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(3900)) /* 65 min */
67 #define LDLM_CTIME_AGE_LIMIT (10)
68 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
71 * LDLM non-error return states
75 ELDLM_LOCK_MATCHED = 1,
77 ELDLM_LOCK_CHANGED = 300,
78 ELDLM_LOCK_ABORTED = 301,
79 ELDLM_LOCK_REPLACED = 302,
80 ELDLM_NO_LOCK_DATA = 303,
81 ELDLM_LOCK_WOULDBLOCK = 304,
83 ELDLM_NAMESPACE_EXISTS = 400,
84 ELDLM_BAD_NAMESPACE = 401,
88 * LDLM namespace type.
89 * The "client" type is actually an indication that this is a narrow local view
90 * into complete namespace on the server. Such namespaces cannot make any
91 * decisions about lack of conflicts or do any autonomous lock granting without
92 * first speaking to a server.
95 LDLM_NAMESPACE_SERVER = 0x01,
96 LDLM_NAMESPACE_CLIENT = 0x02
100 * The blocking callback is overloaded to perform two functions. These flags
101 * indicate which operation should be performed.
103 #define LDLM_CB_BLOCKING 1
104 #define LDLM_CB_CANCELING 2
107 * \name Lock Compatibility Matrix.
109 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
110 * Lock types are described in their respective implementation files:
111 * ldlm_{extent,flock,inodebits,plain}.c.
113 * There are six lock modes along with a compatibility matrix to indicate if
114 * two locks are compatible.
116 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
118 * - PW: Protective Write (normal write) mode. When a client requests a write
119 * lock from an OST, a lock with PW mode will be issued.
120 * - PR: Protective Read (normal read) mode. When a client requests a read from
121 * an OST, a lock with PR mode will be issued. Also, if the client opens a
122 * file for execution, it is granted a lock with PR mode.
123 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
124 * requests a write lock during a file open operation.
125 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
126 * an inodebit lock with the CR mode on the intermediate path component.
140 #define LCK_COMPAT_EX LCK_NL
141 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
142 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
143 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
144 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
145 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
146 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
147 #define LCK_COMPAT_COS (LCK_COS)
148 /** @} Lock Compatibility Matrix */
150 extern enum ldlm_mode lck_compat_array[];
152 static inline void lockmode_verify(enum ldlm_mode mode)
154 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
157 static inline int lockmode_compat(enum ldlm_mode exist_mode,
158 enum ldlm_mode new_mode)
160 return lck_compat_array[exist_mode] & new_mode;
165 * cluster name spaces
169 #define DLM_OST_NAMESPACE 1
170 #define DLM_MDS_NAMESPACE 2
173 - do we just separate this by security domains and use a prefix for
174 multiple namespaces in the same domain?
179 * Locking rules for LDLM:
184 * waiting_locks_spinlock
199 struct ldlm_resource;
200 struct ldlm_namespace;
203 * Operations on LDLM pools.
204 * LDLM pool is a pool of locks in the namespace without any implicitly
206 * Locks in the pool are organized in LRU.
207 * Local memory pressure or server instructions (e.g. mempressure on server)
208 * can trigger freeing of locks from the pool
210 struct ldlm_pool_ops {
211 /** Recalculate pool \a pl usage */
212 int (*po_recalc)(struct ldlm_pool *pl);
213 /** Cancel at least \a nr locks from pool \a pl */
214 int (*po_shrink)(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
215 int (*po_setup)(struct ldlm_pool *pl, int limit);
218 /** One second for pools thread check interval. Each pool has own period. */
219 #define LDLM_POOLS_THREAD_PERIOD (1)
221 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
222 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
224 /** Default recalc period for server side pools in sec. */
225 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
227 /** Default recalc period for client side pools in sec. */
228 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
231 * LDLM pool structure to track granted locks.
232 * For purposes of determining when to release locks on e.g. memory pressure.
233 * This feature is commonly referred to as lru_resize.
236 /** Pool proc directory. */
237 struct proc_dir_entry *pl_proc_dir;
238 /** Pool name, must be long enough to hold compound proc entry name. */
240 /** Lock for protecting SLV/CLV updates. */
242 /** Number of allowed locks in in pool, both, client and server side. */
244 /** Number of granted locks in */
246 /** Grant rate per T. */
247 atomic_t pl_grant_rate;
248 /** Cancel rate per T. */
249 atomic_t pl_cancel_rate;
250 /** Server lock volume (SLV). Protected by pl_lock. */
251 __u64 pl_server_lock_volume;
252 /** Current biggest client lock volume. Protected by pl_lock. */
253 __u64 pl_client_lock_volume;
254 /** Lock volume factor. SLV on client is calculated as following:
255 * server_slv * lock_volume_factor. */
256 atomic_t pl_lock_volume_factor;
257 /** Time when last SLV from server was obtained. */
258 time_t pl_recalc_time;
259 /** Recalculation period for pool. */
260 time_t pl_recalc_period;
261 /** Recalculation and shrink operations. */
262 struct ldlm_pool_ops *pl_ops;
263 /** Number of planned locks for next period. */
265 /** Pool statistics. */
266 struct lprocfs_stats *pl_stats;
269 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
270 void *req_cookie, enum ldlm_mode mode,
271 __u64 flags, void *data);
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 * To ensure delayed LVB initialization, it is highly recommended to use the set
286 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
288 struct ldlm_valblock_ops {
289 int (*lvbo_init)(struct ldlm_resource *res);
290 int (*lvbo_update)(struct ldlm_resource *res,
291 struct ptlrpc_request *r,
293 int (*lvbo_free)(struct ldlm_resource *res);
294 /* Return size of lvb data appropriate RPC size can be reserved */
295 int (*lvbo_size)(struct ldlm_lock *lock);
296 /* Called to fill in lvb data to RPC buffer @buf */
297 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
301 * LDLM pools related, type of lock pool in the namespace.
302 * Greedy means release cached locks aggressively
305 LDLM_NAMESPACE_GREEDY = 1 << 0,
306 LDLM_NAMESPACE_MODEST = 1 << 1
310 * Default values for the "max_nolock_size", "contention_time" and
311 * "contended_locks" namespace tunables.
313 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
314 #define NS_DEFAULT_CONTENTION_SECONDS 2
315 #define NS_DEFAULT_CONTENDED_LOCKS 32
317 struct ldlm_ns_bucket {
318 /** back pointer to namespace */
319 struct ldlm_namespace *nsb_namespace;
321 * Estimated lock callback time. Used by adaptive timeout code to
322 * avoid spurious client evictions due to unresponsiveness when in
323 * fact the network or overall system load is at fault
325 struct adaptive_timeout nsb_at_estimate;
327 * Which res in the bucket should we start with the reclaim.
329 int nsb_reclaim_start;
333 /** LDLM namespace lock stats */
339 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
340 LDLM_NS_TYPE_MDC, /**< MDC namespace */
341 LDLM_NS_TYPE_MDT, /**< MDT namespace */
342 LDLM_NS_TYPE_OSC, /**< OSC namespace */
343 LDLM_NS_TYPE_OST, /**< OST namespace */
344 LDLM_NS_TYPE_MGC, /**< MGC namespace */
345 LDLM_NS_TYPE_MGT, /**< MGT namespace */
351 * Namespace serves to contain locks related to a particular service.
352 * There are two kinds of namespaces:
353 * - Server namespace has knowledge of all locks and is therefore authoritative
354 * to make decisions like what locks could be granted and what conflicts
355 * exist during new lock enqueue.
356 * - Client namespace only has limited knowledge about locks in the namespace,
357 * only seeing locks held by the client.
359 * Every Lustre service has one server namespace present on the server serving
360 * that service. Every client connected to the service has a client namespace
362 * Every lock obtained by client in that namespace is actually represented by
363 * two in-memory locks. One on the server and one on the client. The locks are
364 * linked by a special cookie by which one node can tell to the other which lock
365 * it actually means during communications. Such locks are called remote locks.
366 * The locks held by server only without any reference to a client are called
369 struct ldlm_namespace {
370 /** Backward link to OBD, required for LDLM pool to store new SLV. */
371 struct obd_device *ns_obd;
373 /** Flag indicating if namespace is on client instead of server */
374 enum ldlm_side ns_client;
376 /** Resource hash table for namespace. */
377 struct cfs_hash *ns_rs_hash;
382 /** big refcount (by bucket) */
386 * Namespace connect flags supported by server (may be changed via
387 * /proc, LRU resize may be disabled/enabled).
389 __u64 ns_connect_flags;
391 /** Client side original connect flags supported by server. */
392 __u64 ns_orig_connect_flags;
394 /* namespace proc dir entry */
395 struct proc_dir_entry *ns_proc_dir_entry;
398 * Position in global namespace list linking all namespaces on
401 struct list_head ns_list_chain;
404 * List of unused locks for this namespace. This list is also called
406 * Unused locks are locks with zero reader/writer reference counts.
407 * This list is only used on clients for lock caching purposes.
408 * When we want to release some locks voluntarily or if server wants
409 * us to release some locks due to e.g. memory pressure, we take locks
410 * to release from the head of this list.
411 * Locks are linked via l_lru field in \see struct ldlm_lock.
413 struct list_head ns_unused_list;
414 /** Number of locks in the LRU list above */
418 * Maximum number of locks permitted in the LRU. If 0, means locks
419 * are managed by pools and there is no preset limit, rather it is all
420 * controlled by available memory on this client and on server.
422 unsigned int ns_max_unused;
423 /** Maximum allowed age (last used time) for locks in the LRU */
424 unsigned int ns_max_age;
426 * Server only: number of times we evicted clients due to lack of reply
429 unsigned int ns_timeouts;
431 * Number of seconds since the file change time after which the
432 * MDT will return an UPDATE lock along with a LOOKUP lock.
433 * This allows the client to start caching negative dentries
434 * for a directory and may save an RPC for a later stat.
436 unsigned int ns_ctime_age_limit;
439 * Used to rate-limit ldlm_namespace_dump calls.
440 * \see ldlm_namespace_dump. Increased by 10 seconds every time
443 cfs_time_t ns_next_dump;
445 /** "policy" function that does actual lock conflict determination */
446 ldlm_res_policy ns_policy;
449 * LVB operations for this namespace.
450 * \see struct ldlm_valblock_ops
452 struct ldlm_valblock_ops *ns_lvbo;
455 * Used by filter code to store pointer to OBD of the service.
456 * Should be dropped in favor of \a ns_obd
461 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
462 * a resource is removed.
464 wait_queue_head_t ns_waitq;
465 /** LDLM pool structure for this namespace */
466 struct ldlm_pool ns_pool;
467 /** Definition of how eagerly unused locks will be released from LRU */
468 enum ldlm_appetite ns_appetite;
471 * If more than \a ns_contended_locks are found, the resource is
472 * considered to be contended. Lock enqueues might specify that no
473 * contended locks should be granted
475 unsigned ns_contended_locks;
478 * The resources in this namespace remember contended state during
479 * \a ns_contention_time, in seconds.
481 unsigned ns_contention_time;
484 * Limit size of contended extent locks, in bytes.
485 * If extended lock is requested for more then this many bytes and
486 * caller instructs us not to grant contended locks, we would disregard
489 unsigned ns_max_nolock_size;
491 /** Limit of parallel AST RPC count. */
492 unsigned ns_max_parallel_ast;
495 * Callback to check if a lock is good to be canceled by ELC or
498 ldlm_cancel_cbt ns_cancel;
500 /** LDLM lock stats */
501 struct lprocfs_stats *ns_stats;
504 * Flag to indicate namespace is being freed. Used to determine if
505 * recalculation of LDLM pool statistics should be skipped.
507 unsigned ns_stopping:1;
510 * Which bucket should we start with the lock reclaim.
512 int ns_reclaim_start;
516 * Returns 1 if namespace \a ns is a client namespace.
518 static inline int ns_is_client(struct ldlm_namespace *ns)
521 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
522 LDLM_NAMESPACE_SERVER)));
523 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
524 ns->ns_client == LDLM_NAMESPACE_SERVER);
525 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
529 * Returns 1 if namespace \a ns is a server namespace.
531 static inline int ns_is_server(struct ldlm_namespace *ns)
534 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
535 LDLM_NAMESPACE_SERVER)));
536 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
537 ns->ns_client == LDLM_NAMESPACE_SERVER);
538 return ns->ns_client == LDLM_NAMESPACE_SERVER;
542 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
544 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
547 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
551 * Returns 1 if this namespace supports lru_resize.
553 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
556 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
559 static inline void ns_register_cancel(struct ldlm_namespace *ns,
568 /** Type for blocking callback function of a lock. */
569 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
570 struct ldlm_lock_desc *new, void *data,
572 /** Type for completion callback function of a lock. */
573 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
575 /** Type for glimpse callback function of a lock. */
576 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
578 /** Work list for sending GL ASTs to multiple locks. */
579 struct ldlm_glimpse_work {
580 struct ldlm_lock *gl_lock; /* lock to glimpse */
581 struct list_head gl_list; /* linkage to other gl work structs */
582 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
583 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
584 * glimpse callback request */
587 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
588 #define LDLM_GL_WORK_NOFREE 0x1
590 /** Interval node data for each LDLM_EXTENT lock. */
591 struct ldlm_interval {
592 struct interval_node li_node; /* node for tree management */
593 struct list_head li_group; /* the locks which have the same
594 * policy - group of the policy */
596 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
599 * Interval tree for extent locks.
600 * The interval tree must be accessed under the resource lock.
601 * Interval trees are used for granted extent locks to speed up conflicts
602 * lookup. See ldlm/interval_tree.c for more details.
604 struct ldlm_interval_tree {
607 enum ldlm_mode lit_mode; /* lock mode */
608 struct interval_node *lit_root; /* actual ldlm_interval */
611 /** Whether to track references to exports by LDLM locks. */
612 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
615 enum ldlm_cancel_flags {
616 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
617 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
618 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
625 __u64 blocking_owner;
626 struct obd_export *blocking_export;
627 /* Protected by the hash lock */
632 union ldlm_policy_data {
633 struct ldlm_extent l_extent;
634 struct ldlm_flock l_flock;
635 struct ldlm_inodebits l_inodebits;
638 void ldlm_convert_policy_to_wire(enum ldlm_type type,
639 const union ldlm_policy_data *lpolicy,
640 union ldlm_wire_policy_data *wpolicy);
641 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
642 const union ldlm_wire_policy_data *wpolicy,
643 union ldlm_policy_data *lpolicy);
653 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
655 #define LDLM_GID_ANY ((__u64)-1)
658 * LDLM lock structure
660 * Represents a single LDLM lock and its state in memory. Each lock is
661 * associated with a single ldlm_resource, the object which is being
662 * locked. There may be multiple ldlm_locks on a single resource,
663 * depending on the lock type and whether the locks are conflicting or
669 * When remote side wants to tell us about a lock, they address
670 * it by this opaque handle. The handle does not hold a
671 * reference on the ldlm_lock, so it can be safely passed to
672 * other threads or nodes. When the lock needs to be accessed
673 * from the handle, it is looked up again in the lock table, and
674 * may no longer exist.
676 * Must be first in the structure.
678 struct portals_handle l_handle;
680 * Lock reference count.
681 * This is how many users have pointers to actual structure, so that
682 * we do not accidentally free lock structure that is in use.
686 * Internal spinlock protects l_resource. We should hold this lock
687 * first before taking res_lock.
691 * Pointer to actual resource this lock is in.
692 * ldlm_lock_change_resource() can change this.
694 struct ldlm_resource *l_resource;
696 * List item for client side LRU list.
697 * Protected by ns_lock in struct ldlm_namespace.
699 struct list_head l_lru;
701 * Linkage to resource's lock queues according to current lock state.
702 * (could be granted, waiting or converting)
703 * Protected by lr_lock in struct ldlm_resource.
705 struct list_head l_res_link;
707 * Tree node for ldlm_extent.
709 struct ldlm_interval *l_tree_node;
711 * Per export hash of locks.
712 * Protected by per-bucket exp->exp_lock_hash locks.
714 struct hlist_node l_exp_hash;
716 * Per export hash of flock locks.
717 * Protected by per-bucket exp->exp_flock_hash locks.
719 struct hlist_node l_exp_flock_hash;
722 * Protected by lr_lock.
724 enum ldlm_mode l_req_mode;
726 * Granted mode, also protected by lr_lock.
728 enum ldlm_mode l_granted_mode;
729 /** Lock completion handler pointer. Called when lock is granted. */
730 ldlm_completion_callback l_completion_ast;
732 * Lock blocking AST handler pointer.
733 * It plays two roles:
734 * - as a notification of an attempt to queue a conflicting lock (once)
735 * - as a notification when the lock is being cancelled.
737 * As such it's typically called twice: once for the initial conflict
738 * and then once more when the last user went away and the lock is
739 * cancelled (could happen recursively).
741 ldlm_blocking_callback l_blocking_ast;
743 * Lock glimpse handler.
744 * Glimpse handler is used to obtain LVB updates from a client by
747 ldlm_glimpse_callback l_glimpse_ast;
751 * This is a pointer to actual client export for locks that were granted
752 * to clients. Used server-side.
754 struct obd_export *l_export;
756 * Lock connection export.
757 * Pointer to server export on a client.
759 struct obd_export *l_conn_export;
762 * Remote lock handle.
763 * If the lock is remote, this is the handle of the other side lock
766 struct lustre_handle l_remote_handle;
769 * Representation of private data specific for a lock type.
770 * Examples are: extent range for extent lock or bitmask for ibits locks
772 union ldlm_policy_data l_policy_data;
775 * Lock state flags. Protected by lr_lock.
776 * \see lustre_dlm_flags.h where the bits are defined.
781 * Lock r/w usage counters.
782 * Protected by lr_lock.
787 * If the lock is granted, a process sleeps on this waitq to learn when
788 * it's no longer in use. If the lock is not granted, a process sleeps
789 * on this waitq to learn when it becomes granted.
791 wait_queue_head_t l_waitq;
794 * Seconds. It will be updated if there is any activity related to
795 * the lock, e.g. enqueue the lock or send blocking AST.
797 cfs_time_t l_last_activity;
800 * Time last used by e.g. being matched by lock match.
801 * Jiffies. Should be converted to time if needed.
803 cfs_time_t l_last_used;
805 /** Originally requested extent for the extent lock. */
806 struct ldlm_extent l_req_extent;
809 * Client-side-only members.
812 enum lvb_type l_lvb_type;
815 * Temporary storage for a LVB received during an enqueue operation.
816 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
821 /** Private storage for lock user. Opaque to LDLM. */
825 * Server-side-only members.
829 * Connection cookie for the client originating the operation.
830 * Used by Commit on Share (COS) code. Currently only used for
831 * inodebits locks on MDS.
833 __u64 l_client_cookie;
836 * List item for locks waiting for cancellation from clients.
837 * The lists this could be linked into are:
838 * waiting_locks_list (protected by waiting_locks_spinlock),
839 * then if the lock timed out, it is moved to
840 * expired_lock_thread.elt_expired_locks for further processing.
841 * Protected by elt_lock.
843 struct list_head l_pending_chain;
846 * Set when lock is sent a blocking AST. Time in seconds when timeout
847 * is reached and client holding this lock could be evicted.
848 * This timeout could be further extended by e.g. certain IO activity
850 * \see ost_rw_prolong_locks
852 cfs_time_t l_callback_timeout;
854 /** Local PID of process which created this lock. */
858 * Number of times blocking AST was sent for this lock.
859 * This is for debugging. Valid values are 0 and 1, if there is an
860 * attempt to send blocking AST more than once, an assertion would be
861 * hit. \see ldlm_work_bl_ast_lock
864 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
865 struct list_head l_bl_ast;
866 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
867 struct list_head l_cp_ast;
868 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
869 struct list_head l_rk_ast;
872 * Pointer to a conflicting lock that caused blocking AST to be sent
875 struct ldlm_lock *l_blocking_lock;
878 * Protected by lr_lock, linkages to "skip lists".
879 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
881 struct list_head l_sl_mode;
882 struct list_head l_sl_policy;
884 /** Reference tracking structure to debug leaked locks. */
885 struct lu_ref l_reference;
886 #if LUSTRE_TRACKS_LOCK_EXP_REFS
887 /* Debugging stuff for bug 20498, for tracking export references. */
888 /** number of export references taken */
890 /** link all locks referencing one export */
891 struct list_head l_exp_refs_link;
892 /** referenced export object */
893 struct obd_export *l_exp_refs_target;
896 * export blocking dlm lock list, protected by
897 * l_export->exp_bl_list_lock.
898 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
899 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
901 struct list_head l_exp_list;
904 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
905 #define l_transno l_client_cookie
907 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
908 * which is for server. */
909 #define l_slc_link l_rk_ast
912 * LDLM resource description.
913 * Basically, resource is a representation for a single object.
914 * Object has a name which is currently 4 64-bit integers. LDLM user is
915 * responsible for creation of a mapping between objects it wants to be
916 * protected and resource names.
918 * A resource can only hold locks of a single lock type, though there may be
919 * multiple ldlm_locks on a single resource, depending on the lock type and
920 * whether the locks are conflicting or not.
922 struct ldlm_resource {
923 struct ldlm_ns_bucket *lr_ns_bucket;
926 * List item for list in namespace hash.
927 * protected by ns_lock
929 struct hlist_node lr_hash;
931 /** Reference count for this resource */
932 atomic_t lr_refcount;
934 /** Spinlock to protect locks under this resource. */
938 * protected by lr_lock
940 /** List of locks in granted state */
941 struct list_head lr_granted;
942 /** List of locks waiting to change their granted mode (converted) */
943 struct list_head lr_converting;
945 * List of locks that could not be granted due to conflicts and
946 * that are waiting for conflicts to go away */
947 struct list_head lr_waiting;
951 struct ldlm_res_id lr_name;
954 * Interval trees (only for extent locks) for all modes of this resource
956 struct ldlm_interval_tree *lr_itree;
960 * When the resource was considered as contended,
961 * used only on server side. */
962 cfs_time_t lr_contention_time;
964 * Associated inode, used only on client side.
966 struct inode *lr_lvb_inode;
969 /** Type of locks this resource can hold. Only one type per resource. */
970 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
973 * Server-side-only lock value block elements.
974 * To serialize lvbo_init.
977 struct mutex lr_lvb_mutex;
978 /** protected by lr_lock */
980 /** is lvb initialized ? */
981 bool lr_lvb_initialized;
983 /** List of references to this resource. For debugging. */
984 struct lu_ref lr_reference;
987 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
989 return lock->l_resource->lr_type == LDLM_IBITS &&
990 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
994 ldlm_ns_name(struct ldlm_namespace *ns)
996 return ns->ns_rs_hash->hs_name;
999 static inline struct ldlm_namespace *
1000 ldlm_res_to_ns(struct ldlm_resource *res)
1002 return res->lr_ns_bucket->nsb_namespace;
1005 static inline struct ldlm_namespace *
1006 ldlm_lock_to_ns(struct ldlm_lock *lock)
1008 return ldlm_res_to_ns(lock->l_resource);
1011 static inline char *
1012 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1014 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1017 static inline struct adaptive_timeout *
1018 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1020 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1023 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1025 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1028 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1029 res->lr_lvb_initialized)
1032 mutex_lock(&res->lr_lvb_mutex);
1033 /* Did we lose the race? */
1034 if (res->lr_lvb_initialized) {
1035 mutex_unlock(&res->lr_lvb_mutex);
1038 rc = ns->ns_lvbo->lvbo_init(res);
1040 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1042 if (res->lr_lvb_data != NULL) {
1043 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1044 res->lr_lvb_data = NULL;
1046 res->lr_lvb_len = rc;
1048 res->lr_lvb_initialized = true;
1050 mutex_unlock(&res->lr_lvb_mutex);
1054 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1056 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1058 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1059 return ns->ns_lvbo->lvbo_size(lock);
1064 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1066 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1069 if (ns->ns_lvbo != NULL) {
1070 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1071 /* init lvb now if not already */
1072 rc = ldlm_lvbo_init(lock->l_resource);
1074 CERROR("lock %p: delayed lvb init failed (rc %d)",
1078 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1083 struct ldlm_ast_work {
1084 struct ldlm_lock *w_lock;
1086 struct ldlm_lock_desc w_desc;
1087 struct list_head w_list;
1094 * Common ldlm_enqueue parameters
1096 struct ldlm_enqueue_info {
1097 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1098 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1099 void *ei_cb_bl; /** blocking lock callback */
1100 void *ei_cb_local_bl; /** blocking local lock callback */
1101 void *ei_cb_cp; /** lock completion callback */
1102 void *ei_cb_gl; /** lock glimpse callback */
1103 void *ei_cbdata; /** Data to be passed into callbacks. */
1104 void *ei_namespace; /** lock namespace **/
1105 unsigned int ei_enq_slave:1, /** whether enqueue slave stripes */
1106 ei_nonblock:1; /** non block enqueue */
1109 #define ei_res_id ei_cb_gl
1111 extern struct obd_ops ldlm_obd_ops;
1113 extern char *ldlm_lockname[];
1114 extern char *ldlm_typename[];
1115 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1118 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1119 * For the cases where we do not have actual lock to print along
1120 * with a debugging message that is ldlm-related
1122 #define LDLM_DEBUG_NOLOCK(format, a...) \
1123 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1126 * Support function for lock information printing into debug logs.
1130 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1131 CFS_CHECK_STACK(msgdata, mask, cdls); \
1133 if (((mask) & D_CANTMASK) != 0 || \
1134 ((libcfs_debug & (mask)) != 0 && \
1135 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1136 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1139 void _ldlm_lock_debug(struct ldlm_lock *lock,
1140 struct libcfs_debug_msg_data *data,
1141 const char *fmt, ...)
1142 __attribute__ ((format (printf, 3, 4)));
1145 * Rate-limited version of lock printing function.
1147 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1148 static struct cfs_debug_limit_state _ldlm_cdls; \
1149 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1150 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1153 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1154 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1156 /** Non-rate-limited lock printing function for debugging purposes. */
1157 #define LDLM_DEBUG(lock, fmt, a...) do { \
1158 if (likely(lock != NULL)) { \
1159 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1160 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1161 "### " fmt , ##a); \
1163 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1166 #else /* !LIBCFS_DEBUG */
1167 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1168 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1169 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1172 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1173 int first_enq, enum ldlm_error *err,
1174 struct list_head *work_list);
1177 * Return values for lock iterators.
1178 * Also used during deciding of lock grants and cancellations.
1180 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1181 #define LDLM_ITER_STOP 2 /* stop iterating */
1183 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1184 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1186 /** \defgroup ldlm_iterator Lock iterators
1188 * LDLM provides for a way to iterate through every lock on a resource or
1189 * namespace or every resource in a namespace.
1191 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1193 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1195 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1196 ldlm_iterator_t iter, void *data);
1197 /** @} ldlm_iterator */
1199 int ldlm_replay_locks(struct obd_import *imp);
1202 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1205 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1207 struct ldlm_callback_suite {
1208 ldlm_completion_callback lcs_completion;
1209 ldlm_blocking_callback lcs_blocking;
1210 ldlm_glimpse_callback lcs_glimpse;
1214 #ifdef HAVE_SERVER_SUPPORT
1215 /** \defgroup ldlm_srv_ast Server AST handlers
1216 * These are AST handlers used by server code.
1217 * Their property is that they are just preparing RPCs to be sent to clients.
1220 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1221 void *data, int flag);
1222 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1223 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1224 int ldlm_glimpse_locks(struct ldlm_resource *res,
1225 struct list_head *gl_work_list);
1226 /** @} ldlm_srv_ast */
1228 /** \defgroup ldlm_handlers Server LDLM handlers
1229 * These are handler functions that should be called by "frontends" such as
1230 * MDT or OST to pass through LDLM requests to LDLM for handling
1233 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
1234 ldlm_blocking_callback, ldlm_glimpse_callback);
1235 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1236 const struct ldlm_request *dlm_req,
1237 const struct ldlm_callback_suite *cbs);
1238 int ldlm_handle_convert(struct ptlrpc_request *req);
1239 int ldlm_handle_convert0(struct ptlrpc_request *req,
1240 const struct ldlm_request *dlm_req);
1241 int ldlm_handle_cancel(struct ptlrpc_request *req);
1242 int ldlm_request_cancel(struct ptlrpc_request *req,
1243 const struct ldlm_request *dlm_req,
1244 int first, enum lustre_at_flags flags);
1245 /** @} ldlm_handlers */
1247 void ldlm_revoke_export_locks(struct obd_export *exp);
1248 unsigned int ldlm_bl_timeout(struct ldlm_lock *lock);
1250 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1251 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
1252 int ldlm_get_ref(void);
1253 void ldlm_put_ref(void);
1254 int ldlm_init_export(struct obd_export *exp);
1255 void ldlm_destroy_export(struct obd_export *exp);
1256 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1259 #ifdef HAVE_SERVER_SUPPORT
1260 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1262 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1263 void ldlm_lock2handle(const struct ldlm_lock *lock,
1264 struct lustre_handle *lockh);
1265 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1266 void ldlm_cancel_callback(struct ldlm_lock *);
1267 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1268 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1271 * Obtain a lock reference by its handle.
1273 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1275 return __ldlm_handle2lock(h, 0);
1278 #define LDLM_LOCK_REF_DEL(lock) \
1279 lu_ref_del(&lock->l_reference, "handle", current)
1281 static inline struct ldlm_lock *
1282 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1284 struct ldlm_lock *lock;
1286 lock = __ldlm_handle2lock(h, flags);
1288 LDLM_LOCK_REF_DEL(lock);
1293 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1294 * data from request \a r
1296 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1297 struct ptlrpc_request *req, int increase)
1301 /* delayed lvb init may be required */
1302 rc = ldlm_lvbo_init(res);
1304 CERROR("delayed lvb init failed (rc %d)\n", rc);
1308 if (ldlm_res_to_ns(res)->ns_lvbo &&
1309 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1310 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, req,
1316 int ldlm_error2errno(enum ldlm_error error);
1317 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1318 * confuses user-space. */
1319 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1320 void ldlm_dump_export_locks(struct obd_export *exp);
1324 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1325 * __ldlm_handle2lock().
1327 #define LDLM_LOCK_PUT(lock) \
1329 LDLM_LOCK_REF_DEL(lock); \
1330 /*LDLM_DEBUG((lock), "put");*/ \
1331 ldlm_lock_put(lock); \
1335 * Release a lock reference obtained by some other means (see
1338 #define LDLM_LOCK_RELEASE(lock) \
1340 /*LDLM_DEBUG((lock), "put");*/ \
1341 ldlm_lock_put(lock); \
1344 #define LDLM_LOCK_GET(lock) \
1346 ldlm_lock_get(lock); \
1347 /*LDLM_DEBUG((lock), "get");*/ \
1351 #define ldlm_lock_list_put(head, member, count) \
1353 struct ldlm_lock *_lock, *_next; \
1355 list_for_each_entry_safe(_lock, _next, head, member) { \
1358 list_del_init(&_lock->member); \
1359 LDLM_LOCK_RELEASE(_lock); \
1364 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1365 void ldlm_lock_put(struct ldlm_lock *lock);
1366 void ldlm_lock_destroy(struct ldlm_lock *lock);
1367 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1368 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1369 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1370 enum ldlm_mode mode);
1371 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1372 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1373 enum ldlm_mode mode);
1374 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1375 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1376 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1377 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1378 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1379 const struct ldlm_res_id *, enum ldlm_type type,
1380 union ldlm_policy_data *, enum ldlm_mode mode,
1381 struct lustre_handle *, int unref);
1382 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1384 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock,
1385 enum ldlm_mode new_mode, __u32 *flags);
1386 void ldlm_lock_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1387 void ldlm_lock_cancel(struct ldlm_lock *lock);
1388 void ldlm_reprocess_all(struct ldlm_resource *res);
1389 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1390 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1391 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1394 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1395 enum ldlm_side client,
1396 enum ldlm_appetite apt,
1397 enum ldlm_ns_type ns_type);
1398 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1399 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1400 struct obd_import *imp,
1402 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1403 void ldlm_namespace_free(struct ldlm_namespace *ns,
1404 struct obd_import *imp, int force);
1405 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1406 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1407 enum ldlm_side client);
1408 void ldlm_namespace_get(struct ldlm_namespace *ns);
1409 void ldlm_namespace_put(struct ldlm_namespace *ns);
1410 int ldlm_proc_setup(void);
1411 #ifdef CONFIG_PROC_FS
1412 void ldlm_proc_cleanup(void);
1414 static inline void ldlm_proc_cleanup(void) {}
1417 /* resource.c - internal */
1418 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1419 struct ldlm_resource *parent,
1420 const struct ldlm_res_id *,
1421 enum ldlm_type type, int create);
1422 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1423 int ldlm_resource_putref(struct ldlm_resource *res);
1424 void ldlm_resource_add_lock(struct ldlm_resource *res,
1425 struct list_head *head,
1426 struct ldlm_lock *lock);
1427 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1428 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1429 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1430 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1431 void ldlm_resource_dump(int level, struct ldlm_resource *);
1432 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1433 const struct ldlm_res_id *);
1435 #define LDLM_RESOURCE_ADDREF(res) do { \
1436 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1439 #define LDLM_RESOURCE_DELREF(res) do { \
1440 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1443 /* ldlm_request.c */
1444 int ldlm_expired_completion_wait(void *data);
1445 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1446 * These AST handlers are typically used for server-side local locks and are
1447 * also used by client-side lock handlers to perform minimum level base
1450 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1451 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1452 void *data, int flag);
1453 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1454 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1455 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1456 /** @} ldlm_local_ast */
1458 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1459 * These are typically used by client and server (*_local versions)
1460 * to obtain and release locks.
1462 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1463 struct ldlm_enqueue_info *einfo,
1464 const struct ldlm_res_id *res_id,
1465 union ldlm_policy_data const *policy, __u64 *flags,
1466 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1467 struct lustre_handle *lockh, int async);
1468 int ldlm_prep_enqueue_req(struct obd_export *exp,
1469 struct ptlrpc_request *req,
1470 struct list_head *cancels,
1472 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1473 int version, int opc, int canceloff,
1474 struct list_head *cancels, int count);
1476 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1477 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1478 const struct ldlm_request *dlm_req,
1479 const struct ldlm_callback_suite *cbs);
1480 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1481 enum ldlm_type type, __u8 with_policy,
1482 enum ldlm_mode mode, __u64 *flags, void *lvb,
1484 const struct lustre_handle *lockh, int rc);
1485 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1486 const struct ldlm_res_id *res_id,
1487 enum ldlm_type type, union ldlm_policy_data *policy,
1488 enum ldlm_mode mode, __u64 *flags,
1489 ldlm_blocking_callback blocking,
1490 ldlm_completion_callback completion,
1491 ldlm_glimpse_callback glimpse,
1492 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1493 const __u64 *client_cookie,
1494 struct lustre_handle *lockh);
1495 int ldlm_cli_convert(const struct lustre_handle *lockh, int new_mode,
1497 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1498 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1499 enum ldlm_cancel_flags cancel_flags);
1500 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1501 enum ldlm_cancel_flags flags, void *opaque);
1502 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1503 const struct ldlm_res_id *res_id,
1504 union ldlm_policy_data *policy,
1505 enum ldlm_mode mode,
1506 enum ldlm_cancel_flags flags, void *opaque);
1507 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1508 int count, enum ldlm_cancel_flags flags);
1509 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1510 struct list_head *cancels,
1511 union ldlm_policy_data *policy,
1512 enum ldlm_mode mode, __u64 lock_flags,
1513 enum ldlm_cancel_flags cancel_flags,
1515 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1516 enum ldlm_cancel_flags flags);
1517 int ldlm_cli_cancel_list(struct list_head *head, int count,
1518 struct ptlrpc_request *req,
1519 enum ldlm_cancel_flags flags);
1520 /** @} ldlm_cli_api */
1523 /* This has to be here because recursive inclusion sucks. */
1524 int intent_disposition(struct ldlm_reply *rep, int flag);
1525 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1528 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1529 * than one lock_res is dead-lock safe.
1531 enum lock_res_type {
1536 /** Lock resource. */
1537 static inline void lock_res(struct ldlm_resource *res)
1539 spin_lock(&res->lr_lock);
1542 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1543 static inline void lock_res_nested(struct ldlm_resource *res,
1544 enum lock_res_type mode)
1546 spin_lock_nested(&res->lr_lock, mode);
1549 /** Unlock resource. */
1550 static inline void unlock_res(struct ldlm_resource *res)
1552 spin_unlock(&res->lr_lock);
1555 /** Check if resource is already locked, assert if not. */
1556 static inline void check_res_locked(struct ldlm_resource *res)
1558 assert_spin_locked(&res->lr_lock);
1561 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1562 void unlock_res_and_lock(struct ldlm_lock *lock);
1565 /** \defgroup ldlm_pools Various LDLM pool related functions
1566 * There are not used outside of ldlm.
1569 int ldlm_pools_recalc(enum ldlm_side client);
1570 int ldlm_pools_init(void);
1571 void ldlm_pools_fini(void);
1573 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1574 int idx, enum ldlm_side client);
1575 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1576 void ldlm_pool_fini(struct ldlm_pool *pl);
1577 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1578 int ldlm_pool_recalc(struct ldlm_pool *pl);
1579 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1580 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1581 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1582 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1583 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1584 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1585 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1586 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1587 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);