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
71 #define LDLM_DEFAULT_LRU_SHRINK_BATCH (16)
74 * LDLM non-error return states
78 ELDLM_LOCK_MATCHED = 1,
80 ELDLM_LOCK_CHANGED = 300,
81 ELDLM_LOCK_ABORTED = 301,
82 ELDLM_LOCK_REPLACED = 302,
83 ELDLM_NO_LOCK_DATA = 303,
84 ELDLM_LOCK_WOULDBLOCK = 304,
86 ELDLM_NAMESPACE_EXISTS = 400,
87 ELDLM_BAD_NAMESPACE = 401,
91 * LDLM namespace type.
92 * The "client" type is actually an indication that this is a narrow local view
93 * into complete namespace on the server. Such namespaces cannot make any
94 * decisions about lack of conflicts or do any autonomous lock granting without
95 * first speaking to a server.
98 LDLM_NAMESPACE_SERVER = 0x01,
99 LDLM_NAMESPACE_CLIENT = 0x02
103 * The blocking callback is overloaded to perform two functions. These flags
104 * indicate which operation should be performed.
106 #define LDLM_CB_BLOCKING 1
107 #define LDLM_CB_CANCELING 2
110 * \name Lock Compatibility Matrix.
112 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
113 * Lock types are described in their respective implementation files:
114 * ldlm_{extent,flock,inodebits,plain}.c.
116 * There are six lock modes along with a compatibility matrix to indicate if
117 * two locks are compatible.
119 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
121 * - PW: Protective Write (normal write) mode. When a client requests a write
122 * lock from an OST, a lock with PW mode will be issued.
123 * - PR: Protective Read (normal read) mode. When a client requests a read from
124 * an OST, a lock with PR mode will be issued. Also, if the client opens a
125 * file for execution, it is granted a lock with PR mode.
126 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
127 * requests a write lock during a file open operation.
128 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
129 * an inodebit lock with the CR mode on the intermediate path component.
143 #define LCK_COMPAT_EX LCK_NL
144 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
145 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
146 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
147 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
148 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
149 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
150 #define LCK_COMPAT_COS (LCK_COS)
151 /** @} Lock Compatibility Matrix */
153 extern enum ldlm_mode lck_compat_array[];
155 static inline void lockmode_verify(enum ldlm_mode mode)
157 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
160 static inline int lockmode_compat(enum ldlm_mode exist_mode,
161 enum ldlm_mode new_mode)
163 return lck_compat_array[exist_mode] & new_mode;
168 * cluster name spaces
172 #define DLM_OST_NAMESPACE 1
173 #define DLM_MDS_NAMESPACE 2
176 - do we just separate this by security domains and use a prefix for
177 multiple namespaces in the same domain?
182 * Locking rules for LDLM:
187 * waiting_locks_spinlock
202 struct ldlm_resource;
203 struct ldlm_namespace;
206 * Operations on LDLM pools.
207 * LDLM pool is a pool of locks in the namespace without any implicitly
209 * Locks in the pool are organized in LRU.
210 * Local memory pressure or server instructions (e.g. mempressure on server)
211 * can trigger freeing of locks from the pool
213 struct ldlm_pool_ops {
214 /** Recalculate pool \a pl usage */
215 int (*po_recalc)(struct ldlm_pool *pl);
216 /** Cancel at least \a nr locks from pool \a pl */
217 int (*po_shrink)(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
218 int (*po_setup)(struct ldlm_pool *pl, int limit);
221 /** One second for pools thread check interval. Each pool has own period. */
222 #define LDLM_POOLS_THREAD_PERIOD (1)
224 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
225 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
227 /** Default recalc period for server side pools in sec. */
228 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
230 /** Default recalc period for client side pools in sec. */
231 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
234 * LDLM pool structure to track granted locks.
235 * For purposes of determining when to release locks on e.g. memory pressure.
236 * This feature is commonly referred to as lru_resize.
239 /** Pool debugfs directory. */
240 struct dentry *pl_debugfs_entry;
241 /** Pool name, must be long enough to hold compound proc entry name. */
243 /** Lock for protecting SLV/CLV updates. */
245 /** Number of allowed locks in in pool, both, client and server side. */
247 /** Number of granted locks in */
249 /** Grant rate per T. */
250 atomic_t pl_grant_rate;
251 /** Cancel rate per T. */
252 atomic_t pl_cancel_rate;
253 /** Server lock volume (SLV). Protected by pl_lock. */
254 __u64 pl_server_lock_volume;
255 /** Current biggest client lock volume. Protected by pl_lock. */
256 __u64 pl_client_lock_volume;
257 /** Lock volume factor. SLV on client is calculated as following:
258 * server_slv * lock_volume_factor. */
259 atomic_t pl_lock_volume_factor;
260 /** Time when last SLV from server was obtained. */
261 time64_t pl_recalc_time;
262 /** Recalculation period for pool. */
263 time64_t pl_recalc_period;
264 /** Recalculation and shrink operations. */
265 struct ldlm_pool_ops *pl_ops;
266 /** Number of planned locks for next period. */
268 /** Pool statistics. */
269 struct lprocfs_stats *pl_stats;
272 struct kobject pl_kobj;
273 struct completion pl_kobj_unregister;
276 typedef int (*ldlm_res_policy)(const struct lu_env *env,
277 struct ldlm_namespace *,
278 struct ldlm_lock **, void *req_cookie,
279 enum ldlm_mode mode, __u64 flags, void *data);
281 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
285 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
286 * be associated with an LDLM lock and transferred from client to server and
289 * Currently LVBs are used by:
290 * - OSC-OST code to maintain current object size/times
291 * - layout lock code to return the layout when the layout lock is granted
293 * To ensure delayed LVB initialization, it is highly recommended to use the set
294 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
296 struct ldlm_valblock_ops {
297 int (*lvbo_init)(struct ldlm_resource *res);
298 int (*lvbo_update)(struct ldlm_resource *res, struct ldlm_lock *lock,
299 struct ptlrpc_request *r, int increase);
300 int (*lvbo_free)(struct ldlm_resource *res);
301 /* Return size of lvb data appropriate RPC size can be reserved */
302 int (*lvbo_size)(struct ldlm_lock *lock);
303 /* Called to fill in lvb data to RPC buffer @buf */
304 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int *buflen);
308 * LDLM pools related, type of lock pool in the namespace.
309 * Greedy means release cached locks aggressively
312 LDLM_NAMESPACE_GREEDY = BIT(0),
313 LDLM_NAMESPACE_MODEST = BIT(1),
317 * Default values for the "max_nolock_size", "contention_time" and
318 * "contended_locks" namespace tunables.
320 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
321 #define NS_DEFAULT_CONTENTION_SECONDS 2
322 #define NS_DEFAULT_CONTENDED_LOCKS 32
324 struct ldlm_ns_bucket {
325 /** back pointer to namespace */
326 struct ldlm_namespace *nsb_namespace;
328 * Estimated lock callback time. Used by adaptive timeout code to
329 * avoid spurious client evictions due to unresponsiveness when in
330 * fact the network or overall system load is at fault
332 struct adaptive_timeout nsb_at_estimate;
334 * Which res in the bucket should we start with the reclaim.
336 int nsb_reclaim_start;
337 /* counter of entries in this bucket */
342 /** LDLM namespace lock stats */
348 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
349 LDLM_NS_TYPE_MDC, /**< MDC namespace */
350 LDLM_NS_TYPE_MDT, /**< MDT namespace */
351 LDLM_NS_TYPE_OSC, /**< OSC namespace */
352 LDLM_NS_TYPE_OST, /**< OST namespace */
353 LDLM_NS_TYPE_MGC, /**< MGC namespace */
354 LDLM_NS_TYPE_MGT, /**< MGT namespace */
357 enum ldlm_namespace_flags {
359 * Flag to indicate the LRU cancel is in progress.
360 * Used to limit the process by 1 thread only.
368 * Namespace serves to contain locks related to a particular service.
369 * There are two kinds of namespaces:
370 * - Server namespace has knowledge of all locks and is therefore authoritative
371 * to make decisions like what locks could be granted and what conflicts
372 * exist during new lock enqueue.
373 * - Client namespace only has limited knowledge about locks in the namespace,
374 * only seeing locks held by the client.
376 * Every Lustre service has one server namespace present on the server serving
377 * that service. Every client connected to the service has a client namespace
379 * Every lock obtained by client in that namespace is actually represented by
380 * two in-memory locks. One on the server and one on the client. The locks are
381 * linked by a special cookie by which one node can tell to the other which lock
382 * it actually means during communications. Such locks are called remote locks.
383 * The locks held by server only without any reference to a client are called
386 struct ldlm_namespace {
387 /** Backward link to OBD, required for LDLM pool to store new SLV. */
388 struct obd_device *ns_obd;
390 /** Flag indicating if namespace is on client instead of server */
391 enum ldlm_side ns_client;
393 /** name of this namespace */
396 /** Resource hash table for namespace. */
397 struct cfs_hash *ns_rs_hash;
398 struct ldlm_ns_bucket *ns_rs_buckets;
399 unsigned int ns_bucket_bits;
404 /** big refcount (by bucket) */
408 * Namespace connect flags supported by server (may be changed via
409 * /proc, LRU resize may be disabled/enabled).
411 __u64 ns_connect_flags;
413 /** Client side original connect flags supported by server. */
414 __u64 ns_orig_connect_flags;
416 /* namespace debugfs dir entry */
417 struct dentry *ns_debugfs_entry;
420 * Position in global namespace list linking all namespaces on
423 struct list_head ns_list_chain;
426 * List of unused locks for this namespace. This list is also called
428 * Unused locks are locks with zero reader/writer reference counts.
429 * This list is only used on clients for lock caching purposes.
430 * When we want to release some locks voluntarily or if server wants
431 * us to release some locks due to e.g. memory pressure, we take locks
432 * to release from the head of this list.
433 * Locks are linked via l_lru field in \see struct ldlm_lock.
435 struct list_head ns_unused_list;
436 /** Number of locks in the LRU list above */
438 struct list_head *ns_last_pos;
441 * Maximum number of locks permitted in the LRU. If 0, means locks
442 * are managed by pools and there is no preset limit, rather it is all
443 * controlled by available memory on this client and on server.
445 unsigned int ns_max_unused;
448 * Cancel batch, if unused lock count exceed lru_size
449 * Only be used if LRUR disable.
451 unsigned int ns_cancel_batch;
453 /** Maximum allowed age (last used time) for locks in the LRU. Set in
454 * seconds from userspace, but stored in ns to avoid repeat conversions.
459 * Server only: number of times we evicted clients due to lack of reply
462 unsigned int ns_timeouts;
464 * Number of seconds since the file change time after which
465 * the MDT will return an UPDATE lock along with a LOOKUP lock.
466 * This allows the client to start caching negative dentries
467 * for a directory and may save an RPC for a later stat.
469 timeout_t ns_ctime_age_limit;
471 * Number of (nano)seconds since the lock was last used. The client
472 * may cancel the lock older than this age and flush related data if
473 * another client shows interest in this lock by doing glimpse request.
474 * This allows to cache stat data locally for such files early. Set in
475 * seconds from userspace, but stored in ns to avoid repeat conversions.
477 ktime_t ns_dirty_age_limit;
479 * Used to rate-limit ldlm_namespace_dump calls.
480 * \see ldlm_namespace_dump. Increased by 10 seconds every time
483 time64_t ns_next_dump;
485 /** "policy" function that does actual lock conflict determination */
486 ldlm_res_policy ns_policy;
489 * LVB operations for this namespace.
490 * \see struct ldlm_valblock_ops
492 struct ldlm_valblock_ops *ns_lvbo;
495 * Used by filter code to store pointer to OBD of the service.
496 * Should be dropped in favor of \a ns_obd
501 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
502 * a resource is removed.
504 wait_queue_head_t ns_waitq;
505 /** LDLM pool structure for this namespace */
506 struct ldlm_pool ns_pool;
507 /** Definition of how eagerly unused locks will be released from LRU */
508 enum ldlm_appetite ns_appetite;
511 * If more than \a ns_contended_locks are found, the resource is
512 * considered to be contended. Lock enqueues might specify that no
513 * contended locks should be granted
515 unsigned ns_contended_locks;
518 * The resources in this namespace remember contended state during
519 * \a ns_contention_time, in seconds.
521 timeout_t ns_contention_time;
524 * Limit size of contended extent locks, in bytes.
525 * If extended lock is requested for more then this many bytes and
526 * caller instructs us not to grant contended locks, we would disregard
529 unsigned ns_max_nolock_size;
531 /** Limit of parallel AST RPC count. */
532 unsigned ns_max_parallel_ast;
535 * Callback to check if a lock is good to be canceled by ELC or
538 ldlm_cancel_cbt ns_cancel;
540 /** LDLM lock stats */
541 struct lprocfs_stats *ns_stats;
544 * Flag to indicate namespace is being freed. Used to determine if
545 * recalculation of LDLM pool statistics should be skipped.
547 unsigned ns_stopping:1;
550 * Which bucket should we start with the lock reclaim.
552 int ns_reclaim_start;
554 struct kobject ns_kobj; /* sysfs object */
555 struct completion ns_kobj_unregister;
558 * To avoid another ns_lock usage, a separate bitops field.
560 unsigned long ns_flags;
564 * Returns 1 if namespace \a ns is a client namespace.
566 static inline int ns_is_client(struct ldlm_namespace *ns)
569 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
570 ns->ns_client == LDLM_NAMESPACE_SERVER);
571 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
575 * Returns 1 if namespace \a ns is a server namespace.
577 static inline int ns_is_server(struct ldlm_namespace *ns)
580 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
581 ns->ns_client == LDLM_NAMESPACE_SERVER);
582 return ns->ns_client == LDLM_NAMESPACE_SERVER;
586 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
588 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
591 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
595 * Returns 1 if this namespace supports lru_resize.
597 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
600 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
603 static inline void ns_register_cancel(struct ldlm_namespace *ns,
612 /** Type for blocking callback function of a lock. */
613 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
614 struct ldlm_lock_desc *new, void *data,
616 /** Type for completion callback function of a lock. */
617 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
619 /** Type for glimpse callback function of a lock. */
620 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
622 /** Type for created callback function of a lock. */
623 typedef void (*ldlm_created_callback)(struct ldlm_lock *lock);
625 /** Work list for sending GL ASTs to multiple locks. */
626 struct ldlm_glimpse_work {
627 struct ldlm_lock *gl_lock; /* lock to glimpse */
628 struct list_head gl_list; /* linkage to other gl work structs */
629 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
630 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
631 * glimpse callback request */
632 ptlrpc_interpterer_t gl_interpret_reply;
633 void *gl_interpret_data;
636 struct ldlm_bl_desc {
637 unsigned int bl_same_client:1,
641 struct ldlm_cb_set_arg {
642 struct ptlrpc_request_set *set;
643 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
645 struct list_head *list;
646 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
647 ptlrpc_interpterer_t gl_interpret_reply;
648 void *gl_interpret_data;
649 struct ldlm_bl_desc *bl_desc;
652 struct ldlm_cb_async_args {
653 struct ldlm_cb_set_arg *ca_set_arg;
654 struct ldlm_lock *ca_lock;
657 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
658 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
660 /** Interval node data for each LDLM_EXTENT lock. */
661 struct ldlm_interval {
662 struct interval_node li_node; /* node for tree management */
663 struct list_head li_group; /* the locks which have the same
664 * policy - group of the policy */
666 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
669 * Interval tree for extent locks.
670 * The interval tree must be accessed under the resource lock.
671 * Interval trees are used for granted extent locks to speed up conflicts
672 * lookup. See ldlm/interval_tree.c for more details.
674 struct ldlm_interval_tree {
677 enum ldlm_mode lit_mode; /* lock mode */
678 struct interval_node *lit_root; /* actual ldlm_interval */
682 * Lists of waiting locks for each inodebit type.
683 * A lock can be in several liq_waiting lists and it remains in lr_waiting.
685 struct ldlm_ibits_queues {
686 struct list_head liq_waiting[MDS_INODELOCK_NUMBITS];
689 struct ldlm_ibits_node {
690 struct list_head lin_link[MDS_INODELOCK_NUMBITS];
691 struct ldlm_lock *lock;
694 /** Whether to track references to exports by LDLM locks. */
695 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
698 enum ldlm_cancel_flags {
699 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
700 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
701 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
708 __u64 blocking_owner;
709 struct obd_export *blocking_export;
710 atomic_t blocking_refs;
714 union ldlm_policy_data {
715 struct ldlm_extent l_extent;
716 struct ldlm_flock l_flock;
717 struct ldlm_inodebits l_inodebits;
720 void ldlm_convert_policy_to_wire(enum ldlm_type type,
721 const union ldlm_policy_data *lpolicy,
722 union ldlm_wire_policy_data *wpolicy);
723 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
724 const union ldlm_wire_policy_data *wpolicy,
725 union ldlm_policy_data *lpolicy);
735 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
737 #define LDLM_GID_ANY ((__u64)-1)
740 * LDLM lock structure
742 * Represents a single LDLM lock and its state in memory. Each lock is
743 * associated with a single ldlm_resource, the object which is being
744 * locked. There may be multiple ldlm_locks on a single resource,
745 * depending on the lock type and whether the locks are conflicting or
751 * When remote side wants to tell us about a lock, they address
752 * it by this opaque handle. The handle does not hold a
753 * reference on the ldlm_lock, so it can be safely passed to
754 * other threads or nodes. When the lock needs to be accessed
755 * from the handle, it is looked up again in the lock table, and
756 * may no longer exist.
758 * Must be first in the structure.
760 struct portals_handle l_handle;
762 * Internal spinlock protects l_resource. We should hold this lock
763 * first before taking res_lock.
767 * Pointer to actual resource this lock is in.
768 * ldlm_lock_change_resource() can change this.
770 struct ldlm_resource *l_resource;
772 * List item for client side LRU list.
773 * Protected by ns_lock in struct ldlm_namespace.
775 struct list_head l_lru;
777 * Linkage to resource's lock queues according to current lock state.
778 * (could be granted or waiting)
779 * Protected by lr_lock in struct ldlm_resource.
781 struct list_head l_res_link;
783 * Internal structures per lock type..
786 struct ldlm_interval *l_tree_node;
787 struct ldlm_ibits_node *l_ibits_node;
790 * Per export hash of locks.
791 * Protected by per-bucket exp->exp_lock_hash locks.
793 struct hlist_node l_exp_hash;
795 * Per export hash of flock locks.
796 * Protected by per-bucket exp->exp_flock_hash locks.
798 struct hlist_node l_exp_flock_hash;
801 * Protected by lr_lock.
803 enum ldlm_mode l_req_mode;
805 * Granted mode, also protected by lr_lock.
807 enum ldlm_mode l_granted_mode;
808 /** Lock completion handler pointer. Called when lock is granted. */
809 ldlm_completion_callback l_completion_ast;
811 * Lock blocking AST handler pointer.
812 * It plays two roles:
813 * - as a notification of an attempt to queue a conflicting lock (once)
814 * - as a notification when the lock is being cancelled.
816 * As such it's typically called twice: once for the initial conflict
817 * and then once more when the last user went away and the lock is
818 * cancelled (could happen recursively).
820 ldlm_blocking_callback l_blocking_ast;
822 * Lock glimpse handler.
823 * Glimpse handler is used to obtain LVB updates from a client by
826 ldlm_glimpse_callback l_glimpse_ast;
830 * This is a pointer to actual client export for locks that were granted
831 * to clients. Used server-side.
833 struct obd_export *l_export;
835 * Lock connection export.
836 * Pointer to server export on a client.
838 struct obd_export *l_conn_export;
841 * Remote lock handle.
842 * If the lock is remote, this is the handle of the other side lock
845 struct lustre_handle l_remote_handle;
848 * Representation of private data specific for a lock type.
849 * Examples are: extent range for extent lock or bitmask for ibits locks
851 union ldlm_policy_data l_policy_data;
854 * Lock state flags. Protected by lr_lock.
855 * \see lustre_dlm_flags.h where the bits are defined.
860 * Lock r/w usage counters.
861 * Protected by lr_lock.
866 * If the lock is granted, a process sleeps on this waitq to learn when
867 * it's no longer in use. If the lock is not granted, a process sleeps
868 * on this waitq to learn when it becomes granted.
870 wait_queue_head_t l_waitq;
873 * Time, in nanoseconds, last used by e.g. being matched by lock match.
877 /** Originally requested extent for the extent lock. */
878 struct ldlm_extent l_req_extent;
881 * Client-side-only members.
884 enum lvb_type l_lvb_type;
887 * Temporary storage for a LVB received during an enqueue operation.
888 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
893 /** Private storage for lock user. Opaque to LDLM. */
898 * Seconds. It will be updated if there is any activity related to
899 * the lock at client, e.g. enqueue the lock. For server it is the
900 * time when blocking ast was sent.
903 time64_t l_blast_sent;
906 /* separate ost_lvb used mostly by Data-on-MDT for now.
907 * It is introduced to don't mix with layout lock data. */
908 struct ost_lvb l_ost_lvb;
910 * Server-side-only members.
914 * Connection cookie for the client originating the operation.
915 * Used by Commit on Share (COS) code. Currently only used for
916 * inodebits locks on MDS.
918 __u64 l_client_cookie;
921 * List item for locks waiting for cancellation from clients.
922 * The lists this could be linked into are:
923 * waiting_locks_list (protected by waiting_locks_spinlock),
924 * then if the lock timed out, it is moved to
925 * expired_lock_list for further processing.
927 struct list_head l_pending_chain;
930 * Set when lock is sent a blocking AST. Time in seconds when timeout
931 * is reached and client holding this lock could be evicted.
932 * This timeout could be further extended by e.g. certain IO activity
934 * \see ost_rw_prolong_locks
936 time64_t l_callback_timestamp;
938 /** Local PID of process which created this lock. */
942 * Number of times blocking AST was sent for this lock.
943 * This is for debugging. Valid values are 0 and 1, if there is an
944 * attempt to send blocking AST more than once, an assertion would be
945 * hit. \see ldlm_work_bl_ast_lock
948 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
949 struct list_head l_bl_ast;
950 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
951 struct list_head l_cp_ast;
952 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
953 struct list_head l_rk_ast;
956 * Pointer to a conflicting lock that caused blocking AST to be sent
959 struct ldlm_lock *l_blocking_lock;
962 * Protected by lr_lock, linkages to "skip lists".
963 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
965 struct list_head l_sl_mode;
966 struct list_head l_sl_policy;
968 /** Reference tracking structure to debug leaked locks. */
969 struct lu_ref l_reference;
970 #if LUSTRE_TRACKS_LOCK_EXP_REFS
971 /* Debugging stuff for bug 20498, for tracking export references. */
972 /** number of export references taken */
974 /** link all locks referencing one export */
975 struct list_head l_exp_refs_link;
976 /** referenced export object */
977 struct obd_export *l_exp_refs_target;
980 * export blocking dlm lock list, protected by
981 * l_export->exp_bl_list_lock.
982 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
983 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
985 struct list_head l_exp_list;
988 enum ldlm_match_flags {
989 LDLM_MATCH_UNREF = BIT(0),
990 LDLM_MATCH_AST = BIT(1),
991 LDLM_MATCH_AST_ANY = BIT(2),
995 * Describe the overlap between two locks. itree_overlap_cb data.
997 struct ldlm_match_data {
998 struct ldlm_lock *lmd_old;
999 struct ldlm_lock *lmd_lock;
1000 enum ldlm_mode *lmd_mode;
1001 union ldlm_policy_data *lmd_policy;
1003 __u64 lmd_skip_flags;
1004 enum ldlm_match_flags lmd_match;
1007 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
1008 #define l_transno l_client_cookie
1010 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
1011 * which is for server. */
1012 #define l_slc_link l_rk_ast
1014 #define HANDLE_MAP_SIZE ((LMV_MAX_STRIPE_COUNT + 7) >> 3)
1016 struct lustre_handle_array {
1017 unsigned int ha_count;
1018 /* ha_map is used as bit flag to indicate handle is remote or local */
1019 char ha_map[HANDLE_MAP_SIZE];
1020 struct lustre_handle ha_handles[0];
1024 * LDLM resource description.
1025 * Basically, resource is a representation for a single object.
1026 * Object has a name which is currently 4 64-bit integers. LDLM user is
1027 * responsible for creation of a mapping between objects it wants to be
1028 * protected and resource names.
1030 * A resource can only hold locks of a single lock type, though there may be
1031 * multiple ldlm_locks on a single resource, depending on the lock type and
1032 * whether the locks are conflicting or not.
1034 struct ldlm_resource {
1035 struct ldlm_ns_bucket *lr_ns_bucket;
1038 * List item for list in namespace hash.
1039 * protected by ns_lock
1041 struct hlist_node lr_hash;
1043 /** Reference count for this resource */
1044 atomic_t lr_refcount;
1046 /** Spinlock to protect locks under this resource. */
1050 * protected by lr_lock
1052 /** List of locks in granted state */
1053 struct list_head lr_granted;
1055 * List of locks that could not be granted due to conflicts and
1056 * that are waiting for conflicts to go away */
1057 struct list_head lr_waiting;
1060 /** Resource name */
1061 struct ldlm_res_id lr_name;
1065 * Interval trees (only for extent locks) for all modes of
1068 struct ldlm_interval_tree *lr_itree;
1069 struct ldlm_ibits_queues *lr_ibits_queues;
1074 * When the resource was considered as contended,
1075 * used only on server side.
1077 time64_t lr_contention_time;
1079 * Associated inode, used only on client side.
1081 struct inode *lr_lvb_inode;
1084 /** Type of locks this resource can hold. Only one type per resource. */
1085 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1088 * Server-side-only lock value block elements.
1089 * To serialize lvbo_init.
1092 struct mutex lr_lvb_mutex;
1093 /** protected by lr_lock */
1095 /** is lvb initialized ? */
1096 bool lr_lvb_initialized;
1098 /** List of references to this resource. For debugging. */
1099 struct lu_ref lr_reference;
1102 static inline int ldlm_is_granted(struct ldlm_lock *lock)
1104 return lock->l_req_mode == lock->l_granted_mode;
1107 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1109 return lock->l_resource->lr_type == LDLM_IBITS &&
1110 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1113 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1115 return lock->l_resource->lr_type == LDLM_IBITS &&
1116 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1119 static inline char *
1120 ldlm_ns_name(struct ldlm_namespace *ns)
1125 static inline struct ldlm_namespace *
1126 ldlm_res_to_ns(struct ldlm_resource *res)
1128 return res->lr_ns_bucket->nsb_namespace;
1131 static inline struct ldlm_namespace *
1132 ldlm_lock_to_ns(struct ldlm_lock *lock)
1134 return ldlm_res_to_ns(lock->l_resource);
1137 static inline char *
1138 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1140 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1143 static inline struct adaptive_timeout *
1144 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1146 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1149 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1151 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1154 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1155 res->lr_lvb_initialized)
1158 mutex_lock(&res->lr_lvb_mutex);
1159 /* Did we lose the race? */
1160 if (res->lr_lvb_initialized) {
1161 mutex_unlock(&res->lr_lvb_mutex);
1164 rc = ns->ns_lvbo->lvbo_init(res);
1166 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1168 if (res->lr_lvb_data != NULL) {
1169 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1170 res->lr_lvb_data = NULL;
1172 res->lr_lvb_len = rc;
1174 res->lr_lvb_initialized = true;
1176 mutex_unlock(&res->lr_lvb_mutex);
1180 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1182 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1184 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1185 return ns->ns_lvbo->lvbo_size(lock);
1190 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int *len)
1192 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1195 if (ns->ns_lvbo != NULL) {
1196 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1197 /* init lvb now if not already */
1198 rc = ldlm_lvbo_init(lock->l_resource);
1200 CERROR("lock %p: delayed lvb init failed (rc %d)",
1204 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1209 struct ldlm_ast_work {
1210 struct ldlm_lock *w_lock;
1212 struct ldlm_lock_desc w_desc;
1213 struct list_head w_list;
1220 * Common ldlm_enqueue parameters
1222 struct ldlm_enqueue_info {
1223 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1224 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1225 void *ei_cb_bl; /** blocking lock callback */
1226 void *ei_cb_local_bl; /** blocking local lock callback */
1227 void *ei_cb_cp; /** lock completion callback */
1228 void *ei_cb_gl; /** lock glimpse callback */
1229 ldlm_created_callback ei_cb_created; /** lock created callback */
1230 void *ei_cbdata; /** Data to be passed into callbacks. */
1231 void *ei_namespace; /** lock namespace **/
1232 u64 ei_inodebits; /** lock inode bits **/
1233 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1234 unsigned int ei_enq_slot:1; /** whether acquire rpc slot */
1237 #define ei_res_id ei_cb_gl
1239 extern char *ldlm_lockname[];
1240 extern char *ldlm_typename[];
1241 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1244 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1245 * For the cases where we do not have actual lock to print along
1246 * with a debugging message that is ldlm-related
1248 #define LDLM_DEBUG_NOLOCK(format, a...) \
1249 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1252 * Support function for lock information printing into debug logs.
1256 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1257 CFS_CHECK_STACK(msgdata, mask, cdls); \
1259 if (((mask) & D_CANTMASK) != 0 || \
1260 ((libcfs_debug & (mask)) != 0 && \
1261 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1262 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1265 void _ldlm_lock_debug(struct ldlm_lock *lock,
1266 struct libcfs_debug_msg_data *data,
1267 const char *fmt, ...)
1268 __attribute__ ((format (printf, 3, 4)));
1271 * Rate-limited version of lock printing function.
1273 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1274 static struct cfs_debug_limit_state _ldlm_cdls; \
1275 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1276 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1279 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1280 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1282 /** Non-rate-limited lock printing function for debugging purposes. */
1283 #define LDLM_DEBUG(lock, fmt, a...) do { \
1284 if (likely(lock != NULL)) { \
1285 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1286 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1287 "### " fmt , ##a); \
1289 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1292 #else /* !LIBCFS_DEBUG */
1293 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1294 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1295 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1299 * Three intentions can be used for the policy functions in
1300 * ldlm_processing_policy.
1302 * LDLM_PROCESS_RESCAN:
1304 * It's used when policy functions are called from ldlm_reprocess_queue() to
1305 * reprocess the wait list and try to grant locks, blocking ASTs
1306 * have already been sent in this situation, completion ASTs need be sent for
1307 * the locks being granted.
1309 * LDLM_PROCESS_ENQUEUE:
1311 * It's used when policy functions are called from ldlm_lock_enqueue() to
1312 * process the wait list for handling an enqueue request, blocking
1313 * ASTs have not been sent yet, so list of conflicting locks would be
1314 * collected and ASTs sent.
1316 * LDLM_PROCESS_RECOVERY:
1318 * It's used when policy functions are called from ldlm_reprocess_queue() to
1319 * reprocess the wait list when recovery done. In case of blocking
1320 * ASTs are lost before recovery, it needs not only to grant locks if
1321 * available, but also send blocking ASTs to the locks doesn't have AST sent
1322 * flag. Completion ASTs need be sent for the locks being granted.
1324 enum ldlm_process_intention {
1325 LDLM_PROCESS_RESCAN = 0,
1326 LDLM_PROCESS_ENQUEUE = 1,
1327 LDLM_PROCESS_RECOVERY = 2,
1330 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1331 enum ldlm_process_intention intention,
1332 enum ldlm_error *err,
1333 struct list_head *work_list);
1335 typedef int (*ldlm_reprocessing_policy)(struct ldlm_resource *res,
1336 struct list_head *queue,
1337 struct list_head *work_list,
1338 enum ldlm_process_intention intention,
1339 struct ldlm_lock *hint);
1342 * Return values for lock iterators.
1343 * Also used during deciding of lock grants and cancellations.
1345 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1346 #define LDLM_ITER_STOP 2 /* stop iterating */
1348 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1349 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1351 /** \defgroup ldlm_iterator Lock iterators
1353 * LDLM provides for a way to iterate through every lock on a resource or
1354 * namespace or every resource in a namespace.
1356 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1358 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1360 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1361 ldlm_iterator_t iter, void *data);
1362 /** @} ldlm_iterator */
1364 int ldlm_replay_locks(struct obd_import *imp);
1367 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1370 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1372 struct ldlm_prolong_args {
1373 struct obd_export *lpa_export;
1374 struct ldlm_res_id lpa_resid;
1375 struct ldlm_extent lpa_extent;
1376 enum ldlm_mode lpa_mode;
1377 timeout_t lpa_timeout;
1381 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1382 struct ldlm_prolong_args *arg);
1383 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1385 struct ldlm_callback_suite {
1386 ldlm_completion_callback lcs_completion;
1387 ldlm_blocking_callback lcs_blocking;
1388 ldlm_glimpse_callback lcs_glimpse;
1392 #ifdef HAVE_SERVER_SUPPORT
1393 /** \defgroup ldlm_srv_ast Server AST handlers
1394 * These are AST handlers used by server code.
1395 * Their property is that they are just preparing RPCs to be sent to clients.
1398 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1399 void *data, int flag);
1400 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1401 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1402 int ldlm_glimpse_locks(struct ldlm_resource *res,
1403 struct list_head *gl_work_list);
1404 /** @} ldlm_srv_ast */
1406 /** \defgroup ldlm_handlers Server LDLM handlers
1407 * These are handler functions that should be called by "frontends" such as
1408 * MDT or OST to pass through LDLM requests to LDLM for handling
1411 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1412 const struct ldlm_request *dlm_req,
1413 const struct ldlm_callback_suite *cbs);
1414 int ldlm_handle_convert0(struct ptlrpc_request *req,
1415 const struct ldlm_request *dlm_req);
1416 int ldlm_handle_cancel(struct ptlrpc_request *req);
1417 int ldlm_request_cancel(struct ptlrpc_request *req,
1418 const struct ldlm_request *dlm_req,
1419 int first, enum lustre_at_flags flags);
1420 /** @} ldlm_handlers */
1422 void ldlm_revoke_export_locks(struct obd_export *exp);
1423 timeout_t ldlm_bl_timeout(struct ldlm_lock *lock);
1425 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1426 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, timeout_t timeout);
1427 int ldlm_get_ref(void);
1428 void ldlm_put_ref(void);
1429 int ldlm_init_export(struct obd_export *exp);
1430 void ldlm_destroy_export(struct obd_export *exp);
1431 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1434 #ifdef HAVE_SERVER_SUPPORT
1435 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1436 ldlm_reprocessing_policy
1437 ldlm_get_reprocessing_policy(struct ldlm_resource *res);
1439 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1440 void ldlm_lock2handle(const struct ldlm_lock *lock,
1441 struct lustre_handle *lockh);
1442 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1443 void ldlm_cancel_callback(struct ldlm_lock *);
1444 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1445 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1448 * Obtain a lock reference by its handle.
1450 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1452 return __ldlm_handle2lock(h, 0);
1455 #define LDLM_LOCK_REF_DEL(lock) \
1456 lu_ref_del(&lock->l_reference, "handle", lock)
1458 static inline struct ldlm_lock *
1459 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1461 struct ldlm_lock *lock;
1463 lock = __ldlm_handle2lock(h, flags);
1465 LDLM_LOCK_REF_DEL(lock);
1470 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1471 * data from request \a r
1473 static inline int ldlm_lvbo_update(struct ldlm_resource *res,
1474 struct ldlm_lock *lock,
1475 struct ptlrpc_request *req, int increase)
1477 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1480 /* delayed lvb init may be required */
1481 rc = ldlm_lvbo_init(res);
1483 CERROR("delayed lvb init failed (rc %d)\n", rc);
1487 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1488 return ns->ns_lvbo->lvbo_update(res, lock, req, increase);
1493 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1494 struct ptlrpc_request *req,
1497 return ldlm_lvbo_update(res, NULL, req, increase);
1500 int is_granted_or_cancelled_nolock(struct ldlm_lock *lock);
1502 int ldlm_error2errno(enum ldlm_error error);
1503 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1504 * confuses user-space. */
1505 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1506 void ldlm_dump_export_locks(struct obd_export *exp);
1510 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1511 * __ldlm_handle2lock().
1513 #define LDLM_LOCK_PUT(lock) \
1515 LDLM_LOCK_REF_DEL(lock); \
1516 /*LDLM_DEBUG((lock), "put");*/ \
1517 ldlm_lock_put(lock); \
1521 * Release a lock reference obtained by some other means (see
1524 #define LDLM_LOCK_RELEASE(lock) \
1526 /*LDLM_DEBUG((lock), "put");*/ \
1527 ldlm_lock_put(lock); \
1530 #define LDLM_LOCK_GET(lock) \
1532 ldlm_lock_get(lock); \
1533 /*LDLM_DEBUG((lock), "get");*/ \
1537 #define ldlm_lock_list_put(head, member, count) \
1539 struct ldlm_lock *_lock, *_next; \
1541 list_for_each_entry_safe(_lock, _next, head, member) { \
1544 list_del_init(&_lock->member); \
1545 LDLM_LOCK_RELEASE(_lock); \
1550 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1551 void ldlm_lock_put(struct ldlm_lock *lock);
1552 void ldlm_lock_destroy(struct ldlm_lock *lock);
1553 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1554 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1555 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1556 enum ldlm_mode mode);
1557 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1558 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1559 enum ldlm_mode mode);
1560 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1561 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1562 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1563 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1565 enum ldlm_mode ldlm_lock_match_with_skip(struct ldlm_namespace *ns,
1566 __u64 flags, __u64 skip_flags,
1567 const struct ldlm_res_id *res_id,
1568 enum ldlm_type type,
1569 union ldlm_policy_data *policy,
1570 enum ldlm_mode mode,
1571 struct lustre_handle *lh,
1572 enum ldlm_match_flags match_flags);
1573 static inline enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns,
1575 const struct ldlm_res_id *res_id,
1576 enum ldlm_type type,
1577 union ldlm_policy_data *policy,
1578 enum ldlm_mode mode,
1579 struct lustre_handle *lh)
1581 return ldlm_lock_match_with_skip(ns, flags, 0, res_id, type, policy,
1584 struct ldlm_lock *search_itree(struct ldlm_resource *res,
1585 struct ldlm_match_data *data);
1586 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1588 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1589 void ldlm_lock_cancel(struct ldlm_lock *lock);
1590 void ldlm_reprocess_all(struct ldlm_resource *res, struct ldlm_lock *hint);
1591 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1592 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1593 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1596 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1597 enum ldlm_side client,
1598 enum ldlm_appetite apt,
1599 enum ldlm_ns_type ns_type);
1600 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1601 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1602 struct obd_import *imp,
1604 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1605 void ldlm_namespace_free(struct ldlm_namespace *ns,
1606 struct obd_import *imp, int force);
1607 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1608 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1609 enum ldlm_side client);
1610 void ldlm_namespace_get(struct ldlm_namespace *ns);
1611 void ldlm_namespace_put(struct ldlm_namespace *ns);
1613 int ldlm_debugfs_setup(void);
1614 void ldlm_debugfs_cleanup(void);
1616 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1617 struct lprocfs_stats *srv_stats)
1619 int lock_type = 0, op = 0;
1621 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1623 switch (lock_type) {
1625 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1628 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1631 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1634 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1642 lprocfs_counter_incr(srv_stats, op);
1645 /* resource.c - internal */
1646 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1647 struct ldlm_resource *parent,
1648 const struct ldlm_res_id *,
1649 enum ldlm_type type, int create);
1650 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1651 int ldlm_resource_putref(struct ldlm_resource *res);
1652 void ldlm_resource_add_lock(struct ldlm_resource *res,
1653 struct list_head *head,
1654 struct ldlm_lock *lock);
1655 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1656 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1657 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1658 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1659 void ldlm_resource_dump(int level, struct ldlm_resource *);
1660 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1661 const struct ldlm_res_id *);
1663 #define LDLM_RESOURCE_ADDREF(res) do { \
1664 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1667 #define LDLM_RESOURCE_DELREF(res) do { \
1668 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1671 /* ldlm_request.c */
1672 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1673 * These AST handlers are typically used for server-side local locks and are
1674 * also used by client-side lock handlers to perform minimum level base
1677 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1678 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1679 void *data, int flag);
1680 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1681 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1682 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1683 /** @} ldlm_local_ast */
1685 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1686 * These are typically used by client and server (*_local versions)
1687 * to obtain and release locks.
1689 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1690 struct ldlm_enqueue_info *einfo,
1691 const struct ldlm_res_id *res_id,
1692 union ldlm_policy_data const *policy, __u64 *flags,
1693 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1694 struct lustre_handle *lockh, int async);
1695 int ldlm_prep_enqueue_req(struct obd_export *exp,
1696 struct ptlrpc_request *req,
1697 struct list_head *cancels,
1699 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1700 int version, int opc, int canceloff,
1701 struct list_head *cancels, int count);
1703 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1704 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1705 const struct ldlm_request *dlm_req,
1706 const struct ldlm_callback_suite *cbs);
1707 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1708 enum ldlm_type type, __u8 with_policy,
1709 enum ldlm_mode mode, __u64 *flags, void *lvb,
1711 const struct lustre_handle *lockh, int rc);
1712 int ldlm_cli_enqueue_local(const struct lu_env *env,
1713 struct ldlm_namespace *ns,
1714 const struct ldlm_res_id *res_id,
1715 enum ldlm_type type, union ldlm_policy_data *policy,
1716 enum ldlm_mode mode, __u64 *flags,
1717 ldlm_blocking_callback blocking,
1718 ldlm_completion_callback completion,
1719 ldlm_glimpse_callback glimpse,
1720 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1721 const __u64 *client_cookie,
1722 struct lustre_handle *lockh);
1723 int ldlm_cli_convert_req(struct ldlm_lock *lock, __u32 *flags, __u64 new_bits);
1724 int ldlm_cli_convert(struct ldlm_lock *lock,
1725 enum ldlm_cancel_flags cancel_flags);
1726 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1727 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1728 enum ldlm_cancel_flags cancel_flags);
1729 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1730 enum ldlm_cancel_flags flags, void *opaque);
1731 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1732 const struct ldlm_res_id *res_id,
1733 union ldlm_policy_data *policy,
1734 enum ldlm_mode mode,
1735 enum ldlm_cancel_flags flags, void *opaque);
1736 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1737 int count, enum ldlm_cancel_flags flags);
1738 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1739 struct list_head *cancels,
1740 union ldlm_policy_data *policy,
1741 enum ldlm_mode mode, __u64 lock_flags,
1742 enum ldlm_cancel_flags cancel_flags,
1744 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1745 enum ldlm_cancel_flags flags);
1746 int ldlm_cli_cancel_list(struct list_head *head, int count,
1747 struct ptlrpc_request *req,
1748 enum ldlm_cancel_flags flags);
1750 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);
1751 int ldlm_cli_inodebits_convert(struct ldlm_lock *lock,
1752 enum ldlm_cancel_flags cancel_flags);
1754 /** @} ldlm_cli_api */
1756 extern unsigned int ldlm_enqueue_min;
1759 /* This has to be here because recursive inclusion sucks. */
1760 int intent_disposition(struct ldlm_reply *rep, int flag);
1761 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1764 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1765 * than one lock_res is dead-lock safe.
1767 enum lock_res_type {
1772 /** Lock resource. */
1773 static inline void lock_res(struct ldlm_resource *res)
1775 spin_lock(&res->lr_lock);
1778 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1779 static inline void lock_res_nested(struct ldlm_resource *res,
1780 enum lock_res_type mode)
1782 spin_lock_nested(&res->lr_lock, mode);
1785 /** Unlock resource. */
1786 static inline void unlock_res(struct ldlm_resource *res)
1788 spin_unlock(&res->lr_lock);
1791 /** Check if resource is already locked, assert if not. */
1792 static inline void check_res_locked(struct ldlm_resource *res)
1794 assert_spin_locked(&res->lr_lock);
1797 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1798 void unlock_res_and_lock(struct ldlm_lock *lock);
1801 /** \defgroup ldlm_pools Various LDLM pool related functions
1802 * There are not used outside of ldlm.
1805 int ldlm_pools_init(void);
1806 void ldlm_pools_fini(void);
1808 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1809 int idx, enum ldlm_side client);
1810 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1811 void ldlm_pool_fini(struct ldlm_pool *pl);
1812 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1813 time64_t ldlm_pool_recalc(struct ldlm_pool *pl);
1814 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1815 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1816 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1817 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1818 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1819 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1820 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1821 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1822 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1825 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1826 const struct ldlm_extent *ex2)
1828 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1831 /* check if @ex1 contains @ex2 */
1832 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1833 const struct ldlm_extent *ex2)
1835 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1838 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);