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, 2014, 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
73 typedef enum ldlm_error {
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.
94 typedef enum ldlm_side {
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
304 typedef enum ldlm_appetite {
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 */
338 typedef enum ldlm_ns_type {
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 typedef 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 */
619 } ldlm_cancel_flags_t;
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 typedef union ldlm_policy_data ldlm_policy_data_t;
640 void ldlm_convert_policy_to_wire(enum ldlm_type type,
641 const union ldlm_policy_data *lpolicy,
642 union ldlm_wire_policy_data *wpolicy);
643 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
644 const union ldlm_wire_policy_data *wpolicy,
645 union ldlm_policy_data *lpolicy);
655 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
657 #define LDLM_GID_ANY ((__u64)-1)
660 * LDLM lock structure
662 * Represents a single LDLM lock and its state in memory. Each lock is
663 * associated with a single ldlm_resource, the object which is being
664 * locked. There may be multiple ldlm_locks on a single resource,
665 * depending on the lock type and whether the locks are conflicting or
671 * When remote side wants to tell us about a lock, they address
672 * it by this opaque handle. The handle does not hold a
673 * reference on the ldlm_lock, so it can be safely passed to
674 * other threads or nodes. When the lock needs to be accessed
675 * from the handle, it is looked up again in the lock table, and
676 * may no longer exist.
678 * Must be first in the structure.
680 struct portals_handle l_handle;
682 * Lock reference count.
683 * This is how many users have pointers to actual structure, so that
684 * we do not accidentally free lock structure that is in use.
688 * Internal spinlock protects l_resource. We should hold this lock
689 * first before taking res_lock.
693 * Pointer to actual resource this lock is in.
694 * ldlm_lock_change_resource() can change this.
696 struct ldlm_resource *l_resource;
698 * List item for client side LRU list.
699 * Protected by ns_lock in struct ldlm_namespace.
701 struct list_head l_lru;
703 * Linkage to resource's lock queues according to current lock state.
704 * (could be granted, waiting or converting)
705 * Protected by lr_lock in struct ldlm_resource.
707 struct list_head l_res_link;
709 * Tree node for ldlm_extent.
711 struct ldlm_interval *l_tree_node;
713 * Per export hash of locks.
714 * Protected by per-bucket exp->exp_lock_hash locks.
716 struct hlist_node l_exp_hash;
718 * Per export hash of flock locks.
719 * Protected by per-bucket exp->exp_flock_hash locks.
721 struct hlist_node l_exp_flock_hash;
724 * Protected by lr_lock.
726 enum ldlm_mode l_req_mode;
728 * Granted mode, also protected by lr_lock.
730 enum ldlm_mode l_granted_mode;
731 /** Lock completion handler pointer. Called when lock is granted. */
732 ldlm_completion_callback l_completion_ast;
734 * Lock blocking AST handler pointer.
735 * It plays two roles:
736 * - as a notification of an attempt to queue a conflicting lock (once)
737 * - as a notification when the lock is being cancelled.
739 * As such it's typically called twice: once for the initial conflict
740 * and then once more when the last user went away and the lock is
741 * cancelled (could happen recursively).
743 ldlm_blocking_callback l_blocking_ast;
745 * Lock glimpse handler.
746 * Glimpse handler is used to obtain LVB updates from a client by
749 ldlm_glimpse_callback l_glimpse_ast;
753 * This is a pointer to actual client export for locks that were granted
754 * to clients. Used server-side.
756 struct obd_export *l_export;
758 * Lock connection export.
759 * Pointer to server export on a client.
761 struct obd_export *l_conn_export;
764 * Remote lock handle.
765 * If the lock is remote, this is the handle of the other side lock
768 struct lustre_handle l_remote_handle;
771 * Representation of private data specific for a lock type.
772 * Examples are: extent range for extent lock or bitmask for ibits locks
774 union ldlm_policy_data l_policy_data;
777 * Lock state flags. Protected by lr_lock.
778 * \see lustre_dlm_flags.h where the bits are defined.
783 * Lock r/w usage counters.
784 * Protected by lr_lock.
789 * If the lock is granted, a process sleeps on this waitq to learn when
790 * it's no longer in use. If the lock is not granted, a process sleeps
791 * on this waitq to learn when it becomes granted.
793 wait_queue_head_t l_waitq;
796 * Seconds. It will be updated if there is any activity related to
797 * the lock, e.g. enqueue the lock or send blocking AST.
799 cfs_time_t l_last_activity;
802 * Time last used by e.g. being matched by lock match.
803 * Jiffies. Should be converted to time if needed.
805 cfs_time_t l_last_used;
807 /** Originally requested extent for the extent lock. */
808 struct ldlm_extent l_req_extent;
811 * Client-side-only members.
814 enum lvb_type l_lvb_type;
817 * Temporary storage for a LVB received during an enqueue operation.
822 /** Private storage for lock user. Opaque to LDLM. */
826 * Server-side-only members.
830 * Connection cookie for the client originating the operation.
831 * Used by Commit on Share (COS) code. Currently only used for
832 * inodebits locks on MDS.
834 __u64 l_client_cookie;
837 * List item for locks waiting for cancellation from clients.
838 * The lists this could be linked into are:
839 * waiting_locks_list (protected by waiting_locks_spinlock),
840 * then if the lock timed out, it is moved to
841 * expired_lock_thread.elt_expired_locks for further processing.
842 * Protected by elt_lock.
844 struct list_head l_pending_chain;
847 * Set when lock is sent a blocking AST. Time in seconds when timeout
848 * is reached and client holding this lock could be evicted.
849 * This timeout could be further extended by e.g. certain IO activity
851 * \see ost_rw_prolong_locks
853 cfs_time_t l_callback_timeout;
855 /** Local PID of process which created this lock. */
859 * Number of times blocking AST was sent for this lock.
860 * This is for debugging. Valid values are 0 and 1, if there is an
861 * attempt to send blocking AST more than once, an assertion would be
862 * hit. \see ldlm_work_bl_ast_lock
865 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
866 struct list_head l_bl_ast;
867 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
868 struct list_head l_cp_ast;
869 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
870 struct list_head l_rk_ast;
873 * Pointer to a conflicting lock that caused blocking AST to be sent
876 struct ldlm_lock *l_blocking_lock;
879 * Protected by lr_lock, linkages to "skip lists".
880 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
882 struct list_head l_sl_mode;
883 struct list_head l_sl_policy;
885 /** Reference tracking structure to debug leaked locks. */
886 struct lu_ref l_reference;
887 #if LUSTRE_TRACKS_LOCK_EXP_REFS
888 /* Debugging stuff for bug 20498, for tracking export references. */
889 /** number of export references taken */
891 /** link all locks referencing one export */
892 struct list_head l_exp_refs_link;
893 /** referenced export object */
894 struct obd_export *l_exp_refs_target;
897 * export blocking dlm lock list, protected by
898 * l_export->exp_bl_list_lock.
899 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
900 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
902 struct list_head l_exp_list;
906 * LDLM resource description.
907 * Basically, resource is a representation for a single object.
908 * Object has a name which is currently 4 64-bit integers. LDLM user is
909 * responsible for creation of a mapping between objects it wants to be
910 * protected and resource names.
912 * A resource can only hold locks of a single lock type, though there may be
913 * multiple ldlm_locks on a single resource, depending on the lock type and
914 * whether the locks are conflicting or not.
916 struct ldlm_resource {
917 struct ldlm_ns_bucket *lr_ns_bucket;
920 * List item for list in namespace hash.
921 * protected by ns_lock
923 struct hlist_node lr_hash;
925 /** Reference count for this resource */
926 atomic_t lr_refcount;
928 /** Spinlock to protect locks under this resource. */
932 * protected by lr_lock
934 /** List of locks in granted state */
935 struct list_head lr_granted;
936 /** List of locks waiting to change their granted mode (converted) */
937 struct list_head lr_converting;
939 * List of locks that could not be granted due to conflicts and
940 * that are waiting for conflicts to go away */
941 struct list_head lr_waiting;
945 struct ldlm_res_id lr_name;
948 * Interval trees (only for extent locks) for all modes of this resource
950 struct ldlm_interval_tree *lr_itree;
954 * When the resource was considered as contended,
955 * used only on server side. */
956 cfs_time_t lr_contention_time;
958 * Associated inode, used only on client side.
960 struct inode *lr_lvb_inode;
963 /** Type of locks this resource can hold. Only one type per resource. */
964 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
967 * Server-side-only lock value block elements.
968 * To serialize lvbo_init.
971 struct mutex lr_lvb_mutex;
972 /** protected by lr_lock */
974 /** is lvb initialized ? */
975 bool lr_lvb_initialized;
977 /** List of references to this resource. For debugging. */
978 struct lu_ref lr_reference;
981 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
983 return lock->l_resource->lr_type == LDLM_IBITS &&
984 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
988 ldlm_ns_name(struct ldlm_namespace *ns)
990 return ns->ns_rs_hash->hs_name;
993 static inline struct ldlm_namespace *
994 ldlm_res_to_ns(struct ldlm_resource *res)
996 return res->lr_ns_bucket->nsb_namespace;
999 static inline struct ldlm_namespace *
1000 ldlm_lock_to_ns(struct ldlm_lock *lock)
1002 return ldlm_res_to_ns(lock->l_resource);
1005 static inline char *
1006 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1008 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1011 static inline struct adaptive_timeout *
1012 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1014 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1017 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1019 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1022 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1023 res->lr_lvb_initialized)
1026 mutex_lock(&res->lr_lvb_mutex);
1027 /* Did we lose the race? */
1028 if (res->lr_lvb_initialized) {
1029 mutex_unlock(&res->lr_lvb_mutex);
1032 rc = ns->ns_lvbo->lvbo_init(res);
1034 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1036 if (res->lr_lvb_data != NULL) {
1037 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1038 res->lr_lvb_data = NULL;
1040 res->lr_lvb_len = rc;
1042 res->lr_lvb_initialized = true;
1044 mutex_unlock(&res->lr_lvb_mutex);
1048 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1050 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1052 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1053 return ns->ns_lvbo->lvbo_size(lock);
1058 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1060 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1063 if (ns->ns_lvbo != NULL) {
1064 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1065 /* init lvb now if not already */
1066 rc = ldlm_lvbo_init(lock->l_resource);
1068 CERROR("lock %p: delayed lvb init failed (rc %d)",
1072 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1077 struct ldlm_ast_work {
1078 struct ldlm_lock *w_lock;
1080 struct ldlm_lock_desc w_desc;
1081 struct list_head w_list;
1088 * Common ldlm_enqueue parameters
1090 struct ldlm_enqueue_info {
1091 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1092 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1093 void *ei_cb_bl; /** blocking lock callback */
1094 void *ei_cb_local_bl; /** blocking local lock callback */
1095 void *ei_cb_cp; /** lock completion callback */
1096 void *ei_cb_gl; /** lock glimpse callback */
1097 void *ei_cbdata; /** Data to be passed into callbacks. */
1098 void *ei_namespace; /** lock namespace **/
1099 unsigned int ei_enq_slave:1, /** whether enqueue slave stripes */
1100 ei_nonblock:1; /** non block enqueue */
1103 #define ei_res_id ei_cb_gl
1105 extern struct obd_ops ldlm_obd_ops;
1107 extern char *ldlm_lockname[];
1108 extern char *ldlm_typename[];
1109 extern char *ldlm_it2str(int it);
1112 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1113 * For the cases where we do not have actual lock to print along
1114 * with a debugging message that is ldlm-related
1116 #define LDLM_DEBUG_NOLOCK(format, a...) \
1117 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1120 * Support function for lock information printing into debug logs.
1124 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1125 CFS_CHECK_STACK(msgdata, mask, cdls); \
1127 if (((mask) & D_CANTMASK) != 0 || \
1128 ((libcfs_debug & (mask)) != 0 && \
1129 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1130 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1133 void _ldlm_lock_debug(struct ldlm_lock *lock,
1134 struct libcfs_debug_msg_data *data,
1135 const char *fmt, ...)
1136 __attribute__ ((format (printf, 3, 4)));
1139 * Rate-limited version of lock printing function.
1141 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1142 static cfs_debug_limit_state_t _ldlm_cdls; \
1143 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1144 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1147 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1148 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1150 /** Non-rate-limited lock printing function for debugging purposes. */
1151 #define LDLM_DEBUG(lock, fmt, a...) do { \
1152 if (likely(lock != NULL)) { \
1153 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1154 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1155 "### " fmt , ##a); \
1157 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1160 #else /* !LIBCFS_DEBUG */
1161 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1162 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1163 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1166 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1167 int first_enq, enum ldlm_error *err,
1168 struct list_head *work_list);
1171 * Return values for lock iterators.
1172 * Also used during deciding of lock grants and cancellations.
1174 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1175 #define LDLM_ITER_STOP 2 /* stop iterating */
1177 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1178 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1180 /** \defgroup ldlm_iterator Lock iterators
1182 * LDLM provides for a way to iterate through every lock on a resource or
1183 * namespace or every resource in a namespace.
1185 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1187 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1189 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1190 ldlm_iterator_t iter, void *data);
1191 /** @} ldlm_iterator */
1193 int ldlm_replay_locks(struct obd_import *imp);
1196 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1199 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1201 struct ldlm_callback_suite {
1202 ldlm_completion_callback lcs_completion;
1203 ldlm_blocking_callback lcs_blocking;
1204 ldlm_glimpse_callback lcs_glimpse;
1208 #ifdef HAVE_SERVER_SUPPORT
1209 /** \defgroup ldlm_srv_ast Server AST handlers
1210 * These are AST handlers used by server code.
1211 * Their property is that they are just preparing RPCs to be sent to clients.
1214 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1215 void *data, int flag);
1216 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1217 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1218 int ldlm_glimpse_locks(struct ldlm_resource *res,
1219 struct list_head *gl_work_list);
1220 /** @} ldlm_srv_ast */
1222 /** \defgroup ldlm_handlers Server LDLM handlers
1223 * These are handler functions that should be called by "frontends" such as
1224 * MDT or OST to pass through LDLM requests to LDLM for handling
1227 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
1228 ldlm_blocking_callback, ldlm_glimpse_callback);
1229 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1230 const struct ldlm_request *dlm_req,
1231 const struct ldlm_callback_suite *cbs);
1232 int ldlm_handle_convert(struct ptlrpc_request *req);
1233 int ldlm_handle_convert0(struct ptlrpc_request *req,
1234 const struct ldlm_request *dlm_req);
1235 int ldlm_handle_cancel(struct ptlrpc_request *req);
1236 int ldlm_request_cancel(struct ptlrpc_request *req,
1237 const struct ldlm_request *dlm_req,
1238 int first, enum lustre_at_flags flags);
1239 /** @} ldlm_handlers */
1241 void ldlm_revoke_export_locks(struct obd_export *exp);
1242 unsigned int ldlm_bl_timeout(struct ldlm_lock *lock);
1244 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1245 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
1246 int ldlm_get_ref(void);
1247 void ldlm_put_ref(void);
1248 int ldlm_init_export(struct obd_export *exp);
1249 void ldlm_destroy_export(struct obd_export *exp);
1250 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1253 #ifdef HAVE_SERVER_SUPPORT
1254 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1256 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1257 void ldlm_lock2handle(const struct ldlm_lock *lock,
1258 struct lustre_handle *lockh);
1259 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1260 void ldlm_cancel_callback(struct ldlm_lock *);
1261 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1262 int ldlm_lock_set_data(struct lustre_handle *, void *);
1265 * Obtain a lock reference by its handle.
1267 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1269 return __ldlm_handle2lock(h, 0);
1272 #define LDLM_LOCK_REF_DEL(lock) \
1273 lu_ref_del(&lock->l_reference, "handle", current)
1275 static inline struct ldlm_lock *
1276 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1278 struct ldlm_lock *lock;
1280 lock = __ldlm_handle2lock(h, flags);
1282 LDLM_LOCK_REF_DEL(lock);
1287 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1288 * data from request \a r
1290 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1291 struct ptlrpc_request *req, int increase)
1295 /* delayed lvb init may be required */
1296 rc = ldlm_lvbo_init(res);
1298 CERROR("delayed lvb init failed (rc %d)\n", rc);
1302 if (ldlm_res_to_ns(res)->ns_lvbo &&
1303 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1304 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, req,
1310 int ldlm_error2errno(enum ldlm_error error);
1311 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1312 * confuses user-space. */
1313 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1314 void ldlm_dump_export_locks(struct obd_export *exp);
1318 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1319 * __ldlm_handle2lock().
1321 #define LDLM_LOCK_PUT(lock) \
1323 LDLM_LOCK_REF_DEL(lock); \
1324 /*LDLM_DEBUG((lock), "put");*/ \
1325 ldlm_lock_put(lock); \
1329 * Release a lock reference obtained by some other means (see
1332 #define LDLM_LOCK_RELEASE(lock) \
1334 /*LDLM_DEBUG((lock), "put");*/ \
1335 ldlm_lock_put(lock); \
1338 #define LDLM_LOCK_GET(lock) \
1340 ldlm_lock_get(lock); \
1341 /*LDLM_DEBUG((lock), "get");*/ \
1345 #define ldlm_lock_list_put(head, member, count) \
1347 struct ldlm_lock *_lock, *_next; \
1349 list_for_each_entry_safe(_lock, _next, head, member) { \
1352 list_del_init(&_lock->member); \
1353 LDLM_LOCK_RELEASE(_lock); \
1358 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1359 void ldlm_lock_put(struct ldlm_lock *lock);
1360 void ldlm_lock_destroy(struct ldlm_lock *lock);
1361 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1362 void ldlm_lock_addref(struct lustre_handle *lockh, enum ldlm_mode mode);
1363 int ldlm_lock_addref_try(struct lustre_handle *lockh, enum ldlm_mode mode);
1364 void ldlm_lock_decref(struct lustre_handle *lockh, enum ldlm_mode mode);
1365 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh,
1366 enum ldlm_mode mode);
1367 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1368 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1369 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1370 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1371 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1372 const struct ldlm_res_id *, enum ldlm_type type,
1373 union ldlm_policy_data *, enum ldlm_mode mode,
1374 struct lustre_handle *, int unref);
1375 enum ldlm_mode ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1377 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock,
1378 enum ldlm_mode new_mode, __u32 *flags);
1379 void ldlm_lock_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1380 void ldlm_lock_cancel(struct ldlm_lock *lock);
1381 void ldlm_reprocess_all(struct ldlm_resource *res);
1382 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1383 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1384 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1387 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1388 enum ldlm_side client,
1389 enum ldlm_appetite apt,
1390 enum ldlm_ns_type ns_type);
1391 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1392 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1393 struct obd_import *imp,
1395 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1396 void ldlm_namespace_free(struct ldlm_namespace *ns,
1397 struct obd_import *imp, int force);
1398 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1399 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1400 enum ldlm_side client);
1401 void ldlm_namespace_get(struct ldlm_namespace *ns);
1402 void ldlm_namespace_put(struct ldlm_namespace *ns);
1403 int ldlm_proc_setup(void);
1404 #ifdef CONFIG_PROC_FS
1405 void ldlm_proc_cleanup(void);
1407 static inline void ldlm_proc_cleanup(void) {}
1410 /* resource.c - internal */
1411 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1412 struct ldlm_resource *parent,
1413 const struct ldlm_res_id *,
1414 enum ldlm_type type, int create);
1415 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1416 int ldlm_resource_putref(struct ldlm_resource *res);
1417 void ldlm_resource_add_lock(struct ldlm_resource *res,
1418 struct list_head *head,
1419 struct ldlm_lock *lock);
1420 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1421 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1422 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1423 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1424 void ldlm_resource_dump(int level, struct ldlm_resource *);
1425 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1426 const struct ldlm_res_id *);
1428 #define LDLM_RESOURCE_ADDREF(res) do { \
1429 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1432 #define LDLM_RESOURCE_DELREF(res) do { \
1433 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1436 /* ldlm_request.c */
1437 int ldlm_expired_completion_wait(void *data);
1438 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1439 * These AST handlers are typically used for server-side local locks and are
1440 * also used by client-side lock handlers to perform minimum level base
1443 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1444 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1445 void *data, int flag);
1446 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1447 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1448 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1449 /** @} ldlm_local_ast */
1451 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1452 * These are typically used by client and server (*_local versions)
1453 * to obtain and release locks.
1455 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1456 struct ldlm_enqueue_info *einfo,
1457 const struct ldlm_res_id *res_id,
1458 union ldlm_policy_data const *policy, __u64 *flags,
1459 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1460 struct lustre_handle *lockh, int async);
1461 int ldlm_prep_enqueue_req(struct obd_export *exp,
1462 struct ptlrpc_request *req,
1463 struct list_head *cancels,
1465 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1466 int version, int opc, int canceloff,
1467 struct list_head *cancels, int count);
1469 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1470 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1471 const struct ldlm_request *dlm_req,
1472 const struct ldlm_callback_suite *cbs);
1473 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1474 enum ldlm_type type, __u8 with_policy,
1475 enum ldlm_mode mode, __u64 *flags, void *lvb,
1477 struct lustre_handle *lockh, int rc);
1478 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1479 const struct ldlm_res_id *res_id,
1480 enum ldlm_type type, union ldlm_policy_data *policy,
1481 enum ldlm_mode mode, __u64 *flags,
1482 ldlm_blocking_callback blocking,
1483 ldlm_completion_callback completion,
1484 ldlm_glimpse_callback glimpse,
1485 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1486 const __u64 *client_cookie,
1487 struct lustre_handle *lockh);
1488 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1489 void *data, __u32 data_len);
1490 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1491 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1492 int ldlm_cli_cancel(struct lustre_handle *lockh,
1493 enum ldlm_cancel_flags cancel_flags);
1494 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1495 enum ldlm_cancel_flags flags, void *opaque);
1496 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1497 const struct ldlm_res_id *res_id,
1498 union ldlm_policy_data *policy,
1499 enum ldlm_mode mode,
1500 enum ldlm_cancel_flags flags, void *opaque);
1501 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1502 int count, enum ldlm_cancel_flags flags);
1503 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1504 struct list_head *cancels,
1505 union ldlm_policy_data *policy,
1506 enum ldlm_mode mode, __u64 lock_flags,
1507 enum ldlm_cancel_flags cancel_flags,
1509 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1510 enum ldlm_cancel_flags flags);
1511 int ldlm_cli_cancel_list(struct list_head *head, int count,
1512 struct ptlrpc_request *req,
1513 enum ldlm_cancel_flags flags);
1514 /** @} ldlm_cli_api */
1517 /* This has to be here because recursive inclusion sucks. */
1518 int intent_disposition(struct ldlm_reply *rep, int flag);
1519 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1522 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1523 * than one lock_res is dead-lock safe.
1525 enum lock_res_type {
1530 /** Lock resource. */
1531 static inline void lock_res(struct ldlm_resource *res)
1533 spin_lock(&res->lr_lock);
1536 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1537 static inline void lock_res_nested(struct ldlm_resource *res,
1538 enum lock_res_type mode)
1540 spin_lock_nested(&res->lr_lock, mode);
1543 /** Unlock resource. */
1544 static inline void unlock_res(struct ldlm_resource *res)
1546 spin_unlock(&res->lr_lock);
1549 /** Check if resource is already locked, assert if not. */
1550 static inline void check_res_locked(struct ldlm_resource *res)
1552 assert_spin_locked(&res->lr_lock);
1555 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1556 void unlock_res_and_lock(struct ldlm_lock *lock);
1559 /** \defgroup ldlm_pools Various LDLM pool related functions
1560 * There are not used outside of ldlm.
1563 int ldlm_pools_recalc(enum ldlm_side client);
1564 int ldlm_pools_init(void);
1565 void ldlm_pools_fini(void);
1567 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1568 int idx, enum ldlm_side client);
1569 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1570 void ldlm_pool_fini(struct ldlm_pool *pl);
1571 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1572 int ldlm_pool_recalc(struct ldlm_pool *pl);
1573 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1574 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1575 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1576 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1577 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1578 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1579 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1580 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1581 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);