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 = 1 << 0,
312 LDLM_NAMESPACE_MODEST = 1 << 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;
339 /** LDLM namespace lock stats */
345 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
346 LDLM_NS_TYPE_MDC, /**< MDC namespace */
347 LDLM_NS_TYPE_MDT, /**< MDT namespace */
348 LDLM_NS_TYPE_OSC, /**< OSC namespace */
349 LDLM_NS_TYPE_OST, /**< OST namespace */
350 LDLM_NS_TYPE_MGC, /**< MGC namespace */
351 LDLM_NS_TYPE_MGT, /**< MGT namespace */
357 * Namespace serves to contain locks related to a particular service.
358 * There are two kinds of namespaces:
359 * - Server namespace has knowledge of all locks and is therefore authoritative
360 * to make decisions like what locks could be granted and what conflicts
361 * exist during new lock enqueue.
362 * - Client namespace only has limited knowledge about locks in the namespace,
363 * only seeing locks held by the client.
365 * Every Lustre service has one server namespace present on the server serving
366 * that service. Every client connected to the service has a client namespace
368 * Every lock obtained by client in that namespace is actually represented by
369 * two in-memory locks. One on the server and one on the client. The locks are
370 * linked by a special cookie by which one node can tell to the other which lock
371 * it actually means during communications. Such locks are called remote locks.
372 * The locks held by server only without any reference to a client are called
375 struct ldlm_namespace {
376 /** Backward link to OBD, required for LDLM pool to store new SLV. */
377 struct obd_device *ns_obd;
379 /** Flag indicating if namespace is on client instead of server */
380 enum ldlm_side ns_client;
382 /** name of this namespace */
385 /** Resource hash table for namespace. */
386 struct cfs_hash *ns_rs_hash;
391 /** big refcount (by bucket) */
395 * Namespace connect flags supported by server (may be changed via
396 * /proc, LRU resize may be disabled/enabled).
398 __u64 ns_connect_flags;
400 /** Client side original connect flags supported by server. */
401 __u64 ns_orig_connect_flags;
403 /* namespace debugfs dir entry */
404 struct dentry *ns_debugfs_entry;
407 * Position in global namespace list linking all namespaces on
410 struct list_head ns_list_chain;
413 * List of unused locks for this namespace. This list is also called
415 * Unused locks are locks with zero reader/writer reference counts.
416 * This list is only used on clients for lock caching purposes.
417 * When we want to release some locks voluntarily or if server wants
418 * us to release some locks due to e.g. memory pressure, we take locks
419 * to release from the head of this list.
420 * Locks are linked via l_lru field in \see struct ldlm_lock.
422 struct list_head ns_unused_list;
423 /** Number of locks in the LRU list above */
425 struct list_head *ns_last_pos;
428 * Maximum number of locks permitted in the LRU. If 0, means locks
429 * are managed by pools and there is no preset limit, rather it is all
430 * controlled by available memory on this client and on server.
432 unsigned int ns_max_unused;
434 /** Maximum allowed age (last used time) for locks in the LRU */
438 * Server only: number of times we evicted clients due to lack of reply
441 unsigned int ns_timeouts;
443 * Number of seconds since the file change time after which the
444 * MDT will return an UPDATE lock along with a LOOKUP lock.
445 * This allows the client to start caching negative dentries
446 * for a directory and may save an RPC for a later stat.
448 time64_t ns_ctime_age_limit;
450 * Number of seconds since the lock was last used. The client may
451 * cancel the lock limited by this age and flush related data if
452 * any other client shows interest in it doing glimpse request.
453 * This allows to cache stat data locally for such files early.
455 time64_t ns_dirty_age_limit;
457 * Used to rate-limit ldlm_namespace_dump calls.
458 * \see ldlm_namespace_dump. Increased by 10 seconds every time
461 time64_t ns_next_dump;
463 /** "policy" function that does actual lock conflict determination */
464 ldlm_res_policy ns_policy;
467 * LVB operations for this namespace.
468 * \see struct ldlm_valblock_ops
470 struct ldlm_valblock_ops *ns_lvbo;
473 * Used by filter code to store pointer to OBD of the service.
474 * Should be dropped in favor of \a ns_obd
479 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
480 * a resource is removed.
482 wait_queue_head_t ns_waitq;
483 /** LDLM pool structure for this namespace */
484 struct ldlm_pool ns_pool;
485 /** Definition of how eagerly unused locks will be released from LRU */
486 enum ldlm_appetite ns_appetite;
489 * If more than \a ns_contended_locks are found, the resource is
490 * considered to be contended. Lock enqueues might specify that no
491 * contended locks should be granted
493 unsigned ns_contended_locks;
496 * The resources in this namespace remember contended state during
497 * \a ns_contention_time, in seconds.
499 time64_t ns_contention_time;
502 * Limit size of contended extent locks, in bytes.
503 * If extended lock is requested for more then this many bytes and
504 * caller instructs us not to grant contended locks, we would disregard
507 unsigned ns_max_nolock_size;
509 /** Limit of parallel AST RPC count. */
510 unsigned ns_max_parallel_ast;
513 * Callback to check if a lock is good to be canceled by ELC or
516 ldlm_cancel_cbt ns_cancel;
518 /** LDLM lock stats */
519 struct lprocfs_stats *ns_stats;
522 * Flag to indicate namespace is being freed. Used to determine if
523 * recalculation of LDLM pool statistics should be skipped.
525 unsigned ns_stopping:1;
528 * Which bucket should we start with the lock reclaim.
530 int ns_reclaim_start;
532 struct kobject ns_kobj; /* sysfs object */
533 struct completion ns_kobj_unregister;
537 * Returns 1 if namespace \a ns is a client namespace.
539 static inline int ns_is_client(struct ldlm_namespace *ns)
542 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
543 ns->ns_client == LDLM_NAMESPACE_SERVER);
544 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
548 * Returns 1 if namespace \a ns is a server namespace.
550 static inline int ns_is_server(struct ldlm_namespace *ns)
553 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
554 ns->ns_client == LDLM_NAMESPACE_SERVER);
555 return ns->ns_client == LDLM_NAMESPACE_SERVER;
559 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
561 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
564 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
568 * Returns 1 if this namespace supports lru_resize.
570 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
573 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
576 static inline void ns_register_cancel(struct ldlm_namespace *ns,
585 /** Type for blocking callback function of a lock. */
586 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
587 struct ldlm_lock_desc *new, void *data,
589 /** Type for completion callback function of a lock. */
590 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
592 /** Type for glimpse callback function of a lock. */
593 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
595 /** Type for created callback function of a lock. */
596 typedef void (*ldlm_created_callback)(struct ldlm_lock *lock);
598 /** Work list for sending GL ASTs to multiple locks. */
599 struct ldlm_glimpse_work {
600 struct ldlm_lock *gl_lock; /* lock to glimpse */
601 struct list_head gl_list; /* linkage to other gl work structs */
602 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
603 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
604 * glimpse callback request */
605 ptlrpc_interpterer_t gl_interpret_reply;
606 void *gl_interpret_data;
609 struct ldlm_bl_desc {
610 unsigned int bl_same_client:1,
614 struct ldlm_cb_set_arg {
615 struct ptlrpc_request_set *set;
616 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
618 struct list_head *list;
619 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
620 ptlrpc_interpterer_t gl_interpret_reply;
621 void *gl_interpret_data;
622 struct ldlm_bl_desc *bl_desc;
625 struct ldlm_cb_async_args {
626 struct ldlm_cb_set_arg *ca_set_arg;
627 struct ldlm_lock *ca_lock;
630 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
631 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
633 /** Interval node data for each LDLM_EXTENT lock. */
634 struct ldlm_interval {
635 struct interval_node li_node; /* node for tree management */
636 struct list_head li_group; /* the locks which have the same
637 * policy - group of the policy */
639 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
642 * Interval tree for extent locks.
643 * The interval tree must be accessed under the resource lock.
644 * Interval trees are used for granted extent locks to speed up conflicts
645 * lookup. See ldlm/interval_tree.c for more details.
647 struct ldlm_interval_tree {
650 enum ldlm_mode lit_mode; /* lock mode */
651 struct interval_node *lit_root; /* actual ldlm_interval */
654 /** Whether to track references to exports by LDLM locks. */
655 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
658 enum ldlm_cancel_flags {
659 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
660 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
661 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
662 LCF_CONVERT = 0x8, /* Try to convert IBITS lock before cancel */
669 __u64 blocking_owner;
670 struct obd_export *blocking_export;
671 atomic_t blocking_refs;
675 union ldlm_policy_data {
676 struct ldlm_extent l_extent;
677 struct ldlm_flock l_flock;
678 struct ldlm_inodebits l_inodebits;
681 void ldlm_convert_policy_to_wire(enum ldlm_type type,
682 const union ldlm_policy_data *lpolicy,
683 union ldlm_wire_policy_data *wpolicy);
684 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
685 const union ldlm_wire_policy_data *wpolicy,
686 union ldlm_policy_data *lpolicy);
696 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
698 #define LDLM_GID_ANY ((__u64)-1)
701 * LDLM lock structure
703 * Represents a single LDLM lock and its state in memory. Each lock is
704 * associated with a single ldlm_resource, the object which is being
705 * locked. There may be multiple ldlm_locks on a single resource,
706 * depending on the lock type and whether the locks are conflicting or
712 * When remote side wants to tell us about a lock, they address
713 * it by this opaque handle. The handle does not hold a
714 * reference on the ldlm_lock, so it can be safely passed to
715 * other threads or nodes. When the lock needs to be accessed
716 * from the handle, it is looked up again in the lock table, and
717 * may no longer exist.
719 * Must be first in the structure.
721 struct portals_handle l_handle;
723 * Lock reference count.
724 * This is how many users have pointers to actual structure, so that
725 * we do not accidentally free lock structure that is in use.
729 * Internal spinlock protects l_resource. We should hold this lock
730 * first before taking res_lock.
734 * Pointer to actual resource this lock is in.
735 * ldlm_lock_change_resource() can change this.
737 struct ldlm_resource *l_resource;
739 * List item for client side LRU list.
740 * Protected by ns_lock in struct ldlm_namespace.
742 struct list_head l_lru;
744 * Linkage to resource's lock queues according to current lock state.
745 * (could be granted or waiting)
746 * Protected by lr_lock in struct ldlm_resource.
748 struct list_head l_res_link;
750 * Tree node for ldlm_extent.
752 struct ldlm_interval *l_tree_node;
754 * Per export hash of locks.
755 * Protected by per-bucket exp->exp_lock_hash locks.
757 struct hlist_node l_exp_hash;
759 * Per export hash of flock locks.
760 * Protected by per-bucket exp->exp_flock_hash locks.
762 struct hlist_node l_exp_flock_hash;
765 * Protected by lr_lock.
767 enum ldlm_mode l_req_mode;
769 * Granted mode, also protected by lr_lock.
771 enum ldlm_mode l_granted_mode;
772 /** Lock completion handler pointer. Called when lock is granted. */
773 ldlm_completion_callback l_completion_ast;
775 * Lock blocking AST handler pointer.
776 * It plays two roles:
777 * - as a notification of an attempt to queue a conflicting lock (once)
778 * - as a notification when the lock is being cancelled.
780 * As such it's typically called twice: once for the initial conflict
781 * and then once more when the last user went away and the lock is
782 * cancelled (could happen recursively).
784 ldlm_blocking_callback l_blocking_ast;
786 * Lock glimpse handler.
787 * Glimpse handler is used to obtain LVB updates from a client by
790 ldlm_glimpse_callback l_glimpse_ast;
794 * This is a pointer to actual client export for locks that were granted
795 * to clients. Used server-side.
797 struct obd_export *l_export;
799 * Lock connection export.
800 * Pointer to server export on a client.
802 struct obd_export *l_conn_export;
805 * Remote lock handle.
806 * If the lock is remote, this is the handle of the other side lock
809 struct lustre_handle l_remote_handle;
812 * Representation of private data specific for a lock type.
813 * Examples are: extent range for extent lock or bitmask for ibits locks
815 union ldlm_policy_data l_policy_data;
818 * Lock state flags. Protected by lr_lock.
819 * \see lustre_dlm_flags.h where the bits are defined.
824 * Lock r/w usage counters.
825 * Protected by lr_lock.
830 * If the lock is granted, a process sleeps on this waitq to learn when
831 * it's no longer in use. If the lock is not granted, a process sleeps
832 * on this waitq to learn when it becomes granted.
834 wait_queue_head_t l_waitq;
837 * Time, in nanoseconds, last used by e.g. being matched by lock match.
841 /** Originally requested extent for the extent lock. */
842 struct ldlm_extent l_req_extent;
845 * Client-side-only members.
848 enum lvb_type l_lvb_type;
851 * Temporary storage for a LVB received during an enqueue operation.
852 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
857 /** Private storage for lock user. Opaque to LDLM. */
862 * Seconds. It will be updated if there is any activity related to
863 * the lock at client, e.g. enqueue the lock. For server it is the
864 * time when blocking ast was sent.
867 time64_t l_blast_sent;
870 /* separate ost_lvb used mostly by Data-on-MDT for now.
871 * It is introduced to don't mix with layout lock data. */
872 struct ost_lvb l_ost_lvb;
874 * Server-side-only members.
878 * Connection cookie for the client originating the operation.
879 * Used by Commit on Share (COS) code. Currently only used for
880 * inodebits locks on MDS.
882 __u64 l_client_cookie;
885 * List item for locks waiting for cancellation from clients.
886 * The lists this could be linked into are:
887 * waiting_locks_list (protected by waiting_locks_spinlock),
888 * then if the lock timed out, it is moved to
889 * expired_lock_list for further processing.
891 struct list_head l_pending_chain;
894 * Set when lock is sent a blocking AST. Time in seconds when timeout
895 * is reached and client holding this lock could be evicted.
896 * This timeout could be further extended by e.g. certain IO activity
898 * \see ost_rw_prolong_locks
900 time64_t l_callback_timeout;
902 /** Local PID of process which created this lock. */
906 * Number of times blocking AST was sent for this lock.
907 * This is for debugging. Valid values are 0 and 1, if there is an
908 * attempt to send blocking AST more than once, an assertion would be
909 * hit. \see ldlm_work_bl_ast_lock
912 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
913 struct list_head l_bl_ast;
914 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
915 struct list_head l_cp_ast;
916 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
917 struct list_head l_rk_ast;
920 * Pointer to a conflicting lock that caused blocking AST to be sent
923 struct ldlm_lock *l_blocking_lock;
926 * Protected by lr_lock, linkages to "skip lists".
927 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
929 struct list_head l_sl_mode;
930 struct list_head l_sl_policy;
932 /** Reference tracking structure to debug leaked locks. */
933 struct lu_ref l_reference;
934 #if LUSTRE_TRACKS_LOCK_EXP_REFS
935 /* Debugging stuff for bug 20498, for tracking export references. */
936 /** number of export references taken */
938 /** link all locks referencing one export */
939 struct list_head l_exp_refs_link;
940 /** referenced export object */
941 struct obd_export *l_exp_refs_target;
944 * export blocking dlm lock list, protected by
945 * l_export->exp_bl_list_lock.
946 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
947 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
949 struct list_head l_exp_list;
952 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
953 #define l_transno l_client_cookie
955 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
956 * which is for server. */
957 #define l_slc_link l_rk_ast
959 #define HANDLE_MAP_SIZE ((LMV_MAX_STRIPE_COUNT + 7) >> 3)
961 struct lustre_handle_array {
962 unsigned int ha_count;
963 /* ha_map is used as bit flag to indicate handle is remote or local */
964 char ha_map[HANDLE_MAP_SIZE];
965 struct lustre_handle ha_handles[0];
969 * LDLM resource description.
970 * Basically, resource is a representation for a single object.
971 * Object has a name which is currently 4 64-bit integers. LDLM user is
972 * responsible for creation of a mapping between objects it wants to be
973 * protected and resource names.
975 * A resource can only hold locks of a single lock type, though there may be
976 * multiple ldlm_locks on a single resource, depending on the lock type and
977 * whether the locks are conflicting or not.
979 struct ldlm_resource {
980 struct ldlm_ns_bucket *lr_ns_bucket;
983 * List item for list in namespace hash.
984 * protected by ns_lock
986 struct hlist_node lr_hash;
988 /** Reference count for this resource */
989 atomic_t lr_refcount;
991 /** Spinlock to protect locks under this resource. */
995 * protected by lr_lock
997 /** List of locks in granted state */
998 struct list_head lr_granted;
1000 * List of locks that could not be granted due to conflicts and
1001 * that are waiting for conflicts to go away */
1002 struct list_head lr_waiting;
1005 /** Resource name */
1006 struct ldlm_res_id lr_name;
1009 * Interval trees (only for extent locks) for all modes of this resource
1011 struct ldlm_interval_tree *lr_itree;
1015 * When the resource was considered as contended,
1016 * used only on server side.
1018 time64_t lr_contention_time;
1020 * Associated inode, used only on client side.
1022 struct inode *lr_lvb_inode;
1025 /** Type of locks this resource can hold. Only one type per resource. */
1026 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1029 * Server-side-only lock value block elements.
1030 * To serialize lvbo_init.
1033 struct mutex lr_lvb_mutex;
1034 /** protected by lr_lock */
1036 /** is lvb initialized ? */
1037 bool lr_lvb_initialized;
1039 /** List of references to this resource. For debugging. */
1040 struct lu_ref lr_reference;
1043 static inline int ldlm_is_granted(struct ldlm_lock *lock)
1045 return lock->l_req_mode == lock->l_granted_mode;
1048 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1050 return lock->l_resource->lr_type == LDLM_IBITS &&
1051 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1054 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1056 return lock->l_resource->lr_type == LDLM_IBITS &&
1057 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1060 static inline char *
1061 ldlm_ns_name(struct ldlm_namespace *ns)
1066 static inline struct ldlm_namespace *
1067 ldlm_res_to_ns(struct ldlm_resource *res)
1069 return res->lr_ns_bucket->nsb_namespace;
1072 static inline struct ldlm_namespace *
1073 ldlm_lock_to_ns(struct ldlm_lock *lock)
1075 return ldlm_res_to_ns(lock->l_resource);
1078 static inline char *
1079 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1081 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1084 static inline struct adaptive_timeout *
1085 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1087 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1090 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1092 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1095 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1096 res->lr_lvb_initialized)
1099 mutex_lock(&res->lr_lvb_mutex);
1100 /* Did we lose the race? */
1101 if (res->lr_lvb_initialized) {
1102 mutex_unlock(&res->lr_lvb_mutex);
1105 rc = ns->ns_lvbo->lvbo_init(res);
1107 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1109 if (res->lr_lvb_data != NULL) {
1110 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1111 res->lr_lvb_data = NULL;
1113 res->lr_lvb_len = rc;
1115 res->lr_lvb_initialized = true;
1117 mutex_unlock(&res->lr_lvb_mutex);
1121 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1123 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1125 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1126 return ns->ns_lvbo->lvbo_size(lock);
1131 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int *len)
1133 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1136 if (ns->ns_lvbo != NULL) {
1137 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1138 /* init lvb now if not already */
1139 rc = ldlm_lvbo_init(lock->l_resource);
1141 CERROR("lock %p: delayed lvb init failed (rc %d)",
1145 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1150 struct ldlm_ast_work {
1151 struct ldlm_lock *w_lock;
1153 struct ldlm_lock_desc w_desc;
1154 struct list_head w_list;
1161 * Common ldlm_enqueue parameters
1163 struct ldlm_enqueue_info {
1164 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1165 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1166 void *ei_cb_bl; /** blocking lock callback */
1167 void *ei_cb_local_bl; /** blocking local lock callback */
1168 void *ei_cb_cp; /** lock completion callback */
1169 void *ei_cb_gl; /** lock glimpse callback */
1170 ldlm_created_callback ei_cb_created; /** lock created callback */
1171 void *ei_cbdata; /** Data to be passed into callbacks. */
1172 void *ei_namespace; /** lock namespace **/
1173 u64 ei_inodebits; /** lock inode bits **/
1174 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1177 #define ei_res_id ei_cb_gl
1179 extern struct obd_ops ldlm_obd_ops;
1181 extern char *ldlm_lockname[];
1182 extern char *ldlm_typename[];
1183 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1186 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1187 * For the cases where we do not have actual lock to print along
1188 * with a debugging message that is ldlm-related
1190 #define LDLM_DEBUG_NOLOCK(format, a...) \
1191 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1194 * Support function for lock information printing into debug logs.
1198 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1199 CFS_CHECK_STACK(msgdata, mask, cdls); \
1201 if (((mask) & D_CANTMASK) != 0 || \
1202 ((libcfs_debug & (mask)) != 0 && \
1203 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1204 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1207 void _ldlm_lock_debug(struct ldlm_lock *lock,
1208 struct libcfs_debug_msg_data *data,
1209 const char *fmt, ...)
1210 __attribute__ ((format (printf, 3, 4)));
1213 * Rate-limited version of lock printing function.
1215 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1216 static struct cfs_debug_limit_state _ldlm_cdls; \
1217 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1218 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1221 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1222 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1224 /** Non-rate-limited lock printing function for debugging purposes. */
1225 #define LDLM_DEBUG(lock, fmt, a...) do { \
1226 if (likely(lock != NULL)) { \
1227 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1228 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1229 "### " fmt , ##a); \
1231 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1234 #else /* !LIBCFS_DEBUG */
1235 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1236 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1237 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1241 * Three intentions can be used for the policy functions in
1242 * ldlm_processing_policy.
1244 * LDLM_PROCESS_RESCAN:
1246 * It's used when policy functions are called from ldlm_reprocess_queue() to
1247 * reprocess the wait list and try to grant locks, blocking ASTs
1248 * have already been sent in this situation, completion ASTs need be sent for
1249 * the locks being granted.
1251 * LDLM_PROCESS_ENQUEUE:
1253 * It's used when policy functions are called from ldlm_lock_enqueue() to
1254 * process the wait list for handling an enqueue request, blocking
1255 * ASTs have not been sent yet, so list of conflicting locks would be
1256 * collected and ASTs sent.
1258 * LDLM_PROCESS_RECOVERY:
1260 * It's used when policy functions are called from ldlm_reprocess_queue() to
1261 * reprocess the wait list when recovery done. In case of blocking
1262 * ASTs are lost before recovery, it needs not only to grant locks if
1263 * available, but also send blocking ASTs to the locks doesn't have AST sent
1264 * flag. Completion ASTs need be sent for the locks being granted.
1266 enum ldlm_process_intention {
1267 LDLM_PROCESS_RESCAN = 0,
1268 LDLM_PROCESS_ENQUEUE = 1,
1269 LDLM_PROCESS_RECOVERY = 2,
1272 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1273 enum ldlm_process_intention intention,
1274 enum ldlm_error *err,
1275 struct list_head *work_list);
1278 * Return values for lock iterators.
1279 * Also used during deciding of lock grants and cancellations.
1281 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1282 #define LDLM_ITER_STOP 2 /* stop iterating */
1284 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1285 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1287 /** \defgroup ldlm_iterator Lock iterators
1289 * LDLM provides for a way to iterate through every lock on a resource or
1290 * namespace or every resource in a namespace.
1292 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1294 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1296 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1297 ldlm_iterator_t iter, void *data);
1298 /** @} ldlm_iterator */
1300 int ldlm_replay_locks(struct obd_import *imp);
1303 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1306 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1308 struct ldlm_prolong_args {
1309 struct obd_export *lpa_export;
1310 struct ldlm_res_id lpa_resid;
1311 struct ldlm_extent lpa_extent;
1312 enum ldlm_mode lpa_mode;
1313 time64_t lpa_timeout;
1317 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1318 struct ldlm_prolong_args *arg);
1319 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1321 struct ldlm_callback_suite {
1322 ldlm_completion_callback lcs_completion;
1323 ldlm_blocking_callback lcs_blocking;
1324 ldlm_glimpse_callback lcs_glimpse;
1328 #ifdef HAVE_SERVER_SUPPORT
1329 /** \defgroup ldlm_srv_ast Server AST handlers
1330 * These are AST handlers used by server code.
1331 * Their property is that they are just preparing RPCs to be sent to clients.
1334 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1335 void *data, int flag);
1336 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1337 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1338 int ldlm_glimpse_locks(struct ldlm_resource *res,
1339 struct list_head *gl_work_list);
1340 /** @} ldlm_srv_ast */
1342 /** \defgroup ldlm_handlers Server LDLM handlers
1343 * These are handler functions that should be called by "frontends" such as
1344 * MDT or OST to pass through LDLM requests to LDLM for handling
1347 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1348 const struct ldlm_request *dlm_req,
1349 const struct ldlm_callback_suite *cbs);
1350 int ldlm_handle_convert0(struct ptlrpc_request *req,
1351 const struct ldlm_request *dlm_req);
1352 int ldlm_handle_cancel(struct ptlrpc_request *req);
1353 int ldlm_request_cancel(struct ptlrpc_request *req,
1354 const struct ldlm_request *dlm_req,
1355 int first, enum lustre_at_flags flags);
1356 /** @} ldlm_handlers */
1358 void ldlm_revoke_export_locks(struct obd_export *exp);
1359 time64_t ldlm_bl_timeout(struct ldlm_lock *lock);
1361 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1362 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, time64_t timeout);
1363 int ldlm_get_ref(void);
1364 void ldlm_put_ref(void);
1365 int ldlm_init_export(struct obd_export *exp);
1366 void ldlm_destroy_export(struct obd_export *exp);
1367 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1370 #ifdef HAVE_SERVER_SUPPORT
1371 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1373 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1374 void ldlm_lock2handle(const struct ldlm_lock *lock,
1375 struct lustre_handle *lockh);
1376 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1377 void ldlm_cancel_callback(struct ldlm_lock *);
1378 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1379 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1382 * Obtain a lock reference by its handle.
1384 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1386 return __ldlm_handle2lock(h, 0);
1389 #define LDLM_LOCK_REF_DEL(lock) \
1390 lu_ref_del(&lock->l_reference, "handle", current)
1392 static inline struct ldlm_lock *
1393 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1395 struct ldlm_lock *lock;
1397 lock = __ldlm_handle2lock(h, flags);
1399 LDLM_LOCK_REF_DEL(lock);
1404 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1405 * data from request \a r
1407 static inline int ldlm_lvbo_update(struct ldlm_resource *res,
1408 struct ldlm_lock *lock,
1409 struct ptlrpc_request *req, int increase)
1411 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1414 /* delayed lvb init may be required */
1415 rc = ldlm_lvbo_init(res);
1417 CERROR("delayed lvb init failed (rc %d)\n", rc);
1421 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1422 return ns->ns_lvbo->lvbo_update(res, lock, req, increase);
1427 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1428 struct ptlrpc_request *req,
1431 return ldlm_lvbo_update(res, NULL, req, increase);
1434 int is_granted_or_cancelled_nolock(struct ldlm_lock *lock);
1436 int ldlm_error2errno(enum ldlm_error error);
1437 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1438 * confuses user-space. */
1439 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1440 void ldlm_dump_export_locks(struct obd_export *exp);
1444 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1445 * __ldlm_handle2lock().
1447 #define LDLM_LOCK_PUT(lock) \
1449 LDLM_LOCK_REF_DEL(lock); \
1450 /*LDLM_DEBUG((lock), "put");*/ \
1451 ldlm_lock_put(lock); \
1455 * Release a lock reference obtained by some other means (see
1458 #define LDLM_LOCK_RELEASE(lock) \
1460 /*LDLM_DEBUG((lock), "put");*/ \
1461 ldlm_lock_put(lock); \
1464 #define LDLM_LOCK_GET(lock) \
1466 ldlm_lock_get(lock); \
1467 /*LDLM_DEBUG((lock), "get");*/ \
1471 #define ldlm_lock_list_put(head, member, count) \
1473 struct ldlm_lock *_lock, *_next; \
1475 list_for_each_entry_safe(_lock, _next, head, member) { \
1478 list_del_init(&_lock->member); \
1479 LDLM_LOCK_RELEASE(_lock); \
1484 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1485 void ldlm_lock_put(struct ldlm_lock *lock);
1486 void ldlm_lock_destroy(struct ldlm_lock *lock);
1487 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1488 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1489 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1490 enum ldlm_mode mode);
1491 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1492 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1493 enum ldlm_mode mode);
1494 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1495 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1496 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1497 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1498 enum ldlm_mode ldlm_lock_match_with_skip(struct ldlm_namespace *ns,
1499 __u64 flags, __u64 skip_flags,
1500 const struct ldlm_res_id *res_id,
1501 enum ldlm_type type,
1502 union ldlm_policy_data *policy,
1503 enum ldlm_mode mode,
1504 struct lustre_handle *lh,
1506 static inline enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns,
1508 const struct ldlm_res_id *res_id,
1509 enum ldlm_type type,
1510 union ldlm_policy_data *policy,
1511 enum ldlm_mode mode,
1512 struct lustre_handle *lh,
1515 return ldlm_lock_match_with_skip(ns, flags, 0, res_id, type, policy,
1519 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1521 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1522 void ldlm_lock_cancel(struct ldlm_lock *lock);
1523 void ldlm_reprocess_all(struct ldlm_resource *res);
1524 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1525 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1526 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1529 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1530 enum ldlm_side client,
1531 enum ldlm_appetite apt,
1532 enum ldlm_ns_type ns_type);
1533 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1534 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1535 struct obd_import *imp,
1537 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1538 void ldlm_namespace_free(struct ldlm_namespace *ns,
1539 struct obd_import *imp, int force);
1540 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1541 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1542 enum ldlm_side client);
1543 void ldlm_namespace_get(struct ldlm_namespace *ns);
1544 void ldlm_namespace_put(struct ldlm_namespace *ns);
1546 int ldlm_debugfs_setup(void);
1547 void ldlm_debugfs_cleanup(void);
1549 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1550 struct lprocfs_stats *srv_stats)
1552 int lock_type = 0, op = 0;
1554 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1556 switch (lock_type) {
1558 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1561 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1564 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1567 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1575 lprocfs_counter_incr(srv_stats, op);
1580 /* resource.c - internal */
1581 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1582 struct ldlm_resource *parent,
1583 const struct ldlm_res_id *,
1584 enum ldlm_type type, int create);
1585 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1586 int ldlm_resource_putref(struct ldlm_resource *res);
1587 void ldlm_resource_add_lock(struct ldlm_resource *res,
1588 struct list_head *head,
1589 struct ldlm_lock *lock);
1590 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1591 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1592 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1593 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1594 void ldlm_resource_dump(int level, struct ldlm_resource *);
1595 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1596 const struct ldlm_res_id *);
1598 #define LDLM_RESOURCE_ADDREF(res) do { \
1599 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1602 #define LDLM_RESOURCE_DELREF(res) do { \
1603 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1606 /* ldlm_request.c */
1607 int ldlm_expired_completion_wait(void *data);
1608 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1609 * These AST handlers are typically used for server-side local locks and are
1610 * also used by client-side lock handlers to perform minimum level base
1613 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1614 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1615 void *data, int flag);
1616 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1617 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1618 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1619 /** @} ldlm_local_ast */
1621 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1622 * These are typically used by client and server (*_local versions)
1623 * to obtain and release locks.
1625 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1626 struct ldlm_enqueue_info *einfo,
1627 const struct ldlm_res_id *res_id,
1628 union ldlm_policy_data const *policy, __u64 *flags,
1629 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1630 struct lustre_handle *lockh, int async);
1631 int ldlm_prep_enqueue_req(struct obd_export *exp,
1632 struct ptlrpc_request *req,
1633 struct list_head *cancels,
1635 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1636 int version, int opc, int canceloff,
1637 struct list_head *cancels, int count);
1639 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1640 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1641 const struct ldlm_request *dlm_req,
1642 const struct ldlm_callback_suite *cbs);
1643 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1644 enum ldlm_type type, __u8 with_policy,
1645 enum ldlm_mode mode, __u64 *flags, void *lvb,
1647 const struct lustre_handle *lockh, int rc);
1648 int ldlm_cli_enqueue_local(const struct lu_env *env,
1649 struct ldlm_namespace *ns,
1650 const struct ldlm_res_id *res_id,
1651 enum ldlm_type type, union ldlm_policy_data *policy,
1652 enum ldlm_mode mode, __u64 *flags,
1653 ldlm_blocking_callback blocking,
1654 ldlm_completion_callback completion,
1655 ldlm_glimpse_callback glimpse,
1656 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1657 const __u64 *client_cookie,
1658 struct lustre_handle *lockh);
1659 int ldlm_cli_convert(struct ldlm_lock *lock, __u32 *flags);
1660 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1661 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1662 enum ldlm_cancel_flags cancel_flags);
1663 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1664 enum ldlm_cancel_flags flags, void *opaque);
1665 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1666 const struct ldlm_res_id *res_id,
1667 union ldlm_policy_data *policy,
1668 enum ldlm_mode mode,
1669 enum ldlm_cancel_flags flags, void *opaque);
1670 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1671 int count, enum ldlm_cancel_flags flags);
1672 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1673 struct list_head *cancels,
1674 union ldlm_policy_data *policy,
1675 enum ldlm_mode mode, __u64 lock_flags,
1676 enum ldlm_cancel_flags cancel_flags,
1678 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1679 enum ldlm_cancel_flags flags);
1680 int ldlm_cli_cancel_list(struct list_head *head, int count,
1681 struct ptlrpc_request *req,
1682 enum ldlm_cancel_flags flags);
1684 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);
1685 int ldlm_cli_dropbits(struct ldlm_lock *lock, __u64 drop_bits);
1686 int ldlm_cli_dropbits_list(struct list_head *converts, __u64 drop_bits);
1688 /** @} ldlm_cli_api */
1691 /* This has to be here because recursive inclusion sucks. */
1692 int intent_disposition(struct ldlm_reply *rep, int flag);
1693 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1696 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1697 * than one lock_res is dead-lock safe.
1699 enum lock_res_type {
1704 /** Lock resource. */
1705 static inline void lock_res(struct ldlm_resource *res)
1707 spin_lock(&res->lr_lock);
1710 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1711 static inline void lock_res_nested(struct ldlm_resource *res,
1712 enum lock_res_type mode)
1714 spin_lock_nested(&res->lr_lock, mode);
1717 /** Unlock resource. */
1718 static inline void unlock_res(struct ldlm_resource *res)
1720 spin_unlock(&res->lr_lock);
1723 /** Check if resource is already locked, assert if not. */
1724 static inline void check_res_locked(struct ldlm_resource *res)
1726 assert_spin_locked(&res->lr_lock);
1729 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1730 void unlock_res_and_lock(struct ldlm_lock *lock);
1733 /** \defgroup ldlm_pools Various LDLM pool related functions
1734 * There are not used outside of ldlm.
1737 int ldlm_pools_init(void);
1738 void ldlm_pools_fini(void);
1740 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1741 int idx, enum ldlm_side client);
1742 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1743 void ldlm_pool_fini(struct ldlm_pool *pl);
1744 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1745 time64_t ldlm_pool_recalc(struct ldlm_pool *pl);
1746 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1747 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1748 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1749 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1750 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1751 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1752 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1753 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1754 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1757 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1758 const struct ldlm_extent *ex2)
1760 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1763 /* check if @ex1 contains @ex2 */
1764 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1765 const struct ldlm_extent *ex2)
1767 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1770 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);