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.
69 #define LDLM_DIRTY_AGE_LIMIT (10)
70 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
71 #define LDLM_DEFAULT_LRU_SHRINK_BATCH (16)
72 #define LDLM_DEFAULT_SLV_RECALC_PCT (10)
75 * LDLM non-error return states
79 ELDLM_LOCK_MATCHED = 1,
81 ELDLM_LOCK_CHANGED = 300,
82 ELDLM_LOCK_ABORTED = 301,
83 ELDLM_LOCK_REPLACED = 302,
84 ELDLM_NO_LOCK_DATA = 303,
85 ELDLM_LOCK_WOULDBLOCK = 304,
87 ELDLM_NAMESPACE_EXISTS = 400,
88 ELDLM_BAD_NAMESPACE = 401,
92 * LDLM namespace type.
93 * The "client" type is actually an indication that this is a narrow local view
94 * into complete namespace on the server. Such namespaces cannot make any
95 * decisions about lack of conflicts or do any autonomous lock granting without
96 * first speaking to a server.
99 LDLM_NAMESPACE_SERVER = 0x01,
100 LDLM_NAMESPACE_CLIENT = 0x02
104 * The blocking callback is overloaded to perform two functions. These flags
105 * indicate which operation should be performed.
107 #define LDLM_CB_BLOCKING 1
108 #define LDLM_CB_CANCELING 2
111 * \name Lock Compatibility Matrix.
113 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
114 * Lock types are described in their respective implementation files:
115 * ldlm_{extent,flock,inodebits,plain}.c.
117 * There are nine lock modes along with a compatibility matrix to indicate if
118 * two locks are compatible.
120 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
122 * - PW: Protective Write (normal write) mode. When a client requests a write
123 * lock from an OST, a lock with PW mode will be issued.
124 * - PR: Protective Read (normal read) mode. When a client requests a read from
125 * an OST, a lock with PR mode will be issued. Also, if the client opens a
126 * file for execution, it is granted a lock with PR mode.
127 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
128 * requests a write lock during a file open operation.
129 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
130 * an inodebit lock with the CR mode on the intermediate path component.
132 * - GROUP: Group mode. Locks with the same group ID are compatible with each
134 * - COS: Commit-on-Sharing mode. If Commit-on-Sharing is enabled, PW/EX locks
135 * held in transactions are downgraded to COS mode after transaction stop.
136 * - TXN: Transaction mode. If Commit-on-Sharing is diabled on a DNE system,
137 * PW/EX locks held in transactions are downgraded to TXN mode after
141 * NL CR CW PR PW EX GROUP COS TXN
142 * NL 1 1 1 1 1 1 1 1 1
143 * CR 1 1 1 1 1 0 0 0 1
144 * CW 1 1 1 0 0 0 0 0 0
145 * PR 1 1 0 1 0 0 0 0 1
146 * PW 1 1 0 0 0 0 0 0 0
147 * EX 1 0 0 0 0 0 0 0 0
148 * GROUP 1 0 0 0 0 0 1 0 0
149 * COS 1 0 0 0 0 0 0 1 0
150 * TXN 1 1 0 1 0 0 0 0 1
154 #define LCK_COMPAT_EX LCK_NL
155 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
156 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR | LCK_TXN)
157 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
158 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW | LCK_TXN)
159 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP | LCK_COS)
160 #define LCK_COMPAT_GROUP (LCK_NL | LCK_GROUP)
161 #define LCK_COMPAT_COS (LCK_NL | LCK_COS)
162 #define LCK_COMPAT_TXN LCK_COMPAT_PR
163 /** @} Lock Compatibility Matrix */
165 extern enum ldlm_mode lck_compat_array[];
167 static inline void lockmode_verify(enum ldlm_mode mode)
169 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
172 static inline int lockmode_compat(enum ldlm_mode exist_mode,
173 enum ldlm_mode new_mode)
175 return lck_compat_array[exist_mode] & new_mode;
179 * cluster name spaces
181 #define DLM_OST_NAMESPACE 1
182 #define DLM_MDS_NAMESPACE 2
185 * - do we just separate this by security domains and use a prefix for
186 * multiple namespaces in the same domain?
191 * Locking rules for LDLM:
196 * waiting_locks_spinlock
209 /* Cancel lru flag, it indicates we cancel aged locks. */
210 enum ldlm_lru_flags {
211 LDLM_LRU_FLAG_NO_WAIT = 0x1, /* Cancel locks w/o blocking (neither
212 * sending nor waiting for any RPCs)
214 LDLM_LRU_FLAG_CLEANUP = 0x2, /* Used when clearing lru, tells
215 * prepare_lru_list to set discard flag
216 * on PR extent locks so we don't waste
217 * time saving pages that will be
218 * 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;
476 /* How much SLV should decrease in %% to trigger LRU cancel urgently. */
477 unsigned int ns_recalc_pct;
479 /** Maximum allowed age (last used time) for locks in the LRU. Set in
480 * seconds from userspace, but stored in ns to avoid repeat conversions.
485 * Server only: number of times we evicted clients due to lack of reply
488 unsigned int ns_timeouts;
490 * Number of seconds since the file change time after which
491 * the MDT will return an UPDATE lock along with a LOOKUP lock.
492 * This allows the client to start caching negative dentries
493 * for a directory and may save an RPC for a later stat.
495 timeout_t ns_ctime_age_limit;
497 * Number of (nano)seconds since the lock was last used. The client
498 * may cancel the lock older than this age and flush related data if
499 * another client shows interest in this lock by doing glimpse request.
500 * This allows to cache stat data locally for such files early. Set in
501 * seconds from userspace, but stored in ns to avoid repeat conversions.
503 ktime_t ns_dirty_age_limit;
505 * Used to rate-limit ldlm_namespace_dump calls.
506 * \see ldlm_namespace_dump. Increased by 10 seconds every time
509 time64_t ns_next_dump;
511 /** "policy" function that does actual lock conflict determination */
512 ldlm_res_policy ns_policy;
515 * LVB operations for this namespace.
516 * \see struct ldlm_valblock_ops
518 struct ldlm_valblock_ops *ns_lvbo;
521 * Used by filter code to store pointer to OBD of the service.
522 * Should be dropped in favor of \a ns_obd
527 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
528 * a resource is removed.
530 wait_queue_head_t ns_waitq;
531 /** LDLM pool structure for this namespace */
532 struct ldlm_pool ns_pool;
533 /** Definition of how eagerly unused locks will be released from LRU */
534 enum ldlm_appetite ns_appetite;
537 * If more than \a ns_contended_locks are found, the resource is
538 * considered to be contended. Lock enqueues might specify that no
539 * contended locks should be granted
541 unsigned int ns_contended_locks;
544 * The resources in this namespace remember contended state during
545 * \a ns_contention_time, in seconds.
547 timeout_t ns_contention_time;
550 * Limit size of contended extent locks, in bytes.
551 * If extended lock is requested for more then this many bytes and
552 * caller instructs us not to grant contended locks, we would disregard
555 unsigned int ns_max_nolock_size;
557 /** Limit of parallel AST RPC count. */
558 unsigned int ns_max_parallel_ast;
561 * Callback to check if a lock is good to be canceled by ELC or
564 ldlm_cancel_cbt ns_cancel;
566 /** LDLM lock stats */
567 struct lprocfs_stats *ns_stats;
570 * Flag to indicate namespace is being freed. Used to determine if
571 * recalculation of LDLM pool statistics should be skipped.
573 unsigned int ns_stopping:1,
576 * Flag to indicate the LRU recalc on RPC reply is in progress.
577 * Used to limit the process by 1 thread only.
582 * Which bucket should we start with the lock reclaim.
584 int ns_reclaim_start;
586 struct kobject ns_kobj; /* sysfs object */
587 struct completion ns_kobj_unregister;
590 * To avoid another ns_lock usage, a separate bitops field.
592 unsigned long ns_flags;
596 * Returns 1 if namespace \a ns is a client namespace.
598 static inline int ns_is_client(struct ldlm_namespace *ns)
601 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
602 ns->ns_client == LDLM_NAMESPACE_SERVER);
603 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
607 * Returns 1 if namespace \a ns is a server namespace.
609 static inline int ns_is_server(struct ldlm_namespace *ns)
612 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
613 ns->ns_client == LDLM_NAMESPACE_SERVER);
614 return ns->ns_client == LDLM_NAMESPACE_SERVER;
618 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
620 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
623 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
627 * Returns 1 if this namespace supports lru_resize.
629 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
632 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
635 static inline void ns_register_cancel(struct ldlm_namespace *ns,
644 /** Type for blocking callback function of a lock. */
645 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
646 struct ldlm_lock_desc *new, void *data,
648 /** Type for completion callback function of a lock. */
649 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
651 /** Type for glimpse callback function of a lock. */
652 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
654 /** Type for created callback function of a lock. */
655 typedef void (*ldlm_created_callback)(struct ldlm_lock *lock);
657 /** Work list for sending GL ASTs to multiple locks. */
658 struct ldlm_glimpse_work {
659 struct ldlm_lock *gl_lock; /* lock to glimpse */
660 struct list_head gl_list; /* linkage to other gl work structs */
661 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
662 /* glimpse descriptor to be packed in glimpse callback request */
663 union ldlm_gl_desc *gl_desc;
664 ptlrpc_interpterer_t gl_interpret_reply;
665 void *gl_interpret_data;
668 struct ldlm_bl_desc {
669 unsigned int bl_same_client:1, /* both ops are from the same client. */
670 bl_txn_dependent:1;/* the op that enqueues lock depends on
671 * the op that holds lock.
675 struct ldlm_cb_set_arg {
676 struct ptlrpc_request_set *set;
677 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
679 struct list_head *list;
680 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
681 ptlrpc_interpterer_t gl_interpret_reply;
682 void *gl_interpret_data;
683 struct ldlm_bl_desc *bl_desc;
686 struct ldlm_cb_async_args {
687 struct ldlm_cb_set_arg *ca_set_arg;
688 struct ldlm_lock *ca_lock;
691 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
692 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
694 /** Interval node data for each LDLM_EXTENT lock. */
695 struct ldlm_interval {
696 struct interval_node li_node; /* node for tree management */
697 struct list_head li_group; /* locks having same policy */
699 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
702 * Interval tree for extent locks.
703 * The interval tree must be accessed under the resource lock.
704 * Interval trees are used for granted extent locks to speed up conflicts
705 * lookup. See ldlm/interval_tree.c for more details.
707 struct ldlm_interval_tree {
710 enum ldlm_mode lit_mode; /* lock mode */
711 struct interval_node *lit_root; /* actual ldlm_interval */
715 * Lists of waiting locks for each inodebit type.
716 * A lock can be in several liq_waiting lists and it remains in lr_waiting.
718 struct ldlm_ibits_queues {
719 struct list_head liq_waiting[MDS_INODELOCK_NUMBITS];
722 struct ldlm_ibits_node {
723 struct list_head lin_link[MDS_INODELOCK_NUMBITS];
724 struct ldlm_lock *lock;
727 struct ldlm_flock_node {
728 atomic_t lfn_unlock_pending;
729 bool lfn_needs_reprocess;
730 struct interval_node *lfn_root;
733 /** Whether to track references to exports by LDLM locks. */
734 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
737 enum ldlm_cancel_flags {
738 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
739 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
740 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
747 __u64 blocking_owner;
748 struct obd_export *blocking_export;
749 atomic_t blocking_refs;
753 union ldlm_policy_data {
754 struct ldlm_extent l_extent;
755 struct ldlm_flock l_flock;
756 struct ldlm_inodebits l_inodebits;
759 void ldlm_convert_policy_to_wire(enum ldlm_type type,
760 const union ldlm_policy_data *lpolicy,
761 union ldlm_wire_policy_data *wpolicy);
762 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
763 const union ldlm_wire_policy_data *wpolicy,
764 union ldlm_policy_data *lpolicy);
774 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
776 #define LDLM_GID_ANY ((__u64)-1)
779 * LDLM lock structure
781 * Represents a single LDLM lock and its state in memory. Each lock is
782 * associated with a single ldlm_resource, the object which is being
783 * locked. There may be multiple ldlm_locks on a single resource,
784 * depending on the lock type and whether the locks are conflicting or
790 * When remote side wants to tell us about a lock, they address
791 * it by this opaque handle. The handle does not hold a
792 * reference on the ldlm_lock, so it can be safely passed to
793 * other threads or nodes. When the lock needs to be accessed
794 * from the handle, it is looked up again in the lock table, and
795 * may no longer exist.
797 * Must be first in the structure.
799 struct portals_handle l_handle;
801 * Pointer to actual resource this lock is in.
802 * ldlm_lock_change_resource() can change this on the client.
803 * When this is possible, rcu must be used to stablise
804 * the resource while we lock and check it hasn't been changed.
806 struct ldlm_resource *l_resource;
808 * List item for client side LRU list.
809 * Protected by ns_lock in struct ldlm_namespace.
811 struct list_head l_lru;
813 * Linkage to resource's lock queues according to current lock state.
814 * (could be granted or waiting)
815 * Protected by lr_lock in struct ldlm_resource.
817 struct list_head l_res_link;
819 * Internal structures per lock type..
822 struct ldlm_interval *l_tree_node;
823 struct ldlm_ibits_node *l_ibits_node;
826 * Per export hash of locks.
827 * Protected by per-bucket exp->exp_lock_hash locks.
829 struct hlist_node l_exp_hash;
831 * Per export hash of flock locks.
832 * Protected by per-bucket exp->exp_flock_hash locks.
834 struct hlist_node l_exp_flock_hash;
837 * Protected by lr_lock.
839 enum ldlm_mode l_req_mode;
841 * Granted mode, also protected by lr_lock.
843 enum ldlm_mode l_granted_mode;
844 /** Lock completion handler pointer. Called when lock is granted. */
845 ldlm_completion_callback l_completion_ast;
847 * Lock blocking AST handler pointer.
848 * It plays two roles:
849 * - as a notification of an attempt to queue a conflicting lock (once)
850 * - as a notification when the lock is being cancelled.
852 * As such it's typically called twice: once for the initial conflict
853 * and then once more when the last user went away and the lock is
854 * cancelled (could happen recursively).
856 ldlm_blocking_callback l_blocking_ast;
858 * Lock glimpse handler.
859 * Glimpse handler is used to obtain LVB updates from a client by
862 ldlm_glimpse_callback l_glimpse_ast;
866 * This is a pointer to actual client export for locks that were granted
867 * to clients. Used server-side.
869 struct obd_export *l_export;
871 * Lock connection export.
872 * Pointer to server export on a client.
874 struct obd_export *l_conn_export;
877 * Remote lock handle.
878 * If the lock is remote, this is the handle of the other side lock
881 struct lustre_handle l_remote_handle;
884 * Representation of private data specific for a lock type.
885 * Examples are: extent range for extent lock or bitmask for ibits locks
887 union ldlm_policy_data l_policy_data;
890 * Lock state flags. Protected by lr_lock.
891 * \see lustre_dlm_flags.h where the bits are defined.
896 * Lock r/w usage counters.
897 * Protected by lr_lock.
902 * If the lock is granted, a process sleeps on this waitq to learn when
903 * it's no longer in use. If the lock is not granted, a process sleeps
904 * on this waitq to learn when it becomes granted.
906 wait_queue_head_t l_waitq;
909 * Time, in nanoseconds, last used by e.g. being matched by lock match.
913 /** Originally requested extent for the extent lock. */
914 struct ldlm_extent l_req_extent;
917 * Client-side-only members.
920 enum lvb_type l_lvb_type;
923 * Temporary storage for a LVB received during an enqueue operation.
924 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
929 /** Private storage for lock user. Opaque to LDLM. */
934 * Seconds. It will be updated if there is any activity related to
935 * the lock at client, e.g. enqueue the lock. For server it is the
936 * time when blocking ast was sent.
939 time64_t l_blast_sent;
942 /* separate ost_lvb used mostly by Data-on-MDT for now.
943 * It is introduced to don't mix with layout lock data.
945 struct ost_lvb l_ost_lvb;
947 * Server-side-only members.
951 * Connection cookie for the client originating the operation.
952 * Used by Commit on Share (COS) code. Currently only used for
953 * inodebits locks on MDS.
955 __u64 l_client_cookie;
958 * List item for locks waiting for cancellation from clients.
959 * The lists this could be linked into are:
960 * waiting_locks_list (protected by waiting_locks_spinlock),
961 * then if the lock timed out, it is moved to
962 * expired_lock_list for further processing.
964 struct list_head l_pending_chain;
967 * Set when lock is sent a blocking AST. Time in seconds when timeout
968 * is reached and client holding this lock could be evicted.
969 * This timeout could be further extended by e.g. certain IO activity
971 * \see ost_rw_prolong_locks
973 time64_t l_callback_timestamp;
975 /** Local PID of process which created this lock. */
979 * Number of times blocking AST was sent for this lock.
980 * This is for debugging. Valid values are 0 and 1, if there is an
981 * attempt to send blocking AST more than once, an assertion would be
982 * hit. \see ldlm_work_bl_ast_lock
985 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
986 struct list_head l_bl_ast;
987 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
988 struct list_head l_cp_ast;
989 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
990 struct list_head l_rk_ast;
993 * Pointer to a conflicting lock that caused blocking AST to be sent
996 struct ldlm_lock *l_blocking_lock;
999 * Protected by lr_lock, linkages to "skip lists".
1000 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
1002 struct list_head l_sl_mode;
1003 struct list_head l_sl_policy;
1005 /** Reference tracking structure to debug leaked locks. */
1006 struct lu_ref l_reference;
1007 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1008 /* Debugging stuff for bug 20498, for tracking export references. */
1009 /** number of export references taken */
1011 /** link all locks referencing one export */
1012 struct list_head l_exp_refs_link;
1013 /** referenced export object */
1014 struct obd_export *l_exp_refs_target;
1017 * export blocking dlm lock list, protected by
1018 * l_export->exp_bl_list_lock.
1019 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
1020 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
1022 struct list_head l_exp_list;
1025 enum ldlm_match_flags {
1026 LDLM_MATCH_UNREF = BIT(0),
1027 LDLM_MATCH_AST = BIT(1),
1028 LDLM_MATCH_AST_ANY = BIT(2),
1029 LDLM_MATCH_RIGHT = BIT(3),
1030 LDLM_MATCH_GROUP = BIT(4),
1034 * Describe the overlap between two locks. itree_overlap_cb data.
1036 struct ldlm_match_data {
1037 struct ldlm_lock *lmd_old;
1038 struct ldlm_lock *lmd_lock;
1039 enum ldlm_mode *lmd_mode;
1040 union ldlm_policy_data *lmd_policy;
1042 __u64 lmd_skip_flags;
1043 enum ldlm_match_flags lmd_match;
1046 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
1047 #define l_transno l_client_cookie
1049 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
1050 * 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 refcount_t lr_refcount;
1088 /** Spinlock to protect locks under this resource. */
1091 /* protected by lr_lock */
1093 /** List of locks in granted state */
1094 struct list_head lr_granted;
1096 * List of locks that could not be granted due to conflicts and
1097 * that are waiting for conflicts to go away
1099 struct list_head lr_waiting;
1102 /** Resource name */
1103 struct ldlm_res_id lr_name;
1106 /* Interval trees (for extent locks) all modes of resource */
1107 struct ldlm_interval_tree *lr_itree;
1108 struct ldlm_ibits_queues *lr_ibits_queues;
1109 struct ldlm_flock_node lr_flock_node;
1113 /* resource considered as contended, used only on server side*/
1114 time64_t lr_contention_time;
1116 * Associated inode, used only on client side.
1118 struct inode *lr_lvb_inode;
1121 /** Type of locks this resource can hold. Only one type per resource. */
1122 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1125 * Server-side-only lock value block elements.
1126 * To serialize lvbo_init.
1129 struct mutex lr_lvb_mutex;
1130 /** protected by lr_lock */
1132 /** is lvb initialized ? */
1133 bool lr_lvb_initialized;
1135 /** List of references to this resource. For debugging. */
1136 struct lu_ref lr_reference;
1139 static inline int ldlm_is_granted(struct ldlm_lock *lock)
1141 return lock->l_req_mode == lock->l_granted_mode;
1144 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1146 return lock->l_resource->lr_type == LDLM_IBITS &&
1147 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1150 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1152 return lock->l_resource->lr_type == LDLM_IBITS &&
1153 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1156 static inline char *
1157 ldlm_ns_name(struct ldlm_namespace *ns)
1162 static inline struct ldlm_namespace *
1163 ldlm_res_to_ns(struct ldlm_resource *res)
1165 return res->lr_ns_bucket->nsb_namespace;
1168 static inline struct ldlm_namespace *
1169 ldlm_lock_to_ns(struct ldlm_lock *lock)
1171 return ldlm_res_to_ns(lock->l_resource);
1174 static inline char *
1175 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1177 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1180 static inline struct adaptive_timeout *
1181 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1183 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1186 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1188 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1191 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1192 res->lr_lvb_initialized)
1195 mutex_lock(&res->lr_lvb_mutex);
1196 /* Did we lose the race? */
1197 if (res->lr_lvb_initialized) {
1198 mutex_unlock(&res->lr_lvb_mutex);
1201 rc = ns->ns_lvbo->lvbo_init(res);
1203 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1205 if (res->lr_lvb_data != NULL) {
1206 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1207 res->lr_lvb_data = NULL;
1209 res->lr_lvb_len = rc;
1211 res->lr_lvb_initialized = true;
1213 mutex_unlock(&res->lr_lvb_mutex);
1217 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1219 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1221 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1222 return ns->ns_lvbo->lvbo_size(lock);
1227 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int *len)
1229 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1232 if (ns->ns_lvbo != NULL) {
1233 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1234 /* init lvb now if not already */
1235 rc = ldlm_lvbo_init(lock->l_resource);
1237 CERROR("lock %p: delayed lvb init failed (rc %d)\n",
1241 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1246 struct ldlm_ast_work {
1247 struct ldlm_lock *w_lock;
1249 struct ldlm_lock_desc w_desc;
1250 struct list_head w_list;
1257 * Common ldlm_enqueue parameters
1259 struct ldlm_enqueue_info {
1260 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1261 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1262 void *ei_cb_bl; /** blocking lock callback */
1263 void *ei_cb_local_bl; /** blocking local lock callback */
1264 void *ei_cb_cp; /** lock completion callback */
1265 void *ei_cb_gl; /** lock glimpse callback */
1266 ldlm_created_callback ei_cb_created; /** lock created callback */
1267 void *ei_cbdata; /** Data to be passed into callbacks. */
1268 void *ei_namespace; /** lock namespace **/
1269 u64 ei_inodebits; /** lock inode bits **/
1270 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1271 unsigned int ei_req_slot:1; /** whether acquire rpc slot */
1272 unsigned int ei_mod_slot:1; /** whether acquire mod rpc slot */
1275 #define ei_res_id ei_cb_gl
1277 extern char *ldlm_lockname[];
1278 extern char *ldlm_typename[];
1279 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1282 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1283 * For the cases where we do not have actual lock to print along
1284 * with a debugging message that is ldlm-related
1286 #define LDLM_DEBUG_NOLOCK(format, a...) \
1287 CDEBUG(D_DLMTRACE, "### " format "\n", ##a)
1290 * Support function for lock information printing into debug logs.
1294 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1295 if (((mask) & D_CANTMASK) != 0 || \
1296 ((libcfs_debug & (mask)) != 0 && \
1297 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1298 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1301 __printf(3, 4) /* function attribute */
1302 void _ldlm_lock_debug(struct ldlm_lock *lock,
1303 struct libcfs_debug_msg_data *data,
1304 const char *fmt, ...);
1307 * Rate-limited version of lock printing function.
1309 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1310 static struct cfs_debug_limit_state _ldlm_cdls; \
1311 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1312 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt, ##a); \
1315 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1316 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1318 /** Non-rate-limited lock printing function for debugging purposes. */
1319 #define LDLM_DEBUG(lock, fmt, a...) do { \
1320 if (likely(lock != NULL)) { \
1321 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1322 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1325 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1328 #else /* !LIBCFS_DEBUG */
1329 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1330 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1331 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1335 * Three intentions can be used for the policy functions in
1336 * ldlm_processing_policy.
1338 * LDLM_PROCESS_RESCAN:
1340 * It's used when policy functions are called from ldlm_reprocess_queue() to
1341 * reprocess the wait list and try to grant locks, blocking ASTs
1342 * have already been sent in this situation, completion ASTs need be sent for
1343 * the locks being granted.
1345 * LDLM_PROCESS_ENQUEUE:
1347 * It's used when policy functions are called from ldlm_lock_enqueue() to
1348 * process the wait list for handling an enqueue request, blocking
1349 * ASTs have not been sent yet, so list of conflicting locks would be
1350 * collected and ASTs sent.
1352 * LDLM_PROCESS_RECOVERY:
1354 * It's used when policy functions are called from ldlm_reprocess_queue() to
1355 * reprocess the wait list when recovery done. In case of blocking
1356 * ASTs are lost before recovery, it needs not only to grant locks if
1357 * available, but also send blocking ASTs to the locks doesn't have AST sent
1358 * flag. Completion ASTs need be sent for the locks being granted.
1360 enum ldlm_process_intention {
1361 LDLM_PROCESS_RESCAN = 0,
1362 LDLM_PROCESS_ENQUEUE = 1,
1363 LDLM_PROCESS_RECOVERY = 2,
1366 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1367 enum ldlm_process_intention intention,
1368 enum ldlm_error *err,
1369 struct list_head *work_list);
1371 typedef int (*ldlm_reprocessing_policy)(struct ldlm_resource *res,
1372 struct list_head *queue,
1373 struct list_head *work_list,
1374 enum ldlm_process_intention intention,
1378 * Return values for lock iterators.
1379 * Also used during deciding of lock grants and cancellations.
1381 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1382 #define LDLM_ITER_STOP 2 /* stop iterating */
1384 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1385 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1387 /** \defgroup ldlm_iterator Lock iterators
1389 * LDLM provides for a way to iterate through every lock on a resource or
1390 * namespace or every resource in a namespace.
1393 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1395 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1397 int ldlm_resource_iterate(struct ldlm_namespace *ln,
1398 const struct ldlm_res_id *lri,
1399 ldlm_iterator_t iter, void *data);
1400 /** @} ldlm_iterator */
1402 int ldlm_replay_locks(struct obd_import *imp);
1405 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1408 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1410 struct ldlm_prolong_args {
1411 struct obd_export *lpa_export;
1412 struct ptlrpc_request *lpa_req;
1413 struct ldlm_res_id lpa_resid;
1414 struct ldlm_extent lpa_extent;
1415 enum ldlm_mode lpa_mode;
1419 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1420 struct ldlm_prolong_args *arg);
1421 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1423 struct ldlm_callback_suite {
1424 ldlm_completion_callback lcs_completion;
1425 ldlm_blocking_callback lcs_blocking;
1426 ldlm_glimpse_callback lcs_glimpse;
1430 #ifdef HAVE_SERVER_SUPPORT
1431 /** \defgroup ldlm_srv_ast Server AST handlers
1432 * These are AST handlers used by server code.
1433 * Their property is that they are just preparing RPCs to be sent to clients.
1436 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1437 void *data, int flag);
1438 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1439 void *data, int flag);
1440 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1441 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1442 int ldlm_glimpse_locks(struct ldlm_resource *res,
1443 struct list_head *gl_work_list);
1444 /** @} ldlm_srv_ast */
1446 /** \defgroup ldlm_handlers Server LDLM handlers
1447 * These are handler functions that should be called by "frontends" such as
1448 * MDT or OST to pass through LDLM requests to LDLM for handling
1451 int ldlm_handle_enqueue(struct ldlm_namespace *ns, struct req_capsule *pill,
1452 const struct ldlm_request *dlm_req,
1453 const struct ldlm_callback_suite *cbs);
1454 int ldlm_handle_convert0(struct ptlrpc_request *req,
1455 const struct ldlm_request *dlm_req);
1456 int ldlm_handle_cancel(struct ptlrpc_request *req);
1457 int ldlm_request_cancel(struct ptlrpc_request *req,
1458 const struct ldlm_request *dlm_req,
1459 int first, enum lustre_at_flags flags);
1460 /** @} ldlm_handlers */
1462 void ldlm_revoke_export_locks(struct obd_export *exp);
1463 timeout_t ldlm_bl_timeout(struct ldlm_lock *lock);
1464 timeout_t ldlm_bl_timeout_by_rpc(struct ptlrpc_request *req);
1466 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1467 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, timeout_t timeout);
1468 int ldlm_get_ref(void);
1469 void ldlm_put_ref(void);
1470 int ldlm_init_export(struct obd_export *exp);
1471 void ldlm_destroy_export(struct obd_export *exp);
1472 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1475 #ifdef HAVE_SERVER_SUPPORT
1476 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1477 ldlm_reprocessing_policy
1478 ldlm_get_reprocessing_policy(struct ldlm_resource *res);
1480 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1481 void ldlm_lock2handle(const struct ldlm_lock *lock,
1482 struct lustre_handle *lockh);
1483 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *lh,
1485 void ldlm_cancel_callback(struct ldlm_lock *ll);
1486 int ldlm_lock_remove_from_lru(struct ldlm_lock *ll);
1487 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1490 * Obtain a lock reference by its handle.
1492 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1494 return __ldlm_handle2lock(h, 0);
1497 #define LDLM_LOCK_REF_DEL(lock) \
1498 lu_ref_del(&lock->l_reference, "handle", lock)
1500 static inline struct ldlm_lock *
1501 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1503 struct ldlm_lock *lock;
1505 lock = __ldlm_handle2lock(h, flags);
1507 LDLM_LOCK_REF_DEL(lock);
1512 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1513 * data from request \a r
1515 static inline int ldlm_lvbo_update(struct ldlm_resource *res,
1516 struct ldlm_lock *lock,
1517 struct ptlrpc_request *req, int increase)
1519 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1522 /* delayed lvb init may be required */
1523 rc = ldlm_lvbo_init(res);
1525 CERROR("delayed lvb init failed (rc %d)\n", rc);
1529 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1530 return ns->ns_lvbo->lvbo_update(res, lock, req, increase);
1535 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1536 struct ptlrpc_request *req,
1539 return ldlm_lvbo_update(res, NULL, req, increase);
1542 int is_granted_or_cancelled_nolock(struct ldlm_lock *lock);
1544 int ldlm_error2errno(enum ldlm_error error);
1545 /* don't call it `errno': this confuses user-space. */
1546 enum ldlm_error ldlm_errno2error(int err_no);
1547 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1548 void ldlm_dump_export_locks(struct obd_export *exp);
1551 /* Release temporary lock got by ldlm_handle2lock() or __ldlm_handle2lock() */
1552 #define LDLM_LOCK_PUT(lock) \
1554 LDLM_LOCK_REF_DEL(lock); \
1555 /*LDLM_DEBUG((lock), "put");*/ \
1556 ldlm_lock_put(lock); \
1560 * Release a lock reference obtained by some other means (see
1563 #define LDLM_LOCK_RELEASE(lock) ldlm_lock_put(lock)
1565 #define LDLM_LOCK_GET(lock) \
1567 ldlm_lock_get(lock); \
1568 /*LDLM_DEBUG((lock), "get");*/ \
1572 #define ldlm_lock_list_put(head, member, count) \
1574 struct ldlm_lock *_lock, *_next; \
1576 list_for_each_entry_safe(_lock, _next, head, member) { \
1579 list_del_init(&_lock->member); \
1580 LDLM_LOCK_RELEASE(_lock); \
1585 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1586 void ldlm_lock_put(struct ldlm_lock *lock);
1587 void ldlm_lock_destroy(struct ldlm_lock *lock);
1588 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1589 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1590 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1591 enum ldlm_mode mode);
1592 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1593 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1594 enum ldlm_mode mode);
1595 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1596 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1597 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1598 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1600 enum ldlm_mode ldlm_lock_match_with_skip(struct ldlm_namespace *ns,
1601 __u64 flags, __u64 skip_flags,
1602 const struct ldlm_res_id *res_id,
1603 enum ldlm_type type,
1604 union ldlm_policy_data *policy,
1605 enum ldlm_mode mode,
1606 struct lustre_handle *lh,
1607 enum ldlm_match_flags match_flags);
1608 static inline enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns,
1610 const struct ldlm_res_id *res_id,
1611 enum ldlm_type type,
1612 union ldlm_policy_data *policy,
1613 enum ldlm_mode mode,
1614 struct lustre_handle *lh)
1616 return ldlm_lock_match_with_skip(ns, flags, 0, res_id, type, policy,
1619 struct ldlm_lock *search_itree(struct ldlm_resource *res,
1620 struct ldlm_match_data *data);
1621 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1623 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1624 void ldlm_lock_cancel(struct ldlm_lock *lock);
1625 void ldlm_reprocess_all(struct ldlm_resource *res, __u64 hint);
1626 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1627 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1628 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1631 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1632 enum ldlm_side client,
1633 enum ldlm_appetite apt,
1634 enum ldlm_ns_type ns_type);
1635 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1636 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1637 struct obd_import *imp,
1639 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1640 void ldlm_namespace_free(struct ldlm_namespace *ns,
1641 struct obd_import *imp, int force);
1642 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1643 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1644 enum ldlm_side client);
1645 void ldlm_namespace_get(struct ldlm_namespace *ns);
1646 void ldlm_namespace_put(struct ldlm_namespace *ns);
1648 int ldlm_debugfs_setup(void);
1649 void ldlm_debugfs_cleanup(void);
1651 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1652 struct lprocfs_stats *srv_stats)
1654 int lock_type = 0, op = 0;
1656 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1658 switch (lock_type) {
1660 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1663 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1666 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1669 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1677 lprocfs_counter_incr(srv_stats, op);
1680 /* resource.c - internal */
1681 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1682 const struct ldlm_res_id *,
1683 enum ldlm_type type, int create);
1684 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1685 int ldlm_resource_putref(struct ldlm_resource *res);
1686 void ldlm_resource_add_lock(struct ldlm_resource *res,
1687 struct list_head *head,
1688 struct ldlm_lock *lock);
1689 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1690 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1691 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1692 void ldlm_namespace_dump(int level, struct ldlm_namespace *ln);
1693 void ldlm_resource_dump(int level, struct ldlm_resource *lr);
1694 int ldlm_lock_change_resource(struct ldlm_namespace *ln, struct ldlm_lock *ll,
1695 const struct ldlm_res_id *lri);
1697 #define LDLM_RESOURCE_ADDREF(res) \
1698 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current)
1700 #define LDLM_RESOURCE_DELREF(res) \
1701 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current)
1703 /* ldlm_request.c */
1704 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1705 * These AST handlers are typically used for server-side local locks and are
1706 * also used by client-side lock handlers to perform minimum level base
1709 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1710 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1711 void *data, int flag);
1712 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1713 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1714 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1715 /** @} ldlm_local_ast */
1717 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1718 * These are typically used by client and server (*_local versions)
1719 * to obtain and release locks.
1721 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1722 struct ldlm_enqueue_info *einfo,
1723 const struct ldlm_res_id *res_id,
1724 union ldlm_policy_data const *policy, __u64 *flags,
1725 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1726 struct lustre_handle *lockh, int async);
1727 int ldlm_prep_enqueue_req(struct obd_export *exp,
1728 struct ptlrpc_request *req,
1729 struct list_head *cancels,
1731 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1732 int version, int opc, int canceloff,
1733 struct list_head *cancels, int count);
1735 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1736 int ldlm_handle_enqueue(struct ldlm_namespace *ns, struct req_capsule *pill,
1737 const struct ldlm_request *dlm_req,
1738 const struct ldlm_callback_suite *cbs);
1739 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct req_capsule *pill,
1740 struct ldlm_enqueue_info *einfo, __u8 with_policy,
1741 __u64 *flags, void *lvb, __u32 lvb_len,
1742 const struct lustre_handle *lockh, int rc,
1744 int ldlm_cli_enqueue_local(const struct lu_env *env,
1745 struct ldlm_namespace *ns,
1746 const struct ldlm_res_id *res_id,
1747 enum ldlm_type type, union ldlm_policy_data *policy,
1748 enum ldlm_mode mode, __u64 *flags,
1749 ldlm_blocking_callback blocking,
1750 ldlm_completion_callback completion,
1751 ldlm_glimpse_callback glimpse,
1752 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1753 const __u64 *client_cookie,
1754 struct lustre_handle *lockh);
1755 int ldlm_cli_lock_create_pack(struct obd_export *exp,
1756 struct ldlm_request *dlmreq,
1757 struct ldlm_enqueue_info *einfo,
1758 const struct ldlm_res_id *res_id,
1759 union ldlm_policy_data const *policy,
1760 __u64 *flags, void *lvb, __u32 lvb_len,
1761 enum lvb_type lvb_type,
1762 struct lustre_handle *lockh);
1763 int ldlm_cli_convert_req(struct ldlm_lock *lock, __u32 *flags, __u64 new_bits);
1764 int ldlm_cli_convert(struct ldlm_lock *lock,
1765 enum ldlm_cancel_flags cancel_flags);
1766 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1767 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1768 enum ldlm_cancel_flags cancel_flags);
1769 int ldlm_cli_cancel_unused(struct ldlm_namespace *n,
1770 const struct ldlm_res_id *l,
1771 enum ldlm_cancel_flags flags, void *opaque);
1772 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1773 const struct ldlm_res_id *res_id,
1774 union ldlm_policy_data *policy,
1775 enum ldlm_mode mode,
1776 enum ldlm_cancel_flags flags, void *opaque);
1777 int ldlm_cli_cancel_req(struct obd_export *exp, struct ldlm_lock *lock,
1778 struct list_head *head, int count,
1779 enum ldlm_cancel_flags flags);
1780 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1781 struct list_head *cancels,
1782 union ldlm_policy_data *policy,
1783 enum ldlm_mode mode, __u64 lock_flags,
1784 enum ldlm_cancel_flags cancel_flags,
1786 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1787 enum ldlm_cancel_flags flags);
1788 int ldlm_cli_cancel_list(struct list_head *head, int count,
1789 struct ptlrpc_request *req,
1790 enum ldlm_cancel_flags flags);
1792 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);
1793 int ldlm_cli_inodebits_convert(struct ldlm_lock *lock,
1794 enum ldlm_cancel_flags cancel_flags);
1796 /** @} ldlm_cli_api */
1800 /* This has to be here because recursive inclusion sucks. */
1801 int intent_disposition(struct ldlm_reply *rep, int flag);
1802 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1805 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1806 * than one lock_res is dead-lock safe.
1808 enum lock_res_type {
1813 /** Lock resource. */
1814 static inline void lock_res(struct ldlm_resource *res)
1816 spin_lock(&res->lr_lock);
1819 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1820 static inline void lock_res_nested(struct ldlm_resource *res,
1821 enum lock_res_type mode)
1823 spin_lock_nested(&res->lr_lock, mode);
1826 /** Unlock resource. */
1827 static inline void unlock_res(struct ldlm_resource *res)
1829 spin_unlock(&res->lr_lock);
1832 /** Check if resource is already locked, assert if not. */
1833 static inline void check_res_locked(struct ldlm_resource *res)
1835 assert_spin_locked(&res->lr_lock);
1838 struct ldlm_resource *lock_res_and_lock(struct ldlm_lock *lock);
1839 void unlock_res_and_lock(struct ldlm_lock *lock);
1842 /** \defgroup ldlm_pools Various LDLM pool related functions
1843 * There are not used outside of ldlm.
1846 int ldlm_pools_init(void);
1847 void ldlm_pools_fini(void);
1849 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1850 int idx, enum ldlm_side client);
1851 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1852 void ldlm_pool_fini(struct ldlm_pool *pl);
1853 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1854 time64_t ldlm_pool_recalc(struct ldlm_pool *pl, bool force);
1855 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1856 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1857 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1858 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1859 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1860 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1861 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1862 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1863 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1866 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1867 const struct ldlm_extent *ex2)
1869 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1872 /* check if @ex1 contains @ex2 */
1873 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1874 const struct ldlm_extent *ex2)
1876 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1879 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);