4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2010, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 /** \defgroup LDLM Lustre Distributed Lock Manager
35 * Lustre DLM is based on VAX DLM.
36 * Its two main roles are:
37 * - To provide locking assuring consistency of data on all Lustre nodes.
38 * - To allow clients to cache state protected by a lock by holding the
39 * lock until a conflicting lock is requested or it is expired by the LRU.
44 #ifndef _LUSTRE_DLM_H__
45 #define _LUSTRE_DLM_H__
47 #include <lustre_lib.h>
48 #include <lustre_net.h>
49 #include <lustre_import.h>
50 #include <lustre_handles.h>
51 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
54 #include "lustre_dlm_flags.h"
59 extern struct kset *ldlm_ns_kset;
60 extern struct kset *ldlm_svc_kset;
62 #define OBD_LDLM_DEVICENAME "ldlm"
64 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
65 #define LDLM_DEFAULT_MAX_ALIVE 3900 /* 3900 seconds ~65 min */
66 #define LDLM_CTIME_AGE_LIMIT (10)
67 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
70 * LDLM non-error return states
74 ELDLM_LOCK_MATCHED = 1,
76 ELDLM_LOCK_CHANGED = 300,
77 ELDLM_LOCK_ABORTED = 301,
78 ELDLM_LOCK_REPLACED = 302,
79 ELDLM_NO_LOCK_DATA = 303,
80 ELDLM_LOCK_WOULDBLOCK = 304,
82 ELDLM_NAMESPACE_EXISTS = 400,
83 ELDLM_BAD_NAMESPACE = 401,
87 * LDLM namespace type.
88 * The "client" type is actually an indication that this is a narrow local view
89 * into complete namespace on the server. Such namespaces cannot make any
90 * decisions about lack of conflicts or do any autonomous lock granting without
91 * first speaking to a server.
94 LDLM_NAMESPACE_SERVER = 0x01,
95 LDLM_NAMESPACE_CLIENT = 0x02
99 * The blocking callback is overloaded to perform two functions. These flags
100 * indicate which operation should be performed.
102 #define LDLM_CB_BLOCKING 1
103 #define LDLM_CB_CANCELING 2
106 * \name Lock Compatibility Matrix.
108 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
109 * Lock types are described in their respective implementation files:
110 * ldlm_{extent,flock,inodebits,plain}.c.
112 * There are six lock modes along with a compatibility matrix to indicate if
113 * two locks are compatible.
115 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
117 * - PW: Protective Write (normal write) mode. When a client requests a write
118 * lock from an OST, a lock with PW mode will be issued.
119 * - PR: Protective Read (normal read) mode. When a client requests a read from
120 * an OST, a lock with PR mode will be issued. Also, if the client opens a
121 * file for execution, it is granted a lock with PR mode.
122 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
123 * requests a write lock during a file open operation.
124 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
125 * an inodebit lock with the CR mode on the intermediate path component.
139 #define LCK_COMPAT_EX LCK_NL
140 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
141 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
142 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
143 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
144 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
145 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
146 #define LCK_COMPAT_COS (LCK_COS)
147 /** @} Lock Compatibility Matrix */
149 extern enum ldlm_mode lck_compat_array[];
151 static inline void lockmode_verify(enum ldlm_mode mode)
153 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
156 static inline int lockmode_compat(enum ldlm_mode exist_mode,
157 enum ldlm_mode new_mode)
159 return lck_compat_array[exist_mode] & new_mode;
164 * cluster name spaces
168 #define DLM_OST_NAMESPACE 1
169 #define DLM_MDS_NAMESPACE 2
172 - do we just separate this by security domains and use a prefix for
173 multiple namespaces in the same domain?
178 * Locking rules for LDLM:
183 * waiting_locks_spinlock
198 struct ldlm_resource;
199 struct ldlm_namespace;
202 * Operations on LDLM pools.
203 * LDLM pool is a pool of locks in the namespace without any implicitly
205 * Locks in the pool are organized in LRU.
206 * Local memory pressure or server instructions (e.g. mempressure on server)
207 * can trigger freeing of locks from the pool
209 struct ldlm_pool_ops {
210 /** Recalculate pool \a pl usage */
211 int (*po_recalc)(struct ldlm_pool *pl);
212 /** Cancel at least \a nr locks from pool \a pl */
213 int (*po_shrink)(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
214 int (*po_setup)(struct ldlm_pool *pl, int limit);
217 /** One second for pools thread check interval. Each pool has own period. */
218 #define LDLM_POOLS_THREAD_PERIOD (1)
220 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
221 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
223 /** Default recalc period for server side pools in sec. */
224 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
226 /** Default recalc period for client side pools in sec. */
227 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
230 * LDLM pool structure to track granted locks.
231 * For purposes of determining when to release locks on e.g. memory pressure.
232 * This feature is commonly referred to as lru_resize.
235 /** Pool proc directory. */
236 struct proc_dir_entry *pl_proc_dir;
237 /** Pool name, must be long enough to hold compound proc entry name. */
239 /** Lock for protecting SLV/CLV updates. */
241 /** Number of allowed locks in in pool, both, client and server side. */
243 /** Number of granted locks in */
245 /** Grant rate per T. */
246 atomic_t pl_grant_rate;
247 /** Cancel rate per T. */
248 atomic_t pl_cancel_rate;
249 /** Server lock volume (SLV). Protected by pl_lock. */
250 __u64 pl_server_lock_volume;
251 /** Current biggest client lock volume. Protected by pl_lock. */
252 __u64 pl_client_lock_volume;
253 /** Lock volume factor. SLV on client is calculated as following:
254 * server_slv * lock_volume_factor. */
255 atomic_t pl_lock_volume_factor;
256 /** Time when last SLV from server was obtained. */
257 time64_t pl_recalc_time;
258 /** Recalculation period for pool. */
259 time64_t pl_recalc_period;
260 /** Recalculation and shrink operations. */
261 struct ldlm_pool_ops *pl_ops;
262 /** Number of planned locks for next period. */
264 /** Pool statistics. */
265 struct lprocfs_stats *pl_stats;
268 struct kobject pl_kobj;
269 struct completion pl_kobj_unregister;
272 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
273 void *req_cookie, enum ldlm_mode mode,
274 __u64 flags, void *data);
276 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
280 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
281 * be associated with an LDLM lock and transferred from client to server and
284 * Currently LVBs are used by:
285 * - OSC-OST code to maintain current object size/times
286 * - layout lock code to return the layout when the layout lock is granted
288 * To ensure delayed LVB initialization, it is highly recommended to use the set
289 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
291 struct ldlm_valblock_ops {
292 int (*lvbo_init)(struct ldlm_resource *res);
293 int (*lvbo_update)(struct ldlm_resource *res, struct ldlm_lock *lock,
294 struct ptlrpc_request *r, int increase);
295 int (*lvbo_free)(struct ldlm_resource *res);
296 /* Return size of lvb data appropriate RPC size can be reserved */
297 int (*lvbo_size)(struct ldlm_lock *lock);
298 /* Called to fill in lvb data to RPC buffer @buf */
299 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
303 * LDLM pools related, type of lock pool in the namespace.
304 * Greedy means release cached locks aggressively
307 LDLM_NAMESPACE_GREEDY = 1 << 0,
308 LDLM_NAMESPACE_MODEST = 1 << 1
312 * Default values for the "max_nolock_size", "contention_time" and
313 * "contended_locks" namespace tunables.
315 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
316 #define NS_DEFAULT_CONTENTION_SECONDS 2
317 #define NS_DEFAULT_CONTENDED_LOCKS 32
319 struct ldlm_ns_bucket {
320 /** back pointer to namespace */
321 struct ldlm_namespace *nsb_namespace;
323 * Estimated lock callback time. Used by adaptive timeout code to
324 * avoid spurious client evictions due to unresponsiveness when in
325 * fact the network or overall system load is at fault
327 struct adaptive_timeout nsb_at_estimate;
329 * Which res in the bucket should we start with the reclaim.
331 int nsb_reclaim_start;
335 /** LDLM namespace lock stats */
341 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
342 LDLM_NS_TYPE_MDC, /**< MDC namespace */
343 LDLM_NS_TYPE_MDT, /**< MDT namespace */
344 LDLM_NS_TYPE_OSC, /**< OSC namespace */
345 LDLM_NS_TYPE_OST, /**< OST namespace */
346 LDLM_NS_TYPE_MGC, /**< MGC namespace */
347 LDLM_NS_TYPE_MGT, /**< MGT namespace */
353 * Namespace serves to contain locks related to a particular service.
354 * There are two kinds of namespaces:
355 * - Server namespace has knowledge of all locks and is therefore authoritative
356 * to make decisions like what locks could be granted and what conflicts
357 * exist during new lock enqueue.
358 * - Client namespace only has limited knowledge about locks in the namespace,
359 * only seeing locks held by the client.
361 * Every Lustre service has one server namespace present on the server serving
362 * that service. Every client connected to the service has a client namespace
364 * Every lock obtained by client in that namespace is actually represented by
365 * two in-memory locks. One on the server and one on the client. The locks are
366 * linked by a special cookie by which one node can tell to the other which lock
367 * it actually means during communications. Such locks are called remote locks.
368 * The locks held by server only without any reference to a client are called
371 struct ldlm_namespace {
372 /** Backward link to OBD, required for LDLM pool to store new SLV. */
373 struct obd_device *ns_obd;
375 /** Flag indicating if namespace is on client instead of server */
376 enum ldlm_side ns_client;
378 /** Resource hash table for namespace. */
379 struct cfs_hash *ns_rs_hash;
384 /** big refcount (by bucket) */
388 * Namespace connect flags supported by server (may be changed via
389 * /proc, LRU resize may be disabled/enabled).
391 __u64 ns_connect_flags;
393 /** Client side original connect flags supported by server. */
394 __u64 ns_orig_connect_flags;
396 /* namespace proc dir entry */
397 struct proc_dir_entry *ns_proc_dir_entry;
400 * Position in global namespace list linking all namespaces on
403 struct list_head ns_list_chain;
406 * List of unused locks for this namespace. This list is also called
408 * Unused locks are locks with zero reader/writer reference counts.
409 * This list is only used on clients for lock caching purposes.
410 * When we want to release some locks voluntarily or if server wants
411 * us to release some locks due to e.g. memory pressure, we take locks
412 * to release from the head of this list.
413 * Locks are linked via l_lru field in \see struct ldlm_lock.
415 struct list_head ns_unused_list;
416 /** Number of locks in the LRU list above */
420 * Maximum number of locks permitted in the LRU. If 0, means locks
421 * are managed by pools and there is no preset limit, rather it is all
422 * controlled by available memory on this client and on server.
424 unsigned int ns_max_unused;
426 /** Maximum allowed age (last used time) for locks in the LRU */
430 * Server only: number of times we evicted clients due to lack of reply
433 unsigned int ns_timeouts;
435 * Number of seconds since the file change time after which the
436 * MDT will return an UPDATE lock along with a LOOKUP lock.
437 * This allows the client to start caching negative dentries
438 * for a directory and may save an RPC for a later stat.
440 time64_t ns_ctime_age_limit;
443 * Used to rate-limit ldlm_namespace_dump calls.
444 * \see ldlm_namespace_dump. Increased by 10 seconds every time
447 time64_t ns_next_dump;
449 /** "policy" function that does actual lock conflict determination */
450 ldlm_res_policy ns_policy;
453 * LVB operations for this namespace.
454 * \see struct ldlm_valblock_ops
456 struct ldlm_valblock_ops *ns_lvbo;
459 * Used by filter code to store pointer to OBD of the service.
460 * Should be dropped in favor of \a ns_obd
465 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
466 * a resource is removed.
468 wait_queue_head_t ns_waitq;
469 /** LDLM pool structure for this namespace */
470 struct ldlm_pool ns_pool;
471 /** Definition of how eagerly unused locks will be released from LRU */
472 enum ldlm_appetite ns_appetite;
475 * If more than \a ns_contended_locks are found, the resource is
476 * considered to be contended. Lock enqueues might specify that no
477 * contended locks should be granted
479 unsigned ns_contended_locks;
482 * The resources in this namespace remember contended state during
483 * \a ns_contention_time, in seconds.
485 time64_t ns_contention_time;
488 * Limit size of contended extent locks, in bytes.
489 * If extended lock is requested for more then this many bytes and
490 * caller instructs us not to grant contended locks, we would disregard
493 unsigned ns_max_nolock_size;
495 /** Limit of parallel AST RPC count. */
496 unsigned ns_max_parallel_ast;
499 * Callback to check if a lock is good to be canceled by ELC or
502 ldlm_cancel_cbt ns_cancel;
504 /** LDLM lock stats */
505 struct lprocfs_stats *ns_stats;
508 * Flag to indicate namespace is being freed. Used to determine if
509 * recalculation of LDLM pool statistics should be skipped.
511 unsigned ns_stopping:1;
514 * Which bucket should we start with the lock reclaim.
516 int ns_reclaim_start;
518 struct kobject ns_kobj; /* sysfs object */
519 struct completion ns_kobj_unregister;
523 * Returns 1 if namespace \a ns is a client namespace.
525 static inline int ns_is_client(struct ldlm_namespace *ns)
528 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
529 LDLM_NAMESPACE_SERVER)));
530 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
531 ns->ns_client == LDLM_NAMESPACE_SERVER);
532 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
536 * Returns 1 if namespace \a ns is a server namespace.
538 static inline int ns_is_server(struct ldlm_namespace *ns)
541 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
542 LDLM_NAMESPACE_SERVER)));
543 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
544 ns->ns_client == LDLM_NAMESPACE_SERVER);
545 return ns->ns_client == LDLM_NAMESPACE_SERVER;
549 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
551 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
554 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
558 * Returns 1 if this namespace supports lru_resize.
560 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
563 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
566 static inline void ns_register_cancel(struct ldlm_namespace *ns,
575 /** Type for blocking callback function of a lock. */
576 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
577 struct ldlm_lock_desc *new, void *data,
579 /** Type for completion callback function of a lock. */
580 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
582 /** Type for glimpse callback function of a lock. */
583 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
585 /** Work list for sending GL ASTs to multiple locks. */
586 struct ldlm_glimpse_work {
587 struct ldlm_lock *gl_lock; /* lock to glimpse */
588 struct list_head gl_list; /* linkage to other gl work structs */
589 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
590 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
591 * glimpse callback request */
592 ptlrpc_interpterer_t gl_interpret_reply;
593 void *gl_interpret_data;
596 struct ldlm_cb_set_arg {
597 struct ptlrpc_request_set *set;
598 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
600 struct list_head *list;
601 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
602 ptlrpc_interpterer_t gl_interpret_reply;
603 void *gl_interpret_data;
606 struct ldlm_cb_async_args {
607 struct ldlm_cb_set_arg *ca_set_arg;
608 struct ldlm_lock *ca_lock;
611 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
612 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
614 /** Interval node data for each LDLM_EXTENT lock. */
615 struct ldlm_interval {
616 struct interval_node li_node; /* node for tree management */
617 struct list_head li_group; /* the locks which have the same
618 * policy - group of the policy */
620 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
623 * Interval tree for extent locks.
624 * The interval tree must be accessed under the resource lock.
625 * Interval trees are used for granted extent locks to speed up conflicts
626 * lookup. See ldlm/interval_tree.c for more details.
628 struct ldlm_interval_tree {
631 enum ldlm_mode lit_mode; /* lock mode */
632 struct interval_node *lit_root; /* actual ldlm_interval */
635 /** Whether to track references to exports by LDLM locks. */
636 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
639 enum ldlm_cancel_flags {
640 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
641 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
642 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
649 __u64 blocking_owner;
650 struct obd_export *blocking_export;
651 atomic_t blocking_refs;
655 union ldlm_policy_data {
656 struct ldlm_extent l_extent;
657 struct ldlm_flock l_flock;
658 struct ldlm_inodebits l_inodebits;
661 void ldlm_convert_policy_to_wire(enum ldlm_type type,
662 const union ldlm_policy_data *lpolicy,
663 union ldlm_wire_policy_data *wpolicy);
664 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
665 const union ldlm_wire_policy_data *wpolicy,
666 union ldlm_policy_data *lpolicy);
676 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
678 #define LDLM_GID_ANY ((__u64)-1)
681 * LDLM lock structure
683 * Represents a single LDLM lock and its state in memory. Each lock is
684 * associated with a single ldlm_resource, the object which is being
685 * locked. There may be multiple ldlm_locks on a single resource,
686 * depending on the lock type and whether the locks are conflicting or
692 * When remote side wants to tell us about a lock, they address
693 * it by this opaque handle. The handle does not hold a
694 * reference on the ldlm_lock, so it can be safely passed to
695 * other threads or nodes. When the lock needs to be accessed
696 * from the handle, it is looked up again in the lock table, and
697 * may no longer exist.
699 * Must be first in the structure.
701 struct portals_handle l_handle;
703 * Lock reference count.
704 * This is how many users have pointers to actual structure, so that
705 * we do not accidentally free lock structure that is in use.
709 * Internal spinlock protects l_resource. We should hold this lock
710 * first before taking res_lock.
714 * Pointer to actual resource this lock is in.
715 * ldlm_lock_change_resource() can change this.
717 struct ldlm_resource *l_resource;
719 * List item for client side LRU list.
720 * Protected by ns_lock in struct ldlm_namespace.
722 struct list_head l_lru;
724 * Linkage to resource's lock queues according to current lock state.
725 * (could be granted or waiting)
726 * Protected by lr_lock in struct ldlm_resource.
728 struct list_head l_res_link;
730 * Tree node for ldlm_extent.
732 struct ldlm_interval *l_tree_node;
734 * Per export hash of locks.
735 * Protected by per-bucket exp->exp_lock_hash locks.
737 struct hlist_node l_exp_hash;
739 * Per export hash of flock locks.
740 * Protected by per-bucket exp->exp_flock_hash locks.
742 struct hlist_node l_exp_flock_hash;
745 * Protected by lr_lock.
747 enum ldlm_mode l_req_mode;
749 * Granted mode, also protected by lr_lock.
751 enum ldlm_mode l_granted_mode;
752 /** Lock completion handler pointer. Called when lock is granted. */
753 ldlm_completion_callback l_completion_ast;
755 * Lock blocking AST handler pointer.
756 * It plays two roles:
757 * - as a notification of an attempt to queue a conflicting lock (once)
758 * - as a notification when the lock is being cancelled.
760 * As such it's typically called twice: once for the initial conflict
761 * and then once more when the last user went away and the lock is
762 * cancelled (could happen recursively).
764 ldlm_blocking_callback l_blocking_ast;
766 * Lock glimpse handler.
767 * Glimpse handler is used to obtain LVB updates from a client by
770 ldlm_glimpse_callback l_glimpse_ast;
774 * This is a pointer to actual client export for locks that were granted
775 * to clients. Used server-side.
777 struct obd_export *l_export;
779 * Lock connection export.
780 * Pointer to server export on a client.
782 struct obd_export *l_conn_export;
785 * Remote lock handle.
786 * If the lock is remote, this is the handle of the other side lock
789 struct lustre_handle l_remote_handle;
792 * Representation of private data specific for a lock type.
793 * Examples are: extent range for extent lock or bitmask for ibits locks
795 union ldlm_policy_data l_policy_data;
798 * Lock state flags. Protected by lr_lock.
799 * \see lustre_dlm_flags.h where the bits are defined.
804 * Lock r/w usage counters.
805 * Protected by lr_lock.
810 * If the lock is granted, a process sleeps on this waitq to learn when
811 * it's no longer in use. If the lock is not granted, a process sleeps
812 * on this waitq to learn when it becomes granted.
814 wait_queue_head_t l_waitq;
817 * Seconds. It will be updated if there is any activity related to
818 * the lock, e.g. enqueue the lock or send blocking AST.
820 time64_t l_last_activity;
823 * Time, in nanoseconds, last used by e.g. being matched by lock match.
827 /** Originally requested extent for the extent lock. */
828 struct ldlm_extent l_req_extent;
831 * Client-side-only members.
834 enum lvb_type l_lvb_type;
837 * Temporary storage for a LVB received during an enqueue operation.
838 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
843 /** Private storage for lock user. Opaque to LDLM. */
845 /* separate ost_lvb used mostly by Data-on-MDT for now.
846 * It is introduced to don't mix with layout lock data. */
847 struct ost_lvb l_ost_lvb;
849 * Server-side-only members.
853 * Connection cookie for the client originating the operation.
854 * Used by Commit on Share (COS) code. Currently only used for
855 * inodebits locks on MDS.
857 __u64 l_client_cookie;
860 * List item for locks waiting for cancellation from clients.
861 * The lists this could be linked into are:
862 * waiting_locks_list (protected by waiting_locks_spinlock),
863 * then if the lock timed out, it is moved to
864 * expired_lock_list for further processing.
866 struct list_head l_pending_chain;
869 * Set when lock is sent a blocking AST. Time in seconds when timeout
870 * is reached and client holding this lock could be evicted.
871 * This timeout could be further extended by e.g. certain IO activity
873 * \see ost_rw_prolong_locks
875 time64_t l_callback_timeout;
877 /** Local PID of process which created this lock. */
881 * Number of times blocking AST was sent for this lock.
882 * This is for debugging. Valid values are 0 and 1, if there is an
883 * attempt to send blocking AST more than once, an assertion would be
884 * hit. \see ldlm_work_bl_ast_lock
887 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
888 struct list_head l_bl_ast;
889 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
890 struct list_head l_cp_ast;
891 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
892 struct list_head l_rk_ast;
895 * Pointer to a conflicting lock that caused blocking AST to be sent
898 struct ldlm_lock *l_blocking_lock;
901 * Protected by lr_lock, linkages to "skip lists".
902 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
904 struct list_head l_sl_mode;
905 struct list_head l_sl_policy;
907 /** Reference tracking structure to debug leaked locks. */
908 struct lu_ref l_reference;
909 #if LUSTRE_TRACKS_LOCK_EXP_REFS
910 /* Debugging stuff for bug 20498, for tracking export references. */
911 /** number of export references taken */
913 /** link all locks referencing one export */
914 struct list_head l_exp_refs_link;
915 /** referenced export object */
916 struct obd_export *l_exp_refs_target;
919 * export blocking dlm lock list, protected by
920 * l_export->exp_bl_list_lock.
921 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
922 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
924 struct list_head l_exp_list;
927 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
928 #define l_transno l_client_cookie
930 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
931 * which is for server. */
932 #define l_slc_link l_rk_ast
935 * LDLM resource description.
936 * Basically, resource is a representation for a single object.
937 * Object has a name which is currently 4 64-bit integers. LDLM user is
938 * responsible for creation of a mapping between objects it wants to be
939 * protected and resource names.
941 * A resource can only hold locks of a single lock type, though there may be
942 * multiple ldlm_locks on a single resource, depending on the lock type and
943 * whether the locks are conflicting or not.
945 struct ldlm_resource {
946 struct ldlm_ns_bucket *lr_ns_bucket;
949 * List item for list in namespace hash.
950 * protected by ns_lock
952 struct hlist_node lr_hash;
954 /** Reference count for this resource */
955 atomic_t lr_refcount;
957 /** Spinlock to protect locks under this resource. */
961 * protected by lr_lock
963 /** List of locks in granted state */
964 struct list_head lr_granted;
966 * List of locks that could not be granted due to conflicts and
967 * that are waiting for conflicts to go away */
968 struct list_head lr_waiting;
972 struct ldlm_res_id lr_name;
975 * Interval trees (only for extent locks) for all modes of this resource
977 struct ldlm_interval_tree *lr_itree;
981 * When the resource was considered as contended,
982 * used only on server side.
984 time64_t lr_contention_time;
986 * Associated inode, used only on client side.
988 struct inode *lr_lvb_inode;
991 /** Type of locks this resource can hold. Only one type per resource. */
992 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
995 * Server-side-only lock value block elements.
996 * To serialize lvbo_init.
999 struct mutex lr_lvb_mutex;
1000 /** protected by lr_lock */
1002 /** is lvb initialized ? */
1003 bool lr_lvb_initialized;
1005 /** List of references to this resource. For debugging. */
1006 struct lu_ref lr_reference;
1009 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1011 return lock->l_resource->lr_type == LDLM_IBITS &&
1012 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1015 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1017 return lock->l_resource->lr_type == LDLM_IBITS &&
1018 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1021 static inline char *
1022 ldlm_ns_name(struct ldlm_namespace *ns)
1024 return ns->ns_rs_hash->hs_name;
1027 static inline struct ldlm_namespace *
1028 ldlm_res_to_ns(struct ldlm_resource *res)
1030 return res->lr_ns_bucket->nsb_namespace;
1033 static inline struct ldlm_namespace *
1034 ldlm_lock_to_ns(struct ldlm_lock *lock)
1036 return ldlm_res_to_ns(lock->l_resource);
1039 static inline char *
1040 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1042 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1045 static inline struct adaptive_timeout *
1046 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1048 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1051 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1053 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1056 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1057 res->lr_lvb_initialized)
1060 mutex_lock(&res->lr_lvb_mutex);
1061 /* Did we lose the race? */
1062 if (res->lr_lvb_initialized) {
1063 mutex_unlock(&res->lr_lvb_mutex);
1066 rc = ns->ns_lvbo->lvbo_init(res);
1068 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1070 if (res->lr_lvb_data != NULL) {
1071 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1072 res->lr_lvb_data = NULL;
1074 res->lr_lvb_len = rc;
1076 res->lr_lvb_initialized = true;
1078 mutex_unlock(&res->lr_lvb_mutex);
1082 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1084 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1086 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1087 return ns->ns_lvbo->lvbo_size(lock);
1092 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1094 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1097 if (ns->ns_lvbo != NULL) {
1098 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1099 /* init lvb now if not already */
1100 rc = ldlm_lvbo_init(lock->l_resource);
1102 CERROR("lock %p: delayed lvb init failed (rc %d)",
1106 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1111 struct ldlm_ast_work {
1112 struct ldlm_lock *w_lock;
1114 struct ldlm_lock_desc w_desc;
1115 struct list_head w_list;
1122 * Common ldlm_enqueue parameters
1124 struct ldlm_enqueue_info {
1125 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1126 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1127 void *ei_cb_bl; /** blocking lock callback */
1128 void *ei_cb_local_bl; /** blocking local lock callback */
1129 void *ei_cb_cp; /** lock completion callback */
1130 void *ei_cb_gl; /** lock glimpse callback */
1131 void *ei_cbdata; /** Data to be passed into callbacks. */
1132 void *ei_namespace; /** lock namespace **/
1133 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1136 #define ei_res_id ei_cb_gl
1138 extern struct obd_ops ldlm_obd_ops;
1140 extern char *ldlm_lockname[];
1141 extern char *ldlm_typename[];
1142 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1145 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1146 * For the cases where we do not have actual lock to print along
1147 * with a debugging message that is ldlm-related
1149 #define LDLM_DEBUG_NOLOCK(format, a...) \
1150 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1153 * Support function for lock information printing into debug logs.
1157 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1158 CFS_CHECK_STACK(msgdata, mask, cdls); \
1160 if (((mask) & D_CANTMASK) != 0 || \
1161 ((libcfs_debug & (mask)) != 0 && \
1162 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1163 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1166 void _ldlm_lock_debug(struct ldlm_lock *lock,
1167 struct libcfs_debug_msg_data *data,
1168 const char *fmt, ...)
1169 __attribute__ ((format (printf, 3, 4)));
1172 * Rate-limited version of lock printing function.
1174 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1175 static struct cfs_debug_limit_state _ldlm_cdls; \
1176 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1177 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1180 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1181 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1183 /** Non-rate-limited lock printing function for debugging purposes. */
1184 #define LDLM_DEBUG(lock, fmt, a...) do { \
1185 if (likely(lock != NULL)) { \
1186 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1187 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1188 "### " fmt , ##a); \
1190 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1193 #else /* !LIBCFS_DEBUG */
1194 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1195 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1196 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1200 * Three intentions can be used for the policy functions in
1201 * ldlm_processing_policy.
1203 * LDLM_PROCESS_RESCAN:
1205 * It's used when policy functions are called from ldlm_reprocess_queue() to
1206 * reprocess the wait list and try to grant locks, blocking ASTs
1207 * have already been sent in this situation, completion ASTs need be sent for
1208 * the locks being granted.
1210 * LDLM_PROCESS_ENQUEUE:
1212 * It's used when policy functions are called from ldlm_lock_enqueue() to
1213 * process the wait list for handling an enqueue request, blocking
1214 * ASTs have not been sent yet, so list of conflicting locks would be
1215 * collected and ASTs sent.
1217 * LDLM_PROCESS_RECOVERY:
1219 * It's used when policy functions are called from ldlm_reprocess_queue() to
1220 * reprocess the wait list when recovery done. In case of blocking
1221 * ASTs are lost before recovery, it needs not only to grant locks if
1222 * available, but also send blocking ASTs to the locks doesn't have AST sent
1223 * flag. Completion ASTs need be sent for the locks being granted.
1225 enum ldlm_process_intention {
1226 LDLM_PROCESS_RESCAN = 0,
1227 LDLM_PROCESS_ENQUEUE = 1,
1228 LDLM_PROCESS_RECOVERY = 2,
1231 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1232 enum ldlm_process_intention intention,
1233 enum ldlm_error *err,
1234 struct list_head *work_list);
1237 * Return values for lock iterators.
1238 * Also used during deciding of lock grants and cancellations.
1240 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1241 #define LDLM_ITER_STOP 2 /* stop iterating */
1243 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1244 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1246 /** \defgroup ldlm_iterator Lock iterators
1248 * LDLM provides for a way to iterate through every lock on a resource or
1249 * namespace or every resource in a namespace.
1251 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1253 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1255 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1256 ldlm_iterator_t iter, void *data);
1257 /** @} ldlm_iterator */
1259 int ldlm_replay_locks(struct obd_import *imp);
1262 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1265 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1267 struct ldlm_prolong_args {
1268 struct obd_export *lpa_export;
1269 struct ldlm_res_id lpa_resid;
1270 struct ldlm_extent lpa_extent;
1271 enum ldlm_mode lpa_mode;
1272 time64_t lpa_timeout;
1276 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1277 struct ldlm_prolong_args *arg);
1278 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1280 struct ldlm_callback_suite {
1281 ldlm_completion_callback lcs_completion;
1282 ldlm_blocking_callback lcs_blocking;
1283 ldlm_glimpse_callback lcs_glimpse;
1287 #ifdef HAVE_SERVER_SUPPORT
1288 /** \defgroup ldlm_srv_ast Server AST handlers
1289 * These are AST handlers used by server code.
1290 * Their property is that they are just preparing RPCs to be sent to clients.
1293 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1294 void *data, int flag);
1295 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1296 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1297 int ldlm_glimpse_locks(struct ldlm_resource *res,
1298 struct list_head *gl_work_list);
1299 /** @} ldlm_srv_ast */
1301 /** \defgroup ldlm_handlers Server LDLM handlers
1302 * These are handler functions that should be called by "frontends" such as
1303 * MDT or OST to pass through LDLM requests to LDLM for handling
1306 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1307 const struct ldlm_request *dlm_req,
1308 const struct ldlm_callback_suite *cbs);
1309 int ldlm_handle_convert0(struct ptlrpc_request *req,
1310 const struct ldlm_request *dlm_req);
1311 int ldlm_handle_cancel(struct ptlrpc_request *req);
1312 int ldlm_request_cancel(struct ptlrpc_request *req,
1313 const struct ldlm_request *dlm_req,
1314 int first, enum lustre_at_flags flags);
1315 /** @} ldlm_handlers */
1317 void ldlm_revoke_export_locks(struct obd_export *exp);
1318 time64_t ldlm_bl_timeout(struct ldlm_lock *lock);
1320 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1321 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, time64_t timeout);
1322 int ldlm_get_ref(void);
1323 void ldlm_put_ref(void);
1324 int ldlm_init_export(struct obd_export *exp);
1325 void ldlm_destroy_export(struct obd_export *exp);
1326 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1329 #ifdef HAVE_SERVER_SUPPORT
1330 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1332 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1333 void ldlm_lock2handle(const struct ldlm_lock *lock,
1334 struct lustre_handle *lockh);
1335 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1336 void ldlm_cancel_callback(struct ldlm_lock *);
1337 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1338 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1341 * Obtain a lock reference by its handle.
1343 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1345 return __ldlm_handle2lock(h, 0);
1348 #define LDLM_LOCK_REF_DEL(lock) \
1349 lu_ref_del(&lock->l_reference, "handle", current)
1351 static inline struct ldlm_lock *
1352 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1354 struct ldlm_lock *lock;
1356 lock = __ldlm_handle2lock(h, flags);
1358 LDLM_LOCK_REF_DEL(lock);
1363 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1364 * data from request \a r
1366 static inline int ldlm_lvbo_update(struct ldlm_resource *res,
1367 struct ldlm_lock *lock,
1368 struct ptlrpc_request *req, int increase)
1370 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1373 /* delayed lvb init may be required */
1374 rc = ldlm_lvbo_init(res);
1376 CERROR("delayed lvb init failed (rc %d)\n", rc);
1380 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1381 return ns->ns_lvbo->lvbo_update(res, lock, req, increase);
1386 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1387 struct ptlrpc_request *req, int increase)
1389 return ldlm_lvbo_update(res, NULL, req, increase);
1392 int ldlm_error2errno(enum ldlm_error error);
1393 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1394 * confuses user-space. */
1395 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1396 void ldlm_dump_export_locks(struct obd_export *exp);
1400 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1401 * __ldlm_handle2lock().
1403 #define LDLM_LOCK_PUT(lock) \
1405 LDLM_LOCK_REF_DEL(lock); \
1406 /*LDLM_DEBUG((lock), "put");*/ \
1407 ldlm_lock_put(lock); \
1411 * Release a lock reference obtained by some other means (see
1414 #define LDLM_LOCK_RELEASE(lock) \
1416 /*LDLM_DEBUG((lock), "put");*/ \
1417 ldlm_lock_put(lock); \
1420 #define LDLM_LOCK_GET(lock) \
1422 ldlm_lock_get(lock); \
1423 /*LDLM_DEBUG((lock), "get");*/ \
1427 #define ldlm_lock_list_put(head, member, count) \
1429 struct ldlm_lock *_lock, *_next; \
1431 list_for_each_entry_safe(_lock, _next, head, member) { \
1434 list_del_init(&_lock->member); \
1435 LDLM_LOCK_RELEASE(_lock); \
1440 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1441 void ldlm_lock_put(struct ldlm_lock *lock);
1442 void ldlm_lock_destroy(struct ldlm_lock *lock);
1443 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1444 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1445 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1446 enum ldlm_mode mode);
1447 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1448 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1449 enum ldlm_mode mode);
1450 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1451 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1452 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1453 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1454 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1455 const struct ldlm_res_id *, enum ldlm_type type,
1456 union ldlm_policy_data *, enum ldlm_mode mode,
1457 struct lustre_handle *, int unref);
1458 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1460 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1461 void ldlm_lock_cancel(struct ldlm_lock *lock);
1462 void ldlm_reprocess_all(struct ldlm_resource *res);
1463 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1464 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1465 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1468 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1469 enum ldlm_side client,
1470 enum ldlm_appetite apt,
1471 enum ldlm_ns_type ns_type);
1472 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1473 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1474 struct obd_import *imp,
1476 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1477 void ldlm_namespace_free(struct ldlm_namespace *ns,
1478 struct obd_import *imp, int force);
1479 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1480 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1481 enum ldlm_side client);
1482 void ldlm_namespace_get(struct ldlm_namespace *ns);
1483 void ldlm_namespace_put(struct ldlm_namespace *ns);
1484 int ldlm_proc_setup(void);
1485 #ifdef CONFIG_PROC_FS
1486 void ldlm_proc_cleanup(void);
1488 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1489 struct lprocfs_stats *srv_stats)
1491 int lock_type = 0, op = 0;
1493 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1495 switch (lock_type) {
1497 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1500 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1503 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1506 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1514 lprocfs_counter_incr(srv_stats, op);
1519 static inline void ldlm_proc_cleanup(void) {}
1520 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1521 struct lprocfs_stats *srv_stats) {}
1524 /* resource.c - internal */
1525 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1526 struct ldlm_resource *parent,
1527 const struct ldlm_res_id *,
1528 enum ldlm_type type, int create);
1529 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1530 int ldlm_resource_putref(struct ldlm_resource *res);
1531 void ldlm_resource_add_lock(struct ldlm_resource *res,
1532 struct list_head *head,
1533 struct ldlm_lock *lock);
1534 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1535 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1536 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1537 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1538 void ldlm_resource_dump(int level, struct ldlm_resource *);
1539 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1540 const struct ldlm_res_id *);
1542 #define LDLM_RESOURCE_ADDREF(res) do { \
1543 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1546 #define LDLM_RESOURCE_DELREF(res) do { \
1547 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1550 /* ldlm_request.c */
1551 int ldlm_expired_completion_wait(void *data);
1552 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1553 * These AST handlers are typically used for server-side local locks and are
1554 * also used by client-side lock handlers to perform minimum level base
1557 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1558 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1559 void *data, int flag);
1560 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1561 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1562 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1563 /** @} ldlm_local_ast */
1565 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1566 * These are typically used by client and server (*_local versions)
1567 * to obtain and release locks.
1569 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1570 struct ldlm_enqueue_info *einfo,
1571 const struct ldlm_res_id *res_id,
1572 union ldlm_policy_data const *policy, __u64 *flags,
1573 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1574 struct lustre_handle *lockh, int async);
1575 int ldlm_prep_enqueue_req(struct obd_export *exp,
1576 struct ptlrpc_request *req,
1577 struct list_head *cancels,
1579 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1580 int version, int opc, int canceloff,
1581 struct list_head *cancels, int count);
1583 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1584 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1585 const struct ldlm_request *dlm_req,
1586 const struct ldlm_callback_suite *cbs);
1587 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1588 enum ldlm_type type, __u8 with_policy,
1589 enum ldlm_mode mode, __u64 *flags, void *lvb,
1591 const struct lustre_handle *lockh, int rc);
1592 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1593 const struct ldlm_res_id *res_id,
1594 enum ldlm_type type, union ldlm_policy_data *policy,
1595 enum ldlm_mode mode, __u64 *flags,
1596 ldlm_blocking_callback blocking,
1597 ldlm_completion_callback completion,
1598 ldlm_glimpse_callback glimpse,
1599 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1600 const __u64 *client_cookie,
1601 struct lustre_handle *lockh);
1602 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1603 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1604 enum ldlm_cancel_flags cancel_flags);
1605 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1606 enum ldlm_cancel_flags flags, void *opaque);
1607 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1608 const struct ldlm_res_id *res_id,
1609 union ldlm_policy_data *policy,
1610 enum ldlm_mode mode,
1611 enum ldlm_cancel_flags flags, void *opaque);
1612 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1613 int count, enum ldlm_cancel_flags flags);
1614 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1615 struct list_head *cancels,
1616 union ldlm_policy_data *policy,
1617 enum ldlm_mode mode, __u64 lock_flags,
1618 enum ldlm_cancel_flags cancel_flags,
1620 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1621 enum ldlm_cancel_flags flags);
1622 int ldlm_cli_cancel_list(struct list_head *head, int count,
1623 struct ptlrpc_request *req,
1624 enum ldlm_cancel_flags flags);
1625 /** @} ldlm_cli_api */
1628 /* This has to be here because recursive inclusion sucks. */
1629 int intent_disposition(struct ldlm_reply *rep, int flag);
1630 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1633 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1634 * than one lock_res is dead-lock safe.
1636 enum lock_res_type {
1641 /** Lock resource. */
1642 static inline void lock_res(struct ldlm_resource *res)
1644 spin_lock(&res->lr_lock);
1647 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1648 static inline void lock_res_nested(struct ldlm_resource *res,
1649 enum lock_res_type mode)
1651 spin_lock_nested(&res->lr_lock, mode);
1654 /** Unlock resource. */
1655 static inline void unlock_res(struct ldlm_resource *res)
1657 spin_unlock(&res->lr_lock);
1660 /** Check if resource is already locked, assert if not. */
1661 static inline void check_res_locked(struct ldlm_resource *res)
1663 assert_spin_locked(&res->lr_lock);
1666 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1667 void unlock_res_and_lock(struct ldlm_lock *lock);
1670 /** \defgroup ldlm_pools Various LDLM pool related functions
1671 * There are not used outside of ldlm.
1674 int ldlm_pools_recalc(enum ldlm_side client);
1675 int ldlm_pools_init(void);
1676 void ldlm_pools_fini(void);
1678 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1679 int idx, enum ldlm_side client);
1680 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1681 void ldlm_pool_fini(struct ldlm_pool *pl);
1682 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1683 int ldlm_pool_recalc(struct ldlm_pool *pl);
1684 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1685 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1686 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1687 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1688 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1689 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1690 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1691 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1692 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1695 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1696 const struct ldlm_extent *ex2)
1698 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1701 /* check if @ex1 contains @ex2 */
1702 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1703 const struct ldlm_extent *ex2)
1705 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1708 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);