4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2010, 2015, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 /** \defgroup LDLM Lustre Distributed Lock Manager
35 * Lustre DLM is based on VAX DLM.
36 * Its two main roles are:
37 * - To provide locking assuring consistency of data on all Lustre nodes.
38 * - To allow clients to cache state protected by a lock by holding the
39 * lock until a conflicting lock is requested or it is expired by the LRU.
44 #ifndef _LUSTRE_DLM_H__
45 #define _LUSTRE_DLM_H__
47 #include "lustre_lib.h"
48 #include "lustre_net.h"
49 #include "lustre_import.h"
50 #include "lustre_handles.h"
51 #include "interval_tree.h" /* for interval_node{}, ldlm_extent */
54 #include "lustre_dlm_flags.h"
59 #define OBD_LDLM_DEVICENAME "ldlm"
61 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
62 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
63 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
66 * LDLM non-error return states
70 ELDLM_LOCK_MATCHED = 1,
72 ELDLM_LOCK_CHANGED = 300,
73 ELDLM_LOCK_ABORTED = 301,
74 ELDLM_LOCK_REPLACED = 302,
75 ELDLM_NO_LOCK_DATA = 303,
76 ELDLM_LOCK_WOULDBLOCK = 304,
78 ELDLM_NAMESPACE_EXISTS = 400,
79 ELDLM_BAD_NAMESPACE = 401
83 * LDLM namespace type.
84 * The "client" type is actually an indication that this is a narrow local view
85 * into complete namespace on the server. Such namespaces cannot make any
86 * decisions about lack of conflicts or do any autonomous lock granting without
87 * first speaking to a server.
90 LDLM_NAMESPACE_SERVER = 1 << 0,
91 LDLM_NAMESPACE_CLIENT = 1 << 1
95 * The blocking callback is overloaded to perform two functions. These flags
96 * indicate which operation should be performed.
98 #define LDLM_CB_BLOCKING 1
99 #define LDLM_CB_CANCELING 2
102 * \name Lock Compatibility Matrix.
104 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
105 * Lock types are described in their respective implementation files:
106 * ldlm_{extent,flock,inodebits,plain}.c.
108 * There are six lock modes along with a compatibility matrix to indicate if
109 * two locks are compatible.
111 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
113 * - PW: Protective Write (normal write) mode. When a client requests a write
114 * lock from an OST, a lock with PW mode will be issued.
115 * - PR: Protective Read (normal read) mode. When a client requests a read from
116 * an OST, a lock with PR mode will be issued. Also, if the client opens a
117 * file for execution, it is granted a lock with PR mode.
118 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
119 * requests a write lock during a file open operation.
120 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
121 * an inodebit lock with the CR mode on the intermediate path component.
135 #define LCK_COMPAT_EX LCK_NL
136 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
137 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
138 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
139 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
140 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
141 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
142 #define LCK_COMPAT_COS (LCK_COS)
143 /** @} Lock Compatibility Matrix */
145 extern enum ldlm_mode lck_compat_array[];
147 static inline void lockmode_verify(enum ldlm_mode mode)
149 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
152 static inline int lockmode_compat(enum ldlm_mode exist_mode,
153 enum ldlm_mode new_mode)
155 return (lck_compat_array[exist_mode] & new_mode);
160 * cluster name spaces
164 #define DLM_OST_NAMESPACE 1
165 #define DLM_MDS_NAMESPACE 2
168 - do we just separate this by security domains and use a prefix for
169 multiple namespaces in the same domain?
174 * Locking rules for LDLM:
179 * waiting_locks_spinlock
194 struct ldlm_resource;
195 struct ldlm_namespace;
198 * Operations on LDLM pools.
199 * LDLM pool is a pool of locks in the namespace without any implicitly
201 * Locks in the pool are organized in LRU.
202 * Local memory pressure or server instructions (e.g. mempressure on server)
203 * can trigger freeing of locks from the pool
205 struct ldlm_pool_ops {
206 /** Recalculate pool \a pl usage */
207 int (*po_recalc)(struct ldlm_pool *pl);
208 /** Cancel at least \a nr locks from pool \a pl */
209 int (*po_shrink)(struct ldlm_pool *pl, int nr,
213 /** One second for pools thread check interval. Each pool has own period. */
214 #define LDLM_POOLS_THREAD_PERIOD (1)
216 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
217 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
219 /** Default recalc period for server side pools in sec. */
220 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
222 /** Default recalc period for client side pools in sec. */
223 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
226 * LDLM pool structure to track granted locks.
227 * For purposes of determining when to release locks on e.g. memory pressure.
228 * This feature is commonly referred to as lru_resize.
231 /** Pool debugfs directory. */
232 struct dentry *pl_debugfs_entry;
233 /** Pool name, must be long enough to hold compound proc entry name. */
235 /** Lock for protecting SLV/CLV updates. */
237 /** Number of allowed locks in in pool, both, client and server side. */
239 /** Number of granted locks in */
241 /** Grant rate per T. */
242 atomic_t pl_grant_rate;
243 /** Cancel rate per T. */
244 atomic_t pl_cancel_rate;
245 /** Server lock volume (SLV). Protected by pl_lock. */
246 __u64 pl_server_lock_volume;
247 /** Current biggest client lock volume. Protected by pl_lock. */
248 __u64 pl_client_lock_volume;
249 /** Lock volume factor. SLV on client is calculated as following:
250 * server_slv * lock_volume_factor.
252 atomic_t pl_lock_volume_factor;
253 /** Time when last SLV from server was obtained. */
254 time64_t pl_recalc_time;
255 /** Recalculation period for pool. */
256 time64_t pl_recalc_period;
257 /** Recalculation and shrink operations. */
258 const struct ldlm_pool_ops *pl_ops;
259 /** Number of planned locks for next period. */
261 /** Pool statistics. */
262 struct lprocfs_stats *pl_stats;
265 struct kobject pl_kobj;
266 struct completion pl_kobj_unregister;
269 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
273 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
274 * be associated with an LDLM lock and transferred from client to server and
277 * Currently LVBs are used by:
278 * - OSC-OST code to maintain current object size/times
279 * - layout lock code to return the layout when the layout lock is granted
281 struct ldlm_valblock_ops {
282 int (*lvbo_init)(struct ldlm_resource *res);
283 int (*lvbo_update)(struct ldlm_resource *res,
284 struct ptlrpc_request *r,
286 int (*lvbo_free)(struct ldlm_resource *res);
287 /* Return size of lvb data appropriate RPC size can be reserved */
288 int (*lvbo_size)(struct ldlm_lock *lock);
289 /* Called to fill in lvb data to RPC buffer @buf */
290 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
294 * LDLM pools related, type of lock pool in the namespace.
295 * Greedy means release cached locks aggressively
298 LDLM_NAMESPACE_GREEDY = 1 << 0,
299 LDLM_NAMESPACE_MODEST = 1 << 1
302 struct ldlm_ns_bucket {
303 /** back pointer to namespace */
304 struct ldlm_namespace *nsb_namespace;
306 * Estimated lock callback time. Used by adaptive timeout code to
307 * avoid spurious client evictions due to unresponsiveness when in
308 * fact the network or overall system load is at fault
310 struct adaptive_timeout nsb_at_estimate;
314 /** LDLM namespace lock stats */
321 LDLM_NS_TYPE_UNKNOWN = 0,
339 * Namespace serves to contain locks related to a particular service.
340 * There are two kinds of namespaces:
341 * - Server namespace has knowledge of all locks and is therefore authoritative
342 * to make decisions like what locks could be granted and what conflicts
343 * exist during new lock enqueue.
344 * - Client namespace only has limited knowledge about locks in the namespace,
345 * only seeing locks held by the client.
347 * Every Lustre service has one server namespace present on the server serving
348 * that service. Every client connected to the service has a client namespace
350 * Every lock obtained by client in that namespace is actually represented by
351 * two in-memory locks. One on the server and one on the client. The locks are
352 * linked by a special cookie by which one node can tell to the other which lock
353 * it actually means during communications. Such locks are called remote locks.
354 * The locks held by server only without any reference to a client are called
357 struct ldlm_namespace {
358 /** Backward link to OBD, required for LDLM pool to store new SLV. */
359 struct obd_device *ns_obd;
361 /** Flag indicating if namespace is on client instead of server */
362 ldlm_side_t ns_client;
364 /** Resource hash table for namespace. */
365 struct cfs_hash *ns_rs_hash;
370 /** big refcount (by bucket) */
374 * Namespace connect flags supported by server (may be changed via
375 * sysfs, LRU resize may be disabled/enabled).
377 __u64 ns_connect_flags;
379 /** Client side original connect flags supported by server. */
380 __u64 ns_orig_connect_flags;
382 /* namespace debugfs dir entry */
383 struct dentry *ns_debugfs_entry;
386 * Position in global namespace list linking all namespaces on
389 struct list_head ns_list_chain;
392 * List of unused locks for this namespace. This list is also called
394 * Unused locks are locks with zero reader/writer reference counts.
395 * This list is only used on clients for lock caching purposes.
396 * When we want to release some locks voluntarily or if server wants
397 * us to release some locks due to e.g. memory pressure, we take locks
398 * to release from the head of this list.
399 * Locks are linked via l_lru field in \see struct ldlm_lock.
401 struct list_head ns_unused_list;
402 /** Number of locks in the LRU list above */
406 * Maximum number of locks permitted in the LRU. If 0, means locks
407 * are managed by pools and there is no preset limit, rather it is all
408 * controlled by available memory on this client and on server.
410 unsigned int ns_max_unused;
411 /** Maximum allowed age (last used time) for locks in the LRU */
412 unsigned int ns_max_age;
415 * Used to rate-limit ldlm_namespace_dump calls.
416 * \see ldlm_namespace_dump. Increased by 10 seconds every time
419 unsigned long ns_next_dump;
422 * LVB operations for this namespace.
423 * \see struct ldlm_valblock_ops
425 struct ldlm_valblock_ops *ns_lvbo;
428 * Used by filter code to store pointer to OBD of the service.
429 * Should be dropped in favor of \a ns_obd
434 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
435 * a resource is removed.
437 wait_queue_head_t ns_waitq;
438 /** LDLM pool structure for this namespace */
439 struct ldlm_pool ns_pool;
440 /** Definition of how eagerly unused locks will be released from LRU */
441 enum ldlm_appetite ns_appetite;
443 /** Limit of parallel AST RPC count. */
444 unsigned ns_max_parallel_ast;
447 * Callback to check if a lock is good to be canceled by ELC or
450 ldlm_cancel_cbt ns_cancel;
452 /** LDLM lock stats */
453 struct lprocfs_stats *ns_stats;
456 * Flag to indicate namespace is being freed. Used to determine if
457 * recalculation of LDLM pool statistics should be skipped.
459 unsigned ns_stopping:1;
461 struct kobject ns_kobj; /* sysfs object */
462 struct completion ns_kobj_unregister;
466 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
468 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
470 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
474 * Returns 1 if this namespace supports lru_resize.
476 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
478 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
481 static inline void ns_register_cancel(struct ldlm_namespace *ns,
489 /** Type for blocking callback function of a lock. */
490 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
491 struct ldlm_lock_desc *new, void *data,
493 /** Type for completion callback function of a lock. */
494 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
496 /** Type for glimpse callback function of a lock. */
497 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
499 /** Work list for sending GL ASTs to multiple locks. */
500 struct ldlm_glimpse_work {
501 struct ldlm_lock *gl_lock; /* lock to glimpse */
502 struct list_head gl_list; /* linkage to other gl work structs */
503 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
504 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
505 * glimpse callback request
509 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
510 #define LDLM_GL_WORK_NOFREE 0x1
512 /** Interval node data for each LDLM_EXTENT lock. */
513 struct ldlm_interval {
514 struct interval_node li_node; /* node for tree management */
515 struct list_head li_group; /* the locks which have the same
516 * policy - group of the policy
520 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
523 * Interval tree for extent locks.
524 * The interval tree must be accessed under the resource lock.
525 * Interval trees are used for granted extent locks to speed up conflicts
526 * lookup. See ldlm/interval_tree.c for more details.
528 struct ldlm_interval_tree {
531 enum ldlm_mode lit_mode; /* lock mode */
532 struct interval_node *lit_root; /* actual ldlm_interval */
535 /** Whether to track references to exports by LDLM locks. */
536 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
539 enum ldlm_cancel_flags {
540 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
541 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
542 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
551 __u64 blocking_owner;
552 struct obd_export *blocking_export;
553 /* Protected by the hash lock */
559 struct ldlm_extent l_extent;
560 struct ldlm_flock l_flock;
561 struct ldlm_inodebits l_inodebits;
562 } ldlm_policy_data_t;
564 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
565 const ldlm_wire_policy_data_t *wpolicy,
566 ldlm_policy_data_t *lpolicy);
576 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
578 #define LDLM_GID_ANY ((__u64)-1)
581 * LDLM lock structure
583 * Represents a single LDLM lock and its state in memory. Each lock is
584 * associated with a single ldlm_resource, the object which is being
585 * locked. There may be multiple ldlm_locks on a single resource,
586 * depending on the lock type and whether the locks are conflicting or
592 * When remote side wants to tell us about a lock, they address
593 * it by this opaque handle. The handle does not hold a
594 * reference on the ldlm_lock, so it can be safely passed to
595 * other threads or nodes. When the lock needs to be accessed
596 * from the handle, it is looked up again in the lock table, and
597 * may no longer exist.
599 * Must be first in the structure.
601 struct portals_handle l_handle;
603 * Lock reference count.
604 * This is how many users have pointers to actual structure, so that
605 * we do not accidentally free lock structure that is in use.
609 * Internal spinlock protects l_resource. We should hold this lock
610 * first before taking res_lock.
614 * Pointer to actual resource this lock is in.
615 * ldlm_lock_change_resource() can change this.
617 struct ldlm_resource *l_resource;
619 * List item for client side LRU list.
620 * Protected by ns_lock in struct ldlm_namespace.
622 struct list_head l_lru;
624 * Linkage to resource's lock queues according to current lock state.
625 * (could be granted, waiting or converting)
626 * Protected by lr_lock in struct ldlm_resource.
628 struct list_head l_res_link;
630 * Tree node for ldlm_extent.
632 struct ldlm_interval *l_tree_node;
634 * Per export hash of locks.
635 * Protected by per-bucket exp->exp_lock_hash locks.
637 struct hlist_node l_exp_hash;
639 * Per export hash of flock locks.
640 * Protected by per-bucket exp->exp_flock_hash locks.
642 struct hlist_node l_exp_flock_hash;
645 * Protected by lr_lock.
647 enum ldlm_mode l_req_mode;
649 * Granted mode, also protected by lr_lock.
651 enum ldlm_mode l_granted_mode;
652 /** Lock completion handler pointer. Called when lock is granted. */
653 ldlm_completion_callback l_completion_ast;
655 * Lock blocking AST handler pointer.
656 * It plays two roles:
657 * - as a notification of an attempt to queue a conflicting lock (once)
658 * - as a notification when the lock is being cancelled.
660 * As such it's typically called twice: once for the initial conflict
661 * and then once more when the last user went away and the lock is
662 * cancelled (could happen recursively).
664 ldlm_blocking_callback l_blocking_ast;
666 * Lock glimpse handler.
667 * Glimpse handler is used to obtain LVB updates from a client by
670 ldlm_glimpse_callback l_glimpse_ast;
674 * This is a pointer to actual client export for locks that were granted
675 * to clients. Used server-side.
677 struct obd_export *l_export;
679 * Lock connection export.
680 * Pointer to server export on a client.
682 struct obd_export *l_conn_export;
685 * Remote lock handle.
686 * If the lock is remote, this is the handle of the other side lock
689 struct lustre_handle l_remote_handle;
692 * Representation of private data specific for a lock type.
693 * Examples are: extent range for extent lock or bitmask for ibits locks
695 ldlm_policy_data_t l_policy_data;
698 * Lock state flags. Protected by lr_lock.
699 * \see lustre_dlm_flags.h where the bits are defined.
704 * Lock r/w usage counters.
705 * Protected by lr_lock.
710 * If the lock is granted, a process sleeps on this waitq to learn when
711 * it's no longer in use. If the lock is not granted, a process sleeps
712 * on this waitq to learn when it becomes granted.
714 wait_queue_head_t l_waitq;
717 * Seconds. It will be updated if there is any activity related to
718 * the lock, e.g. enqueue the lock or send blocking AST.
720 time64_t l_last_activity;
723 * Time last used by e.g. being matched by lock match.
724 * Jiffies. Should be converted to time if needed.
726 unsigned long l_last_used;
728 /** Originally requested extent for the extent lock. */
729 struct ldlm_extent l_req_extent;
732 * Client-side-only members.
735 enum lvb_type l_lvb_type;
738 * Temporary storage for a LVB received during an enqueue operation.
743 /** Private storage for lock user. Opaque to LDLM. */
747 * Server-side-only members.
751 * Connection cookie for the client originating the operation.
752 * Used by Commit on Share (COS) code. Currently only used for
753 * inodebits locks on MDS.
755 __u64 l_client_cookie;
758 * List item for locks waiting for cancellation from clients.
759 * The lists this could be linked into are:
760 * waiting_locks_list (protected by waiting_locks_spinlock),
761 * then if the lock timed out, it is moved to
762 * expired_lock_thread.elt_expired_locks for further processing.
763 * Protected by elt_lock.
765 struct list_head l_pending_chain;
768 * Set when lock is sent a blocking AST. Time in seconds when timeout
769 * is reached and client holding this lock could be evicted.
770 * This timeout could be further extended by e.g. certain IO activity
772 * \see ost_rw_prolong_locks
774 unsigned long l_callback_timeout;
776 /** Local PID of process which created this lock. */
780 * Number of times blocking AST was sent for this lock.
781 * This is for debugging. Valid values are 0 and 1, if there is an
782 * attempt to send blocking AST more than once, an assertion would be
783 * hit. \see ldlm_work_bl_ast_lock
786 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
787 struct list_head l_bl_ast;
788 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
789 struct list_head l_cp_ast;
790 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
791 struct list_head l_rk_ast;
794 * Pointer to a conflicting lock that caused blocking AST to be sent
797 struct ldlm_lock *l_blocking_lock;
800 * Protected by lr_lock, linkages to "skip lists".
801 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
803 struct list_head l_sl_mode;
804 struct list_head l_sl_policy;
806 /** Reference tracking structure to debug leaked locks. */
807 struct lu_ref l_reference;
808 #if LUSTRE_TRACKS_LOCK_EXP_REFS
809 /* Debugging stuff for bug 20498, for tracking export references. */
810 /** number of export references taken */
812 /** link all locks referencing one export */
813 struct list_head l_exp_refs_link;
814 /** referenced export object */
815 struct obd_export *l_exp_refs_target;
818 * export blocking dlm lock list, protected by
819 * l_export->exp_bl_list_lock.
820 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
821 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
823 struct list_head l_exp_list;
827 * LDLM resource description.
828 * Basically, resource is a representation for a single object.
829 * Object has a name which is currently 4 64-bit integers. LDLM user is
830 * responsible for creation of a mapping between objects it wants to be
831 * protected and resource names.
833 * A resource can only hold locks of a single lock type, though there may be
834 * multiple ldlm_locks on a single resource, depending on the lock type and
835 * whether the locks are conflicting or not.
837 struct ldlm_resource {
838 struct ldlm_ns_bucket *lr_ns_bucket;
841 * List item for list in namespace hash.
842 * protected by ns_lock
844 struct hlist_node lr_hash;
846 /** Spinlock to protect locks under this resource. */
850 * protected by lr_lock
853 /** List of locks in granted state */
854 struct list_head lr_granted;
856 * List of locks that could not be granted due to conflicts and
857 * that are waiting for conflicts to go away
859 struct list_head lr_waiting;
862 /** Type of locks this resource can hold. Only one type per resource. */
863 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
866 struct ldlm_res_id lr_name;
867 /** Reference count for this resource */
868 atomic_t lr_refcount;
871 * Interval trees (only for extent locks) for all modes of this resource
873 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM];
876 * Server-side-only lock value block elements.
877 * To serialize lvbo_init.
879 struct mutex lr_lvb_mutex;
882 /** When the resource was considered as contended. */
883 unsigned long lr_contention_time;
884 /** List of references to this resource. For debugging. */
885 struct lu_ref lr_reference;
887 struct inode *lr_lvb_inode;
890 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
892 return lock->l_resource->lr_type == LDLM_IBITS &&
893 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
897 ldlm_ns_name(struct ldlm_namespace *ns)
899 return ns->ns_rs_hash->hs_name;
902 static inline struct ldlm_namespace *
903 ldlm_res_to_ns(struct ldlm_resource *res)
905 return res->lr_ns_bucket->nsb_namespace;
908 static inline struct ldlm_namespace *
909 ldlm_lock_to_ns(struct ldlm_lock *lock)
911 return ldlm_res_to_ns(lock->l_resource);
915 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
917 return ldlm_ns_name(ldlm_lock_to_ns(lock));
920 static inline struct adaptive_timeout *
921 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
923 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
926 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
928 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
930 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_init)
931 return ns->ns_lvbo->lvbo_init(res);
936 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
938 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
940 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_size)
941 return ns->ns_lvbo->lvbo_size(lock);
946 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
948 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
951 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
956 struct ldlm_ast_work {
957 struct ldlm_lock *w_lock;
959 struct ldlm_lock_desc w_desc;
960 struct list_head w_list;
967 * Common ldlm_enqueue parameters
969 struct ldlm_enqueue_info {
970 __u32 ei_type; /** Type of the lock being enqueued. */
971 __u32 ei_mode; /** Mode of the lock being enqueued. */
972 void *ei_cb_bl; /** blocking lock callback */
973 void *ei_cb_cp; /** lock completion callback */
974 void *ei_cb_gl; /** lock glimpse callback */
975 void *ei_cbdata; /** Data to be passed into callbacks. */
976 unsigned int ei_enq_slave:1; /* whether enqueue slave stripes */
979 extern struct obd_ops ldlm_obd_ops;
981 extern char *ldlm_lockname[];
982 char *ldlm_it2str(int it);
985 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
986 * For the cases where we do not have actual lock to print along
987 * with a debugging message that is ldlm-related
989 #define LDLM_DEBUG_NOLOCK(format, a...) \
990 CDEBUG(D_DLMTRACE, "### " format "\n", ##a)
993 * Support function for lock information printing into debug logs.
996 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
997 CFS_CHECK_STACK(msgdata, mask, cdls); \
999 if (((mask) & D_CANTMASK) != 0 || \
1000 ((libcfs_debug & (mask)) != 0 && \
1001 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1002 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1005 void _ldlm_lock_debug(struct ldlm_lock *lock,
1006 struct libcfs_debug_msg_data *data,
1007 const char *fmt, ...)
1011 * Rate-limited version of lock printing function.
1013 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1014 static struct cfs_debug_limit_state _ldlm_cdls; \
1015 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1016 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt, ##a);\
1019 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1020 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1022 /** Non-rate-limited lock printing function for debugging purposes. */
1023 #define LDLM_DEBUG(lock, fmt, a...) do { \
1024 if (likely(lock)) { \
1025 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1026 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1029 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1033 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1034 int first_enq, enum ldlm_error *err,
1035 struct list_head *work_list);
1038 * Return values for lock iterators.
1039 * Also used during deciding of lock grants and cancellations.
1041 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1042 #define LDLM_ITER_STOP 2 /* stop iterating */
1044 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1045 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1047 /** \defgroup ldlm_iterator Lock iterators
1049 * LDLM provides for a way to iterate through every lock on a resource or
1050 * namespace or every resource in a namespace.
1053 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1054 ldlm_iterator_t iter, void *data);
1055 /** @} ldlm_iterator */
1057 int ldlm_replay_locks(struct obd_import *imp);
1060 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1063 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1065 struct ldlm_callback_suite {
1066 ldlm_completion_callback lcs_completion;
1067 ldlm_blocking_callback lcs_blocking;
1068 ldlm_glimpse_callback lcs_glimpse;
1072 int ldlm_get_ref(void);
1073 void ldlm_put_ref(void);
1074 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1077 void ldlm_lock2handle(const struct ldlm_lock *lock,
1078 struct lustre_handle *lockh);
1079 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1080 void ldlm_cancel_callback(struct ldlm_lock *);
1081 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1082 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1085 * Obtain a lock reference by its handle.
1087 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1089 return __ldlm_handle2lock(h, 0);
1092 #define LDLM_LOCK_REF_DEL(lock) \
1093 lu_ref_del(&lock->l_reference, "handle", current)
1095 static inline struct ldlm_lock *
1096 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1098 struct ldlm_lock *lock;
1100 lock = __ldlm_handle2lock(h, flags);
1102 LDLM_LOCK_REF_DEL(lock);
1107 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1108 * data from request \a r
1110 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1111 struct ptlrpc_request *r, int increase)
1113 if (ldlm_res_to_ns(res)->ns_lvbo &&
1114 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1115 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1121 int ldlm_error2errno(enum ldlm_error error);
1123 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1124 void ldlm_dump_export_locks(struct obd_export *exp);
1128 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1129 * __ldlm_handle2lock().
1131 #define LDLM_LOCK_PUT(lock) \
1133 LDLM_LOCK_REF_DEL(lock); \
1134 /*LDLM_DEBUG((lock), "put");*/ \
1135 ldlm_lock_put(lock); \
1139 * Release a lock reference obtained by some other means (see
1142 #define LDLM_LOCK_RELEASE(lock) \
1144 /*LDLM_DEBUG((lock), "put");*/ \
1145 ldlm_lock_put(lock); \
1148 #define LDLM_LOCK_GET(lock) \
1150 ldlm_lock_get(lock); \
1151 /*LDLM_DEBUG((lock), "get");*/ \
1155 #define ldlm_lock_list_put(head, member, count) \
1157 struct ldlm_lock *_lock, *_next; \
1159 list_for_each_entry_safe(_lock, _next, head, member) { \
1162 list_del_init(&_lock->member); \
1163 LDLM_LOCK_RELEASE(_lock); \
1168 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1169 void ldlm_lock_put(struct ldlm_lock *lock);
1170 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1171 void ldlm_lock_addref(const struct lustre_handle *lockh, __u32 mode);
1172 int ldlm_lock_addref_try(const struct lustre_handle *lockh, __u32 mode);
1173 void ldlm_lock_decref(const struct lustre_handle *lockh, __u32 mode);
1174 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh, __u32 mode);
1175 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1176 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1177 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1178 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1179 const struct ldlm_res_id *,
1180 enum ldlm_type type, ldlm_policy_data_t *,
1181 enum ldlm_mode mode, struct lustre_handle *,
1183 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1185 void ldlm_lock_cancel(struct ldlm_lock *lock);
1186 void ldlm_lock_dump_handle(int level, const struct lustre_handle *);
1187 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1190 struct ldlm_namespace *
1191 ldlm_namespace_new(struct obd_device *obd, char *name,
1192 ldlm_side_t client, enum ldlm_appetite apt,
1193 enum ldlm_ns_type ns_type);
1194 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1195 void ldlm_namespace_get(struct ldlm_namespace *ns);
1196 void ldlm_namespace_put(struct ldlm_namespace *ns);
1197 int ldlm_debugfs_setup(void);
1198 void ldlm_debugfs_cleanup(void);
1200 /* resource.c - internal */
1201 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1202 struct ldlm_resource *parent,
1203 const struct ldlm_res_id *,
1204 enum ldlm_type type, int create);
1205 int ldlm_resource_putref(struct ldlm_resource *res);
1206 void ldlm_resource_add_lock(struct ldlm_resource *res,
1207 struct list_head *head,
1208 struct ldlm_lock *lock);
1209 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1210 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1211 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1212 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1213 void ldlm_resource_dump(int level, struct ldlm_resource *);
1214 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1215 const struct ldlm_res_id *);
1217 #define LDLM_RESOURCE_ADDREF(res) do { \
1218 lu_ref_add_atomic(&(res)->lr_reference, __func__, current); \
1221 #define LDLM_RESOURCE_DELREF(res) do { \
1222 lu_ref_del(&(res)->lr_reference, __func__, current); \
1225 /* ldlm_request.c */
1226 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1227 * These AST handlers are typically used for server-side local locks and are
1228 * also used by client-side lock handlers to perform minimum level base
1232 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1233 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1234 /** @} ldlm_local_ast */
1236 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1237 * These are typically used by client and server (*_local versions)
1238 * to obtain and release locks.
1241 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1242 struct ldlm_enqueue_info *einfo,
1243 const struct ldlm_res_id *res_id,
1244 ldlm_policy_data_t const *policy, __u64 *flags,
1245 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1246 struct lustre_handle *lockh, int async);
1247 int ldlm_prep_enqueue_req(struct obd_export *exp,
1248 struct ptlrpc_request *req,
1249 struct list_head *cancels,
1251 int ldlm_prep_elc_req(struct obd_export *exp,
1252 struct ptlrpc_request *req,
1253 int version, int opc, int canceloff,
1254 struct list_head *cancels, int count);
1256 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1257 enum ldlm_type type, __u8 with_policy,
1258 enum ldlm_mode mode,
1259 __u64 *flags, void *lvb, __u32 lvb_len,
1260 const struct lustre_handle *lockh, int rc);
1261 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1262 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1263 enum ldlm_cancel_flags cancel_flags);
1264 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1265 enum ldlm_cancel_flags flags, void *opaque);
1266 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1267 const struct ldlm_res_id *res_id,
1268 ldlm_policy_data_t *policy,
1269 enum ldlm_mode mode,
1270 enum ldlm_cancel_flags flags,
1272 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1273 struct list_head *cancels,
1274 ldlm_policy_data_t *policy,
1275 enum ldlm_mode mode, __u64 lock_flags,
1276 enum ldlm_cancel_flags cancel_flags,
1278 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1279 enum ldlm_cancel_flags flags);
1280 int ldlm_cli_cancel_list(struct list_head *head, int count,
1281 struct ptlrpc_request *req,
1282 enum ldlm_cancel_flags flags);
1283 /** @} ldlm_cli_api */
1286 /* This has to be here because recursive inclusion sucks. */
1287 int intent_disposition(struct ldlm_reply *rep, int flag);
1288 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1291 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1292 * than one lock_res is dead-lock safe.
1294 enum lock_res_type {
1299 /** Lock resource. */
1300 static inline void lock_res(struct ldlm_resource *res)
1302 spin_lock(&res->lr_lock);
1305 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1306 static inline void lock_res_nested(struct ldlm_resource *res,
1307 enum lock_res_type mode)
1309 spin_lock_nested(&res->lr_lock, mode);
1312 /** Unlock resource. */
1313 static inline void unlock_res(struct ldlm_resource *res)
1315 spin_unlock(&res->lr_lock);
1318 /** Check if resource is already locked, assert if not. */
1319 static inline void check_res_locked(struct ldlm_resource *res)
1321 assert_spin_locked(&res->lr_lock);
1324 struct ldlm_resource *lock_res_and_lock(struct ldlm_lock *lock);
1325 void unlock_res_and_lock(struct ldlm_lock *lock);
1328 /** \defgroup ldlm_pools Various LDLM pool related functions
1329 * There are not used outside of ldlm.
1332 int ldlm_pools_init(void);
1333 void ldlm_pools_fini(void);
1335 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1336 int idx, ldlm_side_t client);
1337 void ldlm_pool_fini(struct ldlm_pool *pl);
1338 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1339 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);