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(3900)) /* 65 min */
63 #define LDLM_CTIME_AGE_LIMIT (10)
64 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
67 * LDLM non-error return states
71 ELDLM_LOCK_MATCHED = 1,
73 ELDLM_LOCK_CHANGED = 300,
74 ELDLM_LOCK_ABORTED = 301,
75 ELDLM_LOCK_REPLACED = 302,
76 ELDLM_NO_LOCK_DATA = 303,
77 ELDLM_LOCK_WOULDBLOCK = 304,
79 ELDLM_NAMESPACE_EXISTS = 400,
80 ELDLM_BAD_NAMESPACE = 401,
84 * LDLM namespace type.
85 * The "client" type is actually an indication that this is a narrow local view
86 * into complete namespace on the server. Such namespaces cannot make any
87 * decisions about lack of conflicts or do any autonomous lock granting without
88 * first speaking to a server.
91 LDLM_NAMESPACE_SERVER = 0x01,
92 LDLM_NAMESPACE_CLIENT = 0x02
96 * The blocking callback is overloaded to perform two functions. These flags
97 * indicate which operation should be performed.
99 #define LDLM_CB_BLOCKING 1
100 #define LDLM_CB_CANCELING 2
103 * \name Lock Compatibility Matrix.
105 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
106 * Lock types are described in their respective implementation files:
107 * ldlm_{extent,flock,inodebits,plain}.c.
109 * There are six lock modes along with a compatibility matrix to indicate if
110 * two locks are compatible.
112 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
114 * - PW: Protective Write (normal write) mode. When a client requests a write
115 * lock from an OST, a lock with PW mode will be issued.
116 * - PR: Protective Read (normal read) mode. When a client requests a read from
117 * an OST, a lock with PR mode will be issued. Also, if the client opens a
118 * file for execution, it is granted a lock with PR mode.
119 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
120 * requests a write lock during a file open operation.
121 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
122 * an inodebit lock with the CR mode on the intermediate path component.
136 #define LCK_COMPAT_EX LCK_NL
137 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
138 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
139 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
140 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
141 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
142 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
143 #define LCK_COMPAT_COS (LCK_COS)
144 /** @} Lock Compatibility Matrix */
146 extern enum ldlm_mode lck_compat_array[];
148 static inline void lockmode_verify(enum ldlm_mode mode)
150 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
153 static inline int lockmode_compat(enum ldlm_mode exist_mode,
154 enum ldlm_mode new_mode)
156 return lck_compat_array[exist_mode] & new_mode;
161 * cluster name spaces
165 #define DLM_OST_NAMESPACE 1
166 #define DLM_MDS_NAMESPACE 2
169 - do we just separate this by security domains and use a prefix for
170 multiple namespaces in the same domain?
175 * Locking rules for LDLM:
180 * waiting_locks_spinlock
195 struct ldlm_resource;
196 struct ldlm_namespace;
199 * Operations on LDLM pools.
200 * LDLM pool is a pool of locks in the namespace without any implicitly
202 * Locks in the pool are organized in LRU.
203 * Local memory pressure or server instructions (e.g. mempressure on server)
204 * can trigger freeing of locks from the pool
206 struct ldlm_pool_ops {
207 /** Recalculate pool \a pl usage */
208 int (*po_recalc)(struct ldlm_pool *pl);
209 /** Cancel at least \a nr locks from pool \a pl */
210 int (*po_shrink)(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
211 int (*po_setup)(struct ldlm_pool *pl, int limit);
214 /** One second for pools thread check interval. Each pool has own period. */
215 #define LDLM_POOLS_THREAD_PERIOD (1)
217 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
218 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
220 /** Default recalc period for server side pools in sec. */
221 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
223 /** Default recalc period for client side pools in sec. */
224 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
227 * LDLM pool structure to track granted locks.
228 * For purposes of determining when to release locks on e.g. memory pressure.
229 * This feature is commonly referred to as lru_resize.
232 /** Pool proc directory. */
233 struct proc_dir_entry *pl_proc_dir;
234 /** Pool name, must be long enough to hold compound proc entry name. */
236 /** Lock for protecting SLV/CLV updates. */
238 /** Number of allowed locks in in pool, both, client and server side. */
240 /** Number of granted locks in */
242 /** Grant rate per T. */
243 atomic_t pl_grant_rate;
244 /** Cancel rate per T. */
245 atomic_t pl_cancel_rate;
246 /** Server lock volume (SLV). Protected by pl_lock. */
247 __u64 pl_server_lock_volume;
248 /** Current biggest client lock volume. Protected by pl_lock. */
249 __u64 pl_client_lock_volume;
250 /** Lock volume factor. SLV on client is calculated as following:
251 * server_slv * lock_volume_factor. */
252 atomic_t pl_lock_volume_factor;
253 /** Time when last SLV from server was obtained. */
254 time_t pl_recalc_time;
255 /** Recalculation period for pool. */
256 time_t pl_recalc_period;
257 /** Recalculation and shrink operations. */
258 struct ldlm_pool_ops *pl_ops;
259 /** Number of planned locks for next period. */
261 /** Pool statistics. */
262 struct lprocfs_stats *pl_stats;
265 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
266 void *req_cookie, enum ldlm_mode mode,
267 __u64 flags, void *data);
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 * To ensure delayed LVB initialization, it is highly recommended to use the set
282 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
284 struct ldlm_valblock_ops {
285 int (*lvbo_init)(struct ldlm_resource *res);
286 int (*lvbo_update)(struct ldlm_resource *res,
287 struct ptlrpc_request *r,
289 int (*lvbo_free)(struct ldlm_resource *res);
290 /* Return size of lvb data appropriate RPC size can be reserved */
291 int (*lvbo_size)(struct ldlm_lock *lock);
292 /* Called to fill in lvb data to RPC buffer @buf */
293 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
297 * LDLM pools related, type of lock pool in the namespace.
298 * Greedy means release cached locks aggressively
301 LDLM_NAMESPACE_GREEDY = 1 << 0,
302 LDLM_NAMESPACE_MODEST = 1 << 1
306 * Default values for the "max_nolock_size", "contention_time" and
307 * "contended_locks" namespace tunables.
309 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
310 #define NS_DEFAULT_CONTENTION_SECONDS 2
311 #define NS_DEFAULT_CONTENDED_LOCKS 32
313 struct ldlm_ns_bucket {
314 /** back pointer to namespace */
315 struct ldlm_namespace *nsb_namespace;
317 * Estimated lock callback time. Used by adaptive timeout code to
318 * avoid spurious client evictions due to unresponsiveness when in
319 * fact the network or overall system load is at fault
321 struct adaptive_timeout nsb_at_estimate;
323 * Which res in the bucket should we start with the reclaim.
325 int nsb_reclaim_start;
329 /** LDLM namespace lock stats */
335 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
336 LDLM_NS_TYPE_MDC, /**< MDC namespace */
337 LDLM_NS_TYPE_MDT, /**< MDT namespace */
338 LDLM_NS_TYPE_OSC, /**< OSC namespace */
339 LDLM_NS_TYPE_OST, /**< OST namespace */
340 LDLM_NS_TYPE_MGC, /**< MGC namespace */
341 LDLM_NS_TYPE_MGT, /**< MGT namespace */
347 * Namespace serves to contain locks related to a particular service.
348 * There are two kinds of namespaces:
349 * - Server namespace has knowledge of all locks and is therefore authoritative
350 * to make decisions like what locks could be granted and what conflicts
351 * exist during new lock enqueue.
352 * - Client namespace only has limited knowledge about locks in the namespace,
353 * only seeing locks held by the client.
355 * Every Lustre service has one server namespace present on the server serving
356 * that service. Every client connected to the service has a client namespace
358 * Every lock obtained by client in that namespace is actually represented by
359 * two in-memory locks. One on the server and one on the client. The locks are
360 * linked by a special cookie by which one node can tell to the other which lock
361 * it actually means during communications. Such locks are called remote locks.
362 * The locks held by server only without any reference to a client are called
365 struct ldlm_namespace {
366 /** Backward link to OBD, required for LDLM pool to store new SLV. */
367 struct obd_device *ns_obd;
369 /** Flag indicating if namespace is on client instead of server */
370 enum ldlm_side ns_client;
372 /** Resource hash table for namespace. */
373 struct cfs_hash *ns_rs_hash;
378 /** big refcount (by bucket) */
382 * Namespace connect flags supported by server (may be changed via
383 * /proc, LRU resize may be disabled/enabled).
385 __u64 ns_connect_flags;
387 /** Client side original connect flags supported by server. */
388 __u64 ns_orig_connect_flags;
390 /* namespace proc dir entry */
391 struct proc_dir_entry *ns_proc_dir_entry;
394 * Position in global namespace list linking all namespaces on
397 struct list_head ns_list_chain;
400 * List of unused locks for this namespace. This list is also called
402 * Unused locks are locks with zero reader/writer reference counts.
403 * This list is only used on clients for lock caching purposes.
404 * When we want to release some locks voluntarily or if server wants
405 * us to release some locks due to e.g. memory pressure, we take locks
406 * to release from the head of this list.
407 * Locks are linked via l_lru field in \see struct ldlm_lock.
409 struct list_head ns_unused_list;
410 /** Number of locks in the LRU list above */
414 * Maximum number of locks permitted in the LRU. If 0, means locks
415 * are managed by pools and there is no preset limit, rather it is all
416 * controlled by available memory on this client and on server.
418 unsigned int ns_max_unused;
419 /** Maximum allowed age (last used time) for locks in the LRU */
420 unsigned int ns_max_age;
422 * Server only: number of times we evicted clients due to lack of reply
425 unsigned int ns_timeouts;
427 * Number of seconds since the file change time after which the
428 * MDT will return an UPDATE lock along with a LOOKUP lock.
429 * This allows the client to start caching negative dentries
430 * for a directory and may save an RPC for a later stat.
432 unsigned int ns_ctime_age_limit;
435 * Used to rate-limit ldlm_namespace_dump calls.
436 * \see ldlm_namespace_dump. Increased by 10 seconds every time
439 cfs_time_t ns_next_dump;
441 /** "policy" function that does actual lock conflict determination */
442 ldlm_res_policy ns_policy;
445 * LVB operations for this namespace.
446 * \see struct ldlm_valblock_ops
448 struct ldlm_valblock_ops *ns_lvbo;
451 * Used by filter code to store pointer to OBD of the service.
452 * Should be dropped in favor of \a ns_obd
457 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
458 * a resource is removed.
460 wait_queue_head_t ns_waitq;
461 /** LDLM pool structure for this namespace */
462 struct ldlm_pool ns_pool;
463 /** Definition of how eagerly unused locks will be released from LRU */
464 enum ldlm_appetite ns_appetite;
467 * If more than \a ns_contended_locks are found, the resource is
468 * considered to be contended. Lock enqueues might specify that no
469 * contended locks should be granted
471 unsigned ns_contended_locks;
474 * The resources in this namespace remember contended state during
475 * \a ns_contention_time, in seconds.
477 unsigned ns_contention_time;
480 * Limit size of contended extent locks, in bytes.
481 * If extended lock is requested for more then this many bytes and
482 * caller instructs us not to grant contended locks, we would disregard
485 unsigned ns_max_nolock_size;
487 /** Limit of parallel AST RPC count. */
488 unsigned ns_max_parallel_ast;
491 * Callback to check if a lock is good to be canceled by ELC or
494 ldlm_cancel_cbt ns_cancel;
496 /** LDLM lock stats */
497 struct lprocfs_stats *ns_stats;
500 * Flag to indicate namespace is being freed. Used to determine if
501 * recalculation of LDLM pool statistics should be skipped.
503 unsigned ns_stopping:1;
506 * Which bucket should we start with the lock reclaim.
508 int ns_reclaim_start;
512 * Returns 1 if namespace \a ns is a client namespace.
514 static inline int ns_is_client(struct ldlm_namespace *ns)
517 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
518 LDLM_NAMESPACE_SERVER)));
519 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
520 ns->ns_client == LDLM_NAMESPACE_SERVER);
521 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
525 * Returns 1 if namespace \a ns is a server namespace.
527 static inline int ns_is_server(struct ldlm_namespace *ns)
530 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
531 LDLM_NAMESPACE_SERVER)));
532 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
533 ns->ns_client == LDLM_NAMESPACE_SERVER);
534 return ns->ns_client == LDLM_NAMESPACE_SERVER;
538 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
540 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
543 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
547 * Returns 1 if this namespace supports lru_resize.
549 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
552 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
555 static inline void ns_register_cancel(struct ldlm_namespace *ns,
564 /** Type for blocking callback function of a lock. */
565 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
566 struct ldlm_lock_desc *new, void *data,
568 /** Type for completion callback function of a lock. */
569 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
571 /** Type for glimpse callback function of a lock. */
572 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
574 /** Work list for sending GL ASTs to multiple locks. */
575 struct ldlm_glimpse_work {
576 struct ldlm_lock *gl_lock; /* lock to glimpse */
577 struct list_head gl_list; /* linkage to other gl work structs */
578 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
579 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
580 * glimpse callback request */
583 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
584 #define LDLM_GL_WORK_NOFREE 0x1
586 /** Interval node data for each LDLM_EXTENT lock. */
587 struct ldlm_interval {
588 struct interval_node li_node; /* node for tree management */
589 struct list_head li_group; /* the locks which have the same
590 * policy - group of the policy */
592 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
595 * Interval tree for extent locks.
596 * The interval tree must be accessed under the resource lock.
597 * Interval trees are used for granted extent locks to speed up conflicts
598 * lookup. See ldlm/interval_tree.c for more details.
600 struct ldlm_interval_tree {
603 enum ldlm_mode lit_mode; /* lock mode */
604 struct interval_node *lit_root; /* actual ldlm_interval */
607 /** Whether to track references to exports by LDLM locks. */
608 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
611 enum ldlm_cancel_flags {
612 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
613 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
614 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
621 __u64 blocking_owner;
622 struct obd_export *blocking_export;
623 atomic_t blocking_refs;
627 union ldlm_policy_data {
628 struct ldlm_extent l_extent;
629 struct ldlm_flock l_flock;
630 struct ldlm_inodebits l_inodebits;
633 void ldlm_convert_policy_to_wire(enum ldlm_type type,
634 const union ldlm_policy_data *lpolicy,
635 union ldlm_wire_policy_data *wpolicy);
636 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
637 const union ldlm_wire_policy_data *wpolicy,
638 union ldlm_policy_data *lpolicy);
648 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
650 #define LDLM_GID_ANY ((__u64)-1)
653 * LDLM lock structure
655 * Represents a single LDLM lock and its state in memory. Each lock is
656 * associated with a single ldlm_resource, the object which is being
657 * locked. There may be multiple ldlm_locks on a single resource,
658 * depending on the lock type and whether the locks are conflicting or
664 * When remote side wants to tell us about a lock, they address
665 * it by this opaque handle. The handle does not hold a
666 * reference on the ldlm_lock, so it can be safely passed to
667 * other threads or nodes. When the lock needs to be accessed
668 * from the handle, it is looked up again in the lock table, and
669 * may no longer exist.
671 * Must be first in the structure.
673 struct portals_handle l_handle;
675 * Lock reference count.
676 * This is how many users have pointers to actual structure, so that
677 * we do not accidentally free lock structure that is in use.
681 * Internal spinlock protects l_resource. We should hold this lock
682 * first before taking res_lock.
686 * Pointer to actual resource this lock is in.
687 * ldlm_lock_change_resource() can change this.
689 struct ldlm_resource *l_resource;
691 * List item for client side LRU list.
692 * Protected by ns_lock in struct ldlm_namespace.
694 struct list_head l_lru;
696 * Linkage to resource's lock queues according to current lock state.
697 * (could be granted, waiting or converting)
698 * Protected by lr_lock in struct ldlm_resource.
700 struct list_head l_res_link;
702 * Tree node for ldlm_extent.
704 struct ldlm_interval *l_tree_node;
706 * Per export hash of locks.
707 * Protected by per-bucket exp->exp_lock_hash locks.
709 struct hlist_node l_exp_hash;
711 * Per export hash of flock locks.
712 * Protected by per-bucket exp->exp_flock_hash locks.
714 struct hlist_node l_exp_flock_hash;
717 * Protected by lr_lock.
719 enum ldlm_mode l_req_mode;
721 * Granted mode, also protected by lr_lock.
723 enum ldlm_mode l_granted_mode;
724 /** Lock completion handler pointer. Called when lock is granted. */
725 ldlm_completion_callback l_completion_ast;
727 * Lock blocking AST handler pointer.
728 * It plays two roles:
729 * - as a notification of an attempt to queue a conflicting lock (once)
730 * - as a notification when the lock is being cancelled.
732 * As such it's typically called twice: once for the initial conflict
733 * and then once more when the last user went away and the lock is
734 * cancelled (could happen recursively).
736 ldlm_blocking_callback l_blocking_ast;
738 * Lock glimpse handler.
739 * Glimpse handler is used to obtain LVB updates from a client by
742 ldlm_glimpse_callback l_glimpse_ast;
746 * This is a pointer to actual client export for locks that were granted
747 * to clients. Used server-side.
749 struct obd_export *l_export;
751 * Lock connection export.
752 * Pointer to server export on a client.
754 struct obd_export *l_conn_export;
757 * Remote lock handle.
758 * If the lock is remote, this is the handle of the other side lock
761 struct lustre_handle l_remote_handle;
764 * Representation of private data specific for a lock type.
765 * Examples are: extent range for extent lock or bitmask for ibits locks
767 union ldlm_policy_data l_policy_data;
770 * Lock state flags. Protected by lr_lock.
771 * \see lustre_dlm_flags.h where the bits are defined.
776 * Lock r/w usage counters.
777 * Protected by lr_lock.
782 * If the lock is granted, a process sleeps on this waitq to learn when
783 * it's no longer in use. If the lock is not granted, a process sleeps
784 * on this waitq to learn when it becomes granted.
786 wait_queue_head_t l_waitq;
789 * Seconds. It will be updated if there is any activity related to
790 * the lock, e.g. enqueue the lock or send blocking AST.
792 cfs_time_t l_last_activity;
795 * Time last used by e.g. being matched by lock match.
796 * Jiffies. Should be converted to time if needed.
798 cfs_time_t l_last_used;
800 /** Originally requested extent for the extent lock. */
801 struct ldlm_extent l_req_extent;
804 * Client-side-only members.
807 enum lvb_type l_lvb_type;
810 * Temporary storage for a LVB received during an enqueue operation.
811 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
816 /** Private storage for lock user. Opaque to LDLM. */
820 * Server-side-only members.
824 * Connection cookie for the client originating the operation.
825 * Used by Commit on Share (COS) code. Currently only used for
826 * inodebits locks on MDS.
828 __u64 l_client_cookie;
831 * List item for locks waiting for cancellation from clients.
832 * The lists this could be linked into are:
833 * waiting_locks_list (protected by waiting_locks_spinlock),
834 * then if the lock timed out, it is moved to
835 * expired_lock_thread.elt_expired_locks for further processing.
836 * Protected by elt_lock.
838 struct list_head l_pending_chain;
841 * Set when lock is sent a blocking AST. Time in seconds when timeout
842 * is reached and client holding this lock could be evicted.
843 * This timeout could be further extended by e.g. certain IO activity
845 * \see ost_rw_prolong_locks
847 cfs_time_t l_callback_timeout;
849 /** Local PID of process which created this lock. */
853 * Number of times blocking AST was sent for this lock.
854 * This is for debugging. Valid values are 0 and 1, if there is an
855 * attempt to send blocking AST more than once, an assertion would be
856 * hit. \see ldlm_work_bl_ast_lock
859 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
860 struct list_head l_bl_ast;
861 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
862 struct list_head l_cp_ast;
863 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
864 struct list_head l_rk_ast;
867 * Pointer to a conflicting lock that caused blocking AST to be sent
870 struct ldlm_lock *l_blocking_lock;
873 * Protected by lr_lock, linkages to "skip lists".
874 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
876 struct list_head l_sl_mode;
877 struct list_head l_sl_policy;
879 /** Reference tracking structure to debug leaked locks. */
880 struct lu_ref l_reference;
881 #if LUSTRE_TRACKS_LOCK_EXP_REFS
882 /* Debugging stuff for bug 20498, for tracking export references. */
883 /** number of export references taken */
885 /** link all locks referencing one export */
886 struct list_head l_exp_refs_link;
887 /** referenced export object */
888 struct obd_export *l_exp_refs_target;
891 * export blocking dlm lock list, protected by
892 * l_export->exp_bl_list_lock.
893 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
894 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
896 struct list_head l_exp_list;
899 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
900 #define l_transno l_client_cookie
902 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
903 * which is for server. */
904 #define l_slc_link l_rk_ast
907 * LDLM resource description.
908 * Basically, resource is a representation for a single object.
909 * Object has a name which is currently 4 64-bit integers. LDLM user is
910 * responsible for creation of a mapping between objects it wants to be
911 * protected and resource names.
913 * A resource can only hold locks of a single lock type, though there may be
914 * multiple ldlm_locks on a single resource, depending on the lock type and
915 * whether the locks are conflicting or not.
917 struct ldlm_resource {
918 struct ldlm_ns_bucket *lr_ns_bucket;
921 * List item for list in namespace hash.
922 * protected by ns_lock
924 struct hlist_node lr_hash;
926 /** Reference count for this resource */
927 atomic_t lr_refcount;
929 /** Spinlock to protect locks under this resource. */
933 * protected by lr_lock
935 /** List of locks in granted state */
936 struct list_head lr_granted;
937 /** List of locks waiting to change their granted mode (converted) */
938 struct list_head lr_converting;
940 * List of locks that could not be granted due to conflicts and
941 * that are waiting for conflicts to go away */
942 struct list_head lr_waiting;
946 struct ldlm_res_id lr_name;
949 * Interval trees (only for extent locks) for all modes of this resource
951 struct ldlm_interval_tree *lr_itree;
955 * When the resource was considered as contended,
956 * used only on server side. */
957 cfs_time_t lr_contention_time;
959 * Associated inode, used only on client side.
961 struct inode *lr_lvb_inode;
964 /** Type of locks this resource can hold. Only one type per resource. */
965 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
968 * Server-side-only lock value block elements.
969 * To serialize lvbo_init.
972 struct mutex lr_lvb_mutex;
973 /** protected by lr_lock */
975 /** is lvb initialized ? */
976 bool lr_lvb_initialized;
978 /** List of references to this resource. For debugging. */
979 struct lu_ref lr_reference;
982 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
984 return lock->l_resource->lr_type == LDLM_IBITS &&
985 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
989 ldlm_ns_name(struct ldlm_namespace *ns)
991 return ns->ns_rs_hash->hs_name;
994 static inline struct ldlm_namespace *
995 ldlm_res_to_ns(struct ldlm_resource *res)
997 return res->lr_ns_bucket->nsb_namespace;
1000 static inline struct ldlm_namespace *
1001 ldlm_lock_to_ns(struct ldlm_lock *lock)
1003 return ldlm_res_to_ns(lock->l_resource);
1006 static inline char *
1007 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1009 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1012 static inline struct adaptive_timeout *
1013 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1015 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1018 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1020 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1023 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1024 res->lr_lvb_initialized)
1027 mutex_lock(&res->lr_lvb_mutex);
1028 /* Did we lose the race? */
1029 if (res->lr_lvb_initialized) {
1030 mutex_unlock(&res->lr_lvb_mutex);
1033 rc = ns->ns_lvbo->lvbo_init(res);
1035 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1037 if (res->lr_lvb_data != NULL) {
1038 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1039 res->lr_lvb_data = NULL;
1041 res->lr_lvb_len = rc;
1043 res->lr_lvb_initialized = true;
1045 mutex_unlock(&res->lr_lvb_mutex);
1049 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1051 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1053 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1054 return ns->ns_lvbo->lvbo_size(lock);
1059 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1061 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1064 if (ns->ns_lvbo != NULL) {
1065 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1066 /* init lvb now if not already */
1067 rc = ldlm_lvbo_init(lock->l_resource);
1069 CERROR("lock %p: delayed lvb init failed (rc %d)",
1073 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1078 struct ldlm_ast_work {
1079 struct ldlm_lock *w_lock;
1081 struct ldlm_lock_desc w_desc;
1082 struct list_head w_list;
1089 * Common ldlm_enqueue parameters
1091 struct ldlm_enqueue_info {
1092 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1093 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1094 void *ei_cb_bl; /** blocking lock callback */
1095 void *ei_cb_local_bl; /** blocking local lock callback */
1096 void *ei_cb_cp; /** lock completion callback */
1097 void *ei_cb_gl; /** lock glimpse callback */
1098 void *ei_cbdata; /** Data to be passed into callbacks. */
1099 void *ei_namespace; /** lock namespace **/
1100 unsigned int ei_enq_slave:1, /** whether enqueue slave stripes */
1101 ei_nonblock:1; /** non block enqueue */
1104 #define ei_res_id ei_cb_gl
1106 extern struct obd_ops ldlm_obd_ops;
1108 extern char *ldlm_lockname[];
1109 extern char *ldlm_typename[];
1110 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1113 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1114 * For the cases where we do not have actual lock to print along
1115 * with a debugging message that is ldlm-related
1117 #define LDLM_DEBUG_NOLOCK(format, a...) \
1118 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1121 * Support function for lock information printing into debug logs.
1125 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1126 CFS_CHECK_STACK(msgdata, mask, cdls); \
1128 if (((mask) & D_CANTMASK) != 0 || \
1129 ((libcfs_debug & (mask)) != 0 && \
1130 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1131 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1134 void _ldlm_lock_debug(struct ldlm_lock *lock,
1135 struct libcfs_debug_msg_data *data,
1136 const char *fmt, ...)
1137 __attribute__ ((format (printf, 3, 4)));
1140 * Rate-limited version of lock printing function.
1142 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1143 static struct cfs_debug_limit_state _ldlm_cdls; \
1144 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1145 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1148 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1149 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1151 /** Non-rate-limited lock printing function for debugging purposes. */
1152 #define LDLM_DEBUG(lock, fmt, a...) do { \
1153 if (likely(lock != NULL)) { \
1154 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1155 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1156 "### " fmt , ##a); \
1158 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1161 #else /* !LIBCFS_DEBUG */
1162 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1163 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1164 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1167 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1168 int first_enq, enum ldlm_error *err,
1169 struct list_head *work_list);
1172 * Return values for lock iterators.
1173 * Also used during deciding of lock grants and cancellations.
1175 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1176 #define LDLM_ITER_STOP 2 /* stop iterating */
1178 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1179 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1181 /** \defgroup ldlm_iterator Lock iterators
1183 * LDLM provides for a way to iterate through every lock on a resource or
1184 * namespace or every resource in a namespace.
1186 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1188 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1190 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1191 ldlm_iterator_t iter, void *data);
1192 /** @} ldlm_iterator */
1194 int ldlm_replay_locks(struct obd_import *imp);
1197 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1200 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1202 struct ldlm_prolong_args {
1203 struct obd_export *lpa_export;
1204 struct ldlm_res_id lpa_resid;
1205 struct ldlm_extent lpa_extent;
1206 enum ldlm_mode lpa_mode;
1211 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1212 struct ldlm_prolong_args *arg);
1213 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1215 struct ldlm_callback_suite {
1216 ldlm_completion_callback lcs_completion;
1217 ldlm_blocking_callback lcs_blocking;
1218 ldlm_glimpse_callback lcs_glimpse;
1222 #ifdef HAVE_SERVER_SUPPORT
1223 /** \defgroup ldlm_srv_ast Server AST handlers
1224 * These are AST handlers used by server code.
1225 * Their property is that they are just preparing RPCs to be sent to clients.
1228 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1229 void *data, int flag);
1230 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1231 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1232 int ldlm_glimpse_locks(struct ldlm_resource *res,
1233 struct list_head *gl_work_list);
1234 /** @} ldlm_srv_ast */
1236 /** \defgroup ldlm_handlers Server LDLM handlers
1237 * These are handler functions that should be called by "frontends" such as
1238 * MDT or OST to pass through LDLM requests to LDLM for handling
1241 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
1242 ldlm_blocking_callback, ldlm_glimpse_callback);
1243 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1244 const struct ldlm_request *dlm_req,
1245 const struct ldlm_callback_suite *cbs);
1246 int ldlm_handle_convert(struct ptlrpc_request *req);
1247 int ldlm_handle_convert0(struct ptlrpc_request *req,
1248 const struct ldlm_request *dlm_req);
1249 int ldlm_handle_cancel(struct ptlrpc_request *req);
1250 int ldlm_request_cancel(struct ptlrpc_request *req,
1251 const struct ldlm_request *dlm_req,
1252 int first, enum lustre_at_flags flags);
1253 /** @} ldlm_handlers */
1255 void ldlm_revoke_export_locks(struct obd_export *exp);
1256 unsigned int ldlm_bl_timeout(struct ldlm_lock *lock);
1258 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1259 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
1260 int ldlm_get_ref(void);
1261 void ldlm_put_ref(void);
1262 int ldlm_init_export(struct obd_export *exp);
1263 void ldlm_destroy_export(struct obd_export *exp);
1264 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1267 #ifdef HAVE_SERVER_SUPPORT
1268 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1270 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1271 void ldlm_lock2handle(const struct ldlm_lock *lock,
1272 struct lustre_handle *lockh);
1273 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1274 void ldlm_cancel_callback(struct ldlm_lock *);
1275 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1276 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1279 * Obtain a lock reference by its handle.
1281 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1283 return __ldlm_handle2lock(h, 0);
1286 #define LDLM_LOCK_REF_DEL(lock) \
1287 lu_ref_del(&lock->l_reference, "handle", current)
1289 static inline struct ldlm_lock *
1290 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1292 struct ldlm_lock *lock;
1294 lock = __ldlm_handle2lock(h, flags);
1296 LDLM_LOCK_REF_DEL(lock);
1301 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1302 * data from request \a r
1304 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1305 struct ptlrpc_request *req, int increase)
1309 /* delayed lvb init may be required */
1310 rc = ldlm_lvbo_init(res);
1312 CERROR("delayed lvb init failed (rc %d)\n", rc);
1316 if (ldlm_res_to_ns(res)->ns_lvbo &&
1317 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1318 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, req,
1324 int ldlm_error2errno(enum ldlm_error error);
1325 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1326 * confuses user-space. */
1327 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1328 void ldlm_dump_export_locks(struct obd_export *exp);
1332 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1333 * __ldlm_handle2lock().
1335 #define LDLM_LOCK_PUT(lock) \
1337 LDLM_LOCK_REF_DEL(lock); \
1338 /*LDLM_DEBUG((lock), "put");*/ \
1339 ldlm_lock_put(lock); \
1343 * Release a lock reference obtained by some other means (see
1346 #define LDLM_LOCK_RELEASE(lock) \
1348 /*LDLM_DEBUG((lock), "put");*/ \
1349 ldlm_lock_put(lock); \
1352 #define LDLM_LOCK_GET(lock) \
1354 ldlm_lock_get(lock); \
1355 /*LDLM_DEBUG((lock), "get");*/ \
1359 #define ldlm_lock_list_put(head, member, count) \
1361 struct ldlm_lock *_lock, *_next; \
1363 list_for_each_entry_safe(_lock, _next, head, member) { \
1366 list_del_init(&_lock->member); \
1367 LDLM_LOCK_RELEASE(_lock); \
1372 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1373 void ldlm_lock_put(struct ldlm_lock *lock);
1374 void ldlm_lock_destroy(struct ldlm_lock *lock);
1375 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1376 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1377 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1378 enum ldlm_mode mode);
1379 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1380 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1381 enum ldlm_mode mode);
1382 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1383 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1384 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1385 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1386 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1387 const struct ldlm_res_id *, enum ldlm_type type,
1388 union ldlm_policy_data *, enum ldlm_mode mode,
1389 struct lustre_handle *, int unref);
1390 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1392 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock,
1393 enum ldlm_mode new_mode, __u32 *flags);
1394 void ldlm_lock_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1395 void ldlm_lock_cancel(struct ldlm_lock *lock);
1396 void ldlm_reprocess_all(struct ldlm_resource *res);
1397 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1398 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1399 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1402 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1403 enum ldlm_side client,
1404 enum ldlm_appetite apt,
1405 enum ldlm_ns_type ns_type);
1406 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1407 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1408 struct obd_import *imp,
1410 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1411 void ldlm_namespace_free(struct ldlm_namespace *ns,
1412 struct obd_import *imp, int force);
1413 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1414 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1415 enum ldlm_side client);
1416 void ldlm_namespace_get(struct ldlm_namespace *ns);
1417 void ldlm_namespace_put(struct ldlm_namespace *ns);
1418 int ldlm_proc_setup(void);
1419 #ifdef CONFIG_PROC_FS
1420 void ldlm_proc_cleanup(void);
1422 static inline void ldlm_proc_cleanup(void) {}
1425 /* resource.c - internal */
1426 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1427 struct ldlm_resource *parent,
1428 const struct ldlm_res_id *,
1429 enum ldlm_type type, int create);
1430 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1431 int ldlm_resource_putref(struct ldlm_resource *res);
1432 void ldlm_resource_add_lock(struct ldlm_resource *res,
1433 struct list_head *head,
1434 struct ldlm_lock *lock);
1435 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1436 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1437 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1438 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1439 void ldlm_resource_dump(int level, struct ldlm_resource *);
1440 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1441 const struct ldlm_res_id *);
1443 #define LDLM_RESOURCE_ADDREF(res) do { \
1444 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1447 #define LDLM_RESOURCE_DELREF(res) do { \
1448 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1451 /* ldlm_request.c */
1452 int ldlm_expired_completion_wait(void *data);
1453 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1454 * These AST handlers are typically used for server-side local locks and are
1455 * also used by client-side lock handlers to perform minimum level base
1458 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1459 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1460 void *data, int flag);
1461 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1462 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1463 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1464 /** @} ldlm_local_ast */
1466 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1467 * These are typically used by client and server (*_local versions)
1468 * to obtain and release locks.
1470 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1471 struct ldlm_enqueue_info *einfo,
1472 const struct ldlm_res_id *res_id,
1473 union ldlm_policy_data const *policy, __u64 *flags,
1474 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1475 struct lustre_handle *lockh, int async);
1476 int ldlm_prep_enqueue_req(struct obd_export *exp,
1477 struct ptlrpc_request *req,
1478 struct list_head *cancels,
1480 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1481 int version, int opc, int canceloff,
1482 struct list_head *cancels, int count);
1484 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1485 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1486 const struct ldlm_request *dlm_req,
1487 const struct ldlm_callback_suite *cbs);
1488 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1489 enum ldlm_type type, __u8 with_policy,
1490 enum ldlm_mode mode, __u64 *flags, void *lvb,
1492 const struct lustre_handle *lockh, int rc);
1493 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1494 const struct ldlm_res_id *res_id,
1495 enum ldlm_type type, union ldlm_policy_data *policy,
1496 enum ldlm_mode mode, __u64 *flags,
1497 ldlm_blocking_callback blocking,
1498 ldlm_completion_callback completion,
1499 ldlm_glimpse_callback glimpse,
1500 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1501 const __u64 *client_cookie,
1502 struct lustre_handle *lockh);
1503 int ldlm_cli_convert(const struct lustre_handle *lockh, int new_mode,
1505 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1506 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1507 enum ldlm_cancel_flags cancel_flags);
1508 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1509 enum ldlm_cancel_flags flags, void *opaque);
1510 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1511 const struct ldlm_res_id *res_id,
1512 union ldlm_policy_data *policy,
1513 enum ldlm_mode mode,
1514 enum ldlm_cancel_flags flags, void *opaque);
1515 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1516 int count, enum ldlm_cancel_flags flags);
1517 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1518 struct list_head *cancels,
1519 union ldlm_policy_data *policy,
1520 enum ldlm_mode mode, __u64 lock_flags,
1521 enum ldlm_cancel_flags cancel_flags,
1523 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1524 enum ldlm_cancel_flags flags);
1525 int ldlm_cli_cancel_list(struct list_head *head, int count,
1526 struct ptlrpc_request *req,
1527 enum ldlm_cancel_flags flags);
1528 /** @} ldlm_cli_api */
1531 /* This has to be here because recursive inclusion sucks. */
1532 int intent_disposition(struct ldlm_reply *rep, int flag);
1533 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1536 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1537 * than one lock_res is dead-lock safe.
1539 enum lock_res_type {
1544 /** Lock resource. */
1545 static inline void lock_res(struct ldlm_resource *res)
1547 spin_lock(&res->lr_lock);
1550 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1551 static inline void lock_res_nested(struct ldlm_resource *res,
1552 enum lock_res_type mode)
1554 spin_lock_nested(&res->lr_lock, mode);
1557 /** Unlock resource. */
1558 static inline void unlock_res(struct ldlm_resource *res)
1560 spin_unlock(&res->lr_lock);
1563 /** Check if resource is already locked, assert if not. */
1564 static inline void check_res_locked(struct ldlm_resource *res)
1566 assert_spin_locked(&res->lr_lock);
1569 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1570 void unlock_res_and_lock(struct ldlm_lock *lock);
1573 /** \defgroup ldlm_pools Various LDLM pool related functions
1574 * There are not used outside of ldlm.
1577 int ldlm_pools_recalc(enum ldlm_side client);
1578 int ldlm_pools_init(void);
1579 void ldlm_pools_fini(void);
1581 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1582 int idx, enum ldlm_side client);
1583 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1584 void ldlm_pool_fini(struct ldlm_pool *pl);
1585 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1586 int ldlm_pool_recalc(struct ldlm_pool *pl);
1587 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1588 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1589 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1590 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1591 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1592 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1593 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1594 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1595 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1598 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1599 const struct ldlm_extent *ex2)
1601 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1604 /* check if @ex1 contains @ex2 */
1605 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1606 const struct ldlm_extent *ex2)
1608 return ex1->start <= ex2->start && ex1->end >= ex2->end;