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 /* if client lock is unused for that time it can be cancelled if any other
68 * client shows interest in that lock, e.g. glimpse is occured. */
69 #define LDLM_DIRTY_AGE_LIMIT (10)
70 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
73 * LDLM non-error return states
77 ELDLM_LOCK_MATCHED = 1,
79 ELDLM_LOCK_CHANGED = 300,
80 ELDLM_LOCK_ABORTED = 301,
81 ELDLM_LOCK_REPLACED = 302,
82 ELDLM_NO_LOCK_DATA = 303,
83 ELDLM_LOCK_WOULDBLOCK = 304,
85 ELDLM_NAMESPACE_EXISTS = 400,
86 ELDLM_BAD_NAMESPACE = 401,
90 * LDLM namespace type.
91 * The "client" type is actually an indication that this is a narrow local view
92 * into complete namespace on the server. Such namespaces cannot make any
93 * decisions about lack of conflicts or do any autonomous lock granting without
94 * first speaking to a server.
97 LDLM_NAMESPACE_SERVER = 0x01,
98 LDLM_NAMESPACE_CLIENT = 0x02
102 * The blocking callback is overloaded to perform two functions. These flags
103 * indicate which operation should be performed.
105 #define LDLM_CB_BLOCKING 1
106 #define LDLM_CB_CANCELING 2
109 * \name Lock Compatibility Matrix.
111 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
112 * Lock types are described in their respective implementation files:
113 * ldlm_{extent,flock,inodebits,plain}.c.
115 * There are six lock modes along with a compatibility matrix to indicate if
116 * two locks are compatible.
118 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
120 * - PW: Protective Write (normal write) mode. When a client requests a write
121 * lock from an OST, a lock with PW mode will be issued.
122 * - PR: Protective Read (normal read) mode. When a client requests a read from
123 * an OST, a lock with PR mode will be issued. Also, if the client opens a
124 * file for execution, it is granted a lock with PR mode.
125 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
126 * requests a write lock during a file open operation.
127 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
128 * an inodebit lock with the CR mode on the intermediate path component.
142 #define LCK_COMPAT_EX LCK_NL
143 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
144 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
145 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
146 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
147 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
148 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
149 #define LCK_COMPAT_COS (LCK_COS)
150 /** @} Lock Compatibility Matrix */
152 extern enum ldlm_mode lck_compat_array[];
154 static inline void lockmode_verify(enum ldlm_mode mode)
156 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
159 static inline int lockmode_compat(enum ldlm_mode exist_mode,
160 enum ldlm_mode new_mode)
162 return lck_compat_array[exist_mode] & new_mode;
167 * cluster name spaces
171 #define DLM_OST_NAMESPACE 1
172 #define DLM_MDS_NAMESPACE 2
175 - do we just separate this by security domains and use a prefix for
176 multiple namespaces in the same domain?
181 * Locking rules for LDLM:
186 * waiting_locks_spinlock
201 struct ldlm_resource;
202 struct ldlm_namespace;
205 * Operations on LDLM pools.
206 * LDLM pool is a pool of locks in the namespace without any implicitly
208 * Locks in the pool are organized in LRU.
209 * Local memory pressure or server instructions (e.g. mempressure on server)
210 * can trigger freeing of locks from the pool
212 struct ldlm_pool_ops {
213 /** Recalculate pool \a pl usage */
214 int (*po_recalc)(struct ldlm_pool *pl);
215 /** Cancel at least \a nr locks from pool \a pl */
216 int (*po_shrink)(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
217 int (*po_setup)(struct ldlm_pool *pl, int limit);
220 /** One second for pools thread check interval. Each pool has own period. */
221 #define LDLM_POOLS_THREAD_PERIOD (1)
223 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
224 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
226 /** Default recalc period for server side pools in sec. */
227 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
229 /** Default recalc period for client side pools in sec. */
230 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
233 * LDLM pool structure to track granted locks.
234 * For purposes of determining when to release locks on e.g. memory pressure.
235 * This feature is commonly referred to as lru_resize.
238 /** Pool debugfs directory. */
239 struct dentry *pl_debugfs_entry;
240 /** Pool name, must be long enough to hold compound proc entry name. */
242 /** Lock for protecting SLV/CLV updates. */
244 /** Number of allowed locks in in pool, both, client and server side. */
246 /** Number of granted locks in */
248 /** Grant rate per T. */
249 atomic_t pl_grant_rate;
250 /** Cancel rate per T. */
251 atomic_t pl_cancel_rate;
252 /** Server lock volume (SLV). Protected by pl_lock. */
253 __u64 pl_server_lock_volume;
254 /** Current biggest client lock volume. Protected by pl_lock. */
255 __u64 pl_client_lock_volume;
256 /** Lock volume factor. SLV on client is calculated as following:
257 * server_slv * lock_volume_factor. */
258 atomic_t pl_lock_volume_factor;
259 /** Time when last SLV from server was obtained. */
260 time64_t pl_recalc_time;
261 /** Recalculation period for pool. */
262 time64_t pl_recalc_period;
263 /** Recalculation and shrink operations. */
264 struct ldlm_pool_ops *pl_ops;
265 /** Number of planned locks for next period. */
267 /** Pool statistics. */
268 struct lprocfs_stats *pl_stats;
271 struct kobject pl_kobj;
272 struct completion pl_kobj_unregister;
275 typedef int (*ldlm_res_policy)(const struct lu_env *env,
276 struct ldlm_namespace *,
277 struct ldlm_lock **, void *req_cookie,
278 enum ldlm_mode mode, __u64 flags, void *data);
280 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
284 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
285 * be associated with an LDLM lock and transferred from client to server and
288 * Currently LVBs are used by:
289 * - OSC-OST code to maintain current object size/times
290 * - layout lock code to return the layout when the layout lock is granted
292 * To ensure delayed LVB initialization, it is highly recommended to use the set
293 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
295 struct ldlm_valblock_ops {
296 int (*lvbo_init)(struct ldlm_resource *res);
297 int (*lvbo_update)(struct ldlm_resource *res, struct ldlm_lock *lock,
298 struct ptlrpc_request *r, int increase);
299 int (*lvbo_free)(struct ldlm_resource *res);
300 /* Return size of lvb data appropriate RPC size can be reserved */
301 int (*lvbo_size)(struct ldlm_lock *lock);
302 /* Called to fill in lvb data to RPC buffer @buf */
303 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int *buflen);
307 * LDLM pools related, type of lock pool in the namespace.
308 * Greedy means release cached locks aggressively
311 LDLM_NAMESPACE_GREEDY = BIT(0),
312 LDLM_NAMESPACE_MODEST = BIT(1),
316 * Default values for the "max_nolock_size", "contention_time" and
317 * "contended_locks" namespace tunables.
319 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
320 #define NS_DEFAULT_CONTENTION_SECONDS 2
321 #define NS_DEFAULT_CONTENDED_LOCKS 32
323 struct ldlm_ns_bucket {
324 /** back pointer to namespace */
325 struct ldlm_namespace *nsb_namespace;
327 * Estimated lock callback time. Used by adaptive timeout code to
328 * avoid spurious client evictions due to unresponsiveness when in
329 * fact the network or overall system load is at fault
331 struct adaptive_timeout nsb_at_estimate;
333 * Which res in the bucket should we start with the reclaim.
335 int nsb_reclaim_start;
336 /* counter of entries in this bucket */
341 /** LDLM namespace lock stats */
347 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
348 LDLM_NS_TYPE_MDC, /**< MDC namespace */
349 LDLM_NS_TYPE_MDT, /**< MDT namespace */
350 LDLM_NS_TYPE_OSC, /**< OSC namespace */
351 LDLM_NS_TYPE_OST, /**< OST namespace */
352 LDLM_NS_TYPE_MGC, /**< MGC namespace */
353 LDLM_NS_TYPE_MGT, /**< MGT namespace */
359 * Namespace serves to contain locks related to a particular service.
360 * There are two kinds of namespaces:
361 * - Server namespace has knowledge of all locks and is therefore authoritative
362 * to make decisions like what locks could be granted and what conflicts
363 * exist during new lock enqueue.
364 * - Client namespace only has limited knowledge about locks in the namespace,
365 * only seeing locks held by the client.
367 * Every Lustre service has one server namespace present on the server serving
368 * that service. Every client connected to the service has a client namespace
370 * Every lock obtained by client in that namespace is actually represented by
371 * two in-memory locks. One on the server and one on the client. The locks are
372 * linked by a special cookie by which one node can tell to the other which lock
373 * it actually means during communications. Such locks are called remote locks.
374 * The locks held by server only without any reference to a client are called
377 struct ldlm_namespace {
378 /** Backward link to OBD, required for LDLM pool to store new SLV. */
379 struct obd_device *ns_obd;
381 /** Flag indicating if namespace is on client instead of server */
382 enum ldlm_side ns_client;
384 /** name of this namespace */
387 /** Resource hash table for namespace. */
388 struct cfs_hash *ns_rs_hash;
389 struct ldlm_ns_bucket *ns_rs_buckets;
390 unsigned int ns_bucket_bits;
395 /** big refcount (by bucket) */
399 * Namespace connect flags supported by server (may be changed via
400 * /proc, LRU resize may be disabled/enabled).
402 __u64 ns_connect_flags;
404 /** Client side original connect flags supported by server. */
405 __u64 ns_orig_connect_flags;
407 /* namespace debugfs dir entry */
408 struct dentry *ns_debugfs_entry;
411 * Position in global namespace list linking all namespaces on
414 struct list_head ns_list_chain;
417 * List of unused locks for this namespace. This list is also called
419 * Unused locks are locks with zero reader/writer reference counts.
420 * This list is only used on clients for lock caching purposes.
421 * When we want to release some locks voluntarily or if server wants
422 * us to release some locks due to e.g. memory pressure, we take locks
423 * to release from the head of this list.
424 * Locks are linked via l_lru field in \see struct ldlm_lock.
426 struct list_head ns_unused_list;
427 /** Number of locks in the LRU list above */
429 struct list_head *ns_last_pos;
432 * Maximum number of locks permitted in the LRU. If 0, means locks
433 * are managed by pools and there is no preset limit, rather it is all
434 * controlled by available memory on this client and on server.
436 unsigned int ns_max_unused;
438 /** Maximum allowed age (last used time) for locks in the LRU. Set in
439 * seconds from userspace, but stored in ns to avoid repeat conversions.
444 * Server only: number of times we evicted clients due to lack of reply
447 unsigned int ns_timeouts;
449 * Number of seconds since the file change time after which
450 * the MDT will return an UPDATE lock along with a LOOKUP lock.
451 * This allows the client to start caching negative dentries
452 * for a directory and may save an RPC for a later stat.
454 timeout_t ns_ctime_age_limit;
456 * Number of (nano)seconds since the lock was last used. The client
457 * may cancel the lock older than this age and flush related data if
458 * another client shows interest in this lock by doing glimpse request.
459 * This allows to cache stat data locally for such files early. Set in
460 * seconds from userspace, but stored in ns to avoid repeat conversions.
462 ktime_t ns_dirty_age_limit;
464 * Used to rate-limit ldlm_namespace_dump calls.
465 * \see ldlm_namespace_dump. Increased by 10 seconds every time
468 time64_t ns_next_dump;
470 /** "policy" function that does actual lock conflict determination */
471 ldlm_res_policy ns_policy;
474 * LVB operations for this namespace.
475 * \see struct ldlm_valblock_ops
477 struct ldlm_valblock_ops *ns_lvbo;
480 * Used by filter code to store pointer to OBD of the service.
481 * Should be dropped in favor of \a ns_obd
486 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
487 * a resource is removed.
489 wait_queue_head_t ns_waitq;
490 /** LDLM pool structure for this namespace */
491 struct ldlm_pool ns_pool;
492 /** Definition of how eagerly unused locks will be released from LRU */
493 enum ldlm_appetite ns_appetite;
496 * If more than \a ns_contended_locks are found, the resource is
497 * considered to be contended. Lock enqueues might specify that no
498 * contended locks should be granted
500 unsigned ns_contended_locks;
503 * The resources in this namespace remember contended state during
504 * \a ns_contention_time, in seconds.
506 timeout_t ns_contention_time;
509 * Limit size of contended extent locks, in bytes.
510 * If extended lock is requested for more then this many bytes and
511 * caller instructs us not to grant contended locks, we would disregard
514 unsigned ns_max_nolock_size;
516 /** Limit of parallel AST RPC count. */
517 unsigned ns_max_parallel_ast;
520 * Callback to check if a lock is good to be canceled by ELC or
523 ldlm_cancel_cbt ns_cancel;
525 /** LDLM lock stats */
526 struct lprocfs_stats *ns_stats;
529 * Flag to indicate namespace is being freed. Used to determine if
530 * recalculation of LDLM pool statistics should be skipped.
532 unsigned ns_stopping:1;
535 * Which bucket should we start with the lock reclaim.
537 int ns_reclaim_start;
539 struct kobject ns_kobj; /* sysfs object */
540 struct completion ns_kobj_unregister;
544 * Returns 1 if namespace \a ns is a client namespace.
546 static inline int ns_is_client(struct ldlm_namespace *ns)
549 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
550 ns->ns_client == LDLM_NAMESPACE_SERVER);
551 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
555 * Returns 1 if namespace \a ns is a server namespace.
557 static inline int ns_is_server(struct ldlm_namespace *ns)
560 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
561 ns->ns_client == LDLM_NAMESPACE_SERVER);
562 return ns->ns_client == LDLM_NAMESPACE_SERVER;
566 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
568 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
571 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
575 * Returns 1 if this namespace supports lru_resize.
577 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
580 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
583 static inline void ns_register_cancel(struct ldlm_namespace *ns,
592 /** Type for blocking callback function of a lock. */
593 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
594 struct ldlm_lock_desc *new, void *data,
596 /** Type for completion callback function of a lock. */
597 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
599 /** Type for glimpse callback function of a lock. */
600 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
602 /** Type for created callback function of a lock. */
603 typedef void (*ldlm_created_callback)(struct ldlm_lock *lock);
605 /** Work list for sending GL ASTs to multiple locks. */
606 struct ldlm_glimpse_work {
607 struct ldlm_lock *gl_lock; /* lock to glimpse */
608 struct list_head gl_list; /* linkage to other gl work structs */
609 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
610 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
611 * glimpse callback request */
612 ptlrpc_interpterer_t gl_interpret_reply;
613 void *gl_interpret_data;
616 struct ldlm_bl_desc {
617 unsigned int bl_same_client:1,
621 struct ldlm_cb_set_arg {
622 struct ptlrpc_request_set *set;
623 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
625 struct list_head *list;
626 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
627 ptlrpc_interpterer_t gl_interpret_reply;
628 void *gl_interpret_data;
629 struct ldlm_bl_desc *bl_desc;
632 struct ldlm_cb_async_args {
633 struct ldlm_cb_set_arg *ca_set_arg;
634 struct ldlm_lock *ca_lock;
637 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
638 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
640 /** Interval node data for each LDLM_EXTENT lock. */
641 struct ldlm_interval {
642 struct interval_node li_node; /* node for tree management */
643 struct list_head li_group; /* the locks which have the same
644 * policy - group of the policy */
646 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
649 * Interval tree for extent locks.
650 * The interval tree must be accessed under the resource lock.
651 * Interval trees are used for granted extent locks to speed up conflicts
652 * lookup. See ldlm/interval_tree.c for more details.
654 struct ldlm_interval_tree {
657 enum ldlm_mode lit_mode; /* lock mode */
658 struct interval_node *lit_root; /* actual ldlm_interval */
662 * Lists of waiting locks for each inodebit type.
663 * A lock can be in several liq_waiting lists and it remains in lr_waiting.
665 struct ldlm_ibits_queues {
666 struct list_head liq_waiting[MDS_INODELOCK_NUMBITS];
669 struct ldlm_ibits_node {
670 struct list_head lin_link[MDS_INODELOCK_NUMBITS];
671 struct ldlm_lock *lock;
674 /** Whether to track references to exports by LDLM locks. */
675 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
678 enum ldlm_cancel_flags {
679 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
680 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
681 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
688 __u64 blocking_owner;
689 struct obd_export *blocking_export;
690 atomic_t blocking_refs;
694 union ldlm_policy_data {
695 struct ldlm_extent l_extent;
696 struct ldlm_flock l_flock;
697 struct ldlm_inodebits l_inodebits;
700 void ldlm_convert_policy_to_wire(enum ldlm_type type,
701 const union ldlm_policy_data *lpolicy,
702 union ldlm_wire_policy_data *wpolicy);
703 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
704 const union ldlm_wire_policy_data *wpolicy,
705 union ldlm_policy_data *lpolicy);
715 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
717 #define LDLM_GID_ANY ((__u64)-1)
720 * LDLM lock structure
722 * Represents a single LDLM lock and its state in memory. Each lock is
723 * associated with a single ldlm_resource, the object which is being
724 * locked. There may be multiple ldlm_locks on a single resource,
725 * depending on the lock type and whether the locks are conflicting or
731 * When remote side wants to tell us about a lock, they address
732 * it by this opaque handle. The handle does not hold a
733 * reference on the ldlm_lock, so it can be safely passed to
734 * other threads or nodes. When the lock needs to be accessed
735 * from the handle, it is looked up again in the lock table, and
736 * may no longer exist.
738 * Must be first in the structure.
740 struct portals_handle l_handle;
742 * Internal spinlock protects l_resource. We should hold this lock
743 * first before taking res_lock.
747 * Pointer to actual resource this lock is in.
748 * ldlm_lock_change_resource() can change this.
750 struct ldlm_resource *l_resource;
752 * List item for client side LRU list.
753 * Protected by ns_lock in struct ldlm_namespace.
755 struct list_head l_lru;
757 * Linkage to resource's lock queues according to current lock state.
758 * (could be granted or waiting)
759 * Protected by lr_lock in struct ldlm_resource.
761 struct list_head l_res_link;
763 * Internal structures per lock type..
766 struct ldlm_interval *l_tree_node;
767 struct ldlm_ibits_node *l_ibits_node;
770 * Per export hash of locks.
771 * Protected by per-bucket exp->exp_lock_hash locks.
773 struct hlist_node l_exp_hash;
775 * Per export hash of flock locks.
776 * Protected by per-bucket exp->exp_flock_hash locks.
778 struct hlist_node l_exp_flock_hash;
781 * Protected by lr_lock.
783 enum ldlm_mode l_req_mode;
785 * Granted mode, also protected by lr_lock.
787 enum ldlm_mode l_granted_mode;
788 /** Lock completion handler pointer. Called when lock is granted. */
789 ldlm_completion_callback l_completion_ast;
791 * Lock blocking AST handler pointer.
792 * It plays two roles:
793 * - as a notification of an attempt to queue a conflicting lock (once)
794 * - as a notification when the lock is being cancelled.
796 * As such it's typically called twice: once for the initial conflict
797 * and then once more when the last user went away and the lock is
798 * cancelled (could happen recursively).
800 ldlm_blocking_callback l_blocking_ast;
802 * Lock glimpse handler.
803 * Glimpse handler is used to obtain LVB updates from a client by
806 ldlm_glimpse_callback l_glimpse_ast;
810 * This is a pointer to actual client export for locks that were granted
811 * to clients. Used server-side.
813 struct obd_export *l_export;
815 * Lock connection export.
816 * Pointer to server export on a client.
818 struct obd_export *l_conn_export;
821 * Remote lock handle.
822 * If the lock is remote, this is the handle of the other side lock
825 struct lustre_handle l_remote_handle;
828 * Representation of private data specific for a lock type.
829 * Examples are: extent range for extent lock or bitmask for ibits locks
831 union ldlm_policy_data l_policy_data;
834 * Lock state flags. Protected by lr_lock.
835 * \see lustre_dlm_flags.h where the bits are defined.
840 * Lock r/w usage counters.
841 * Protected by lr_lock.
846 * If the lock is granted, a process sleeps on this waitq to learn when
847 * it's no longer in use. If the lock is not granted, a process sleeps
848 * on this waitq to learn when it becomes granted.
850 wait_queue_head_t l_waitq;
853 * Time, in nanoseconds, last used by e.g. being matched by lock match.
857 /** Originally requested extent for the extent lock. */
858 struct ldlm_extent l_req_extent;
861 * Client-side-only members.
864 enum lvb_type l_lvb_type;
867 * Temporary storage for a LVB received during an enqueue operation.
868 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
873 /** Private storage for lock user. Opaque to LDLM. */
878 * Seconds. It will be updated if there is any activity related to
879 * the lock at client, e.g. enqueue the lock. For server it is the
880 * time when blocking ast was sent.
883 time64_t l_blast_sent;
886 /* separate ost_lvb used mostly by Data-on-MDT for now.
887 * It is introduced to don't mix with layout lock data. */
888 struct ost_lvb l_ost_lvb;
890 * Server-side-only members.
894 * Connection cookie for the client originating the operation.
895 * Used by Commit on Share (COS) code. Currently only used for
896 * inodebits locks on MDS.
898 __u64 l_client_cookie;
901 * List item for locks waiting for cancellation from clients.
902 * The lists this could be linked into are:
903 * waiting_locks_list (protected by waiting_locks_spinlock),
904 * then if the lock timed out, it is moved to
905 * expired_lock_list for further processing.
907 struct list_head l_pending_chain;
910 * Set when lock is sent a blocking AST. Time in seconds when timeout
911 * is reached and client holding this lock could be evicted.
912 * This timeout could be further extended by e.g. certain IO activity
914 * \see ost_rw_prolong_locks
916 time64_t l_callback_timestamp;
918 /** Local PID of process which created this lock. */
922 * Number of times blocking AST was sent for this lock.
923 * This is for debugging. Valid values are 0 and 1, if there is an
924 * attempt to send blocking AST more than once, an assertion would be
925 * hit. \see ldlm_work_bl_ast_lock
928 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
929 struct list_head l_bl_ast;
930 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
931 struct list_head l_cp_ast;
932 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
933 struct list_head l_rk_ast;
936 * Pointer to a conflicting lock that caused blocking AST to be sent
939 struct ldlm_lock *l_blocking_lock;
942 * Protected by lr_lock, linkages to "skip lists".
943 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
945 struct list_head l_sl_mode;
946 struct list_head l_sl_policy;
948 /** Reference tracking structure to debug leaked locks. */
949 struct lu_ref l_reference;
950 #if LUSTRE_TRACKS_LOCK_EXP_REFS
951 /* Debugging stuff for bug 20498, for tracking export references. */
952 /** number of export references taken */
954 /** link all locks referencing one export */
955 struct list_head l_exp_refs_link;
956 /** referenced export object */
957 struct obd_export *l_exp_refs_target;
960 * export blocking dlm lock list, protected by
961 * l_export->exp_bl_list_lock.
962 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
963 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
965 struct list_head l_exp_list;
969 * Describe the overlap between two locks. itree_overlap_cb data.
971 struct ldlm_match_data {
972 struct ldlm_lock *lmd_old;
973 struct ldlm_lock *lmd_lock;
974 enum ldlm_mode *lmd_mode;
975 union ldlm_policy_data *lmd_policy;
977 __u64 lmd_skip_flags;
979 bool lmd_has_ast_data;
982 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
983 #define l_transno l_client_cookie
985 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
986 * which is for server. */
987 #define l_slc_link l_rk_ast
989 #define HANDLE_MAP_SIZE ((LMV_MAX_STRIPE_COUNT + 7) >> 3)
991 struct lustre_handle_array {
992 unsigned int ha_count;
993 /* ha_map is used as bit flag to indicate handle is remote or local */
994 char ha_map[HANDLE_MAP_SIZE];
995 struct lustre_handle ha_handles[0];
999 * LDLM resource description.
1000 * Basically, resource is a representation for a single object.
1001 * Object has a name which is currently 4 64-bit integers. LDLM user is
1002 * responsible for creation of a mapping between objects it wants to be
1003 * protected and resource names.
1005 * A resource can only hold locks of a single lock type, though there may be
1006 * multiple ldlm_locks on a single resource, depending on the lock type and
1007 * whether the locks are conflicting or not.
1009 struct ldlm_resource {
1010 struct ldlm_ns_bucket *lr_ns_bucket;
1013 * List item for list in namespace hash.
1014 * protected by ns_lock
1016 struct hlist_node lr_hash;
1018 /** Reference count for this resource */
1019 atomic_t lr_refcount;
1021 /** Spinlock to protect locks under this resource. */
1025 * protected by lr_lock
1027 /** List of locks in granted state */
1028 struct list_head lr_granted;
1030 * List of locks that could not be granted due to conflicts and
1031 * that are waiting for conflicts to go away */
1032 struct list_head lr_waiting;
1035 /** Resource name */
1036 struct ldlm_res_id lr_name;
1040 * Interval trees (only for extent locks) for all modes of
1043 struct ldlm_interval_tree *lr_itree;
1044 struct ldlm_ibits_queues *lr_ibits_queues;
1049 * When the resource was considered as contended,
1050 * used only on server side.
1052 time64_t lr_contention_time;
1054 * Associated inode, used only on client side.
1056 struct inode *lr_lvb_inode;
1059 /** Type of locks this resource can hold. Only one type per resource. */
1060 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1063 * Server-side-only lock value block elements.
1064 * To serialize lvbo_init.
1067 struct mutex lr_lvb_mutex;
1068 /** protected by lr_lock */
1070 /** is lvb initialized ? */
1071 bool lr_lvb_initialized;
1073 /** List of references to this resource. For debugging. */
1074 struct lu_ref lr_reference;
1077 static inline int ldlm_is_granted(struct ldlm_lock *lock)
1079 return lock->l_req_mode == lock->l_granted_mode;
1082 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1084 return lock->l_resource->lr_type == LDLM_IBITS &&
1085 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1088 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1090 return lock->l_resource->lr_type == LDLM_IBITS &&
1091 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1094 static inline char *
1095 ldlm_ns_name(struct ldlm_namespace *ns)
1100 static inline struct ldlm_namespace *
1101 ldlm_res_to_ns(struct ldlm_resource *res)
1103 return res->lr_ns_bucket->nsb_namespace;
1106 static inline struct ldlm_namespace *
1107 ldlm_lock_to_ns(struct ldlm_lock *lock)
1109 return ldlm_res_to_ns(lock->l_resource);
1112 static inline char *
1113 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1115 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1118 static inline struct adaptive_timeout *
1119 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1121 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1124 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1126 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1129 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1130 res->lr_lvb_initialized)
1133 mutex_lock(&res->lr_lvb_mutex);
1134 /* Did we lose the race? */
1135 if (res->lr_lvb_initialized) {
1136 mutex_unlock(&res->lr_lvb_mutex);
1139 rc = ns->ns_lvbo->lvbo_init(res);
1141 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1143 if (res->lr_lvb_data != NULL) {
1144 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1145 res->lr_lvb_data = NULL;
1147 res->lr_lvb_len = rc;
1149 res->lr_lvb_initialized = true;
1151 mutex_unlock(&res->lr_lvb_mutex);
1155 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1157 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1159 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1160 return ns->ns_lvbo->lvbo_size(lock);
1165 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int *len)
1167 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1170 if (ns->ns_lvbo != NULL) {
1171 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1172 /* init lvb now if not already */
1173 rc = ldlm_lvbo_init(lock->l_resource);
1175 CERROR("lock %p: delayed lvb init failed (rc %d)",
1179 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1184 struct ldlm_ast_work {
1185 struct ldlm_lock *w_lock;
1187 struct ldlm_lock_desc w_desc;
1188 struct list_head w_list;
1195 * Common ldlm_enqueue parameters
1197 struct ldlm_enqueue_info {
1198 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1199 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1200 void *ei_cb_bl; /** blocking lock callback */
1201 void *ei_cb_local_bl; /** blocking local lock callback */
1202 void *ei_cb_cp; /** lock completion callback */
1203 void *ei_cb_gl; /** lock glimpse callback */
1204 ldlm_created_callback ei_cb_created; /** lock created callback */
1205 void *ei_cbdata; /** Data to be passed into callbacks. */
1206 void *ei_namespace; /** lock namespace **/
1207 u64 ei_inodebits; /** lock inode bits **/
1208 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1209 unsigned int ei_enq_slot:1; /** whether acquire rpc slot */
1212 #define ei_res_id ei_cb_gl
1214 extern char *ldlm_lockname[];
1215 extern char *ldlm_typename[];
1216 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1219 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1220 * For the cases where we do not have actual lock to print along
1221 * with a debugging message that is ldlm-related
1223 #define LDLM_DEBUG_NOLOCK(format, a...) \
1224 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1227 * Support function for lock information printing into debug logs.
1231 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1232 CFS_CHECK_STACK(msgdata, mask, cdls); \
1234 if (((mask) & D_CANTMASK) != 0 || \
1235 ((libcfs_debug & (mask)) != 0 && \
1236 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1237 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1240 void _ldlm_lock_debug(struct ldlm_lock *lock,
1241 struct libcfs_debug_msg_data *data,
1242 const char *fmt, ...)
1243 __attribute__ ((format (printf, 3, 4)));
1246 * Rate-limited version of lock printing function.
1248 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1249 static struct cfs_debug_limit_state _ldlm_cdls; \
1250 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1251 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1254 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1255 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1257 /** Non-rate-limited lock printing function for debugging purposes. */
1258 #define LDLM_DEBUG(lock, fmt, a...) do { \
1259 if (likely(lock != NULL)) { \
1260 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1261 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1262 "### " fmt , ##a); \
1264 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1267 #else /* !LIBCFS_DEBUG */
1268 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1269 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1270 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1274 * Three intentions can be used for the policy functions in
1275 * ldlm_processing_policy.
1277 * LDLM_PROCESS_RESCAN:
1279 * It's used when policy functions are called from ldlm_reprocess_queue() to
1280 * reprocess the wait list and try to grant locks, blocking ASTs
1281 * have already been sent in this situation, completion ASTs need be sent for
1282 * the locks being granted.
1284 * LDLM_PROCESS_ENQUEUE:
1286 * It's used when policy functions are called from ldlm_lock_enqueue() to
1287 * process the wait list for handling an enqueue request, blocking
1288 * ASTs have not been sent yet, so list of conflicting locks would be
1289 * collected and ASTs sent.
1291 * LDLM_PROCESS_RECOVERY:
1293 * It's used when policy functions are called from ldlm_reprocess_queue() to
1294 * reprocess the wait list when recovery done. In case of blocking
1295 * ASTs are lost before recovery, it needs not only to grant locks if
1296 * available, but also send blocking ASTs to the locks doesn't have AST sent
1297 * flag. Completion ASTs need be sent for the locks being granted.
1299 enum ldlm_process_intention {
1300 LDLM_PROCESS_RESCAN = 0,
1301 LDLM_PROCESS_ENQUEUE = 1,
1302 LDLM_PROCESS_RECOVERY = 2,
1305 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1306 enum ldlm_process_intention intention,
1307 enum ldlm_error *err,
1308 struct list_head *work_list);
1310 typedef int (*ldlm_reprocessing_policy)(struct ldlm_resource *res,
1311 struct list_head *queue,
1312 struct list_head *work_list,
1313 enum ldlm_process_intention intention,
1314 struct ldlm_lock *hint);
1317 * Return values for lock iterators.
1318 * Also used during deciding of lock grants and cancellations.
1320 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1321 #define LDLM_ITER_STOP 2 /* stop iterating */
1323 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1324 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1326 /** \defgroup ldlm_iterator Lock iterators
1328 * LDLM provides for a way to iterate through every lock on a resource or
1329 * namespace or every resource in a namespace.
1331 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1333 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1335 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1336 ldlm_iterator_t iter, void *data);
1337 /** @} ldlm_iterator */
1339 int ldlm_replay_locks(struct obd_import *imp);
1342 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1345 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1347 struct ldlm_prolong_args {
1348 struct obd_export *lpa_export;
1349 struct ldlm_res_id lpa_resid;
1350 struct ldlm_extent lpa_extent;
1351 enum ldlm_mode lpa_mode;
1352 timeout_t lpa_timeout;
1356 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1357 struct ldlm_prolong_args *arg);
1358 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1360 struct ldlm_callback_suite {
1361 ldlm_completion_callback lcs_completion;
1362 ldlm_blocking_callback lcs_blocking;
1363 ldlm_glimpse_callback lcs_glimpse;
1367 #ifdef HAVE_SERVER_SUPPORT
1368 /** \defgroup ldlm_srv_ast Server AST handlers
1369 * These are AST handlers used by server code.
1370 * Their property is that they are just preparing RPCs to be sent to clients.
1373 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1374 void *data, int flag);
1375 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1376 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1377 int ldlm_glimpse_locks(struct ldlm_resource *res,
1378 struct list_head *gl_work_list);
1379 /** @} ldlm_srv_ast */
1381 /** \defgroup ldlm_handlers Server LDLM handlers
1382 * These are handler functions that should be called by "frontends" such as
1383 * MDT or OST to pass through LDLM requests to LDLM for handling
1386 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1387 const struct ldlm_request *dlm_req,
1388 const struct ldlm_callback_suite *cbs);
1389 int ldlm_handle_convert0(struct ptlrpc_request *req,
1390 const struct ldlm_request *dlm_req);
1391 int ldlm_handle_cancel(struct ptlrpc_request *req);
1392 int ldlm_request_cancel(struct ptlrpc_request *req,
1393 const struct ldlm_request *dlm_req,
1394 int first, enum lustre_at_flags flags);
1395 /** @} ldlm_handlers */
1397 void ldlm_revoke_export_locks(struct obd_export *exp);
1398 timeout_t ldlm_bl_timeout(struct ldlm_lock *lock);
1400 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1401 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, timeout_t timeout);
1402 int ldlm_get_ref(void);
1403 void ldlm_put_ref(void);
1404 int ldlm_init_export(struct obd_export *exp);
1405 void ldlm_destroy_export(struct obd_export *exp);
1406 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1409 #ifdef HAVE_SERVER_SUPPORT
1410 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1411 ldlm_reprocessing_policy
1412 ldlm_get_reprocessing_policy(struct ldlm_resource *res);
1414 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1415 void ldlm_lock2handle(const struct ldlm_lock *lock,
1416 struct lustre_handle *lockh);
1417 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1418 void ldlm_cancel_callback(struct ldlm_lock *);
1419 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1420 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1423 * Obtain a lock reference by its handle.
1425 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1427 return __ldlm_handle2lock(h, 0);
1430 #define LDLM_LOCK_REF_DEL(lock) \
1431 lu_ref_del(&lock->l_reference, "handle", lock)
1433 static inline struct ldlm_lock *
1434 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1436 struct ldlm_lock *lock;
1438 lock = __ldlm_handle2lock(h, flags);
1440 LDLM_LOCK_REF_DEL(lock);
1445 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1446 * data from request \a r
1448 static inline int ldlm_lvbo_update(struct ldlm_resource *res,
1449 struct ldlm_lock *lock,
1450 struct ptlrpc_request *req, int increase)
1452 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1455 /* delayed lvb init may be required */
1456 rc = ldlm_lvbo_init(res);
1458 CERROR("delayed lvb init failed (rc %d)\n", rc);
1462 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1463 return ns->ns_lvbo->lvbo_update(res, lock, req, increase);
1468 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1469 struct ptlrpc_request *req,
1472 return ldlm_lvbo_update(res, NULL, req, increase);
1475 int is_granted_or_cancelled_nolock(struct ldlm_lock *lock);
1477 int ldlm_error2errno(enum ldlm_error error);
1478 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1479 * confuses user-space. */
1480 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1481 void ldlm_dump_export_locks(struct obd_export *exp);
1485 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1486 * __ldlm_handle2lock().
1488 #define LDLM_LOCK_PUT(lock) \
1490 LDLM_LOCK_REF_DEL(lock); \
1491 /*LDLM_DEBUG((lock), "put");*/ \
1492 ldlm_lock_put(lock); \
1496 * Release a lock reference obtained by some other means (see
1499 #define LDLM_LOCK_RELEASE(lock) \
1501 /*LDLM_DEBUG((lock), "put");*/ \
1502 ldlm_lock_put(lock); \
1505 #define LDLM_LOCK_GET(lock) \
1507 ldlm_lock_get(lock); \
1508 /*LDLM_DEBUG((lock), "get");*/ \
1512 #define ldlm_lock_list_put(head, member, count) \
1514 struct ldlm_lock *_lock, *_next; \
1516 list_for_each_entry_safe(_lock, _next, head, member) { \
1519 list_del_init(&_lock->member); \
1520 LDLM_LOCK_RELEASE(_lock); \
1525 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1526 void ldlm_lock_put(struct ldlm_lock *lock);
1527 void ldlm_lock_destroy(struct ldlm_lock *lock);
1528 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1529 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1530 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1531 enum ldlm_mode mode);
1532 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1533 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1534 enum ldlm_mode mode);
1535 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1536 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1537 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1538 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1539 enum ldlm_mode ldlm_lock_match_with_skip(struct ldlm_namespace *ns,
1540 __u64 flags, __u64 skip_flags,
1541 const struct ldlm_res_id *res_id,
1542 enum ldlm_type type,
1543 union ldlm_policy_data *policy,
1544 enum ldlm_mode mode,
1545 struct lustre_handle *lh,
1547 static inline enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns,
1549 const struct ldlm_res_id *res_id,
1550 enum ldlm_type type,
1551 union ldlm_policy_data *policy,
1552 enum ldlm_mode mode,
1553 struct lustre_handle *lh,
1556 return ldlm_lock_match_with_skip(ns, flags, 0, res_id, type, policy,
1559 struct ldlm_lock *search_itree(struct ldlm_resource *res,
1560 struct ldlm_match_data *data);
1561 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1563 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1564 void ldlm_lock_cancel(struct ldlm_lock *lock);
1565 void ldlm_reprocess_all(struct ldlm_resource *res, struct ldlm_lock *hint);
1566 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1567 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1568 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1571 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1572 enum ldlm_side client,
1573 enum ldlm_appetite apt,
1574 enum ldlm_ns_type ns_type);
1575 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1576 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1577 struct obd_import *imp,
1579 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1580 void ldlm_namespace_free(struct ldlm_namespace *ns,
1581 struct obd_import *imp, int force);
1582 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1583 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1584 enum ldlm_side client);
1585 void ldlm_namespace_get(struct ldlm_namespace *ns);
1586 void ldlm_namespace_put(struct ldlm_namespace *ns);
1588 int ldlm_debugfs_setup(void);
1589 void ldlm_debugfs_cleanup(void);
1591 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1592 struct lprocfs_stats *srv_stats)
1594 int lock_type = 0, op = 0;
1596 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1598 switch (lock_type) {
1600 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1603 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1606 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1609 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1617 lprocfs_counter_incr(srv_stats, op);
1620 /* resource.c - internal */
1621 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1622 struct ldlm_resource *parent,
1623 const struct ldlm_res_id *,
1624 enum ldlm_type type, int create);
1625 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1626 int ldlm_resource_putref(struct ldlm_resource *res);
1627 void ldlm_resource_add_lock(struct ldlm_resource *res,
1628 struct list_head *head,
1629 struct ldlm_lock *lock);
1630 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1631 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1632 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1633 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1634 void ldlm_resource_dump(int level, struct ldlm_resource *);
1635 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1636 const struct ldlm_res_id *);
1638 #define LDLM_RESOURCE_ADDREF(res) do { \
1639 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1642 #define LDLM_RESOURCE_DELREF(res) do { \
1643 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1646 /* ldlm_request.c */
1647 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1648 * These AST handlers are typically used for server-side local locks and are
1649 * also used by client-side lock handlers to perform minimum level base
1652 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1653 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1654 void *data, int flag);
1655 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1656 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1657 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1658 /** @} ldlm_local_ast */
1660 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1661 * These are typically used by client and server (*_local versions)
1662 * to obtain and release locks.
1664 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1665 struct ldlm_enqueue_info *einfo,
1666 const struct ldlm_res_id *res_id,
1667 union ldlm_policy_data const *policy, __u64 *flags,
1668 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1669 struct lustre_handle *lockh, int async);
1670 int ldlm_prep_enqueue_req(struct obd_export *exp,
1671 struct ptlrpc_request *req,
1672 struct list_head *cancels,
1674 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1675 int version, int opc, int canceloff,
1676 struct list_head *cancels, int count);
1678 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1679 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1680 const struct ldlm_request *dlm_req,
1681 const struct ldlm_callback_suite *cbs);
1682 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1683 enum ldlm_type type, __u8 with_policy,
1684 enum ldlm_mode mode, __u64 *flags, void *lvb,
1686 const struct lustre_handle *lockh, int rc);
1687 int ldlm_cli_enqueue_local(const struct lu_env *env,
1688 struct ldlm_namespace *ns,
1689 const struct ldlm_res_id *res_id,
1690 enum ldlm_type type, union ldlm_policy_data *policy,
1691 enum ldlm_mode mode, __u64 *flags,
1692 ldlm_blocking_callback blocking,
1693 ldlm_completion_callback completion,
1694 ldlm_glimpse_callback glimpse,
1695 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1696 const __u64 *client_cookie,
1697 struct lustre_handle *lockh);
1698 int ldlm_cli_convert_req(struct ldlm_lock *lock, __u32 *flags, __u64 new_bits);
1699 int ldlm_cli_convert(struct ldlm_lock *lock,
1700 enum ldlm_cancel_flags cancel_flags);
1701 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1702 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1703 enum ldlm_cancel_flags cancel_flags);
1704 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1705 enum ldlm_cancel_flags flags, void *opaque);
1706 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1707 const struct ldlm_res_id *res_id,
1708 union ldlm_policy_data *policy,
1709 enum ldlm_mode mode,
1710 enum ldlm_cancel_flags flags, void *opaque);
1711 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1712 int count, enum ldlm_cancel_flags flags);
1713 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1714 struct list_head *cancels,
1715 union ldlm_policy_data *policy,
1716 enum ldlm_mode mode, __u64 lock_flags,
1717 enum ldlm_cancel_flags cancel_flags,
1719 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1720 enum ldlm_cancel_flags flags);
1721 int ldlm_cli_cancel_list(struct list_head *head, int count,
1722 struct ptlrpc_request *req,
1723 enum ldlm_cancel_flags flags);
1725 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);
1726 int ldlm_cli_inodebits_convert(struct ldlm_lock *lock,
1727 enum ldlm_cancel_flags cancel_flags);
1729 /** @} ldlm_cli_api */
1731 extern unsigned int ldlm_enqueue_min;
1734 /* This has to be here because recursive inclusion sucks. */
1735 int intent_disposition(struct ldlm_reply *rep, int flag);
1736 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1739 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1740 * than one lock_res is dead-lock safe.
1742 enum lock_res_type {
1747 /** Lock resource. */
1748 static inline void lock_res(struct ldlm_resource *res)
1750 spin_lock(&res->lr_lock);
1753 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1754 static inline void lock_res_nested(struct ldlm_resource *res,
1755 enum lock_res_type mode)
1757 spin_lock_nested(&res->lr_lock, mode);
1760 /** Unlock resource. */
1761 static inline void unlock_res(struct ldlm_resource *res)
1763 spin_unlock(&res->lr_lock);
1766 /** Check if resource is already locked, assert if not. */
1767 static inline void check_res_locked(struct ldlm_resource *res)
1769 assert_spin_locked(&res->lr_lock);
1772 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1773 void unlock_res_and_lock(struct ldlm_lock *lock);
1776 /** \defgroup ldlm_pools Various LDLM pool related functions
1777 * There are not used outside of ldlm.
1780 int ldlm_pools_init(void);
1781 void ldlm_pools_fini(void);
1783 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1784 int idx, enum ldlm_side client);
1785 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1786 void ldlm_pool_fini(struct ldlm_pool *pl);
1787 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1788 time64_t ldlm_pool_recalc(struct ldlm_pool *pl);
1789 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1790 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1791 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1792 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1793 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1794 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1795 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1796 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1797 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1800 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1801 const struct ldlm_extent *ex2)
1803 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1806 /* check if @ex1 contains @ex2 */
1807 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1808 const struct ldlm_extent *ex2)
1810 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1813 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);