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 */
51 #include <linux/interval_tree_generic.h>
54 #include "lustre_dlm_flags.h"
59 extern struct kset *ldlm_ns_kset;
60 extern struct kset *ldlm_svc_kset;
62 #define OBD_LDLM_DEVICENAME "ldlm"
64 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
65 #define LDLM_DEFAULT_MAX_ALIVE 3900 /* 3900 seconds ~65 min */
66 #define LDLM_CTIME_AGE_LIMIT (10)
67 /* if client lock is unused for that time it can be cancelled if any other
68 * client shows interest in that lock, e.g. glimpse is occured.
70 #define LDLM_DIRTY_AGE_LIMIT (10)
71 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
72 #define LDLM_DEFAULT_LRU_SHRINK_BATCH (16)
73 #define LDLM_DEFAULT_SLV_RECALC_PCT (10)
76 * LDLM non-error return states
80 ELDLM_LOCK_MATCHED = 1,
82 ELDLM_LOCK_CHANGED = 300,
83 ELDLM_LOCK_ABORTED = 301,
84 ELDLM_LOCK_REPLACED = 302,
85 ELDLM_NO_LOCK_DATA = 303,
86 ELDLM_LOCK_WOULDBLOCK = 304,
88 ELDLM_NAMESPACE_EXISTS = 400,
89 ELDLM_BAD_NAMESPACE = 401,
93 * LDLM namespace type.
94 * The "client" type is actually an indication that this is a narrow local view
95 * into complete namespace on the server. Such namespaces cannot make any
96 * decisions about lack of conflicts or do any autonomous lock granting without
97 * first speaking to a server.
100 LDLM_NAMESPACE_SERVER = 0x01,
101 LDLM_NAMESPACE_CLIENT = 0x02
105 * The blocking callback is overloaded to perform two functions. These flags
106 * indicate which operation should be performed.
108 #define LDLM_CB_BLOCKING 1
109 #define LDLM_CB_CANCELING 2
112 * \name Lock Compatibility Matrix.
114 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
115 * Lock types are described in their respective implementation files:
116 * ldlm_{extent,flock,inodebits,plain}.c.
118 * There are nine lock modes along with a compatibility matrix to indicate if
119 * two locks are compatible.
121 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
123 * - PW: Protective Write (normal write) mode. When a client requests a write
124 * lock from an OST, a lock with PW mode will be issued.
125 * - PR: Protective Read (normal read) mode. When a client requests a read from
126 * an OST, a lock with PR mode will be issued. Also, if the client opens a
127 * file for execution, it is granted a lock with PR mode.
128 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
129 * requests a write lock during a file open operation.
130 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
131 * an inodebit lock with the CR mode on the intermediate path component.
133 * - GROUP: Group mode. Locks with the same group ID are compatible with each
135 * - COS: Commit-on-Sharing mode. If Commit-on-Sharing is enabled, PW/EX locks
136 * held in transactions are downgraded to COS mode after transaction stop.
137 * - TXN: Transaction mode. If Commit-on-Sharing is diabled on a DNE system,
138 * PW/EX locks held in transactions are downgraded to TXN mode after
142 * NL CR CW PR PW EX GROUP COS TXN
143 * NL 1 1 1 1 1 1 1 1 1
144 * CR 1 1 1 1 1 0 0 0 1
145 * CW 1 1 1 0 0 0 0 0 0
146 * PR 1 1 0 1 0 0 0 0 1
147 * PW 1 1 0 0 0 0 0 0 0
148 * EX 1 0 0 0 0 0 0 0 0
149 * GROUP 1 0 0 0 0 0 1 0 0
150 * COS 1 0 0 0 0 0 0 1 0
151 * TXN 1 1 0 1 0 0 0 0 1
155 #define LCK_COMPAT_EX LCK_NL
156 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
157 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR | LCK_TXN)
158 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
159 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW | LCK_TXN)
160 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP | LCK_COS)
161 #define LCK_COMPAT_GROUP (LCK_NL | LCK_GROUP)
162 #define LCK_COMPAT_COS (LCK_NL | LCK_COS)
163 #define LCK_COMPAT_TXN LCK_COMPAT_PR
164 /** @} Lock Compatibility Matrix */
166 extern enum ldlm_mode lck_compat_array[];
168 static inline void lockmode_verify(enum ldlm_mode mode)
170 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
173 static inline int lockmode_compat(enum ldlm_mode exist_mode,
174 enum ldlm_mode new_mode)
176 return lck_compat_array[exist_mode] & new_mode;
180 * cluster name spaces
182 #define DLM_OST_NAMESPACE 1
183 #define DLM_MDS_NAMESPACE 2
186 * - do we just separate this by security domains and use a prefix for
187 * multiple namespaces in the same domain?
192 * Locking rules for LDLM:
197 * waiting_locks_spinlock
210 /* Cancel lru flag, it indicates we cancel aged locks. */
211 enum ldlm_lru_flags {
212 LDLM_LRU_FLAG_NO_WAIT = 0x1, /* Cancel locks w/o blocking (neither
213 * 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
225 struct ldlm_resource;
226 struct ldlm_namespace;
229 * Operations on LDLM pools.
230 * LDLM pool is a pool of locks in the namespace without any implicitly
232 * Locks in the pool are organized in LRU.
233 * Local memory pressure or server instructions (e.g. mempressure on server)
234 * can trigger freeing of locks from the pool
236 struct ldlm_pool_ops {
237 /** Recalculate pool \a pl usage */
238 int (*po_recalc)(struct ldlm_pool *pl, bool force);
239 /** Cancel at least \a nr locks from pool \a pl */
240 int (*po_shrink)(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
241 int (*po_setup)(struct ldlm_pool *pl, int limit);
244 /** One second for pools thread check interval. Each pool has own period. */
245 #define LDLM_POOLS_THREAD_PERIOD (1)
247 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
248 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
250 /** Default recalc period for server side pools in sec. */
251 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
253 /** Default recalc period for client side pools in sec. */
254 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
257 * LDLM pool structure to track granted locks.
258 * For purposes of determining when to release locks on e.g. memory pressure.
259 * This feature is commonly referred to as lru_resize.
262 /** Pool debugfs directory. */
263 struct dentry *pl_debugfs_entry;
264 /** Pool name, must be long enough to hold compound proc entry name. */
266 /** Lock for protecting SLV/CLV updates. */
268 /** Number of allowed locks in in pool, both, client and server side. */
270 /** Number of granted locks in */
272 /** Grant rate per T. */
273 atomic_t pl_grant_rate;
274 /** Cancel rate per T. */
275 atomic_t pl_cancel_rate;
276 /** Server lock volume (SLV). Protected by pl_lock. */
277 __u64 pl_server_lock_volume;
278 /** Current biggest client lock volume. Protected by pl_lock. */
279 __u64 pl_client_lock_volume;
280 /** Lock volume factor, shown in percents in procfs, but internally
281 * Client SLV calculated as: server_slv * lock_volume_factor >> 8.
283 atomic_t pl_lock_volume_factor;
284 /** Time when last SLV from server was obtained. */
285 time64_t pl_recalc_time;
286 /** Recalculation period for pool. */
287 time64_t pl_recalc_period;
288 /** Recalculation and shrink operations. */
289 struct ldlm_pool_ops *pl_ops;
290 /** Number of planned locks for next period. */
292 /** Pool statistics. */
293 struct lprocfs_stats *pl_stats;
296 struct kobject pl_kobj;
297 struct completion pl_kobj_unregister;
300 typedef int (*ldlm_res_policy)(const struct lu_env *env,
301 struct ldlm_namespace *,
302 struct ldlm_lock **, void *req_cookie,
303 enum ldlm_mode mode, __u64 flags, void *data);
305 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
309 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
310 * be associated with an LDLM lock and transferred from client to server and
313 * Currently LVBs are used by:
314 * - OSC-OST code to maintain current object size/times
315 * - layout lock code to return the layout when the layout lock is granted
317 * To ensure delayed LVB initialization, it is highly recommended to use the set
318 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
320 struct ldlm_valblock_ops {
321 int (*lvbo_init)(struct ldlm_resource *res);
322 int (*lvbo_update)(struct ldlm_resource *res, struct ldlm_lock *lock,
323 struct ptlrpc_request *r, int increase);
324 int (*lvbo_free)(struct ldlm_resource *res);
325 /* Return size of lvb data appropriate RPC size can be reserved */
326 int (*lvbo_size)(struct ldlm_lock *lock);
327 /* Called to fill in lvb data to RPC buffer @buf */
328 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int *buflen);
332 * LDLM pools related, type of lock pool in the namespace.
333 * Greedy means release cached locks aggressively
336 LDLM_NAMESPACE_GREEDY = BIT(0),
337 LDLM_NAMESPACE_MODEST = BIT(1),
341 * Default values for the "max_nolock_size", "contention_time" and
342 * "contended_locks" namespace tunables.
344 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
345 #define NS_DEFAULT_CONTENTION_SECONDS 2
346 #define NS_DEFAULT_CONTENDED_LOCKS 32
348 struct ldlm_ns_bucket {
349 /** back pointer to namespace */
350 struct ldlm_namespace *nsb_namespace;
352 * Estimated lock callback time. Used by adaptive timeout code to
353 * avoid spurious client evictions due to unresponsiveness when in
354 * fact the network or overall system load is at fault
356 struct adaptive_timeout nsb_at_estimate;
358 * Which res in the bucket should we start with the reclaim.
360 int nsb_reclaim_start;
361 /* counter of entries in this bucket */
366 /** LDLM namespace lock stats */
372 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
373 LDLM_NS_TYPE_MDC, /**< MDC namespace */
374 LDLM_NS_TYPE_MDT, /**< MDT namespace */
375 LDLM_NS_TYPE_OSC, /**< OSC namespace */
376 LDLM_NS_TYPE_OST, /**< OST namespace */
377 LDLM_NS_TYPE_MGC, /**< MGC namespace */
378 LDLM_NS_TYPE_MGT, /**< MGT namespace */
381 enum ldlm_namespace_flags {
383 * Flag to indicate the LRU cancel is in progress.
384 * Used to limit the process by 1 thread only.
392 * Namespace serves to contain locks related to a particular service.
393 * There are two kinds of namespaces:
394 * - Server namespace has knowledge of all locks and is therefore authoritative
395 * to make decisions like what locks could be granted and what conflicts
396 * exist during new lock enqueue.
397 * - Client namespace only has limited knowledge about locks in the namespace,
398 * only seeing locks held by the client.
400 * Every Lustre service has one server namespace present on the server serving
401 * that service. Every client connected to the service has a client namespace
403 * Every lock obtained by client in that namespace is actually represented by
404 * two in-memory locks. One on the server and one on the client. The locks are
405 * linked by a special cookie by which one node can tell to the other which lock
406 * it actually means during communications. Such locks are called remote locks.
407 * The locks held by server only without any reference to a client are called
410 struct ldlm_namespace {
411 /** Backward link to OBD, required for LDLM pool to store new SLV. */
412 struct obd_device *ns_obd;
414 /** Flag indicating if namespace is on client instead of server */
415 enum ldlm_side ns_client;
417 /** name of this namespace */
420 /** Resource hash table for namespace. */
421 struct cfs_hash *ns_rs_hash;
422 struct ldlm_ns_bucket *ns_rs_buckets;
423 unsigned int ns_bucket_bits;
428 /** big refcount (by bucket) */
432 * Namespace connect flags supported by server (may be changed via
433 * /proc, LRU resize may be disabled/enabled).
435 __u64 ns_connect_flags;
437 /** Client side original connect flags supported by server. */
438 __u64 ns_orig_connect_flags;
440 /* namespace debugfs dir entry */
441 struct dentry *ns_debugfs_entry;
444 * Position in global namespace list linking all namespaces on
447 struct list_head ns_list_chain;
450 * List of unused locks for this namespace. This list is also called
452 * Unused locks are locks with zero reader/writer reference counts.
453 * This list is only used on clients for lock caching purposes.
454 * When we want to release some locks voluntarily or if server wants
455 * us to release some locks due to e.g. memory pressure, we take locks
456 * to release from the head of this list.
457 * Locks are linked via l_lru field in \see struct ldlm_lock.
459 struct list_head ns_unused_list;
460 /** Number of locks in the LRU list above */
462 struct list_head *ns_last_pos;
465 * Maximum number of locks permitted in the LRU. If 0, means locks
466 * are managed by pools and there is no preset limit, rather it is all
467 * controlled by available memory on this client and on server.
469 unsigned int ns_max_unused;
472 * Cancel batch, if unused lock count exceed lru_size
473 * Only be used if LRUR disable.
475 unsigned int ns_cancel_batch;
477 /* How much SLV should decrease in %% to trigger LRU cancel urgently. */
478 unsigned int ns_recalc_pct;
480 /** Maximum allowed age (last used time) for locks in the LRU. Set in
481 * seconds from userspace, but stored in ns to avoid repeat conversions.
486 * Server only: number of times we evicted clients due to lack of reply
489 unsigned int ns_timeouts;
491 * Number of seconds since the file change time after which
492 * the MDT will return an UPDATE lock along with a LOOKUP lock.
493 * This allows the client to start caching negative dentries
494 * for a directory and may save an RPC for a later stat.
496 timeout_t ns_ctime_age_limit;
498 * Number of (nano)seconds since the lock was last used. The client
499 * may cancel the lock older than this age and flush related data if
500 * another client shows interest in this lock by doing glimpse request.
501 * This allows to cache stat data locally for such files early. Set in
502 * seconds from userspace, but stored in ns to avoid repeat conversions.
504 ktime_t ns_dirty_age_limit;
506 * Used to rate-limit ldlm_namespace_dump calls.
507 * \see ldlm_namespace_dump. Increased by 10 seconds every time
510 time64_t ns_next_dump;
512 /** "policy" function that does actual lock conflict determination */
513 ldlm_res_policy ns_policy;
516 * LVB operations for this namespace.
517 * \see struct ldlm_valblock_ops
519 struct ldlm_valblock_ops *ns_lvbo;
522 * Used by filter code to store pointer to OBD of the service.
523 * Should be dropped in favor of \a ns_obd
528 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
529 * a resource is removed.
531 wait_queue_head_t ns_waitq;
532 /** LDLM pool structure for this namespace */
533 struct ldlm_pool ns_pool;
534 /** Definition of how eagerly unused locks will be released from LRU */
535 enum ldlm_appetite ns_appetite;
538 * If more than \a ns_contended_locks are found, the resource is
539 * considered to be contended. Lock enqueues might specify that no
540 * contended locks should be granted
542 unsigned int ns_contended_locks;
545 * The resources in this namespace remember contended state during
546 * \a ns_contention_time, in seconds.
548 timeout_t ns_contention_time;
551 * Limit size of contended extent locks, in bytes.
552 * If extended lock is requested for more then this many bytes and
553 * caller instructs us not to grant contended locks, we would disregard
556 unsigned int ns_max_nolock_size;
558 /** Limit of parallel AST RPC count. */
559 unsigned int ns_max_parallel_ast;
562 * Callback to check if a lock is good to be canceled by ELC or
565 ldlm_cancel_cbt ns_cancel;
567 /** LDLM lock stats */
568 struct lprocfs_stats *ns_stats;
571 * Flag to indicate namespace is being freed. Used to determine if
572 * recalculation of LDLM pool statistics should be skipped.
574 unsigned int ns_stopping:1,
577 * Flag to indicate the LRU recalc on RPC reply is in progress.
578 * Used to limit the process by 1 thread only.
583 * Which bucket should we start with the lock reclaim.
585 int ns_reclaim_start;
587 struct kobject ns_kobj; /* sysfs object */
588 struct completion ns_kobj_unregister;
591 * To avoid another ns_lock usage, a separate bitops field.
593 unsigned long ns_flags;
597 * Returns 1 if namespace \a ns is a client namespace.
599 static inline int ns_is_client(struct ldlm_namespace *ns)
602 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
603 ns->ns_client == LDLM_NAMESPACE_SERVER);
604 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
608 * Returns 1 if namespace \a ns is a server namespace.
610 static inline int ns_is_server(struct ldlm_namespace *ns)
613 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
614 ns->ns_client == LDLM_NAMESPACE_SERVER);
615 return ns->ns_client == LDLM_NAMESPACE_SERVER;
619 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
621 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
624 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
628 * Returns 1 if this namespace supports lru_resize.
630 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
633 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
636 static inline void ns_register_cancel(struct ldlm_namespace *ns,
645 /** Type for blocking callback function of a lock. */
646 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
647 struct ldlm_lock_desc *new, void *data,
649 /** Type for completion callback function of a lock. */
650 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
652 /** Type for glimpse callback function of a lock. */
653 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
655 /** Type for created callback function of a lock. */
656 typedef void (*ldlm_created_callback)(struct ldlm_lock *lock);
658 /** Work list for sending GL ASTs to multiple locks. */
659 struct ldlm_glimpse_work {
660 struct ldlm_lock *gl_lock; /* lock to glimpse */
661 struct list_head gl_list; /* linkage to other gl work structs */
662 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
663 /* glimpse descriptor to be packed in glimpse callback request */
664 union ldlm_gl_desc *gl_desc;
665 ptlrpc_interpterer_t gl_interpret_reply;
666 void *gl_interpret_data;
669 struct ldlm_bl_desc {
670 unsigned int bl_same_client:1, /* both ops are from the same client. */
671 bl_txn_dependent:1;/* the op that enqueues lock depends on
672 * the op that holds lock.
676 struct ldlm_cb_set_arg {
677 struct ptlrpc_request_set *set;
678 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
680 struct list_head *list;
681 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
682 ptlrpc_interpterer_t gl_interpret_reply;
683 void *gl_interpret_data;
684 struct ldlm_bl_desc *bl_desc;
687 struct ldlm_cb_async_args {
688 struct ldlm_cb_set_arg *ca_set_arg;
689 struct ldlm_lock *ca_lock;
692 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
693 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
696 * Interval tree for extent locks.
697 * The interval tree must be accessed under the resource lock.
698 * Interval trees are used for granted extent locks to speed up conflicts
699 * lookup. See ldlm/interval_tree.c for more details.
701 struct ldlm_interval_tree {
704 enum ldlm_mode lit_mode; /* lock mode */
705 struct interval_tree_root lit_root; /* actual interval tree */
709 * Lists of waiting locks for each inodebit type.
710 * A lock can be in several liq_waiting lists and it remains in lr_waiting.
712 struct ldlm_ibits_queues {
713 struct list_head liq_waiting[MDS_INODELOCK_NUMBITS];
716 struct ldlm_ibits_node {
717 struct list_head lin_link[MDS_INODELOCK_NUMBITS];
718 struct ldlm_lock *lock;
721 struct ldlm_flock_node {
722 atomic_t lfn_unlock_pending;
723 bool lfn_needs_reprocess;
724 struct interval_tree_root lfn_root;
727 /** Whether to track references to exports by LDLM locks. */
728 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
731 enum ldlm_cancel_flags {
732 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
733 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
734 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
741 __u64 blocking_owner;
742 struct obd_export *blocking_export;
743 atomic_t blocking_refs;
747 union ldlm_policy_data {
748 struct ldlm_extent l_extent;
749 struct ldlm_flock l_flock;
750 struct ldlm_inodebits l_inodebits;
753 void ldlm_convert_policy_to_wire(enum ldlm_type type,
754 const union ldlm_policy_data *lpolicy,
755 union ldlm_wire_policy_data *wpolicy);
756 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
757 const union ldlm_wire_policy_data *wpolicy,
758 union ldlm_policy_data *lpolicy);
768 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
770 #define LDLM_GID_ANY ((__u64)-1)
773 * LDLM lock structure
775 * Represents a single LDLM lock and its state in memory. Each lock is
776 * associated with a single ldlm_resource, the object which is being
777 * locked. There may be multiple ldlm_locks on a single resource,
778 * depending on the lock type and whether the locks are conflicting or
784 * When remote side wants to tell us about a lock, they address
785 * it by this opaque handle. The handle does not hold a
786 * reference on the ldlm_lock, so it can be safely passed to
787 * other threads or nodes. When the lock needs to be accessed
788 * from the handle, it is looked up again in the lock table, and
789 * may no longer exist.
791 * Must be first in the structure.
793 struct portals_handle l_handle;
795 * Pointer to actual resource this lock is in.
796 * ldlm_lock_change_resource() can change this on the client.
797 * When this is possible, rcu must be used to stablise
798 * the resource while we lock and check it hasn't been changed.
800 struct ldlm_resource *l_resource;
802 * List item for client side LRU list.
803 * Protected by ns_lock in struct ldlm_namespace.
805 struct list_head l_lru;
807 * Linkage to resource's lock queues according to current lock state.
808 * (could be granted or waiting)
809 * Protected by lr_lock in struct ldlm_resource.
811 struct list_head l_res_link;
813 * Internal structures per lock type..
816 /* LDLM_EXTENT locks only */
818 /* Originally requested extent for the extent lock. */
819 struct ldlm_extent l_req_extent;
822 struct list_head l_same_extent;
824 /* LDLM_PLAIN and LDLM_IBITS locks */
827 * Protected by lr_lock, linkages to "skip lists".
828 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
830 struct list_head l_sl_mode;
831 struct list_head l_sl_policy;
833 struct ldlm_ibits_node *l_ibits_node;
835 /* LDLM_FLOCK locks */
838 * Per export hash of flock locks.
839 * Protected by per-bucket exp->exp_flock_hash locks.
841 struct hlist_node l_exp_flock_hash;
842 struct ldlm_lock *l_same_owner;
844 struct rb_node l_fl_rb;
845 u64 l_fl_subtree_last;
849 * Per export hash of locks.
850 * Protected by per-bucket exp->exp_lock_hash locks.
852 struct hlist_node l_exp_hash;
855 * Protected by lr_lock.
857 enum ldlm_mode l_req_mode;
859 * Granted mode, also protected by lr_lock.
861 enum ldlm_mode l_granted_mode;
862 /** Lock completion handler pointer. Called when lock is granted. */
863 ldlm_completion_callback l_completion_ast;
865 * Lock blocking AST handler pointer.
866 * It plays two roles:
867 * - as a notification of an attempt to queue a conflicting lock (once)
868 * - as a notification when the lock is being cancelled.
870 * As such it's typically called twice: once for the initial conflict
871 * and then once more when the last user went away and the lock is
872 * cancelled (could happen recursively).
874 ldlm_blocking_callback l_blocking_ast;
876 * Lock glimpse handler.
877 * Glimpse handler is used to obtain LVB updates from a client by
880 ldlm_glimpse_callback l_glimpse_ast;
884 * This is a pointer to actual client export for locks that were granted
885 * to clients. Used server-side.
887 struct obd_export *l_export;
889 * Lock connection export.
890 * Pointer to server export on a client.
892 struct obd_export *l_conn_export;
895 * Remote lock handle.
896 * If the lock is remote, this is the handle of the other side lock
899 struct lustre_handle l_remote_handle;
902 * Representation of private data specific for a lock type.
903 * Examples are: extent range for extent lock or bitmask for ibits locks
905 union ldlm_policy_data l_policy_data;
908 * Lock state flags. Protected by lr_lock.
909 * \see lustre_dlm_flags.h where the bits are defined.
914 * Lock r/w usage counters.
915 * Protected by lr_lock.
920 * If the lock is granted, a process sleeps on this waitq to learn when
921 * it's no longer in use. If the lock is not granted, a process sleeps
922 * on this waitq to learn when it becomes granted.
924 wait_queue_head_t l_waitq;
927 * Time, in nanoseconds, last used by e.g. being matched by lock match.
932 * Client-side-only members.
935 enum lvb_type l_lvb_type;
938 * Temporary storage for a LVB received during an enqueue operation.
939 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
944 /** Private storage for lock user. Opaque to LDLM. */
949 * Seconds. It will be updated if there is any activity related to
950 * the lock at client, e.g. enqueue the lock. For server it is the
951 * time when blocking ast was sent.
954 time64_t l_blast_sent;
957 /* separate ost_lvb used mostly by Data-on-MDT for now.
958 * It is introduced to don't mix with layout lock data.
960 struct ost_lvb l_ost_lvb;
962 * Server-side-only members.
966 * Connection cookie for the client originating the operation.
967 * Used by Commit on Share (COS) code. Currently only used for
968 * inodebits locks on MDS.
970 __u64 l_client_cookie;
973 * List item for locks waiting for cancellation from clients.
974 * The lists this could be linked into are:
975 * waiting_locks_list (protected by waiting_locks_spinlock),
976 * then if the lock timed out, it is moved to
977 * expired_lock_list for further processing.
979 struct list_head l_pending_chain;
982 * Set when lock is sent a blocking AST. Time in seconds when timeout
983 * is reached and client holding this lock could be evicted.
984 * This timeout could be further extended by e.g. certain IO activity
986 * \see ost_rw_prolong_locks
988 time64_t l_callback_timestamp;
990 /** Local PID of process which created this lock. */
994 * Number of times blocking AST was sent for this lock.
995 * This is for debugging. Valid values are 0 and 1, if there is an
996 * attempt to send blocking AST more than once, an assertion would be
997 * hit. \see ldlm_work_bl_ast_lock
1000 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
1001 struct list_head l_bl_ast;
1002 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
1003 struct list_head l_cp_ast;
1004 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
1005 struct list_head l_rk_ast;
1008 * Pointer to a conflicting lock that caused blocking AST to be sent
1011 struct ldlm_lock *l_blocking_lock;
1013 /** Reference tracking structure to debug leaked locks. */
1014 struct lu_ref l_reference;
1015 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1016 /* Debugging stuff for bug 20498, for tracking export references. */
1017 /** number of export references taken */
1019 /** link all locks referencing one export */
1020 struct list_head l_exp_refs_link;
1021 /** referenced export object */
1022 struct obd_export *l_exp_refs_target;
1025 * export blocking dlm lock list, protected by
1026 * l_export->exp_bl_list_lock.
1027 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
1028 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
1030 struct list_head l_exp_list;
1033 enum ldlm_match_flags {
1034 LDLM_MATCH_UNREF = BIT(0),
1035 LDLM_MATCH_AST = BIT(1),
1036 LDLM_MATCH_AST_ANY = BIT(2),
1037 LDLM_MATCH_RIGHT = BIT(3),
1038 LDLM_MATCH_GROUP = BIT(4),
1041 #ifdef HAVE_INTERVAL_TREE_CACHED
1042 #define extent_last(tree) rb_entry_safe(rb_last(&tree->lit_root.rb_root),\
1043 struct ldlm_lock, l_rb)
1044 #define extent_first(tree) rb_entry_safe(rb_first(&tree->lit_root.rb_root),\
1045 struct ldlm_lock, l_rb)
1046 #define extent_top(tree) rb_entry_safe(tree->lit_root.rb_root.rb_node, \
1047 struct ldlm_lock, l_rb)
1049 #define extent_last(tree) rb_entry_safe(rb_last(&tree->lit_root), \
1050 struct ldlm_lock, l_rb)
1051 #define extent_first(tree) rb_entry_safe(rb_first(&tree->lit_root), \
1052 struct ldlm_lock, l_rb)
1053 #define extent_top(tree) rb_entry_safe(tree->lit_root.rb_node, \
1054 struct ldlm_lock, l_rb)
1057 #define extent_next(lock) rb_entry_safe(rb_next(&lock->l_rb), \
1058 struct ldlm_lock, l_rb)
1059 #define extent_prev(lock) rb_entry_safe(rb_prev(&lock->l_rb), \
1060 struct ldlm_lock, l_rb)
1063 * Describe the overlap between two locks. itree_overlap_cb data.
1065 struct ldlm_match_data {
1066 struct ldlm_lock *lmd_old;
1067 struct ldlm_lock *lmd_lock;
1068 enum ldlm_mode *lmd_mode;
1069 union ldlm_policy_data *lmd_policy;
1071 __u64 lmd_skip_flags;
1072 enum ldlm_match_flags lmd_match;
1075 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
1076 #define l_transno l_client_cookie
1078 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
1079 * which is for server.
1081 #define l_slc_link l_rk_ast
1083 struct lustre_handle_array {
1084 unsigned int ha_count;
1085 /* ha_map is used as bit flag to indicate handle is remote or local */
1086 DECLARE_BITMAP(ha_map, LMV_MAX_STRIPE_COUNT);
1087 struct lustre_handle ha_handles[0];
1091 * LDLM resource description.
1092 * Basically, resource is a representation for a single object.
1093 * Object has a name which is currently 4 64-bit integers. LDLM user is
1094 * responsible for creation of a mapping between objects it wants to be
1095 * protected and resource names.
1097 * A resource can only hold locks of a single lock type, though there may be
1098 * multiple ldlm_locks on a single resource, depending on the lock type and
1099 * whether the locks are conflicting or not.
1101 struct ldlm_resource {
1102 struct ldlm_ns_bucket *lr_ns_bucket;
1105 * List item for list in namespace hash.
1106 * protected by ns_lock.
1107 * Shared with linkage for RCU-delayed free.
1110 struct hlist_node lr_hash;
1111 struct rcu_head lr_rcu;
1114 /** Reference count for this resource */
1115 refcount_t lr_refcount;
1117 /** Spinlock to protect locks under this resource. */
1120 /* protected by lr_lock */
1122 /** List of locks in granted state */
1123 struct list_head lr_granted;
1125 * List of locks that could not be granted due to conflicts and
1126 * that are waiting for conflicts to go away
1128 struct list_head lr_waiting;
1131 /** Resource name */
1132 struct ldlm_res_id lr_name;
1135 /* Interval trees (for extent locks) all modes of resource */
1136 struct ldlm_interval_tree *lr_itree;
1137 struct ldlm_ibits_queues *lr_ibits_queues;
1138 struct ldlm_flock_node lr_flock_node;
1142 /* resource considered as contended, used only on server side*/
1143 time64_t lr_contention_time;
1145 * Associated inode, used only on client side.
1147 struct inode *lr_lvb_inode;
1150 /** Type of locks this resource can hold. Only one type per resource. */
1151 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1154 * Server-side-only lock value block elements.
1155 * To serialize lvbo_init.
1158 struct mutex lr_lvb_mutex;
1159 /** protected by lr_lock */
1161 /** is lvb initialized ? */
1162 bool lr_lvb_initialized;
1164 /** List of references to this resource. For debugging. */
1165 struct lu_ref lr_reference;
1168 static inline int ldlm_is_granted(struct ldlm_lock *lock)
1170 return lock->l_req_mode == lock->l_granted_mode;
1173 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1175 return lock->l_resource->lr_type == LDLM_IBITS &&
1176 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1179 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1181 return lock->l_resource->lr_type == LDLM_IBITS &&
1182 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1185 static inline char *
1186 ldlm_ns_name(struct ldlm_namespace *ns)
1191 static inline struct ldlm_namespace *
1192 ldlm_res_to_ns(struct ldlm_resource *res)
1194 return res->lr_ns_bucket->nsb_namespace;
1197 static inline struct ldlm_namespace *
1198 ldlm_lock_to_ns(struct ldlm_lock *lock)
1200 return ldlm_res_to_ns(lock->l_resource);
1203 static inline char *
1204 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1206 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1209 static inline struct adaptive_timeout *
1210 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1212 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1215 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1217 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1220 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1221 res->lr_lvb_initialized)
1224 mutex_lock(&res->lr_lvb_mutex);
1225 /* Did we lose the race? */
1226 if (res->lr_lvb_initialized) {
1227 mutex_unlock(&res->lr_lvb_mutex);
1230 rc = ns->ns_lvbo->lvbo_init(res);
1232 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1234 if (res->lr_lvb_data != NULL) {
1235 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1236 res->lr_lvb_data = NULL;
1238 res->lr_lvb_len = rc;
1240 res->lr_lvb_initialized = true;
1242 mutex_unlock(&res->lr_lvb_mutex);
1246 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1248 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1250 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1251 return ns->ns_lvbo->lvbo_size(lock);
1256 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int *len)
1258 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1261 if (ns->ns_lvbo != NULL) {
1262 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1263 /* init lvb now if not already */
1264 rc = ldlm_lvbo_init(lock->l_resource);
1266 CERROR("lock %p: delayed lvb init failed (rc %d)\n",
1270 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1275 struct ldlm_ast_work {
1276 struct ldlm_lock *w_lock;
1278 struct ldlm_lock_desc w_desc;
1279 struct list_head w_list;
1286 * Common ldlm_enqueue parameters
1288 struct ldlm_enqueue_info {
1289 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1290 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1291 void *ei_cb_bl; /** blocking lock callback */
1292 void *ei_cb_local_bl; /** blocking local lock callback */
1293 void *ei_cb_cp; /** lock completion callback */
1294 void *ei_cb_gl; /** lock glimpse callback */
1295 ldlm_created_callback ei_cb_created; /** lock created callback */
1296 void *ei_cbdata; /** Data to be passed into callbacks. */
1297 void *ei_namespace; /** lock namespace **/
1298 u64 ei_inodebits; /** lock inode bits **/
1299 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1300 unsigned int ei_req_slot:1; /** whether acquire rpc slot */
1301 unsigned int ei_mod_slot:1; /** whether acquire mod rpc slot */
1304 #define ei_res_id ei_cb_gl
1306 extern char *ldlm_lockname[];
1307 extern char *ldlm_typename[];
1308 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1311 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1312 * For the cases where we do not have actual lock to print along
1313 * with a debugging message that is ldlm-related
1315 #define LDLM_DEBUG_NOLOCK(format, a...) \
1316 CDEBUG(D_DLMTRACE, "### " format "\n", ##a)
1319 * Support function for lock information printing into debug logs.
1323 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1324 if (((mask) & D_CANTMASK) != 0 || \
1325 ((libcfs_debug & (mask)) != 0 && \
1326 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1327 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1330 __printf(3, 4) /* function attribute */
1331 void _ldlm_lock_debug(struct ldlm_lock *lock,
1332 struct libcfs_debug_msg_data *data,
1333 const char *fmt, ...);
1336 * Rate-limited version of lock printing function.
1338 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1339 static struct cfs_debug_limit_state _ldlm_cdls; \
1340 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1341 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt, ##a); \
1344 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1345 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1347 /** Non-rate-limited lock printing function for debugging purposes. */
1348 #define LDLM_DEBUG(lock, fmt, a...) do { \
1349 if (likely(lock != NULL)) { \
1350 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1351 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1354 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1357 #else /* !LIBCFS_DEBUG */
1358 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1359 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1360 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1364 * Three intentions can be used for the policy functions in
1365 * ldlm_processing_policy.
1367 * LDLM_PROCESS_RESCAN:
1369 * It's used when policy functions are called from ldlm_reprocess_queue() to
1370 * reprocess the wait list and try to grant locks, blocking ASTs
1371 * have already been sent in this situation, completion ASTs need be sent for
1372 * the locks being granted.
1374 * LDLM_PROCESS_ENQUEUE:
1376 * It's used when policy functions are called from ldlm_lock_enqueue() to
1377 * process the wait list for handling an enqueue request, blocking
1378 * ASTs have not been sent yet, so list of conflicting locks would be
1379 * collected and ASTs sent.
1381 * LDLM_PROCESS_RECOVERY:
1383 * It's used when policy functions are called from ldlm_reprocess_queue() to
1384 * reprocess the wait list when recovery done. In case of blocking
1385 * ASTs are lost before recovery, it needs not only to grant locks if
1386 * available, but also send blocking ASTs to the locks doesn't have AST sent
1387 * flag. Completion ASTs need be sent for the locks being granted.
1389 enum ldlm_process_intention {
1390 LDLM_PROCESS_RESCAN = 0,
1391 LDLM_PROCESS_ENQUEUE = 1,
1392 LDLM_PROCESS_RECOVERY = 2,
1395 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1396 enum ldlm_process_intention intention,
1397 enum ldlm_error *err,
1398 struct list_head *work_list);
1400 typedef int (*ldlm_reprocessing_policy)(struct ldlm_resource *res,
1401 struct list_head *queue,
1402 struct list_head *work_list,
1403 enum ldlm_process_intention intention,
1407 * Return values for lock iterators.
1408 * Also used during deciding of lock grants and cancellations.
1410 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1411 #define LDLM_ITER_STOP 2 /* stop iterating */
1413 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1414 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1416 /** \defgroup ldlm_iterator Lock iterators
1418 * LDLM provides for a way to iterate through every lock on a resource or
1419 * namespace or every resource in a namespace.
1422 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1424 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1426 int ldlm_resource_iterate(struct ldlm_namespace *ln,
1427 const struct ldlm_res_id *lri,
1428 ldlm_iterator_t iter, void *data);
1429 /** @} ldlm_iterator */
1431 int ldlm_replay_locks(struct obd_import *imp);
1434 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1437 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1439 struct ldlm_prolong_args {
1440 struct obd_export *lpa_export;
1441 struct ptlrpc_request *lpa_req;
1442 struct ldlm_res_id lpa_resid;
1443 struct ldlm_extent lpa_extent;
1444 enum ldlm_mode lpa_mode;
1448 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1449 struct ldlm_prolong_args *arg);
1450 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1452 struct ldlm_callback_suite {
1453 ldlm_completion_callback lcs_completion;
1454 ldlm_blocking_callback lcs_blocking;
1455 ldlm_glimpse_callback lcs_glimpse;
1459 #ifdef HAVE_SERVER_SUPPORT
1460 /** \defgroup ldlm_srv_ast Server AST handlers
1461 * These are AST handlers used by server code.
1462 * Their property is that they are just preparing RPCs to be sent to clients.
1465 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1466 void *data, int flag);
1467 int tgt_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1468 void *data, int flag);
1469 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1470 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1471 int ldlm_glimpse_locks(struct ldlm_resource *res,
1472 struct list_head *gl_work_list);
1473 /** @} ldlm_srv_ast */
1475 /** \defgroup ldlm_handlers Server LDLM handlers
1476 * These are handler functions that should be called by "frontends" such as
1477 * MDT or OST to pass through LDLM requests to LDLM for handling
1480 int ldlm_handle_enqueue(struct ldlm_namespace *ns, struct req_capsule *pill,
1481 const struct ldlm_request *dlm_req,
1482 const struct ldlm_callback_suite *cbs);
1483 int ldlm_handle_convert0(struct ptlrpc_request *req,
1484 const struct ldlm_request *dlm_req);
1485 int ldlm_handle_cancel(struct ptlrpc_request *req);
1486 int ldlm_request_cancel(struct ptlrpc_request *req,
1487 const struct ldlm_request *dlm_req,
1488 int first, enum lustre_at_flags flags);
1489 /** @} ldlm_handlers */
1491 void ldlm_revoke_export_locks(struct obd_export *exp);
1492 timeout_t ldlm_bl_timeout(struct ldlm_lock *lock);
1493 timeout_t ldlm_bl_timeout_by_rpc(struct ptlrpc_request *req);
1495 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1496 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, timeout_t timeout);
1497 int ldlm_get_ref(void);
1498 void ldlm_put_ref(void);
1499 int ldlm_init_export(struct obd_export *exp);
1500 void ldlm_destroy_export(struct obd_export *exp);
1501 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1504 #ifdef HAVE_SERVER_SUPPORT
1505 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1506 ldlm_reprocessing_policy
1507 ldlm_get_reprocessing_policy(struct ldlm_resource *res);
1509 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1510 void ldlm_lock2handle(const struct ldlm_lock *lock,
1511 struct lustre_handle *lockh);
1512 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *lh,
1514 void ldlm_cancel_callback(struct ldlm_lock *ll);
1515 int ldlm_lock_remove_from_lru(struct ldlm_lock *ll);
1516 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1519 * Obtain a lock reference by its handle.
1521 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1523 return __ldlm_handle2lock(h, 0);
1526 #define LDLM_LOCK_REF_DEL(lock) \
1527 lu_ref_del(&lock->l_reference, "handle", lock)
1529 static inline struct ldlm_lock *
1530 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1532 struct ldlm_lock *lock;
1534 lock = __ldlm_handle2lock(h, flags);
1536 LDLM_LOCK_REF_DEL(lock);
1541 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1542 * data from request \a r
1544 static inline int ldlm_lvbo_update(struct ldlm_resource *res,
1545 struct ldlm_lock *lock,
1546 struct ptlrpc_request *req, int increase)
1548 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1551 /* delayed lvb init may be required */
1552 rc = ldlm_lvbo_init(res);
1554 CERROR("delayed lvb init failed (rc %d)\n", rc);
1558 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1559 return ns->ns_lvbo->lvbo_update(res, lock, req, increase);
1564 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1565 struct ptlrpc_request *req,
1568 return ldlm_lvbo_update(res, NULL, req, increase);
1571 int is_granted_or_cancelled_nolock(struct ldlm_lock *lock);
1573 int ldlm_error2errno(enum ldlm_error error);
1574 /* don't call it `errno': this confuses user-space. */
1575 enum ldlm_error ldlm_errno2error(int err_no);
1576 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1577 void ldlm_dump_export_locks(struct obd_export *exp);
1580 /* Release temporary lock got by ldlm_handle2lock() or __ldlm_handle2lock() */
1581 #define LDLM_LOCK_PUT(lock) \
1583 LDLM_LOCK_REF_DEL(lock); \
1584 /*LDLM_DEBUG((lock), "put");*/ \
1585 ldlm_lock_put(lock); \
1589 * Release a lock reference obtained by some other means (see
1592 #define LDLM_LOCK_RELEASE(lock) ldlm_lock_put(lock)
1594 #define LDLM_LOCK_GET(lock) \
1596 ldlm_lock_get(lock); \
1597 /*LDLM_DEBUG((lock), "get");*/ \
1601 #define ldlm_lock_list_put(head, member, count) \
1603 struct ldlm_lock *_lock, *_next; \
1605 list_for_each_entry_safe(_lock, _next, head, member) { \
1608 list_del_init(&_lock->member); \
1609 LDLM_LOCK_RELEASE(_lock); \
1614 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1615 void ldlm_lock_put(struct ldlm_lock *lock);
1616 void ldlm_lock_destroy(struct ldlm_lock *lock);
1617 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1618 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1619 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1620 enum ldlm_mode mode);
1621 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1622 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1623 enum ldlm_mode mode);
1624 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1625 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1626 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1627 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1629 enum ldlm_mode ldlm_lock_match_with_skip(struct ldlm_namespace *ns,
1630 __u64 flags, __u64 skip_flags,
1631 const struct ldlm_res_id *res_id,
1632 enum ldlm_type type,
1633 union ldlm_policy_data *policy,
1634 enum ldlm_mode mode,
1635 struct lustre_handle *lh,
1636 enum ldlm_match_flags match_flags);
1637 static inline enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns,
1639 const struct ldlm_res_id *res_id,
1640 enum ldlm_type type,
1641 union ldlm_policy_data *policy,
1642 enum ldlm_mode mode,
1643 struct lustre_handle *lh)
1645 return ldlm_lock_match_with_skip(ns, flags, 0, res_id, type, policy,
1648 struct ldlm_lock *search_itree(struct ldlm_resource *res,
1649 struct ldlm_match_data *data);
1650 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1652 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1653 void ldlm_lock_cancel(struct ldlm_lock *lock);
1654 void ldlm_reprocess_all(struct ldlm_resource *res, __u64 hint);
1655 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1656 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1657 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1660 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1661 enum ldlm_side client,
1662 enum ldlm_appetite apt,
1663 enum ldlm_ns_type ns_type);
1664 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1665 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1666 struct obd_import *imp,
1668 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1669 void ldlm_namespace_free(struct ldlm_namespace *ns,
1670 struct obd_import *imp, int force);
1671 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1672 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1673 enum ldlm_side client);
1674 void ldlm_namespace_get(struct ldlm_namespace *ns);
1675 void ldlm_namespace_put(struct ldlm_namespace *ns);
1677 int ldlm_debugfs_setup(void);
1678 void ldlm_debugfs_cleanup(void);
1680 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1681 struct lprocfs_stats *srv_stats)
1683 int lock_type = 0, op = 0;
1685 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1687 switch (lock_type) {
1689 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1692 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1695 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1698 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1706 lprocfs_counter_incr(srv_stats, op);
1709 /* resource.c - internal */
1710 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1711 const struct ldlm_res_id *,
1712 enum ldlm_type type, int create);
1713 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1714 int ldlm_resource_putref(struct ldlm_resource *res);
1715 void ldlm_resource_add_lock(struct ldlm_resource *res,
1716 struct list_head *head,
1717 struct ldlm_lock *lock);
1718 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1719 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1720 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1721 void ldlm_namespace_dump(int level, struct ldlm_namespace *ln);
1722 void ldlm_resource_dump(int level, struct ldlm_resource *lr);
1723 int ldlm_lock_change_resource(struct ldlm_namespace *ln, struct ldlm_lock *ll,
1724 const struct ldlm_res_id *lri);
1726 #define LDLM_RESOURCE_ADDREF(res) \
1727 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current)
1729 #define LDLM_RESOURCE_DELREF(res) \
1730 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current)
1732 /* ldlm_request.c */
1733 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1734 * These AST handlers are typically used for server-side local locks and are
1735 * also used by client-side lock handlers to perform minimum level base
1738 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1739 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1740 void *data, int flag);
1741 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1742 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1743 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1744 /** @} ldlm_local_ast */
1746 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1747 * These are typically used by client and server (*_local versions)
1748 * to obtain and release locks.
1750 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1751 struct ldlm_enqueue_info *einfo,
1752 const struct ldlm_res_id *res_id,
1753 union ldlm_policy_data const *policy, __u64 *flags,
1754 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1755 struct lustre_handle *lockh, int async);
1756 int ldlm_prep_enqueue_req(struct obd_export *exp,
1757 struct ptlrpc_request *req,
1758 struct list_head *cancels,
1760 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1761 int version, int opc, int canceloff,
1762 struct list_head *cancels, int count);
1764 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1765 int ldlm_handle_enqueue(struct ldlm_namespace *ns, struct req_capsule *pill,
1766 const struct ldlm_request *dlm_req,
1767 const struct ldlm_callback_suite *cbs);
1768 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct req_capsule *pill,
1769 struct ldlm_enqueue_info *einfo, __u8 with_policy,
1770 __u64 *flags, void *lvb, __u32 lvb_len,
1771 const struct lustre_handle *lockh, int rc,
1773 int ldlm_cli_enqueue_local(const struct lu_env *env,
1774 struct ldlm_namespace *ns,
1775 const struct ldlm_res_id *res_id,
1776 enum ldlm_type type, union ldlm_policy_data *policy,
1777 enum ldlm_mode mode, __u64 *flags,
1778 ldlm_blocking_callback blocking,
1779 ldlm_completion_callback completion,
1780 ldlm_glimpse_callback glimpse,
1781 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1782 const __u64 *client_cookie,
1783 struct lustre_handle *lockh);
1784 int ldlm_cli_lock_create_pack(struct obd_export *exp,
1785 struct ldlm_request *dlmreq,
1786 struct ldlm_enqueue_info *einfo,
1787 const struct ldlm_res_id *res_id,
1788 union ldlm_policy_data const *policy,
1789 __u64 *flags, void *lvb, __u32 lvb_len,
1790 enum lvb_type lvb_type,
1791 struct lustre_handle *lockh);
1792 int ldlm_cli_convert_req(struct ldlm_lock *lock, __u32 *flags, __u64 new_bits);
1793 int ldlm_cli_convert(struct ldlm_lock *lock,
1794 enum ldlm_cancel_flags cancel_flags);
1795 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1796 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1797 enum ldlm_cancel_flags cancel_flags);
1798 int ldlm_cli_cancel_unused(struct ldlm_namespace *n,
1799 const struct ldlm_res_id *l,
1800 enum ldlm_cancel_flags flags, void *opaque);
1801 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1802 const struct ldlm_res_id *res_id,
1803 union ldlm_policy_data *policy,
1804 enum ldlm_mode mode,
1805 enum ldlm_cancel_flags flags, void *opaque);
1806 int ldlm_cli_cancel_req(struct obd_export *exp, struct ldlm_lock *lock,
1807 struct list_head *head, int count,
1808 enum ldlm_cancel_flags flags);
1809 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1810 struct list_head *cancels,
1811 union ldlm_policy_data *policy,
1812 enum ldlm_mode mode, __u64 lock_flags,
1813 enum ldlm_cancel_flags cancel_flags,
1815 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1816 enum ldlm_cancel_flags flags);
1817 int ldlm_cli_cancel_list(struct list_head *head, int count,
1818 struct ptlrpc_request *req,
1819 enum ldlm_cancel_flags flags);
1821 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);
1822 int ldlm_cli_inodebits_convert(struct ldlm_lock *lock,
1823 enum ldlm_cancel_flags cancel_flags);
1825 /** @} ldlm_cli_api */
1829 /* This has to be here because recursive inclusion sucks. */
1830 int intent_disposition(struct ldlm_reply *rep, int flag);
1831 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1834 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1835 * than one lock_res is dead-lock safe.
1837 enum lock_res_type {
1842 /** Lock resource. */
1843 static inline void lock_res(struct ldlm_resource *res)
1845 spin_lock(&res->lr_lock);
1848 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1849 static inline void lock_res_nested(struct ldlm_resource *res,
1850 enum lock_res_type mode)
1852 spin_lock_nested(&res->lr_lock, mode);
1855 /** Unlock resource. */
1856 static inline void unlock_res(struct ldlm_resource *res)
1858 spin_unlock(&res->lr_lock);
1861 /** Check if resource is already locked, assert if not. */
1862 static inline void check_res_locked(struct ldlm_resource *res)
1864 assert_spin_locked(&res->lr_lock);
1867 struct ldlm_resource *lock_res_and_lock(struct ldlm_lock *lock);
1868 void unlock_res_and_lock(struct ldlm_lock *lock);
1871 /** \defgroup ldlm_pools Various LDLM pool related functions
1872 * There are not used outside of ldlm.
1875 int ldlm_pools_init(void);
1876 void ldlm_pools_fini(void);
1878 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1879 int idx, enum ldlm_side client);
1880 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1881 void ldlm_pool_fini(struct ldlm_pool *pl);
1882 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1883 time64_t ldlm_pool_recalc(struct ldlm_pool *pl, bool force);
1884 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1885 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1886 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1887 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1888 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1889 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1890 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1891 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1892 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1895 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1896 const struct ldlm_extent *ex2)
1898 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1901 /* check if @ex1 contains @ex2 */
1902 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1903 const struct ldlm_extent *ex2)
1905 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1908 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);