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)(struct ldlm_namespace *, struct ldlm_lock **,
276 void *req_cookie, enum ldlm_mode mode,
277 __u64 flags, void *data);
279 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
283 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
284 * be associated with an LDLM lock and transferred from client to server and
287 * Currently LVBs are used by:
288 * - OSC-OST code to maintain current object size/times
289 * - layout lock code to return the layout when the layout lock is granted
291 * To ensure delayed LVB initialization, it is highly recommended to use the set
292 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
294 struct ldlm_valblock_ops {
295 int (*lvbo_init)(struct ldlm_resource *res);
296 int (*lvbo_update)(struct ldlm_resource *res, struct ldlm_lock *lock,
297 struct ptlrpc_request *r, int increase);
298 int (*lvbo_free)(struct ldlm_resource *res);
299 /* Return size of lvb data appropriate RPC size can be reserved */
300 int (*lvbo_size)(struct ldlm_lock *lock);
301 /* Called to fill in lvb data to RPC buffer @buf */
302 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
306 * LDLM pools related, type of lock pool in the namespace.
307 * Greedy means release cached locks aggressively
310 LDLM_NAMESPACE_GREEDY = 1 << 0,
311 LDLM_NAMESPACE_MODEST = 1 << 1
315 * Default values for the "max_nolock_size", "contention_time" and
316 * "contended_locks" namespace tunables.
318 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
319 #define NS_DEFAULT_CONTENTION_SECONDS 2
320 #define NS_DEFAULT_CONTENDED_LOCKS 32
322 struct ldlm_ns_bucket {
323 /** back pointer to namespace */
324 struct ldlm_namespace *nsb_namespace;
326 * Estimated lock callback time. Used by adaptive timeout code to
327 * avoid spurious client evictions due to unresponsiveness when in
328 * fact the network or overall system load is at fault
330 struct adaptive_timeout nsb_at_estimate;
332 * Which res in the bucket should we start with the reclaim.
334 int nsb_reclaim_start;
338 /** LDLM namespace lock stats */
344 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
345 LDLM_NS_TYPE_MDC, /**< MDC namespace */
346 LDLM_NS_TYPE_MDT, /**< MDT namespace */
347 LDLM_NS_TYPE_OSC, /**< OSC namespace */
348 LDLM_NS_TYPE_OST, /**< OST namespace */
349 LDLM_NS_TYPE_MGC, /**< MGC namespace */
350 LDLM_NS_TYPE_MGT, /**< MGT namespace */
356 * Namespace serves to contain locks related to a particular service.
357 * There are two kinds of namespaces:
358 * - Server namespace has knowledge of all locks and is therefore authoritative
359 * to make decisions like what locks could be granted and what conflicts
360 * exist during new lock enqueue.
361 * - Client namespace only has limited knowledge about locks in the namespace,
362 * only seeing locks held by the client.
364 * Every Lustre service has one server namespace present on the server serving
365 * that service. Every client connected to the service has a client namespace
367 * Every lock obtained by client in that namespace is actually represented by
368 * two in-memory locks. One on the server and one on the client. The locks are
369 * linked by a special cookie by which one node can tell to the other which lock
370 * it actually means during communications. Such locks are called remote locks.
371 * The locks held by server only without any reference to a client are called
374 struct ldlm_namespace {
375 /** Backward link to OBD, required for LDLM pool to store new SLV. */
376 struct obd_device *ns_obd;
378 /** Flag indicating if namespace is on client instead of server */
379 enum ldlm_side ns_client;
381 /** Resource hash table for namespace. */
382 struct cfs_hash *ns_rs_hash;
387 /** big refcount (by bucket) */
391 * Namespace connect flags supported by server (may be changed via
392 * /proc, LRU resize may be disabled/enabled).
394 __u64 ns_connect_flags;
396 /** Client side original connect flags supported by server. */
397 __u64 ns_orig_connect_flags;
399 /* namespace debugfs dir entry */
400 struct dentry *ns_debugfs_entry;
403 * Position in global namespace list linking all namespaces on
406 struct list_head ns_list_chain;
409 * List of unused locks for this namespace. This list is also called
411 * Unused locks are locks with zero reader/writer reference counts.
412 * This list is only used on clients for lock caching purposes.
413 * When we want to release some locks voluntarily or if server wants
414 * us to release some locks due to e.g. memory pressure, we take locks
415 * to release from the head of this list.
416 * Locks are linked via l_lru field in \see struct ldlm_lock.
418 struct list_head ns_unused_list;
419 /** Number of locks in the LRU list above */
423 * Maximum number of locks permitted in the LRU. If 0, means locks
424 * are managed by pools and there is no preset limit, rather it is all
425 * controlled by available memory on this client and on server.
427 unsigned int ns_max_unused;
429 /** Maximum allowed age (last used time) for locks in the LRU */
433 * Server only: number of times we evicted clients due to lack of reply
436 unsigned int ns_timeouts;
438 * Number of seconds since the file change time after which the
439 * MDT will return an UPDATE lock along with a LOOKUP lock.
440 * This allows the client to start caching negative dentries
441 * for a directory and may save an RPC for a later stat.
443 time64_t ns_ctime_age_limit;
445 * Number of seconds since the lock was last used. The client may
446 * cancel the lock limited by this age and flush related data if
447 * any other client shows interest in it doing glimpse request.
448 * This allows to cache stat data locally for such files early.
450 time64_t ns_dirty_age_limit;
452 * Used to rate-limit ldlm_namespace_dump calls.
453 * \see ldlm_namespace_dump. Increased by 10 seconds every time
456 time64_t ns_next_dump;
458 /** "policy" function that does actual lock conflict determination */
459 ldlm_res_policy ns_policy;
462 * LVB operations for this namespace.
463 * \see struct ldlm_valblock_ops
465 struct ldlm_valblock_ops *ns_lvbo;
468 * Used by filter code to store pointer to OBD of the service.
469 * Should be dropped in favor of \a ns_obd
474 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
475 * a resource is removed.
477 wait_queue_head_t ns_waitq;
478 /** LDLM pool structure for this namespace */
479 struct ldlm_pool ns_pool;
480 /** Definition of how eagerly unused locks will be released from LRU */
481 enum ldlm_appetite ns_appetite;
484 * If more than \a ns_contended_locks are found, the resource is
485 * considered to be contended. Lock enqueues might specify that no
486 * contended locks should be granted
488 unsigned ns_contended_locks;
491 * The resources in this namespace remember contended state during
492 * \a ns_contention_time, in seconds.
494 time64_t ns_contention_time;
497 * Limit size of contended extent locks, in bytes.
498 * If extended lock is requested for more then this many bytes and
499 * caller instructs us not to grant contended locks, we would disregard
502 unsigned ns_max_nolock_size;
504 /** Limit of parallel AST RPC count. */
505 unsigned ns_max_parallel_ast;
508 * Callback to check if a lock is good to be canceled by ELC or
511 ldlm_cancel_cbt ns_cancel;
513 /** LDLM lock stats */
514 struct lprocfs_stats *ns_stats;
517 * Flag to indicate namespace is being freed. Used to determine if
518 * recalculation of LDLM pool statistics should be skipped.
520 unsigned ns_stopping:1;
523 * Which bucket should we start with the lock reclaim.
525 int ns_reclaim_start;
527 struct kobject ns_kobj; /* sysfs object */
528 struct completion ns_kobj_unregister;
532 * Returns 1 if namespace \a ns is a client namespace.
534 static inline int ns_is_client(struct ldlm_namespace *ns)
537 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
538 LDLM_NAMESPACE_SERVER)));
539 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
540 ns->ns_client == LDLM_NAMESPACE_SERVER);
541 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
545 * Returns 1 if namespace \a ns is a server namespace.
547 static inline int ns_is_server(struct ldlm_namespace *ns)
550 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
551 LDLM_NAMESPACE_SERVER)));
552 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
553 ns->ns_client == LDLM_NAMESPACE_SERVER);
554 return ns->ns_client == LDLM_NAMESPACE_SERVER;
558 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
560 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
563 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
567 * Returns 1 if this namespace supports lru_resize.
569 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
572 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
575 static inline void ns_register_cancel(struct ldlm_namespace *ns,
584 /** Type for blocking callback function of a lock. */
585 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
586 struct ldlm_lock_desc *new, void *data,
588 /** Type for completion callback function of a lock. */
589 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
591 /** Type for glimpse callback function of a lock. */
592 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
594 /** Work list for sending GL ASTs to multiple locks. */
595 struct ldlm_glimpse_work {
596 struct ldlm_lock *gl_lock; /* lock to glimpse */
597 struct list_head gl_list; /* linkage to other gl work structs */
598 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
599 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
600 * glimpse callback request */
601 ptlrpc_interpterer_t gl_interpret_reply;
602 void *gl_interpret_data;
605 struct ldlm_cb_set_arg {
606 struct ptlrpc_request_set *set;
607 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
609 struct list_head *list;
610 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
611 ptlrpc_interpterer_t gl_interpret_reply;
612 void *gl_interpret_data;
615 struct ldlm_cb_async_args {
616 struct ldlm_cb_set_arg *ca_set_arg;
617 struct ldlm_lock *ca_lock;
620 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
621 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
623 /** Interval node data for each LDLM_EXTENT lock. */
624 struct ldlm_interval {
625 struct interval_node li_node; /* node for tree management */
626 struct list_head li_group; /* the locks which have the same
627 * policy - group of the policy */
629 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
632 * Interval tree for extent locks.
633 * The interval tree must be accessed under the resource lock.
634 * Interval trees are used for granted extent locks to speed up conflicts
635 * lookup. See ldlm/interval_tree.c for more details.
637 struct ldlm_interval_tree {
640 enum ldlm_mode lit_mode; /* lock mode */
641 struct interval_node *lit_root; /* actual ldlm_interval */
644 /** Whether to track references to exports by LDLM locks. */
645 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
648 enum ldlm_cancel_flags {
649 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
650 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
651 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
658 __u64 blocking_owner;
659 struct obd_export *blocking_export;
660 atomic_t blocking_refs;
664 union ldlm_policy_data {
665 struct ldlm_extent l_extent;
666 struct ldlm_flock l_flock;
667 struct ldlm_inodebits l_inodebits;
670 void ldlm_convert_policy_to_wire(enum ldlm_type type,
671 const union ldlm_policy_data *lpolicy,
672 union ldlm_wire_policy_data *wpolicy);
673 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
674 const union ldlm_wire_policy_data *wpolicy,
675 union ldlm_policy_data *lpolicy);
685 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
687 #define LDLM_GID_ANY ((__u64)-1)
690 * LDLM lock structure
692 * Represents a single LDLM lock and its state in memory. Each lock is
693 * associated with a single ldlm_resource, the object which is being
694 * locked. There may be multiple ldlm_locks on a single resource,
695 * depending on the lock type and whether the locks are conflicting or
701 * When remote side wants to tell us about a lock, they address
702 * it by this opaque handle. The handle does not hold a
703 * reference on the ldlm_lock, so it can be safely passed to
704 * other threads or nodes. When the lock needs to be accessed
705 * from the handle, it is looked up again in the lock table, and
706 * may no longer exist.
708 * Must be first in the structure.
710 struct portals_handle l_handle;
712 * Lock reference count.
713 * This is how many users have pointers to actual structure, so that
714 * we do not accidentally free lock structure that is in use.
718 * Internal spinlock protects l_resource. We should hold this lock
719 * first before taking res_lock.
723 * Pointer to actual resource this lock is in.
724 * ldlm_lock_change_resource() can change this.
726 struct ldlm_resource *l_resource;
728 * List item for client side LRU list.
729 * Protected by ns_lock in struct ldlm_namespace.
731 struct list_head l_lru;
733 * Linkage to resource's lock queues according to current lock state.
734 * (could be granted or waiting)
735 * Protected by lr_lock in struct ldlm_resource.
737 struct list_head l_res_link;
739 * Tree node for ldlm_extent.
741 struct ldlm_interval *l_tree_node;
743 * Per export hash of locks.
744 * Protected by per-bucket exp->exp_lock_hash locks.
746 struct hlist_node l_exp_hash;
748 * Per export hash of flock locks.
749 * Protected by per-bucket exp->exp_flock_hash locks.
751 struct hlist_node l_exp_flock_hash;
754 * Protected by lr_lock.
756 enum ldlm_mode l_req_mode;
758 * Granted mode, also protected by lr_lock.
760 enum ldlm_mode l_granted_mode;
761 /** Lock completion handler pointer. Called when lock is granted. */
762 ldlm_completion_callback l_completion_ast;
764 * Lock blocking AST handler pointer.
765 * It plays two roles:
766 * - as a notification of an attempt to queue a conflicting lock (once)
767 * - as a notification when the lock is being cancelled.
769 * As such it's typically called twice: once for the initial conflict
770 * and then once more when the last user went away and the lock is
771 * cancelled (could happen recursively).
773 ldlm_blocking_callback l_blocking_ast;
775 * Lock glimpse handler.
776 * Glimpse handler is used to obtain LVB updates from a client by
779 ldlm_glimpse_callback l_glimpse_ast;
783 * This is a pointer to actual client export for locks that were granted
784 * to clients. Used server-side.
786 struct obd_export *l_export;
788 * Lock connection export.
789 * Pointer to server export on a client.
791 struct obd_export *l_conn_export;
794 * Remote lock handle.
795 * If the lock is remote, this is the handle of the other side lock
798 struct lustre_handle l_remote_handle;
801 * Representation of private data specific for a lock type.
802 * Examples are: extent range for extent lock or bitmask for ibits locks
804 union ldlm_policy_data l_policy_data;
807 * Lock state flags. Protected by lr_lock.
808 * \see lustre_dlm_flags.h where the bits are defined.
813 * Lock r/w usage counters.
814 * Protected by lr_lock.
819 * If the lock is granted, a process sleeps on this waitq to learn when
820 * it's no longer in use. If the lock is not granted, a process sleeps
821 * on this waitq to learn when it becomes granted.
823 wait_queue_head_t l_waitq;
826 * Seconds. It will be updated if there is any activity related to
827 * the lock, e.g. enqueue the lock or send blocking AST.
829 time64_t l_last_activity;
832 * Time, in nanoseconds, last used by e.g. being matched by lock match.
836 /** Originally requested extent for the extent lock. */
837 struct ldlm_extent l_req_extent;
840 * Client-side-only members.
843 enum lvb_type l_lvb_type;
846 * Temporary storage for a LVB received during an enqueue operation.
847 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
852 /** Private storage for lock user. Opaque to LDLM. */
854 /* separate ost_lvb used mostly by Data-on-MDT for now.
855 * It is introduced to don't mix with layout lock data. */
856 struct ost_lvb l_ost_lvb;
858 * Server-side-only members.
862 * Connection cookie for the client originating the operation.
863 * Used by Commit on Share (COS) code. Currently only used for
864 * inodebits locks on MDS.
866 __u64 l_client_cookie;
869 * List item for locks waiting for cancellation from clients.
870 * The lists this could be linked into are:
871 * waiting_locks_list (protected by waiting_locks_spinlock),
872 * then if the lock timed out, it is moved to
873 * expired_lock_list for further processing.
875 struct list_head l_pending_chain;
878 * Set when lock is sent a blocking AST. Time in seconds when timeout
879 * is reached and client holding this lock could be evicted.
880 * This timeout could be further extended by e.g. certain IO activity
882 * \see ost_rw_prolong_locks
884 time64_t l_callback_timeout;
886 /** Local PID of process which created this lock. */
890 * Number of times blocking AST was sent for this lock.
891 * This is for debugging. Valid values are 0 and 1, if there is an
892 * attempt to send blocking AST more than once, an assertion would be
893 * hit. \see ldlm_work_bl_ast_lock
896 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
897 struct list_head l_bl_ast;
898 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
899 struct list_head l_cp_ast;
900 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
901 struct list_head l_rk_ast;
904 * Pointer to a conflicting lock that caused blocking AST to be sent
907 struct ldlm_lock *l_blocking_lock;
910 * Protected by lr_lock, linkages to "skip lists".
911 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
913 struct list_head l_sl_mode;
914 struct list_head l_sl_policy;
916 /** Reference tracking structure to debug leaked locks. */
917 struct lu_ref l_reference;
918 #if LUSTRE_TRACKS_LOCK_EXP_REFS
919 /* Debugging stuff for bug 20498, for tracking export references. */
920 /** number of export references taken */
922 /** link all locks referencing one export */
923 struct list_head l_exp_refs_link;
924 /** referenced export object */
925 struct obd_export *l_exp_refs_target;
928 * export blocking dlm lock list, protected by
929 * l_export->exp_bl_list_lock.
930 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
931 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
933 struct list_head l_exp_list;
936 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
937 #define l_transno l_client_cookie
939 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
940 * which is for server. */
941 #define l_slc_link l_rk_ast
944 * LDLM resource description.
945 * Basically, resource is a representation for a single object.
946 * Object has a name which is currently 4 64-bit integers. LDLM user is
947 * responsible for creation of a mapping between objects it wants to be
948 * protected and resource names.
950 * A resource can only hold locks of a single lock type, though there may be
951 * multiple ldlm_locks on a single resource, depending on the lock type and
952 * whether the locks are conflicting or not.
954 struct ldlm_resource {
955 struct ldlm_ns_bucket *lr_ns_bucket;
958 * List item for list in namespace hash.
959 * protected by ns_lock
961 struct hlist_node lr_hash;
963 /** Reference count for this resource */
964 atomic_t lr_refcount;
966 /** Spinlock to protect locks under this resource. */
970 * protected by lr_lock
972 /** List of locks in granted state */
973 struct list_head lr_granted;
975 * List of locks that could not be granted due to conflicts and
976 * that are waiting for conflicts to go away */
977 struct list_head lr_waiting;
981 struct ldlm_res_id lr_name;
984 * Interval trees (only for extent locks) for all modes of this resource
986 struct ldlm_interval_tree *lr_itree;
990 * When the resource was considered as contended,
991 * used only on server side.
993 time64_t lr_contention_time;
995 * Associated inode, used only on client side.
997 struct inode *lr_lvb_inode;
1000 /** Type of locks this resource can hold. Only one type per resource. */
1001 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1004 * Server-side-only lock value block elements.
1005 * To serialize lvbo_init.
1008 struct mutex lr_lvb_mutex;
1009 /** protected by lr_lock */
1011 /** is lvb initialized ? */
1012 bool lr_lvb_initialized;
1014 /** List of references to this resource. For debugging. */
1015 struct lu_ref lr_reference;
1018 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1020 return lock->l_resource->lr_type == LDLM_IBITS &&
1021 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1024 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1026 return lock->l_resource->lr_type == LDLM_IBITS &&
1027 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1030 static inline char *
1031 ldlm_ns_name(struct ldlm_namespace *ns)
1033 return ns->ns_rs_hash->hs_name;
1036 static inline struct ldlm_namespace *
1037 ldlm_res_to_ns(struct ldlm_resource *res)
1039 return res->lr_ns_bucket->nsb_namespace;
1042 static inline struct ldlm_namespace *
1043 ldlm_lock_to_ns(struct ldlm_lock *lock)
1045 return ldlm_res_to_ns(lock->l_resource);
1048 static inline char *
1049 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1051 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1054 static inline struct adaptive_timeout *
1055 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1057 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1060 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1062 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1065 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1066 res->lr_lvb_initialized)
1069 mutex_lock(&res->lr_lvb_mutex);
1070 /* Did we lose the race? */
1071 if (res->lr_lvb_initialized) {
1072 mutex_unlock(&res->lr_lvb_mutex);
1075 rc = ns->ns_lvbo->lvbo_init(res);
1077 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1079 if (res->lr_lvb_data != NULL) {
1080 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1081 res->lr_lvb_data = NULL;
1083 res->lr_lvb_len = rc;
1085 res->lr_lvb_initialized = true;
1087 mutex_unlock(&res->lr_lvb_mutex);
1091 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1093 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1095 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1096 return ns->ns_lvbo->lvbo_size(lock);
1101 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1103 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1106 if (ns->ns_lvbo != NULL) {
1107 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1108 /* init lvb now if not already */
1109 rc = ldlm_lvbo_init(lock->l_resource);
1111 CERROR("lock %p: delayed lvb init failed (rc %d)",
1115 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1120 struct ldlm_ast_work {
1121 struct ldlm_lock *w_lock;
1123 struct ldlm_lock_desc w_desc;
1124 struct list_head w_list;
1131 * Common ldlm_enqueue parameters
1133 struct ldlm_enqueue_info {
1134 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1135 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1136 void *ei_cb_bl; /** blocking lock callback */
1137 void *ei_cb_local_bl; /** blocking local lock callback */
1138 void *ei_cb_cp; /** lock completion callback */
1139 void *ei_cb_gl; /** lock glimpse callback */
1140 void *ei_cbdata; /** Data to be passed into callbacks. */
1141 void *ei_namespace; /** lock namespace **/
1142 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1145 #define ei_res_id ei_cb_gl
1147 extern struct obd_ops ldlm_obd_ops;
1149 extern char *ldlm_lockname[];
1150 extern char *ldlm_typename[];
1151 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1154 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1155 * For the cases where we do not have actual lock to print along
1156 * with a debugging message that is ldlm-related
1158 #define LDLM_DEBUG_NOLOCK(format, a...) \
1159 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1162 * Support function for lock information printing into debug logs.
1166 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1167 CFS_CHECK_STACK(msgdata, mask, cdls); \
1169 if (((mask) & D_CANTMASK) != 0 || \
1170 ((libcfs_debug & (mask)) != 0 && \
1171 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1172 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1175 void _ldlm_lock_debug(struct ldlm_lock *lock,
1176 struct libcfs_debug_msg_data *data,
1177 const char *fmt, ...)
1178 __attribute__ ((format (printf, 3, 4)));
1181 * Rate-limited version of lock printing function.
1183 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1184 static struct cfs_debug_limit_state _ldlm_cdls; \
1185 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1186 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1189 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1190 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1192 /** Non-rate-limited lock printing function for debugging purposes. */
1193 #define LDLM_DEBUG(lock, fmt, a...) do { \
1194 if (likely(lock != NULL)) { \
1195 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1196 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1197 "### " fmt , ##a); \
1199 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1202 #else /* !LIBCFS_DEBUG */
1203 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1204 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1205 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1209 * Three intentions can be used for the policy functions in
1210 * ldlm_processing_policy.
1212 * LDLM_PROCESS_RESCAN:
1214 * It's used when policy functions are called from ldlm_reprocess_queue() to
1215 * reprocess the wait list and try to grant locks, blocking ASTs
1216 * have already been sent in this situation, completion ASTs need be sent for
1217 * the locks being granted.
1219 * LDLM_PROCESS_ENQUEUE:
1221 * It's used when policy functions are called from ldlm_lock_enqueue() to
1222 * process the wait list for handling an enqueue request, blocking
1223 * ASTs have not been sent yet, so list of conflicting locks would be
1224 * collected and ASTs sent.
1226 * LDLM_PROCESS_RECOVERY:
1228 * It's used when policy functions are called from ldlm_reprocess_queue() to
1229 * reprocess the wait list when recovery done. In case of blocking
1230 * ASTs are lost before recovery, it needs not only to grant locks if
1231 * available, but also send blocking ASTs to the locks doesn't have AST sent
1232 * flag. Completion ASTs need be sent for the locks being granted.
1234 enum ldlm_process_intention {
1235 LDLM_PROCESS_RESCAN = 0,
1236 LDLM_PROCESS_ENQUEUE = 1,
1237 LDLM_PROCESS_RECOVERY = 2,
1240 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1241 enum ldlm_process_intention intention,
1242 enum ldlm_error *err,
1243 struct list_head *work_list);
1246 * Return values for lock iterators.
1247 * Also used during deciding of lock grants and cancellations.
1249 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1250 #define LDLM_ITER_STOP 2 /* stop iterating */
1252 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1253 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1255 /** \defgroup ldlm_iterator Lock iterators
1257 * LDLM provides for a way to iterate through every lock on a resource or
1258 * namespace or every resource in a namespace.
1260 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1262 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1264 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1265 ldlm_iterator_t iter, void *data);
1266 /** @} ldlm_iterator */
1268 int ldlm_replay_locks(struct obd_import *imp);
1271 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1274 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1276 struct ldlm_prolong_args {
1277 struct obd_export *lpa_export;
1278 struct ldlm_res_id lpa_resid;
1279 struct ldlm_extent lpa_extent;
1280 enum ldlm_mode lpa_mode;
1281 time64_t lpa_timeout;
1285 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1286 struct ldlm_prolong_args *arg);
1287 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1289 struct ldlm_callback_suite {
1290 ldlm_completion_callback lcs_completion;
1291 ldlm_blocking_callback lcs_blocking;
1292 ldlm_glimpse_callback lcs_glimpse;
1296 #ifdef HAVE_SERVER_SUPPORT
1297 /** \defgroup ldlm_srv_ast Server AST handlers
1298 * These are AST handlers used by server code.
1299 * Their property is that they are just preparing RPCs to be sent to clients.
1302 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1303 void *data, int flag);
1304 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1305 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1306 int ldlm_glimpse_locks(struct ldlm_resource *res,
1307 struct list_head *gl_work_list);
1308 /** @} ldlm_srv_ast */
1310 /** \defgroup ldlm_handlers Server LDLM handlers
1311 * These are handler functions that should be called by "frontends" such as
1312 * MDT or OST to pass through LDLM requests to LDLM for handling
1315 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1316 const struct ldlm_request *dlm_req,
1317 const struct ldlm_callback_suite *cbs);
1318 int ldlm_handle_convert0(struct ptlrpc_request *req,
1319 const struct ldlm_request *dlm_req);
1320 int ldlm_handle_cancel(struct ptlrpc_request *req);
1321 int ldlm_request_cancel(struct ptlrpc_request *req,
1322 const struct ldlm_request *dlm_req,
1323 int first, enum lustre_at_flags flags);
1324 /** @} ldlm_handlers */
1326 void ldlm_revoke_export_locks(struct obd_export *exp);
1327 time64_t ldlm_bl_timeout(struct ldlm_lock *lock);
1329 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1330 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, time64_t timeout);
1331 int ldlm_get_ref(void);
1332 void ldlm_put_ref(void);
1333 int ldlm_init_export(struct obd_export *exp);
1334 void ldlm_destroy_export(struct obd_export *exp);
1335 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1338 #ifdef HAVE_SERVER_SUPPORT
1339 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1341 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1342 void ldlm_lock2handle(const struct ldlm_lock *lock,
1343 struct lustre_handle *lockh);
1344 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1345 void ldlm_cancel_callback(struct ldlm_lock *);
1346 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1347 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1350 * Obtain a lock reference by its handle.
1352 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1354 return __ldlm_handle2lock(h, 0);
1357 #define LDLM_LOCK_REF_DEL(lock) \
1358 lu_ref_del(&lock->l_reference, "handle", current)
1360 static inline struct ldlm_lock *
1361 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1363 struct ldlm_lock *lock;
1365 lock = __ldlm_handle2lock(h, flags);
1367 LDLM_LOCK_REF_DEL(lock);
1372 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1373 * data from request \a r
1375 static inline int ldlm_lvbo_update(struct ldlm_resource *res,
1376 struct ldlm_lock *lock,
1377 struct ptlrpc_request *req, int increase)
1379 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1382 /* delayed lvb init may be required */
1383 rc = ldlm_lvbo_init(res);
1385 CERROR("delayed lvb init failed (rc %d)\n", rc);
1389 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1390 return ns->ns_lvbo->lvbo_update(res, lock, req, increase);
1395 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1396 struct ptlrpc_request *req, int increase)
1398 return ldlm_lvbo_update(res, NULL, req, increase);
1401 int ldlm_error2errno(enum ldlm_error error);
1402 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1403 * confuses user-space. */
1404 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1405 void ldlm_dump_export_locks(struct obd_export *exp);
1409 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1410 * __ldlm_handle2lock().
1412 #define LDLM_LOCK_PUT(lock) \
1414 LDLM_LOCK_REF_DEL(lock); \
1415 /*LDLM_DEBUG((lock), "put");*/ \
1416 ldlm_lock_put(lock); \
1420 * Release a lock reference obtained by some other means (see
1423 #define LDLM_LOCK_RELEASE(lock) \
1425 /*LDLM_DEBUG((lock), "put");*/ \
1426 ldlm_lock_put(lock); \
1429 #define LDLM_LOCK_GET(lock) \
1431 ldlm_lock_get(lock); \
1432 /*LDLM_DEBUG((lock), "get");*/ \
1436 #define ldlm_lock_list_put(head, member, count) \
1438 struct ldlm_lock *_lock, *_next; \
1440 list_for_each_entry_safe(_lock, _next, head, member) { \
1443 list_del_init(&_lock->member); \
1444 LDLM_LOCK_RELEASE(_lock); \
1449 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1450 void ldlm_lock_put(struct ldlm_lock *lock);
1451 void ldlm_lock_destroy(struct ldlm_lock *lock);
1452 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1453 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1454 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1455 enum ldlm_mode mode);
1456 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1457 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1458 enum ldlm_mode mode);
1459 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1460 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1461 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1462 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1463 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1464 const struct ldlm_res_id *, enum ldlm_type type,
1465 union ldlm_policy_data *, enum ldlm_mode mode,
1466 struct lustre_handle *, int unref);
1467 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1469 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1470 void ldlm_lock_cancel(struct ldlm_lock *lock);
1471 void ldlm_reprocess_all(struct ldlm_resource *res);
1472 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1473 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1474 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1477 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1478 enum ldlm_side client,
1479 enum ldlm_appetite apt,
1480 enum ldlm_ns_type ns_type);
1481 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1482 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1483 struct obd_import *imp,
1485 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1486 void ldlm_namespace_free(struct ldlm_namespace *ns,
1487 struct obd_import *imp, int force);
1488 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1489 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1490 enum ldlm_side client);
1491 void ldlm_namespace_get(struct ldlm_namespace *ns);
1492 void ldlm_namespace_put(struct ldlm_namespace *ns);
1494 int ldlm_debugfs_setup(void);
1495 void ldlm_debugfs_cleanup(void);
1497 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1498 struct lprocfs_stats *srv_stats)
1500 int lock_type = 0, op = 0;
1502 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1504 switch (lock_type) {
1506 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1509 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1512 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1515 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1523 lprocfs_counter_incr(srv_stats, op);
1528 /* resource.c - internal */
1529 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1530 struct ldlm_resource *parent,
1531 const struct ldlm_res_id *,
1532 enum ldlm_type type, int create);
1533 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1534 int ldlm_resource_putref(struct ldlm_resource *res);
1535 void ldlm_resource_add_lock(struct ldlm_resource *res,
1536 struct list_head *head,
1537 struct ldlm_lock *lock);
1538 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1539 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1540 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1541 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1542 void ldlm_resource_dump(int level, struct ldlm_resource *);
1543 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1544 const struct ldlm_res_id *);
1546 #define LDLM_RESOURCE_ADDREF(res) do { \
1547 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1550 #define LDLM_RESOURCE_DELREF(res) do { \
1551 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1554 /* ldlm_request.c */
1555 int ldlm_expired_completion_wait(void *data);
1556 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1557 * These AST handlers are typically used for server-side local locks and are
1558 * also used by client-side lock handlers to perform minimum level base
1561 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1562 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1563 void *data, int flag);
1564 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1565 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1566 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1567 /** @} ldlm_local_ast */
1569 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1570 * These are typically used by client and server (*_local versions)
1571 * to obtain and release locks.
1573 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1574 struct ldlm_enqueue_info *einfo,
1575 const struct ldlm_res_id *res_id,
1576 union ldlm_policy_data const *policy, __u64 *flags,
1577 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1578 struct lustre_handle *lockh, int async);
1579 int ldlm_prep_enqueue_req(struct obd_export *exp,
1580 struct ptlrpc_request *req,
1581 struct list_head *cancels,
1583 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1584 int version, int opc, int canceloff,
1585 struct list_head *cancels, int count);
1587 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1588 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1589 const struct ldlm_request *dlm_req,
1590 const struct ldlm_callback_suite *cbs);
1591 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1592 enum ldlm_type type, __u8 with_policy,
1593 enum ldlm_mode mode, __u64 *flags, void *lvb,
1595 const struct lustre_handle *lockh, int rc);
1596 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1597 const struct ldlm_res_id *res_id,
1598 enum ldlm_type type, union ldlm_policy_data *policy,
1599 enum ldlm_mode mode, __u64 *flags,
1600 ldlm_blocking_callback blocking,
1601 ldlm_completion_callback completion,
1602 ldlm_glimpse_callback glimpse,
1603 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1604 const __u64 *client_cookie,
1605 struct lustre_handle *lockh);
1606 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1607 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1608 enum ldlm_cancel_flags cancel_flags);
1609 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1610 enum ldlm_cancel_flags flags, void *opaque);
1611 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1612 const struct ldlm_res_id *res_id,
1613 union ldlm_policy_data *policy,
1614 enum ldlm_mode mode,
1615 enum ldlm_cancel_flags flags, void *opaque);
1616 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1617 int count, enum ldlm_cancel_flags flags);
1618 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1619 struct list_head *cancels,
1620 union ldlm_policy_data *policy,
1621 enum ldlm_mode mode, __u64 lock_flags,
1622 enum ldlm_cancel_flags cancel_flags,
1624 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1625 enum ldlm_cancel_flags flags);
1626 int ldlm_cli_cancel_list(struct list_head *head, int count,
1627 struct ptlrpc_request *req,
1628 enum ldlm_cancel_flags flags);
1629 /** @} ldlm_cli_api */
1632 /* This has to be here because recursive inclusion sucks. */
1633 int intent_disposition(struct ldlm_reply *rep, int flag);
1634 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1637 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1638 * than one lock_res is dead-lock safe.
1640 enum lock_res_type {
1645 /** Lock resource. */
1646 static inline void lock_res(struct ldlm_resource *res)
1648 spin_lock(&res->lr_lock);
1651 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1652 static inline void lock_res_nested(struct ldlm_resource *res,
1653 enum lock_res_type mode)
1655 spin_lock_nested(&res->lr_lock, mode);
1658 /** Unlock resource. */
1659 static inline void unlock_res(struct ldlm_resource *res)
1661 spin_unlock(&res->lr_lock);
1664 /** Check if resource is already locked, assert if not. */
1665 static inline void check_res_locked(struct ldlm_resource *res)
1667 assert_spin_locked(&res->lr_lock);
1670 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1671 void unlock_res_and_lock(struct ldlm_lock *lock);
1674 /** \defgroup ldlm_pools Various LDLM pool related functions
1675 * There are not used outside of ldlm.
1678 time64_t ldlm_pools_recalc(enum ldlm_side client);
1679 int ldlm_pools_init(void);
1680 void ldlm_pools_fini(void);
1682 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1683 int idx, enum ldlm_side client);
1684 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1685 void ldlm_pool_fini(struct ldlm_pool *pl);
1686 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1687 time64_t ldlm_pool_recalc(struct ldlm_pool *pl);
1688 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1689 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1690 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1691 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1692 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1693 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1694 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1695 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1696 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1699 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1700 const struct ldlm_extent *ex2)
1702 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1705 /* check if @ex1 contains @ex2 */
1706 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1707 const struct ldlm_extent *ex2)
1709 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1712 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);