server. Such namespaces cannot make any
90 * decisions about lack of conflicts or do any autonomous lock granting without
91 * first speaking to a server.
94 LDLM_NAMESPACE_SERVER = 1 << 0,
95 LDLM_NAMESPACE_CLIENT = 1 << 1
99 * The blocking callback is overloaded to perform two functions. These flags
100 * indicate which operation should be performed.
102 #define LDLM_CB_BLOCKING 1
103 #define LDLM_CB_CANCELING 2
106 * \name Lock Compatibility Matrix.
108 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
109 * Lock types are described in their respective implementation files:
110 * ldlm_{extent,flock,inodebits,plain}.c.
112 * There are six lock modes along with a compatibility matrix to indicate if
113 * two locks are compatible.
115 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
117 * - PW: Protective Write (normal write) mode. When a client requests a write
118 * lock from an OST, a lock with PW mode will be issued.
119 * - PR: Protective Read (normal read) mode. When a client requests a read from
120 * an OST, a lock with PR mode will be issued. Also, if the client opens a
121 * file for execution, it is granted a lock with PR mode.
122 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
123 * requests a write lock during a file open operation.
124 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
125 * an inodebit lock with the CR mode on the intermediate path component.
139 #define LCK_COMPAT_EX LCK_NL
140 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
141 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
142 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
143 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
144 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
145 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
146 #define LCK_COMPAT_COS (LCK_COS)
147 /** @} Lock Compatibility Matrix */
149 extern enum ldlm_mode lck_compat_array[];
151 static inline void lockmode_verify(enum ldlm_mode mode)
153 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
156 static inline int lockmode_compat(enum ldlm_mode exist_mode,
157 enum ldlm_mode new_mode)
159 return (lck_compat_array[exist_mode] & new_mode);
164 * cluster name spaces
168 #define DLM_OST_NAMESPACE 1
169 #define DLM_MDS_NAMESPACE 2
172 - do we just separate this by security domains and use a prefix for
173 multiple namespaces in the same domain?
178 * Locking rules for LDLM:
183 * waiting_locks_spinlock
198 struct ldlm_resource;
199 struct ldlm_namespace;
202 * Operations on LDLM pools.
203 * LDLM pool is a pool of locks in the namespace without any implicitly
205 * Locks in the pool are organized in LRU.
206 * Local memory pressure or server instructions (e.g. mempressure on server)
207 * can trigger freeing of locks from the pool
209 struct ldlm_pool_ops {
210 /** Recalculate pool \a pl usage */
211 int (*po_recalc)(struct ldlm_pool *pl);
212 /** Cancel at least \a nr locks from pool \a pl */
213 int (*po_shrink)(struct ldlm_pool *pl, int nr,
217 /** One second for pools thread check interval. Each pool has own period. */
218 #define LDLM_POOLS_THREAD_PERIOD (1)
220 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
221 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
223 /** Default recalc period for server side pools in sec. */
224 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
226 /** Default recalc period for client side pools in sec. */
227 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
230 * LDLM pool structure to track granted locks.
231 * For purposes of determining when to release locks on e.g. memory pressure.
232 * This feature is commonly referred to as lru_resize.
235 /** Pool debugfs directory. */
236 struct dentry *pl_debugfs_entry;
237 /** Pool name, must be long enough to hold compound proc entry name. */
239 /** Lock for protecting SLV/CLV updates. */
241 /** Number of allowed locks in in pool, both, client and server side. */
243 /** Number of granted locks in */
245 /** Grant rate per T. */
246 atomic_t pl_grant_rate;
247 /** Cancel rate per T. */
248 atomic_t pl_cancel_rate;
249 /** Server lock volume (SLV). Protected by pl_lock. */
250 __u64 pl_server_lock_volume;
251 /** Current biggest client lock volume. Protected by pl_lock. */
252 __u64 pl_client_lock_volume;
253 /** Lock volume factor. SLV on client is calculated as following:
254 * server_slv * lock_volume_factor.
256 atomic_t pl_lock_volume_factor;
257 /** Time when last SLV from server was obtained. */
258 time64_t pl_recalc_time;
259 /** Recalculation period for pool. */
260 time64_t pl_recalc_period;
261 /** Recalculation and shrink operations. */
262 const struct ldlm_pool_ops *pl_ops;
263 /** Number of planned locks for next period. */
265 /** Pool statistics. */
266 struct lprocfs_stats *pl_stats;
269 struct kobject pl_kobj;
270 struct completion pl_kobj_unregister;
273 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
277 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
278 * be associated with an LDLM lock and transferred from client to server and
281 * Currently LVBs are used by:
282 * - OSC-OST code to maintain current object size/times
283 * - layout lock code to return the layout when the layout lock is granted
285 struct ldlm_valblock_ops {
286 int (*lvbo_init)(struct ldlm_resource *res);
287 int (*lvbo_update)(struct ldlm_resource *res,
288 struct ptlrpc_request *r,
290 int (*lvbo_free)(struct ldlm_resource *res);
291 /* Return size of lvb data appropriate RPC size can be reserved */
292 int (*lvbo_size)(struct ldlm_lock *lock);
293 /* Called to fill in lvb data to RPC buffer @buf */
294 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
298 * LDLM pools related, type of lock pool in the namespace.
299 * Greedy means release cached locks aggressively
302 LDLM_NAMESPACE_GREEDY = 1 << 0,
303 LDLM_NAMESPACE_MODEST = 1 << 1
306 struct ldlm_ns_bucket {
307 /** back pointer to namespace */
308 struct ldlm_namespace *nsb_namespace;
310 * Estimated lock callback time. Used by adaptive timeout code to
311 * avoid spurious client evictions due to unresponsiveness when in
312 * fact the network or overall system load is at fault
314 struct adaptive_timeout nsb_at_estimate;
318 /** LDLM namespace lock stats */
325 LDLM_NS_TYPE_UNKNOWN = 0,
343 * Namespace serves to contain locks related to a particular service.
344 * There are two kinds of namespaces:
345 * - Server namespace has knowledge of all locks and is therefore authoritative
346 * to make decisions like what locks could be granted and what conflicts
347 * exist during new lock enqueue.
348 * - Client namespace only has limited knowledge about locks in the namespace,
349 * only seeing locks held by the client.
351 * Every Lustre service has one server namespace present on the server serving
352 * that service. Every client connected to the service has a client namespace
354 * Every lock obtained by client in that namespace is actually represented by
355 * two in-memory locks. One on the server and one on the client. The locks are
356 * linked by a special cookie by which one node can tell to the other which lock
357 * it actually means during communications. Such locks are called remote locks.
358 * The locks held by server only without any reference to a client are called
361 struct ldlm_namespace {
362 /** Backward link to OBD, required for LDLM pool to store new SLV. */
363 struct obd_device *ns_obd;
365 /** Flag indicating if namespace is on client instead of server */
366 ldlm_side_t ns_client;
368 /** Resource hash table for namespace. */
369 struct cfs_hash *ns_rs_hash;
374 /** big refcount (by bucket) */
378 * Namespace connect flags supported by server (may be changed via
379 * sysfs, LRU resize may be disabled/enabled).
381 __u64 ns_connect_flags;
383 /** Client side original connect flags supported by server. */
384 __u64 ns_orig_connect_flags;
386 /* namespace debugfs dir entry */
387 struct dentry *ns_debugfs_entry;
390 * Position in global namespace list linking all namespaces on
393 struct list_head ns_list_chain;
396 * List of unused locks for this namespace. This list is also called
398 * Unused locks are locks with zero reader/writer reference counts.
399 * This list is only used on clients for lock caching purposes.
400 * When we want to release some locks voluntarily or if server wants
401 * us to release some locks due to e.g. memory pressure, we take locks
402 * to release from the head of this list.
403 * Locks are linked via l_lru field in \see struct ldlm_lock.
405 struct list_head ns_unused_list;
406 /** Number of locks in the LRU list above */
410 * Maximum number of locks permitted in the LRU. If 0, means locks
411 * are managed by pools and there is no preset limit, rather it is all
412 * controlled by available memory on this client and on server.
414 unsigned int ns_max_unused;
415 /** Maximum allowed age (last used time) for locks in the LRU */
416 unsigned int ns_max_age;
419 * Used to rate-limit ldlm_namespace_dump calls.
420 * \see ldlm_namespace_dump. Increased by 10 seconds every time
423 unsigned long ns_next_dump;
426 * LVB operations for this namespace.
427 * \see struct ldlm_valblock_ops
429 struct ldlm_valblock_ops *ns_lvbo;
432 * Used by filter code to store pointer to OBD of the service.
433 * Should be dropped in favor of \a ns_obd
438 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
439 * a resource is removed.
441 wait_queue_head_t ns_waitq;
442 /** LDLM pool structure for this namespace */
443 struct ldlm_pool ns_pool;
444 /** Definition of how eagerly unused locks will be released from LRU */
445 enum ldlm_appetite ns_appetite;
447 /** Limit of parallel AST RPC count. */
448 unsigned ns_max_parallel_ast;
451 * Callback to check if a lock is good to be canceled by ELC or
454 ldlm_cancel_cbt ns_cancel;
456 /** LDLM lock stats */
457 struct lprocfs_stats *ns_stats;
460 * Flag to indicate namespace is being freed. Used to determine if
461 * recalculation of LDLM pool statistics should be skipped.
463 unsigned ns_stopping:1;
465 struct kobject ns_kobj; /* sysfs object */
466 struct completion ns_kobj_unregister;
470 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
472 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
474 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
478 * Returns 1 if this namespace supports lru_resize.
480 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
482 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
485 static inline void ns_register_cancel(struct ldlm_namespace *ns,
493 /** Type for blocking callback function of a lock. */
494 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
495 struct ldlm_lock_desc *new, void *data,
497 /** Type for completion callback function of a lock. */
498 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
500 /** Type for glimpse callback function of a lock. */
501 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
503 /** Work list for sending GL ASTs to multiple locks. */
504 struct ldlm_glimpse_work {
505 struct ldlm_lock *gl_lock; /* lock to glimpse */
506 struct list_head gl_list; /* linkage to other gl work structs */
507 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
508 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
509 * glimpse callback request
513 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
514 #define LDLM_GL_WORK_NOFREE 0x1
516 /** Interval node data for each LDLM_EXTENT lock. */
517 struct ldlm_interval {
518 struct interval_node li_node; /* node for tree management */
519 struct list_head li_group; /* the locks which have the same
520 * policy - group of the policy
524 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
527 * Interval tree for extent locks.
528 * The interval tree must be accessed under the resource lock.
529 * Interval trees are used for granted extent locks to speed up conflicts
530 * lookup. See ldlm/interval_tree.c for more details.
532 struct ldlm_interval_tree {
535 enum ldlm_mode lit_mode; /* lock mode */
536 struct interval_node *lit_root; /* actual ldlm_interval */
539 /** Whether to track references to exports by LDLM locks. */
540 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
543 enum ldlm_cancel_flags {
544 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
545 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
546 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
555 __u64 blocking_owner;
556 struct obd_export *blocking_export;
557 /* Protected by the hash lock */
563 struct ldlm_extent l_extent;
564 struct ldlm_flock l_flock;
565 struct ldlm_inodebits l_inodebits;
566 } ldlm_policy_data_t;
568 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
569 const ldlm_wire_policy_data_t *wpolicy,
570 ldlm_policy_data_t *lpolicy);
580 * LDLM lock structure
582 * Represents a single LDLM lock and its state in memory. Each lock is
583 * associated with a single ldlm_resource, the object which is being
584 * locked. There may be multiple ldlm_locks on a single resource,
585 * depending on the lock type and whether the locks are conflicting or
591 * When remote side wants to tell us about a lock, they address
592 * it by this opaque handle. The handle does not hold a
593 * reference on the ldlm_lock, so it can be safely passed to
594 * other threads or nodes. When the lock needs to be accessed
595 * from the handle, it is looked up again in the lock table, and
596 * may no longer exist.
598 * Must be first in the structure.
600 struct portals_handle l_handle;
602 * Lock reference count.
603 * This is how many users have pointers to actual structure, so that
604 * we do not accidentally free lock structure that is in use.
608 * Internal spinlock protects l_resource. We should hold this lock
609 * first before taking res_lock.
613 * Pointer to actual resource this lock is in.
614 * ldlm_lock_change_resource() can change this.
616 struct ldlm_resource *l_resource;
618 * List item for client side LRU list.
619 * Protected by ns_lock in struct ldlm_namespace.
621 struct list_head l_lru;
623 * Linkage to resource's lock queues according to current lock state.
624 * (could be granted, waiting or converting)
625 * Protected by lr_lock in struct ldlm_resource.
627 struct list_head l_res_link;
629 * Tree node for ldlm_extent.
631 struct ldlm_interval *l_tree_node;
633 * Per export hash of locks.
634 * Protected by per-bucket exp->exp_lock_hash locks.
636 struct hlist_node l_exp_hash;
638 * Per export hash of flock locks.
639 * Protected by per-bucket exp->exp_flock_hash locks.
641 struct hlist_node l_exp_flock_hash;
644 * Protected by lr_lock.
646 enum ldlm_mode l_req_mode;
648 * Granted mode, also protected by lr_lock.
650 enum ldlm_mode l_granted_mode;
651 /** Lock completion handler pointer. Called when lock is granted. */
652 ldlm_completion_callback l_completion_ast;
654 * Lock blocking AST handler pointer.
655 * It plays two roles:
656 * - as a notification of an attempt to queue a conflicting lock (once)
657 * - as a notification when the lock is being cancelled.
659 * As such it's typically called twice: once for the initial conflict
660 * and then once more when the last user went away and the lock is
661 * cancelled (could happen recursively).
663 ldlm_blocking_callback l_blocking_ast;
665 * Lock glimpse handler.
666 * Glimpse handler is used to obtain LVB updates from a client by
669 ldlm_glimpse_callback l_glimpse_ast;
673 * This is a pointer to actual client export for locks that were granted
674 * to clients. Used server-side.
676 struct obd_export *l_export;
678 * Lock connection export.
679 * Pointer to server export on a client.
681 struct obd_export *l_conn_export;
684 * Remote lock handle.
685 * If the lock is remote, this is the handle of the other side lock
688 struct lustre_handle l_remote_handle;
691 * Representation of private data specific for a lock type.
692 * Examples are: extent range for extent lock or bitmask for ibits locks
694 ldlm_policy_data_t l_policy_data;
697 * Lock state flags. Protected by lr_lock.
698 * \see lustre_dlm_flags.h where the bits are defined.
703 * Lock r/w usage counters.
704 * Protected by lr_lock.
709 * If the lock is granted, a process sleeps on this waitq to learn when
710 * it's no longer in use. If the lock is not granted, a process sleeps
711 * on this waitq to learn when it becomes granted.
713 wait_queue_head_t l_waitq;
716 * Seconds. It will be updated if there is any activity related to
717 * the lock, e.g. enqueue the lock or send blocking AST.
719 time64_t l_last_activity;
722 * Time last used by e.g. being matched by lock match.
723 * Jiffies. Should be converted to time if needed.
725 unsigned long l_last_used;
727 /** Originally requested extent for the extent lock. */
728 struct ldlm_extent l_req_extent;
731 * Client-side-only members.
734 enum lvb_type l_lvb_type;
737 * Temporary storage for a LVB received during an enqueue operation.
742 /** Private storage for lock user. Opaque to LDLM. */
746 * Server-side-only members.
750 * Connection cookie for the client originating the operation.
751 * Used by Commit on Share (COS) code. Currently only used for
752 * inodebits locks on MDS.
754 __u64 l_client_cookie;
757 * List item for locks waiting for cancellation from clients.
758 * The lists this could be linked into are:
759 * waiting_locks_list (protected by waiting_locks_spinlock),
760 * then if the lock timed out, it is moved to
761 * expired_lock_thread.elt_expired_locks for further processing.
762 * Protected by elt_lock.
764 struct list_head l_pending_chain;
767 * Set when lock is sent a blocking AST. Time in seconds when timeout
768 * is reached and client holding this lock could be evicted.
769 * This timeout could be further extended by e.g. certain IO activity
771 * \see ost_rw_prolong_locks
773 unsigned long l_callback_timeout;
775 /** Local PID of process which created this lock. */
779 * Number of times blocking AST was sent for this lock.
780 * This is for debugging. Valid values are 0 and 1, if there is an
781 * attempt to send blocking AST more than once, an assertion would be
782 * hit. \see ldlm_work_bl_ast_lock
785 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
786 struct list_head l_bl_ast;
787 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
788 struct list_head l_cp_ast;
789 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
790 struct list_head l_rk_ast;
793 * Pointer to a conflicting lock that caused blocking AST to be sent
796 struct ldlm_lock *l_blocking_lock;
799 * Protected by lr_lock, linkages to "skip lists".
800 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
802 struct list_head l_sl_mode;
803 struct list_head l_sl_policy;
805 /** Reference tracking structure to debug leaked locks. */
806 struct lu_ref l_reference;
807 #if LUSTRE_TRACKS_LOCK_EXP_REFS
808 /* Debugging stuff for bug 20498, for tracking export references. */
809 /** number of export references taken */
811 /** link all locks referencing one export */
812 struct list_head l_exp_refs_link;
813 /** referenced export object */
814 struct obd_export *l_exp_refs_target;
817 * export blocking dlm lock list, protected by
818 * l_export->exp_bl_list_lock.
819 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
820 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
822 struct list_head l_exp_list;
826 * LDLM resource description.
827 * Basically, resource is a representation for a single object.
828 * Object has a name which is currently 4 64-bit integers. LDLM user is
829 * responsible for creation of a mapping between objects it wants to be
830 * protected and resource names.
832 * A resource can only hold locks of a single lock type, though there may be
833 * multiple ldlm_locks on a single resource, depending on the lock type and
834 * whether the locks are conflicting or not.
836 struct ldlm_resource {
837 struct ldlm_ns_bucket *lr_ns_bucket;
840 * List item for list in namespace hash.
841 * protected by ns_lock
843 struct hlist_node lr_hash;
845 /** Spinlock to protect locks under this resource. */
849 * protected by lr_lock
852 /** List of locks in granted state */
853 struct list_head lr_granted;
855 * List of locks that could not be granted due to conflicts and
856 * that are waiting for conflicts to go away
858 struct list_head lr_waiting;
861 /** Type of locks this resource can hold. Only one type per resource. */
862 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
865 struct ldlm_res_id lr_name;
866 /** Reference count for this resource */
867 atomic_t lr_refcount;
870 * Interval trees (only for extent locks) for all modes of this resource
872 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM];
875 * Server-side-only lock value block elements.
876 * To serialize lvbo_init.
878 struct mutex lr_lvb_mutex;
881 /** When the resource was considered as contended. */
882 unsigned long lr_contention_time;
883 /** List of references to this resource. For debugging. */
884 struct lu_ref lr_reference;
886 struct inode *lr_lvb_inode;
889 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
891 return lock->l_resource->lr_type == LDLM_IBITS &&
892 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
896 ldlm_ns_name(struct ldlm_namespace *ns)
898 return ns->ns_rs_hash->hs_name;
901 static inline struct ldlm_namespace *
902 ldlm_res_to_ns(struct ldlm_resource *res)
904 return res->lr_ns_bucket->nsb_namespace;
907 static inline struct ldlm_namespace *
908 ldlm_lock_to_ns(struct ldlm_lock *lock)
910 return ldlm_res_to_ns(lock->l_resource);
914 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
916 return ldlm_ns_name(ldlm_lock_to_ns(lock));
919 static inline struct adaptive_timeout *
920 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
922 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
925 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
927 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
929 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init)
930 return ns->ns_lvbo->lvbo_init(res);
935 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
937 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
939 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_size)
940 return ns->ns_lvbo->lvbo_size(lock);
945 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
947 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
950 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
955 struct ldlm_ast_work {
956 struct ldlm_lock *w_lock;
958 struct ldlm_lock_desc w_desc;
959 struct list_head w_list;
966 * Common ldlm_enqueue parameters
968 struct ldlm_enqueue_info {
969 __u32 ei_type; /** Type of the lock being enqueued. */
970 __u32 ei_mode; /** Mode of the lock being enqueued. */
971 void *ei_cb_bl; /** blocking lock callback */
972 void *ei_cb_cp; /** lock completion callback */
973 void *ei_cb_gl; /** lock glimpse callback */
974 void *ei_cbdata; /** Data to be passed into callbacks. */
977 extern struct obd_ops ldlm_obd_ops;
979 extern char *ldlm_lockname[];
980 char *ldlm_it2str(int it);
983 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
984 * For the cases where we do not have actual lock to print along
985 * with a debugging message that is ldlm-related
987 #define LDLM_DEBUG_NOLOCK(format, a...) \
988 CDEBUG(D_DLMTRACE, "### " format "\n", ##a)
991 * Support function for lock information printing into debug logs.
994 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
995 CFS_CHECK_STACK(msgdata, mask, cdls); \
997 if (((mask) & D_CANTMASK) != 0 || \
998 ((libcfs_debug & (mask)) != 0 && \
999 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1000 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1003 void _ldlm_lock_debug(struct ldlm_lock *lock,
1004 struct libcfs_debug_msg_data *data,
1005 const char *fmt, ...)
1009 * Rate-limited version of lock printing function.
1011 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1012 static struct cfs_debug_limit_state _ldlm_cdls; \
1013 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1014 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt, ##a);\
1017 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1018 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1020 /** Non-rate-limited lock printing function for debugging purposes. */
1021 #define LDLM_DEBUG(lock, fmt, a...) do { \
1022 if (likely(lock)) { \
1023 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1024 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1027 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1031 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1032 int first_enq, enum ldlm_error *err,
1033 struct list_head *work_list);
1036 * Return values for lock iterators.
1037 * Also used during deciding of lock grants and cancellations.
1039 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1040 #define LDLM_ITER_STOP 2 /* stop iterating */
1042 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1043 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1045 /** \defgroup ldlm_iterator Lock iterators
1047 * LDLM provides for a way to iterate through every lock on a resource or
1048 * namespace or every resource in a namespace.
1051 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1052 ldlm_iterator_t iter, void *data);
1053 /** @} ldlm_iterator */
1055 int ldlm_replay_locks(struct obd_import *imp);
1058 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1061 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1063 struct ldlm_callback_suite {
1064 ldlm_completion_callback lcs_completion;
1065 ldlm_blocking_callback lcs_blocking;
1066 ldlm_glimpse_callback lcs_glimpse;
1070 int ldlm_get_ref(void);
1071 void ldlm_put_ref(void);
1072 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1075 void ldlm_lock2handle(const struct ldlm_lock *lock,
1076 struct lustre_handle *lockh);
1077 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1078 void ldlm_cancel_callback(struct ldlm_lock *);
1079 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1080 int ldlm_lock_set_data(struct lustre_handle *, void *);
1083 * Obtain a lock reference by its handle.
1085 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1087 return __ldlm_handle2lock(h, 0);
1090 #define LDLM_LOCK_REF_DEL(lock) \
1091 lu_ref_del(&lock->l_reference, "handle", current)
1093 static inline struct ldlm_lock *
1094 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1096 struct ldlm_lock *lock;
1098 lock = __ldlm_handle2lock(h, flags);
1100 LDLM_LOCK_REF_DEL(lock);
1105 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1106 * data from request \a r
1108 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1109 struct ptlrpc_request *r, int increase)
1111 if (ldlm_res_to_ns(res)->ns_lvbo &&
1112 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1113 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1119 int ldlm_error2errno(enum ldlm_error error);
1121 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1122 void ldlm_dump_export_locks(struct obd_export *exp);
1126 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1127 * __ldlm_handle2lock().
1129 #define LDLM_LOCK_PUT(lock) \
1131 LDLM_LOCK_REF_DEL(lock); \
1132 /*LDLM_DEBUG((lock), "put");*/ \
1133 ldlm_lock_put(lock); \
1137 * Release a lock reference obtained by some other means (see
1140 #define LDLM_LOCK_RELEASE(lock) \
1142 /*LDLM_DEBUG((lock), "put");*/ \
1143 ldlm_lock_put(lock); \
1146 #define LDLM_LOCK_GET(lock) \
1148 ldlm_lock_get(lock); \
1149 /*LDLM_DEBUG((lock), "get");*/ \
1153 #define ldlm_lock_list_put(head, member, count) \
1155 struct ldlm_lock *_lock, *_next; \
1157 list_for_each_entry_safe(_lock, _next, head, member) { \
1160 list_del_init(&_lock->member); \
1161 LDLM_LOCK_RELEASE(_lock); \
1166 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1167 void ldlm_lock_put(struct ldlm_lock *lock);
1168 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1169 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1170 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1171 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1172 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1173 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1174 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1175 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1176 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1177 const struct ldlm_res_id *,
1178 enum ldlm_type type, ldlm_policy_data_t *,
1179 enum ldlm_mode mode, struct lustre_handle *,
1181 enum ldlm_mode ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1183 void ldlm_lock_cancel(struct ldlm_lock *lock);
1184 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1185 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1188 struct ldlm_namespace *
1189 ldlm_namespace_new(struct obd_device *obd, char *name,
1190 ldlm_side_t client, enum ldlm_appetite apt,
1191 enum ldlm_ns_type ns_type);
1192 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1193 void ldlm_namespace_get(struct ldlm_namespace *ns);
1194 void ldlm_namespace_put(struct ldlm_namespace *ns);
1195 int ldlm_debugfs_setup(void);
1196 void ldlm_debugfs_cleanup(void);
1198 /* resource.c - internal */
1199 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1200 struct ldlm_resource *parent,
1201 const struct ldlm_res_id *,
1202 enum ldlm_type type, int create);
1203 int ldlm_resource_putref(struct ldlm_resource *res);
1204 void ldlm_resource_add_lock(struct ldlm_resource *res,
1205 struct list_head *head,
1206 struct ldlm_lock *lock);
1207 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1208 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1209 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1210 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1211 void ldlm_resource_dump(int level, struct ldlm_resource *);
1212 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1213 const struct ldlm_res_id *);
1215 #define LDLM_RESOURCE_ADDREF(res) do { \
1216 lu_ref_add_atomic(&(res)->lr_reference, __func__, current); \
1219 #define LDLM_RESOURCE_DELREF(res) do { \
1220 lu_ref_del(&(res)->lr_reference, __func__, current); \
1223 /* ldlm_request.c */
1224 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1225 * These AST handlers are typically used for server-side local locks and are
1226 * also used by client-side lock handlers to perform minimum level base
1230 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1231 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1232 /** @} ldlm_local_ast */
1234 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1235 * These are typically used by client and server (*_local versions)
1236 * to obtain and release locks.
1239 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1240 struct ldlm_enqueue_info *einfo,
1241 const struct ldlm_res_id *res_id,
1242 ldlm_policy_data_t const *policy, __u64 *flags,
1243 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1244 struct lustre_handle *lockh, int async);
1245 int ldlm_prep_enqueue_req(struct obd_export *exp,
1246 struct ptlrpc_request *req,
1247 struct list_head *cancels,
1249 int ldlm_prep_elc_req(struct obd_export *exp,
1250 struct ptlrpc_request *req,
1251 int version, int opc, int canceloff,
1252 struct list_head *cancels, int count);
1254 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1255 enum ldlm_type type, __u8 with_policy,
1256 enum ldlm_mode mode,
1257 __u64 *flags, void *lvb, __u32 lvb_len,
1258 struct lustre_handle *lockh, int rc);
1259 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1260 int ldlm_cli_cancel(struct lustre_handle *lockh,
1261 enum ldlm_cancel_flags cancel_flags);
1262 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1263 enum ldlm_cancel_flags flags, void *opaque);
1264 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1265 const struct ldlm_res_id *res_id,
1266 ldlm_policy_data_t *policy,
1267 enum ldlm_mode mode,
1268 enum ldlm_cancel_flags flags,
1270 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1271 struct list_head *cancels,
1272 ldlm_policy_data_t *policy,
1273 enum ldlm_mode mode, __u64 lock_flags,
1274 enum ldlm_cancel_flags cancel_flags,
1276 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1277 enum ldlm_cancel_flags flags);
1278 int ldlm_cli_cancel_list(struct list_head *head, int count,
1279 struct ptlrpc_request *req,
1280 enum ldlm_cancel_flags flags);
1281 /** @} ldlm_cli_api */
1284 /* This has to be here because recursive inclusion sucks. */
1285 int intent_disposition(struct ldlm_reply *rep, int flag);
1286 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1288 /* ioctls for trying requests */
1289 #define IOC_LDLM_TYPE 'f'
1290 #define IOC_LDLM_MIN_NR 40
1292 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1293 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1294 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1295 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1296 #define IOC_LDLM_MAX_NR 43
1299 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1300 * than one lock_res is dead-lock safe.
1302 enum lock_res_type {
1307 /** Lock resource. */
1308 static inline void lock_res(struct ldlm_resource *res)
1310 spin_lock(&res->lr_lock);
1313 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1314 static inline void lock_res_nested(struct ldlm_resource *res,
1315 enum lock_res_type mode)
1317 spin_lock_nested(&res->lr_lock, mode);
1320 /** Unlock resource. */
1321 static inline void unlock_res(struct ldlm_resource *res)
1323 spin_unlock(&res->lr_lock);
1326 /** Check if resource is already locked, assert if not. */
1327 static inline void check_res_locked(struct ldlm_resource *res)
1329 assert_spin_locked(&res->lr_lock);
1332 struct ldlm_resource *lock_res_and_lock(struct ldlm_lock *lock);
1333 void unlock_res_and_lock(struct ldlm_lock *lock);
1336 /** \defgroup ldlm_pools Various LDLM pool related functions
1337 * There are not used outside of ldlm.
1340 int ldlm_pools_init(void);
1341 void ldlm_pools_fini(void);
1343 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1344 int idx, ldlm_side_t client);
1345 void ldlm_pool_fini(struct ldlm_pool *pl);
1346 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1347 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);