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/
32 /** \defgroup LDLM Lustre Distributed Lock Manager
34 * Lustre DLM is based on VAX DLM.
35 * Its two main roles are:
36 * - To provide locking assuring consistency of data on all Lustre nodes.
37 * - To allow clients to cache state protected by a lock by holding the
38 * lock until a conflicting lock is requested or it is expired by the LRU.
43 #ifndef _LUSTRE_DLM_H__
44 #define _LUSTRE_DLM_H__
46 #include <lustre_lib.h>
47 #include <lustre_net.h>
48 #include <lustre_import.h>
49 #include <lustre_handles.h>
50 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
53 #include "lustre_dlm_flags.h"
58 extern struct kset *ldlm_ns_kset;
59 extern struct kset *ldlm_svc_kset;
61 #define OBD_LDLM_DEVICENAME "ldlm"
63 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
64 #define LDLM_DEFAULT_MAX_ALIVE 3900 /* 3900 seconds ~65 min */
65 #define LDLM_CTIME_AGE_LIMIT (10)
66 /* if client lock is unused for that time it can be cancelled if any other
67 * client shows interest in that lock, e.g. glimpse is occured. */
68 #define LDLM_DIRTY_AGE_LIMIT (10)
69 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
70 #define LDLM_DEFAULT_LRU_SHRINK_BATCH (16)
71 #define LDLM_DEFAULT_SLV_RECALC_PCT (10)
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 nine 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.
131 * - GROUP: Group mode. Locks with the same group ID are compatible with each
133 * - COS: Commit-on-Sharing mode. If Commit-on-Sharing is enabled, PW/EX locks
134 * held in transactions are downgraded to COS mode after transaction stop.
135 * - TXN: Transaction mode. If Commit-on-Sharing is diabled on a DNE system,
136 * PW/EX locks held in transactions are downgraded to TXN mode after
140 * NL CR CW PR PW EX GROUP COS TXN
141 * NL 1 1 1 1 1 1 1 1 1
142 * CR 1 1 1 1 1 0 0 0 1
143 * CW 1 1 1 0 0 0 0 0 1
144 * PR 1 1 0 1 0 0 0 0 1
145 * PW 1 1 0 0 0 0 0 0 0
146 * EX 1 0 0 0 0 0 0 0 0
147 * GROUP 1 0 0 0 0 0 1 0 0
148 * COS 1 0 0 0 0 0 0 1 0
149 * TXN 1 1 1 1 0 0 0 0 1
153 #define LCK_COMPAT_EX LCK_NL
154 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
155 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR | LCK_TXN)
156 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW | LCK_TXN)
157 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW | LCK_TXN)
158 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP | LCK_COS)
159 #define LCK_COMPAT_GROUP (LCK_NL | LCK_GROUP)
160 #define LCK_COMPAT_COS (LCK_NL | LCK_COS)
161 #define LCK_COMPAT_TXN (LCK_COMPAT_PR | LCK_CW)
162 /** @} Lock Compatibility Matrix */
164 extern enum ldlm_mode lck_compat_array[];
166 static inline void lockmode_verify(enum ldlm_mode mode)
168 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
171 static inline int lockmode_compat(enum ldlm_mode exist_mode,
172 enum ldlm_mode new_mode)
174 return lck_compat_array[exist_mode] & new_mode;
179 * cluster name spaces
183 #define DLM_OST_NAMESPACE 1
184 #define DLM_MDS_NAMESPACE 2
187 - do we just separate this by security domains and use a prefix for
188 multiple namespaces in the same domain?
193 * Locking rules for LDLM:
198 * waiting_locks_spinlock
211 /* Cancel lru flag, it indicates we cancel aged locks. */
212 enum ldlm_lru_flags {
213 LDLM_LRU_FLAG_NO_WAIT = 0x1, /* Cancel locks w/o blocking (neither
214 * sending nor waiting for any RPCs) */
215 LDLM_LRU_FLAG_CLEANUP = 0x2, /* Used when clearing lru, tells
216 * prepare_lru_list to set discard flag
217 * on PR extent locks so we don't waste
218 * time saving pages that will be
219 * discarded momentarily */
224 struct ldlm_resource;
225 struct ldlm_namespace;
228 * Operations on LDLM pools.
229 * LDLM pool is a pool of locks in the namespace without any implicitly
231 * Locks in the pool are organized in LRU.
232 * Local memory pressure or server instructions (e.g. mempressure on server)
233 * can trigger freeing of locks from the pool
235 struct ldlm_pool_ops {
236 /** Recalculate pool \a pl usage */
237 int (*po_recalc)(struct ldlm_pool *pl, bool force);
238 /** Cancel at least \a nr locks from pool \a pl */
239 int (*po_shrink)(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
240 int (*po_setup)(struct ldlm_pool *pl, int limit);
243 /** One second for pools thread check interval. Each pool has own period. */
244 #define LDLM_POOLS_THREAD_PERIOD (1)
246 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
247 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
249 /** Default recalc period for server side pools in sec. */
250 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
252 /** Default recalc period for client side pools in sec. */
253 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
256 * LDLM pool structure to track granted locks.
257 * For purposes of determining when to release locks on e.g. memory pressure.
258 * This feature is commonly referred to as lru_resize.
261 /** Pool debugfs directory. */
262 struct dentry *pl_debugfs_entry;
263 /** Pool name, must be long enough to hold compound proc entry name. */
265 /** Lock for protecting SLV/CLV updates. */
267 /** Number of allowed locks in in pool, both, client and server side. */
269 /** Number of granted locks in */
271 /** Grant rate per T. */
272 atomic_t pl_grant_rate;
273 /** Cancel rate per T. */
274 atomic_t pl_cancel_rate;
275 /** Server lock volume (SLV). Protected by pl_lock. */
276 __u64 pl_server_lock_volume;
277 /** Current biggest client lock volume. Protected by pl_lock. */
278 __u64 pl_client_lock_volume;
279 /** Lock volume factor, shown in percents in procfs, but internally
280 * Client SLV calculated as: server_slv * lock_volume_factor >> 8.
282 atomic_t pl_lock_volume_factor;
283 /** Time when last SLV from server was obtained. */
284 time64_t pl_recalc_time;
285 /** Recalculation period for pool. */
286 time64_t pl_recalc_period;
287 /** Recalculation and shrink operations. */
288 struct ldlm_pool_ops *pl_ops;
289 /** Number of planned locks for next period. */
291 /** Pool statistics. */
292 struct lprocfs_stats *pl_stats;
295 struct kobject pl_kobj;
296 struct completion pl_kobj_unregister;
299 typedef int (*ldlm_res_policy)(const struct lu_env *env,
300 struct ldlm_namespace *,
301 struct ldlm_lock **, void *req_cookie,
302 enum ldlm_mode mode, __u64 flags, void *data);
304 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
308 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
309 * be associated with an LDLM lock and transferred from client to server and
312 * Currently LVBs are used by:
313 * - OSC-OST code to maintain current object size/times
314 * - layout lock code to return the layout when the layout lock is granted
316 * To ensure delayed LVB initialization, it is highly recommended to use the set
317 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
319 struct ldlm_valblock_ops {
320 int (*lvbo_init)(struct ldlm_resource *res);
321 int (*lvbo_update)(struct ldlm_resource *res, struct ldlm_lock *lock,
322 struct ptlrpc_request *r, int increase);
323 int (*lvbo_free)(struct ldlm_resource *res);
324 /* Return size of lvb data appropriate RPC size can be reserved */
325 int (*lvbo_size)(struct ldlm_lock *lock);
326 /* Called to fill in lvb data to RPC buffer @buf */
327 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int *buflen);
331 * LDLM pools related, type of lock pool in the namespace.
332 * Greedy means release cached locks aggressively
335 LDLM_NAMESPACE_GREEDY = BIT(0),
336 LDLM_NAMESPACE_MODEST = BIT(1),
340 * Default values for the "max_nolock_size", "contention_time" and
341 * "contended_locks" namespace tunables.
343 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
344 #define NS_DEFAULT_CONTENTION_SECONDS 2
345 #define NS_DEFAULT_CONTENDED_LOCKS 32
347 struct ldlm_ns_bucket {
348 /** back pointer to namespace */
349 struct ldlm_namespace *nsb_namespace;
351 * Estimated lock callback time. Used by adaptive timeout code to
352 * avoid spurious client evictions due to unresponsiveness when in
353 * fact the network or overall system load is at fault
355 struct adaptive_timeout nsb_at_estimate;
357 * Which res in the bucket should we start with the reclaim.
359 int nsb_reclaim_start;
360 /* counter of entries in this bucket */
365 /** LDLM namespace lock stats */
371 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
372 LDLM_NS_TYPE_MDC, /**< MDC namespace */
373 LDLM_NS_TYPE_MDT, /**< MDT namespace */
374 LDLM_NS_TYPE_OSC, /**< OSC namespace */
375 LDLM_NS_TYPE_OST, /**< OST namespace */
376 LDLM_NS_TYPE_MGC, /**< MGC namespace */
377 LDLM_NS_TYPE_MGT, /**< MGT namespace */
380 enum ldlm_namespace_flags {
382 * Flag to indicate the LRU cancel is in progress.
383 * Used to limit the process by 1 thread only.
391 * Namespace serves to contain locks related to a particular service.
392 * There are two kinds of namespaces:
393 * - Server namespace has knowledge of all locks and is therefore authoritative
394 * to make decisions like what locks could be granted and what conflicts
395 * exist during new lock enqueue.
396 * - Client namespace only has limited knowledge about locks in the namespace,
397 * only seeing locks held by the client.
399 * Every Lustre service has one server namespace present on the server serving
400 * that service. Every client connected to the service has a client namespace
402 * Every lock obtained by client in that namespace is actually represented by
403 * two in-memory locks. One on the server and one on the client. The locks are
404 * linked by a special cookie by which one node can tell to the other which lock
405 * it actually means during communications. Such locks are called remote locks.
406 * The locks held by server only without any reference to a client are called
409 struct ldlm_namespace {
410 /** Backward link to OBD, required for LDLM pool to store new SLV. */
411 struct obd_device *ns_obd;
413 /** Flag indicating if namespace is on client instead of server */
414 enum ldlm_side ns_client;
416 /** name of this namespace */
419 /** Resource hash table for namespace. */
420 struct cfs_hash *ns_rs_hash;
421 struct ldlm_ns_bucket *ns_rs_buckets;
422 unsigned int ns_bucket_bits;
427 /** big refcount (by bucket) */
431 * Namespace connect flags supported by server (may be changed via
432 * /proc, LRU resize may be disabled/enabled).
434 __u64 ns_connect_flags;
436 /** Client side original connect flags supported by server. */
437 __u64 ns_orig_connect_flags;
439 /* namespace debugfs dir entry */
440 struct dentry *ns_debugfs_entry;
443 * Position in global namespace list linking all namespaces on
446 struct list_head ns_list_chain;
449 * List of unused locks for this namespace. This list is also called
451 * Unused locks are locks with zero reader/writer reference counts.
452 * This list is only used on clients for lock caching purposes.
453 * When we want to release some locks voluntarily or if server wants
454 * us to release some locks due to e.g. memory pressure, we take locks
455 * to release from the head of this list.
456 * Locks are linked via l_lru field in \see struct ldlm_lock.
458 struct list_head ns_unused_list;
459 /** Number of locks in the LRU list above */
461 struct list_head *ns_last_pos;
464 * Maximum number of locks permitted in the LRU. If 0, means locks
465 * are managed by pools and there is no preset limit, rather it is all
466 * controlled by available memory on this client and on server.
468 unsigned int ns_max_unused;
471 * Cancel batch, if unused lock count exceed lru_size
472 * Only be used if LRUR disable.
474 unsigned int ns_cancel_batch;
477 * How much the SLV should decrease in %% to trigger LRU cancel urgently.
479 unsigned int ns_recalc_pct;
481 /** Maximum allowed age (last used time) for locks in the LRU. Set in
482 * seconds from userspace, but stored in ns to avoid repeat conversions.
487 * Server only: number of times we evicted clients due to lack of reply
490 unsigned int ns_timeouts;
492 * Number of seconds since the file change time after which
493 * the MDT will return an UPDATE lock along with a LOOKUP lock.
494 * This allows the client to start caching negative dentries
495 * for a directory and may save an RPC for a later stat.
497 timeout_t ns_ctime_age_limit;
499 * Number of (nano)seconds since the lock was last used. The client
500 * may cancel the lock older than this age and flush related data if
501 * another client shows interest in this lock by doing glimpse request.
502 * This allows to cache stat data locally for such files early. Set in
503 * seconds from userspace, but stored in ns to avoid repeat conversions.
505 ktime_t ns_dirty_age_limit;
507 * Used to rate-limit ldlm_namespace_dump calls.
508 * \see ldlm_namespace_dump. Increased by 10 seconds every time
511 time64_t ns_next_dump;
513 /** "policy" function that does actual lock conflict determination */
514 ldlm_res_policy ns_policy;
517 * LVB operations for this namespace.
518 * \see struct ldlm_valblock_ops
520 struct ldlm_valblock_ops *ns_lvbo;
523 * Used by filter code to store pointer to OBD of the service.
524 * Should be dropped in favor of \a ns_obd
529 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
530 * a resource is removed.
532 wait_queue_head_t ns_waitq;
533 /** LDLM pool structure for this namespace */
534 struct ldlm_pool ns_pool;
535 /** Definition of how eagerly unused locks will be released from LRU */
536 enum ldlm_appetite ns_appetite;
539 * If more than \a ns_contended_locks are found, the resource is
540 * considered to be contended. Lock enqueues might specify that no
541 * contended locks should be granted
543 unsigned ns_contended_locks;
546 * The resources in this namespace remember contended state during
547 * \a ns_contention_time, in seconds.
549 timeout_t ns_contention_time;
552 * Limit size of contended extent locks, in bytes.
553 * If extended lock is requested for more then this many bytes and
554 * caller instructs us not to grant contended locks, we would disregard
557 unsigned ns_max_nolock_size;
559 /** Limit of parallel AST RPC count. */
560 unsigned ns_max_parallel_ast;
563 * Callback to check if a lock is good to be canceled by ELC or
566 ldlm_cancel_cbt ns_cancel;
568 /** LDLM lock stats */
569 struct lprocfs_stats *ns_stats;
572 * Flag to indicate namespace is being freed. Used to determine if
573 * recalculation of LDLM pool statistics should be skipped.
575 unsigned ns_stopping:1,
578 * Flag to indicate the LRU recalc on RPC reply is in progress.
579 * Used to limit the process by 1 thread only.
584 * Which bucket should we start with the lock reclaim.
586 int ns_reclaim_start;
588 struct kobject ns_kobj; /* sysfs object */
589 struct completion ns_kobj_unregister;
592 * To avoid another ns_lock usage, a separate bitops field.
594 unsigned long ns_flags;
598 * Returns 1 if namespace \a ns is a client namespace.
600 static inline int ns_is_client(struct ldlm_namespace *ns)
603 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
604 ns->ns_client == LDLM_NAMESPACE_SERVER);
605 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
609 * Returns 1 if namespace \a ns is a server namespace.
611 static inline int ns_is_server(struct ldlm_namespace *ns)
614 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
615 ns->ns_client == LDLM_NAMESPACE_SERVER);
616 return ns->ns_client == LDLM_NAMESPACE_SERVER;
620 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
622 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
625 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
629 * Returns 1 if this namespace supports lru_resize.
631 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
634 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
637 static inline void ns_register_cancel(struct ldlm_namespace *ns,
646 /** Type for blocking callback function of a lock. */
647 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
648 struct ldlm_lock_desc *new, void *data,
650 /** Type for completion callback function of a lock. */
651 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
653 /** Type for glimpse callback function of a lock. */
654 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
656 /** Type for created callback function of a lock. */
657 typedef void (*ldlm_created_callback)(struct ldlm_lock *lock);
659 /** Work list for sending GL ASTs to multiple locks. */
660 struct ldlm_glimpse_work {
661 struct ldlm_lock *gl_lock; /* lock to glimpse */
662 struct list_head gl_list; /* linkage to other gl work structs */
663 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
664 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
665 * glimpse callback request */
666 ptlrpc_interpterer_t gl_interpret_reply;
667 void *gl_interpret_data;
670 struct ldlm_bl_desc {
671 unsigned int bl_same_client:1, /* both ops are from the same client. */
672 bl_txn_dependent:1;/* the op that enqueues lock depends on
673 * the op that holds lock.
677 struct ldlm_cb_set_arg {
678 struct ptlrpc_request_set *set;
679 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
681 struct list_head *list;
682 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
683 ptlrpc_interpterer_t gl_interpret_reply;
684 void *gl_interpret_data;
685 struct ldlm_bl_desc *bl_desc;
688 struct ldlm_cb_async_args {
689 struct ldlm_cb_set_arg *ca_set_arg;
690 struct ldlm_lock *ca_lock;
693 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
694 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
696 /** Interval node data for each LDLM_EXTENT lock. */
697 struct ldlm_interval {
698 struct interval_node li_node; /* node for tree management */
699 struct list_head li_group; /* the locks which have the same
700 * policy - group of the policy */
702 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
705 * Interval tree for extent locks.
706 * The interval tree must be accessed under the resource lock.
707 * Interval trees are used for granted extent locks to speed up conflicts
708 * lookup. See ldlm/interval_tree.c for more details.
710 struct ldlm_interval_tree {
713 enum ldlm_mode lit_mode; /* lock mode */
714 struct interval_node *lit_root; /* actual ldlm_interval */
718 * Lists of waiting locks for each inodebit type.
719 * A lock can be in several liq_waiting lists and it remains in lr_waiting.
721 struct ldlm_ibits_queues {
722 struct list_head liq_waiting[MDS_INODELOCK_NUMBITS];
725 struct ldlm_ibits_node {
726 struct list_head lin_link[MDS_INODELOCK_NUMBITS];
727 struct ldlm_lock *lock;
730 struct ldlm_flock_node {
731 atomic_t lfn_unlock_pending;
732 bool lfn_needs_reprocess;
735 /** Whether to track references to exports by LDLM locks. */
736 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
739 enum ldlm_cancel_flags {
740 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
741 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
742 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
749 __u64 blocking_owner;
750 struct obd_export *blocking_export;
751 atomic_t blocking_refs;
755 union ldlm_policy_data {
756 struct ldlm_extent l_extent;
757 struct ldlm_flock l_flock;
758 struct ldlm_inodebits l_inodebits;
761 void ldlm_convert_policy_to_wire(enum ldlm_type type,
762 const union ldlm_policy_data *lpolicy,
763 union ldlm_wire_policy_data *wpolicy);
764 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
765 const union ldlm_wire_policy_data *wpolicy,
766 union ldlm_policy_data *lpolicy);
776 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
778 #define LDLM_GID_ANY ((__u64)-1)
781 * LDLM lock structure
783 * Represents a single LDLM lock and its state in memory. Each lock is
784 * associated with a single ldlm_resource, the object which is being
785 * locked. There may be multiple ldlm_locks on a single resource,
786 * depending on the lock type and whether the locks are conflicting or
792 * When remote side wants to tell us about a lock, they address
793 * it by this opaque handle. The handle does not hold a
794 * reference on the ldlm_lock, so it can be safely passed to
795 * other threads or nodes. When the lock needs to be accessed
796 * from the handle, it is looked up again in the lock table, and
797 * may no longer exist.
799 * Must be first in the structure.
801 struct portals_handle l_handle;
803 * Pointer to actual resource this lock is in.
804 * ldlm_lock_change_resource() can change this on the client.
805 * When this is possible, rcu must be used to stablise
806 * the resource while we lock and check it hasn't been changed.
808 struct ldlm_resource *l_resource;
810 * List item for client side LRU list.
811 * Protected by ns_lock in struct ldlm_namespace.
813 struct list_head l_lru;
815 * Linkage to resource's lock queues according to current lock state.
816 * (could be granted or waiting)
817 * Protected by lr_lock in struct ldlm_resource.
819 struct list_head l_res_link;
821 * Internal structures per lock type..
824 struct ldlm_interval *l_tree_node;
825 struct ldlm_ibits_node *l_ibits_node;
828 * Per export hash of locks.
829 * Protected by per-bucket exp->exp_lock_hash locks.
831 struct hlist_node l_exp_hash;
833 * Per export hash of flock locks.
834 * Protected by per-bucket exp->exp_flock_hash locks.
836 struct hlist_node l_exp_flock_hash;
839 * Protected by lr_lock.
841 enum ldlm_mode l_req_mode;
843 * Granted mode, also protected by lr_lock.
845 enum ldlm_mode l_granted_mode;
846 /** Lock completion handler pointer. Called when lock is granted. */
847 ldlm_completion_callback l_completion_ast;
849 * Lock blocking AST handler pointer.
850 * It plays two roles:
851 * - as a notification of an attempt to queue a conflicting lock (once)
852 * - as a notification when the lock is being cancelled.
854 * As such it's typically called twice: once for the initial conflict
855 * and then once more when the last user went away and the lock is
856 * cancelled (could happen recursively).
858 ldlm_blocking_callback l_blocking_ast;
860 * Lock glimpse handler.
861 * Glimpse handler is used to obtain LVB updates from a client by
864 ldlm_glimpse_callback l_glimpse_ast;
868 * This is a pointer to actual client export for locks that were granted
869 * to clients. Used server-side.
871 struct obd_export *l_export;
873 * Lock connection export.
874 * Pointer to server export on a client.
876 struct obd_export *l_conn_export;
879 * Remote lock handle.
880 * If the lock is remote, this is the handle of the other side lock
883 struct lustre_handle l_remote_handle;
886 * Representation of private data specific for a lock type.
887 * Examples are: extent range for extent lock or bitmask for ibits locks
889 union ldlm_policy_data l_policy_data;
892 * Lock state flags. Protected by lr_lock.
893 * \see lustre_dlm_flags.h where the bits are defined.
898 * Lock r/w usage counters.
899 * Protected by lr_lock.
904 * If the lock is granted, a process sleeps on this waitq to learn when
905 * it's no longer in use. If the lock is not granted, a process sleeps
906 * on this waitq to learn when it becomes granted.
908 wait_queue_head_t l_waitq;
911 * Time, in nanoseconds, last used by e.g. being matched by lock match.
915 /** Originally requested extent for the extent lock. */
916 struct ldlm_extent l_req_extent;
919 * Client-side-only members.
922 enum lvb_type l_lvb_type;
925 * Temporary storage for a LVB received during an enqueue operation.
926 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
931 /** Private storage for lock user. Opaque to LDLM. */
936 * Seconds. It will be updated if there is any activity related to
937 * the lock at client, e.g. enqueue the lock. For server it is the
938 * time when blocking ast was sent.
941 time64_t l_blast_sent;
944 /* separate ost_lvb used mostly by Data-on-MDT for now.
945 * It is introduced to don't mix with layout lock data. */
946 struct ost_lvb l_ost_lvb;
948 * Server-side-only members.
952 * Connection cookie for the client originating the operation.
953 * Used by Commit on Share (COS) code. Currently only used for
954 * inodebits locks on MDS.
956 __u64 l_client_cookie;
959 * List item for locks waiting for cancellation from clients.
960 * The lists this could be linked into are:
961 * waiting_locks_list (protected by waiting_locks_spinlock),
962 * then if the lock timed out, it is moved to
963 * expired_lock_list for further processing.
965 struct list_head l_pending_chain;
968 * Set when lock is sent a blocking AST. Time in seconds when timeout
969 * is reached and client holding this lock could be evicted.
970 * This timeout could be further extended by e.g. certain IO activity
972 * \see ost_rw_prolong_locks
974 time64_t l_callback_timestamp;
976 /** Local PID of process which created this lock. */
980 * Number of times blocking AST was sent for this lock.
981 * This is for debugging. Valid values are 0 and 1, if there is an
982 * attempt to send blocking AST more than once, an assertion would be
983 * hit. \see ldlm_work_bl_ast_lock
986 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
987 struct list_head l_bl_ast;
988 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
989 struct list_head l_cp_ast;
990 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
991 struct list_head l_rk_ast;
994 * Pointer to a conflicting lock that caused blocking AST to be sent
997 struct ldlm_lock *l_blocking_lock;
1000 * Protected by lr_lock, linkages to "skip lists".
1001 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
1003 struct list_head l_sl_mode;
1004 struct list_head l_sl_policy;
1006 /** Reference tracking structure to debug leaked locks. */
1007 struct lu_ref l_reference;
1008 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1009 /* Debugging stuff for bug 20498, for tracking export references. */
1010 /** number of export references taken */
1012 /** link all locks referencing one export */
1013 struct list_head l_exp_refs_link;
1014 /** referenced export object */
1015 struct obd_export *l_exp_refs_target;
1018 * export blocking dlm lock list, protected by
1019 * l_export->exp_bl_list_lock.
1020 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
1021 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
1023 struct list_head l_exp_list;
1026 enum ldlm_match_flags {
1027 LDLM_MATCH_UNREF = BIT(0),
1028 LDLM_MATCH_AST = BIT(1),
1029 LDLM_MATCH_AST_ANY = BIT(2),
1030 LDLM_MATCH_RIGHT = BIT(3),
1031 LDLM_MATCH_GROUP = BIT(4),
1035 * Describe the overlap between two locks. itree_overlap_cb data.
1037 struct ldlm_match_data {
1038 struct ldlm_lock *lmd_old;
1039 struct ldlm_lock *lmd_lock;
1040 enum ldlm_mode *lmd_mode;
1041 union ldlm_policy_data *lmd_policy;
1043 __u64 lmd_skip_flags;
1044 enum ldlm_match_flags lmd_match;
1047 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
1048 #define l_transno l_client_cookie
1050 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
1051 * which is for server. */
1052 #define l_slc_link l_rk_ast
1054 struct lustre_handle_array {
1055 unsigned int ha_count;
1056 /* ha_map is used as bit flag to indicate handle is remote or local */
1057 DECLARE_BITMAP(ha_map, LMV_MAX_STRIPE_COUNT);
1058 struct lustre_handle ha_handles[0];
1062 * LDLM resource description.
1063 * Basically, resource is a representation for a single object.
1064 * Object has a name which is currently 4 64-bit integers. LDLM user is
1065 * responsible for creation of a mapping between objects it wants to be
1066 * protected and resource names.
1068 * A resource can only hold locks of a single lock type, though there may be
1069 * multiple ldlm_locks on a single resource, depending on the lock type and
1070 * whether the locks are conflicting or not.
1072 struct ldlm_resource {
1073 struct ldlm_ns_bucket *lr_ns_bucket;
1076 * List item for list in namespace hash.
1077 * protected by ns_lock.
1078 * Shared with linkage for RCU-delayed free.
1081 struct hlist_node lr_hash;
1082 struct rcu_head lr_rcu;
1085 /** Reference count for this resource */
1086 atomic_t lr_refcount;
1088 /** Spinlock to protect locks under this resource. */
1092 * protected by lr_lock
1094 /** List of locks in granted state */
1095 struct list_head lr_granted;
1097 * List of locks that could not be granted due to conflicts and
1098 * that are waiting for conflicts to go away */
1099 struct list_head lr_waiting;
1102 /** Resource name */
1103 struct ldlm_res_id lr_name;
1107 * Interval trees (only for extent locks) for all modes of
1110 struct ldlm_interval_tree *lr_itree;
1111 struct ldlm_ibits_queues *lr_ibits_queues;
1112 struct ldlm_flock_node lr_flock_node;
1117 * When the resource was considered as contended,
1118 * used only on server side.
1120 time64_t lr_contention_time;
1122 * Associated inode, used only on client side.
1124 struct inode *lr_lvb_inode;
1127 /** Type of locks this resource can hold. Only one type per resource. */
1128 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1131 * Server-side-only lock value block elements.
1132 * To serialize lvbo_init.
1135 struct mutex lr_lvb_mutex;
1136 /** protected by lr_lock */
1138 /** is lvb initialized ? */
1139 bool lr_lvb_initialized;
1141 /** List of references to this resource. For debugging. */
1142 struct lu_ref lr_reference;
1145 static inline int ldlm_is_granted(struct ldlm_lock *lock)
1147 return lock->l_req_mode == lock->l_granted_mode;
1150 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1152 return lock->l_resource->lr_type == LDLM_IBITS &&
1153 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1156 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1158 return lock->l_resource->lr_type == LDLM_IBITS &&
1159 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1162 static inline char *
1163 ldlm_ns_name(struct ldlm_namespace *ns)
1168 static inline struct ldlm_namespace *
1169 ldlm_res_to_ns(struct ldlm_resource *res)
1171 return res->lr_ns_bucket->nsb_namespace;
1174 static inline struct ldlm_namespace *
1175 ldlm_lock_to_ns(struct ldlm_lock *lock)
1177 return ldlm_res_to_ns(lock->l_resource);
1180 static inline char *
1181 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1183 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1186 static inline struct adaptive_timeout *
1187 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1189 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1192 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1194 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1197 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1198 res->lr_lvb_initialized)
1201 mutex_lock(&res->lr_lvb_mutex);
1202 /* Did we lose the race? */
1203 if (res->lr_lvb_initialized) {
1204 mutex_unlock(&res->lr_lvb_mutex);
1207 rc = ns->ns_lvbo->lvbo_init(res);
1209 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1211 if (res->lr_lvb_data != NULL) {
1212 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1213 res->lr_lvb_data = NULL;
1215 res->lr_lvb_len = rc;
1217 res->lr_lvb_initialized = true;
1219 mutex_unlock(&res->lr_lvb_mutex);
1223 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1225 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1227 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1228 return ns->ns_lvbo->lvbo_size(lock);
1233 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int *len)
1235 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1238 if (ns->ns_lvbo != NULL) {
1239 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1240 /* init lvb now if not already */
1241 rc = ldlm_lvbo_init(lock->l_resource);
1243 CERROR("lock %p: delayed lvb init failed (rc %d)\n",
1247 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1252 struct ldlm_ast_work {
1253 struct ldlm_lock *w_lock;
1255 struct ldlm_lock_desc w_desc;
1256 struct list_head w_list;
1263 * Common ldlm_enqueue parameters
1265 struct ldlm_enqueue_info {
1266 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1267 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1268 void *ei_cb_bl; /** blocking lock callback */
1269 void *ei_cb_local_bl; /** blocking local lock callback */
1270 void *ei_cb_cp; /** lock completion callback */
1271 void *ei_cb_gl; /** lock glimpse callback */
1272 ldlm_created_callback ei_cb_created; /** lock created callback */
1273 void *ei_cbdata; /** Data to be passed into callbacks. */
1274 void *ei_namespace; /** lock namespace **/
1275 u64 ei_inodebits; /** lock inode bits **/
1276 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1277 unsigned int ei_req_slot:1; /** whether acquire rpc slot */
1278 unsigned int ei_mod_slot:1; /** whether acquire mod rpc slot */
1281 #define ei_res_id ei_cb_gl
1283 extern char *ldlm_lockname[];
1284 extern char *ldlm_typename[];
1285 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1288 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1289 * For the cases where we do not have actual lock to print along
1290 * with a debugging message that is ldlm-related
1292 #define LDLM_DEBUG_NOLOCK(format, a...) \
1293 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1296 * Support function for lock information printing into debug logs.
1300 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1301 CFS_CHECK_STACK(msgdata, mask, cdls); \
1303 if (((mask) & D_CANTMASK) != 0 || \
1304 ((libcfs_debug & (mask)) != 0 && \
1305 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1306 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1309 void _ldlm_lock_debug(struct ldlm_lock *lock,
1310 struct libcfs_debug_msg_data *data,
1311 const char *fmt, ...)
1312 __attribute__ ((format (printf, 3, 4)));
1315 * Rate-limited version of lock printing function.
1317 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1318 static struct cfs_debug_limit_state _ldlm_cdls; \
1319 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1320 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1323 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1324 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1326 /** Non-rate-limited lock printing function for debugging purposes. */
1327 #define LDLM_DEBUG(lock, fmt, a...) do { \
1328 if (likely(lock != NULL)) { \
1329 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1330 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1331 "### " fmt , ##a); \
1333 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1336 #else /* !LIBCFS_DEBUG */
1337 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1338 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1339 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1343 * Three intentions can be used for the policy functions in
1344 * ldlm_processing_policy.
1346 * LDLM_PROCESS_RESCAN:
1348 * It's used when policy functions are called from ldlm_reprocess_queue() to
1349 * reprocess the wait list and try to grant locks, blocking ASTs
1350 * have already been sent in this situation, completion ASTs need be sent for
1351 * the locks being granted.
1353 * LDLM_PROCESS_ENQUEUE:
1355 * It's used when policy functions are called from ldlm_lock_enqueue() to
1356 * process the wait list for handling an enqueue request, blocking
1357 * ASTs have not been sent yet, so list of conflicting locks would be
1358 * collected and ASTs sent.
1360 * LDLM_PROCESS_RECOVERY:
1362 * It's used when policy functions are called from ldlm_reprocess_queue() to
1363 * reprocess the wait list when recovery done. In case of blocking
1364 * ASTs are lost before recovery, it needs not only to grant locks if
1365 * available, but also send blocking ASTs to the locks doesn't have AST sent
1366 * flag. Completion ASTs need be sent for the locks being granted.
1368 enum ldlm_process_intention {
1369 LDLM_PROCESS_RESCAN = 0,
1370 LDLM_PROCESS_ENQUEUE = 1,
1371 LDLM_PROCESS_RECOVERY = 2,
1374 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1375 enum ldlm_process_intention intention,
1376 enum ldlm_error *err,
1377 struct list_head *work_list);
1379 typedef int (*ldlm_reprocessing_policy)(struct ldlm_resource *res,
1380 struct list_head *queue,
1381 struct list_head *work_list,
1382 enum ldlm_process_intention intention,
1386 * Return values for lock iterators.
1387 * Also used during deciding of lock grants and cancellations.
1389 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1390 #define LDLM_ITER_STOP 2 /* stop iterating */
1392 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1393 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1395 /** \defgroup ldlm_iterator Lock iterators
1397 * LDLM provides for a way to iterate through every lock on a resource or
1398 * namespace or every resource in a namespace.
1400 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1402 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1404 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1405 ldlm_iterator_t iter, void *data);
1406 /** @} ldlm_iterator */
1408 int ldlm_replay_locks(struct obd_import *imp);
1411 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1414 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1416 struct ldlm_prolong_args {
1417 struct obd_export *lpa_export;
1418 struct ptlrpc_request *lpa_req;
1419 struct ldlm_res_id lpa_resid;
1420 struct ldlm_extent lpa_extent;
1421 enum ldlm_mode lpa_mode;
1425 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1426 struct ldlm_prolong_args *arg);
1427 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1429 struct ldlm_callback_suite {
1430 ldlm_completion_callback lcs_completion;
1431 ldlm_blocking_callback lcs_blocking;
1432 ldlm_glimpse_callback lcs_glimpse;
1436 #ifdef HAVE_SERVER_SUPPORT
1437 /** \defgroup ldlm_srv_ast Server AST handlers
1438 * These are AST handlers used by server code.
1439 * Their property is that they are just preparing RPCs to be sent to clients.
1442 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1443 void *data, int flag);
1444 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1445 void *data, int flag);
1446 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1447 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1448 int ldlm_glimpse_locks(struct ldlm_resource *res,
1449 struct list_head *gl_work_list);
1450 /** @} ldlm_srv_ast */
1452 /** \defgroup ldlm_handlers Server LDLM handlers
1453 * These are handler functions that should be called by "frontends" such as
1454 * MDT or OST to pass through LDLM requests to LDLM for handling
1457 int ldlm_handle_enqueue(struct ldlm_namespace *ns, struct req_capsule *pill,
1458 const struct ldlm_request *dlm_req,
1459 const struct ldlm_callback_suite *cbs);
1460 int ldlm_handle_convert0(struct ptlrpc_request *req,
1461 const struct ldlm_request *dlm_req);
1462 int ldlm_handle_cancel(struct ptlrpc_request *req);
1463 int ldlm_request_cancel(struct ptlrpc_request *req,
1464 const struct ldlm_request *dlm_req,
1465 int first, enum lustre_at_flags flags);
1466 /** @} ldlm_handlers */
1468 void ldlm_revoke_export_locks(struct obd_export *exp);
1469 timeout_t ldlm_bl_timeout(struct ldlm_lock *lock);
1470 timeout_t ldlm_bl_timeout_by_rpc(struct ptlrpc_request *req);
1472 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1473 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, timeout_t timeout);
1474 int ldlm_get_ref(void);
1475 void ldlm_put_ref(void);
1476 int ldlm_init_export(struct obd_export *exp);
1477 void ldlm_destroy_export(struct obd_export *exp);
1478 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1481 #ifdef HAVE_SERVER_SUPPORT
1482 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1483 ldlm_reprocessing_policy
1484 ldlm_get_reprocessing_policy(struct ldlm_resource *res);
1486 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1487 void ldlm_lock2handle(const struct ldlm_lock *lock,
1488 struct lustre_handle *lockh);
1489 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1490 void ldlm_cancel_callback(struct ldlm_lock *);
1491 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1492 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1495 * Obtain a lock reference by its handle.
1497 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1499 return __ldlm_handle2lock(h, 0);
1502 #define LDLM_LOCK_REF_DEL(lock) \
1503 lu_ref_del(&lock->l_reference, "handle", lock)
1505 static inline struct ldlm_lock *
1506 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1508 struct ldlm_lock *lock;
1510 lock = __ldlm_handle2lock(h, flags);
1512 LDLM_LOCK_REF_DEL(lock);
1517 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1518 * data from request \a r
1520 static inline int ldlm_lvbo_update(struct ldlm_resource *res,
1521 struct ldlm_lock *lock,
1522 struct ptlrpc_request *req, int increase)
1524 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1527 /* delayed lvb init may be required */
1528 rc = ldlm_lvbo_init(res);
1530 CERROR("delayed lvb init failed (rc %d)\n", rc);
1534 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1535 return ns->ns_lvbo->lvbo_update(res, lock, req, increase);
1540 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1541 struct ptlrpc_request *req,
1544 return ldlm_lvbo_update(res, NULL, req, increase);
1547 int is_granted_or_cancelled_nolock(struct ldlm_lock *lock);
1549 int ldlm_error2errno(enum ldlm_error error);
1550 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1551 * confuses user-space. */
1552 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1553 void ldlm_dump_export_locks(struct obd_export *exp);
1557 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1558 * __ldlm_handle2lock().
1560 #define LDLM_LOCK_PUT(lock) \
1562 LDLM_LOCK_REF_DEL(lock); \
1563 /*LDLM_DEBUG((lock), "put");*/ \
1564 ldlm_lock_put(lock); \
1568 * Release a lock reference obtained by some other means (see
1571 #define LDLM_LOCK_RELEASE(lock) \
1573 /*LDLM_DEBUG((lock), "put");*/ \
1574 ldlm_lock_put(lock); \
1577 #define LDLM_LOCK_GET(lock) \
1579 ldlm_lock_get(lock); \
1580 /*LDLM_DEBUG((lock), "get");*/ \
1584 #define ldlm_lock_list_put(head, member, count) \
1586 struct ldlm_lock *_lock, *_next; \
1588 list_for_each_entry_safe(_lock, _next, head, member) { \
1591 list_del_init(&_lock->member); \
1592 LDLM_LOCK_RELEASE(_lock); \
1597 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1598 void ldlm_lock_put(struct ldlm_lock *lock);
1599 void ldlm_lock_destroy(struct ldlm_lock *lock);
1600 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1601 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1602 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1603 enum ldlm_mode mode);
1604 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1605 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1606 enum ldlm_mode mode);
1607 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1608 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1609 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1610 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1612 enum ldlm_mode ldlm_lock_match_with_skip(struct ldlm_namespace *ns,
1613 __u64 flags, __u64 skip_flags,
1614 const struct ldlm_res_id *res_id,
1615 enum ldlm_type type,
1616 union ldlm_policy_data *policy,
1617 enum ldlm_mode mode,
1618 struct lustre_handle *lh,
1619 enum ldlm_match_flags match_flags);
1620 static inline enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns,
1622 const struct ldlm_res_id *res_id,
1623 enum ldlm_type type,
1624 union ldlm_policy_data *policy,
1625 enum ldlm_mode mode,
1626 struct lustre_handle *lh)
1628 return ldlm_lock_match_with_skip(ns, flags, 0, res_id, type, policy,
1631 struct ldlm_lock *search_itree(struct ldlm_resource *res,
1632 struct ldlm_match_data *data);
1633 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1635 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1636 void ldlm_lock_cancel(struct ldlm_lock *lock);
1637 void ldlm_reprocess_all(struct ldlm_resource *res, __u64 hint);
1638 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1639 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1640 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1643 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1644 enum ldlm_side client,
1645 enum ldlm_appetite apt,
1646 enum ldlm_ns_type ns_type);
1647 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1648 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1649 struct obd_import *imp,
1651 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1652 void ldlm_namespace_free(struct ldlm_namespace *ns,
1653 struct obd_import *imp, int force);
1654 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1655 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1656 enum ldlm_side client);
1657 void ldlm_namespace_get(struct ldlm_namespace *ns);
1658 void ldlm_namespace_put(struct ldlm_namespace *ns);
1660 int ldlm_debugfs_setup(void);
1661 void ldlm_debugfs_cleanup(void);
1663 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1664 struct lprocfs_stats *srv_stats)
1666 int lock_type = 0, op = 0;
1668 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1670 switch (lock_type) {
1672 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1675 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1678 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1681 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1689 lprocfs_counter_incr(srv_stats, op);
1692 /* resource.c - internal */
1693 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1694 const struct ldlm_res_id *,
1695 enum ldlm_type type, int create);
1696 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1697 int ldlm_resource_putref(struct ldlm_resource *res);
1698 void ldlm_resource_add_lock(struct ldlm_resource *res,
1699 struct list_head *head,
1700 struct ldlm_lock *lock);
1701 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1702 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1703 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1704 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1705 void ldlm_resource_dump(int level, struct ldlm_resource *);
1706 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1707 const struct ldlm_res_id *);
1709 #define LDLM_RESOURCE_ADDREF(res) do { \
1710 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1713 #define LDLM_RESOURCE_DELREF(res) do { \
1714 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1717 /* ldlm_request.c */
1718 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1719 * These AST handlers are typically used for server-side local locks and are
1720 * also used by client-side lock handlers to perform minimum level base
1723 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1724 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1725 void *data, int flag);
1726 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1727 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1728 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1729 /** @} ldlm_local_ast */
1731 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1732 * These are typically used by client and server (*_local versions)
1733 * to obtain and release locks.
1735 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1736 struct ldlm_enqueue_info *einfo,
1737 const struct ldlm_res_id *res_id,
1738 union ldlm_policy_data const *policy, __u64 *flags,
1739 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1740 struct lustre_handle *lockh, int async);
1741 int ldlm_prep_enqueue_req(struct obd_export *exp,
1742 struct ptlrpc_request *req,
1743 struct list_head *cancels,
1745 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1746 int version, int opc, int canceloff,
1747 struct list_head *cancels, int count);
1749 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1750 int ldlm_handle_enqueue(struct ldlm_namespace *ns, struct req_capsule *pill,
1751 const struct ldlm_request *dlm_req,
1752 const struct ldlm_callback_suite *cbs);
1753 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct req_capsule *pill,
1754 struct ldlm_enqueue_info *einfo, __u8 with_policy,
1755 __u64 *flags, void *lvb, __u32 lvb_len,
1756 const struct lustre_handle *lockh, int rc,
1758 int ldlm_cli_enqueue_local(const struct lu_env *env,
1759 struct ldlm_namespace *ns,
1760 const struct ldlm_res_id *res_id,
1761 enum ldlm_type type, union ldlm_policy_data *policy,
1762 enum ldlm_mode mode, __u64 *flags,
1763 ldlm_blocking_callback blocking,
1764 ldlm_completion_callback completion,
1765 ldlm_glimpse_callback glimpse,
1766 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1767 const __u64 *client_cookie,
1768 struct lustre_handle *lockh);
1769 int ldlm_cli_lock_create_pack(struct obd_export *exp,
1770 struct ldlm_request *dlmreq,
1771 struct ldlm_enqueue_info *einfo,
1772 const struct ldlm_res_id *res_id,
1773 union ldlm_policy_data const *policy,
1774 __u64 *flags, void *lvb, __u32 lvb_len,
1775 enum lvb_type lvb_type,
1776 struct lustre_handle *lockh);
1777 int ldlm_cli_convert_req(struct ldlm_lock *lock, __u32 *flags, __u64 new_bits);
1778 int ldlm_cli_convert(struct ldlm_lock *lock,
1779 enum ldlm_cancel_flags cancel_flags);
1780 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1781 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1782 enum ldlm_cancel_flags cancel_flags);
1783 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1784 enum ldlm_cancel_flags flags, void *opaque);
1785 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1786 const struct ldlm_res_id *res_id,
1787 union ldlm_policy_data *policy,
1788 enum ldlm_mode mode,
1789 enum ldlm_cancel_flags flags, void *opaque);
1790 int ldlm_cli_cancel_req(struct obd_export *exp, struct ldlm_lock *lock,
1791 struct list_head *head, int count,
1792 enum ldlm_cancel_flags flags);
1793 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1794 struct list_head *cancels,
1795 union ldlm_policy_data *policy,
1796 enum ldlm_mode mode, __u64 lock_flags,
1797 enum ldlm_cancel_flags cancel_flags,
1799 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1800 enum ldlm_cancel_flags flags);
1801 int ldlm_cli_cancel_list(struct list_head *head, int count,
1802 struct ptlrpc_request *req,
1803 enum ldlm_cancel_flags flags);
1805 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);
1806 int ldlm_cli_inodebits_convert(struct ldlm_lock *lock,
1807 enum ldlm_cancel_flags cancel_flags);
1809 /** @} ldlm_cli_api */
1811 extern unsigned int ldlm_enqueue_min;
1814 /* This has to be here because recursive inclusion sucks. */
1815 int intent_disposition(struct ldlm_reply *rep, int flag);
1816 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1819 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1820 * than one lock_res is dead-lock safe.
1822 enum lock_res_type {
1827 /** Lock resource. */
1828 static inline void lock_res(struct ldlm_resource *res)
1830 spin_lock(&res->lr_lock);
1833 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1834 static inline void lock_res_nested(struct ldlm_resource *res,
1835 enum lock_res_type mode)
1837 spin_lock_nested(&res->lr_lock, mode);
1840 /** Unlock resource. */
1841 static inline void unlock_res(struct ldlm_resource *res)
1843 spin_unlock(&res->lr_lock);
1846 /** Check if resource is already locked, assert if not. */
1847 static inline void check_res_locked(struct ldlm_resource *res)
1849 assert_spin_locked(&res->lr_lock);
1852 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1853 void unlock_res_and_lock(struct ldlm_lock *lock);
1856 /** \defgroup ldlm_pools Various LDLM pool related functions
1857 * There are not used outside of ldlm.
1860 int ldlm_pools_init(void);
1861 void ldlm_pools_fini(void);
1863 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1864 int idx, enum ldlm_side client);
1865 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1866 void ldlm_pool_fini(struct ldlm_pool *pl);
1867 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1868 time64_t ldlm_pool_recalc(struct ldlm_pool *pl, bool force);
1869 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1870 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1871 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1872 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1873 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1874 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1875 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1876 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1877 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1880 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1881 const struct ldlm_extent *ex2)
1883 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1886 /* check if @ex1 contains @ex2 */
1887 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1888 const struct ldlm_extent *ex2)
1890 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1893 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);