4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.gnu.org/licenses/gpl-2.0.html
23 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Use is subject to license terms.
26 * Copyright (c) 2010, 2017, Intel Corporation.
29 * This file is part of Lustre, http://www.lustre.org/
30 * Lustre is a trademark of Sun Microsystems, Inc.
33 /** \defgroup LDLM Lustre Distributed Lock Manager
35 * Lustre DLM is based on VAX DLM.
36 * Its two main roles are:
37 * - To provide locking assuring consistency of data on all Lustre nodes.
38 * - To allow clients to cache state protected by a lock by holding the
39 * lock until a conflicting lock is requested or it is expired by the LRU.
44 #ifndef _LUSTRE_DLM_H__
45 #define _LUSTRE_DLM_H__
47 #include <lustre_lib.h>
48 #include <lustre_net.h>
49 #include <lustre_import.h>
50 #include <lustre_handles.h>
51 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
54 #include "lustre_dlm_flags.h"
59 extern struct kset *ldlm_ns_kset;
60 extern struct kset *ldlm_svc_kset;
62 #define OBD_LDLM_DEVICENAME "ldlm"
64 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
65 #define LDLM_DEFAULT_MAX_ALIVE 3900 /* 3900 seconds ~65 min */
66 #define LDLM_CTIME_AGE_LIMIT (10)
67 /* if client lock is unused for that time it can be cancelled if any other
68 * client shows interest in that lock, e.g. glimpse is occured. */
69 #define LDLM_DIRTY_AGE_LIMIT (10)
70 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
71 #define LDLM_DEFAULT_LRU_SHRINK_BATCH (16)
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 six 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.
144 #define LCK_COMPAT_EX LCK_NL
145 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
146 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
147 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
148 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
149 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
150 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
151 #define LCK_COMPAT_COS (LCK_COS)
152 /** @} Lock Compatibility Matrix */
154 extern enum ldlm_mode lck_compat_array[];
156 static inline void lockmode_verify(enum ldlm_mode mode)
158 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
161 static inline int lockmode_compat(enum ldlm_mode exist_mode,
162 enum ldlm_mode new_mode)
164 return lck_compat_array[exist_mode] & new_mode;
169 * cluster name spaces
173 #define DLM_OST_NAMESPACE 1
174 #define DLM_MDS_NAMESPACE 2
177 - do we just separate this by security domains and use a prefix for
178 multiple namespaces in the same domain?
183 * Locking rules for LDLM:
188 * waiting_locks_spinlock
201 /* Cancel lru flag, it indicates we cancel aged locks. */
202 enum ldlm_lru_flags {
203 LDLM_LRU_FLAG_NO_WAIT = 0x1, /* Cancel locks w/o blocking (neither
204 * sending nor waiting for any RPCs) */
205 LDLM_LRU_FLAG_CLEANUP = 0x2, /* Used when clearing lru, tells
206 * prepare_lru_list to set discard flag
207 * on PR extent locks so we don't waste
208 * time saving pages that will be
209 * discarded momentarily */
214 struct ldlm_resource;
215 struct ldlm_namespace;
218 * Operations on LDLM pools.
219 * LDLM pool is a pool of locks in the namespace without any implicitly
221 * Locks in the pool are organized in LRU.
222 * Local memory pressure or server instructions (e.g. mempressure on server)
223 * can trigger freeing of locks from the pool
225 struct ldlm_pool_ops {
226 /** Recalculate pool \a pl usage */
227 int (*po_recalc)(struct ldlm_pool *pl, bool force);
228 /** Cancel at least \a nr locks from pool \a pl */
229 int (*po_shrink)(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
230 int (*po_setup)(struct ldlm_pool *pl, int limit);
233 /** One second for pools thread check interval. Each pool has own period. */
234 #define LDLM_POOLS_THREAD_PERIOD (1)
236 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
237 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
239 /** Default recalc period for server side pools in sec. */
240 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
242 /** Default recalc period for client side pools in sec. */
243 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
246 * LDLM pool structure to track granted locks.
247 * For purposes of determining when to release locks on e.g. memory pressure.
248 * This feature is commonly referred to as lru_resize.
251 /** Pool debugfs directory. */
252 struct dentry *pl_debugfs_entry;
253 /** Pool name, must be long enough to hold compound proc entry name. */
255 /** Lock for protecting SLV/CLV updates. */
257 /** Number of allowed locks in in pool, both, client and server side. */
259 /** Number of granted locks in */
261 /** Grant rate per T. */
262 atomic_t pl_grant_rate;
263 /** Cancel rate per T. */
264 atomic_t pl_cancel_rate;
265 /** Server lock volume (SLV). Protected by pl_lock. */
266 __u64 pl_server_lock_volume;
267 /** Current biggest client lock volume. Protected by pl_lock. */
268 __u64 pl_client_lock_volume;
269 /** Lock volume factor, shown in percents in procfs, but internally
270 * Client SLV calculated as: server_slv * lock_volume_factor >> 8.
272 atomic_t pl_lock_volume_factor;
273 /** Time when last SLV from server was obtained. */
274 time64_t pl_recalc_time;
275 /** Recalculation period for pool. */
276 time64_t pl_recalc_period;
277 /** Recalculation and shrink operations. */
278 struct ldlm_pool_ops *pl_ops;
279 /** Number of planned locks for next period. */
281 /** Pool statistics. */
282 struct lprocfs_stats *pl_stats;
285 struct kobject pl_kobj;
286 struct completion pl_kobj_unregister;
289 typedef int (*ldlm_res_policy)(const struct lu_env *env,
290 struct ldlm_namespace *,
291 struct ldlm_lock **, void *req_cookie,
292 enum ldlm_mode mode, __u64 flags, void *data);
294 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
298 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
299 * be associated with an LDLM lock and transferred from client to server and
302 * Currently LVBs are used by:
303 * - OSC-OST code to maintain current object size/times
304 * - layout lock code to return the layout when the layout lock is granted
306 * To ensure delayed LVB initialization, it is highly recommended to use the set
307 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
309 struct ldlm_valblock_ops {
310 int (*lvbo_init)(struct ldlm_resource *res);
311 int (*lvbo_update)(struct ldlm_resource *res, struct ldlm_lock *lock,
312 struct ptlrpc_request *r, int increase);
313 int (*lvbo_free)(struct ldlm_resource *res);
314 /* Return size of lvb data appropriate RPC size can be reserved */
315 int (*lvbo_size)(struct ldlm_lock *lock);
316 /* Called to fill in lvb data to RPC buffer @buf */
317 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int *buflen);
321 * LDLM pools related, type of lock pool in the namespace.
322 * Greedy means release cached locks aggressively
325 LDLM_NAMESPACE_GREEDY = BIT(0),
326 LDLM_NAMESPACE_MODEST = BIT(1),
330 * Default values for the "max_nolock_size", "contention_time" and
331 * "contended_locks" namespace tunables.
333 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
334 #define NS_DEFAULT_CONTENTION_SECONDS 2
335 #define NS_DEFAULT_CONTENDED_LOCKS 32
337 struct ldlm_ns_bucket {
338 /** back pointer to namespace */
339 struct ldlm_namespace *nsb_namespace;
341 * Estimated lock callback time. Used by adaptive timeout code to
342 * avoid spurious client evictions due to unresponsiveness when in
343 * fact the network or overall system load is at fault
345 struct adaptive_timeout nsb_at_estimate;
347 * Which res in the bucket should we start with the reclaim.
349 int nsb_reclaim_start;
350 /* counter of entries in this bucket */
355 /** LDLM namespace lock stats */
361 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
362 LDLM_NS_TYPE_MDC, /**< MDC namespace */
363 LDLM_NS_TYPE_MDT, /**< MDT namespace */
364 LDLM_NS_TYPE_OSC, /**< OSC namespace */
365 LDLM_NS_TYPE_OST, /**< OST namespace */
366 LDLM_NS_TYPE_MGC, /**< MGC namespace */
367 LDLM_NS_TYPE_MGT, /**< MGT namespace */
370 enum ldlm_namespace_flags {
372 * Flag to indicate the LRU cancel is in progress.
373 * Used to limit the process by 1 thread only.
381 * Namespace serves to contain locks related to a particular service.
382 * There are two kinds of namespaces:
383 * - Server namespace has knowledge of all locks and is therefore authoritative
384 * to make decisions like what locks could be granted and what conflicts
385 * exist during new lock enqueue.
386 * - Client namespace only has limited knowledge about locks in the namespace,
387 * only seeing locks held by the client.
389 * Every Lustre service has one server namespace present on the server serving
390 * that service. Every client connected to the service has a client namespace
392 * Every lock obtained by client in that namespace is actually represented by
393 * two in-memory locks. One on the server and one on the client. The locks are
394 * linked by a special cookie by which one node can tell to the other which lock
395 * it actually means during communications. Such locks are called remote locks.
396 * The locks held by server only without any reference to a client are called
399 struct ldlm_namespace {
400 /** Backward link to OBD, required for LDLM pool to store new SLV. */
401 struct obd_device *ns_obd;
403 /** Flag indicating if namespace is on client instead of server */
404 enum ldlm_side ns_client;
406 /** name of this namespace */
409 /** Resource hash table for namespace. */
410 struct cfs_hash *ns_rs_hash;
411 struct ldlm_ns_bucket *ns_rs_buckets;
412 unsigned int ns_bucket_bits;
417 /** big refcount (by bucket) */
421 * Namespace connect flags supported by server (may be changed via
422 * /proc, LRU resize may be disabled/enabled).
424 __u64 ns_connect_flags;
426 /** Client side original connect flags supported by server. */
427 __u64 ns_orig_connect_flags;
429 /* namespace debugfs dir entry */
430 struct dentry *ns_debugfs_entry;
433 * Position in global namespace list linking all namespaces on
436 struct list_head ns_list_chain;
439 * List of unused locks for this namespace. This list is also called
441 * Unused locks are locks with zero reader/writer reference counts.
442 * This list is only used on clients for lock caching purposes.
443 * When we want to release some locks voluntarily or if server wants
444 * us to release some locks due to e.g. memory pressure, we take locks
445 * to release from the head of this list.
446 * Locks are linked via l_lru field in \see struct ldlm_lock.
448 struct list_head ns_unused_list;
449 /** Number of locks in the LRU list above */
451 struct list_head *ns_last_pos;
454 * Maximum number of locks permitted in the LRU. If 0, means locks
455 * are managed by pools and there is no preset limit, rather it is all
456 * controlled by available memory on this client and on server.
458 unsigned int ns_max_unused;
461 * Cancel batch, if unused lock count exceed lru_size
462 * Only be used if LRUR disable.
464 unsigned int ns_cancel_batch;
467 * How much the SLV should decrease in %% to trigger LRU cancel urgently.
469 unsigned int ns_recalc_pct;
471 /** Maximum allowed age (last used time) for locks in the LRU. Set in
472 * seconds from userspace, but stored in ns to avoid repeat conversions.
477 * Server only: number of times we evicted clients due to lack of reply
480 unsigned int ns_timeouts;
482 * Number of seconds since the file change time after which
483 * the MDT will return an UPDATE lock along with a LOOKUP lock.
484 * This allows the client to start caching negative dentries
485 * for a directory and may save an RPC for a later stat.
487 timeout_t ns_ctime_age_limit;
489 * Number of (nano)seconds since the lock was last used. The client
490 * may cancel the lock older than this age and flush related data if
491 * another client shows interest in this lock by doing glimpse request.
492 * This allows to cache stat data locally for such files early. Set in
493 * seconds from userspace, but stored in ns to avoid repeat conversions.
495 ktime_t ns_dirty_age_limit;
497 * Used to rate-limit ldlm_namespace_dump calls.
498 * \see ldlm_namespace_dump. Increased by 10 seconds every time
501 time64_t ns_next_dump;
503 /** "policy" function that does actual lock conflict determination */
504 ldlm_res_policy ns_policy;
507 * LVB operations for this namespace.
508 * \see struct ldlm_valblock_ops
510 struct ldlm_valblock_ops *ns_lvbo;
513 * Used by filter code to store pointer to OBD of the service.
514 * Should be dropped in favor of \a ns_obd
519 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
520 * a resource is removed.
522 wait_queue_head_t ns_waitq;
523 /** LDLM pool structure for this namespace */
524 struct ldlm_pool ns_pool;
525 /** Definition of how eagerly unused locks will be released from LRU */
526 enum ldlm_appetite ns_appetite;
529 * If more than \a ns_contended_locks are found, the resource is
530 * considered to be contended. Lock enqueues might specify that no
531 * contended locks should be granted
533 unsigned ns_contended_locks;
536 * The resources in this namespace remember contended state during
537 * \a ns_contention_time, in seconds.
539 timeout_t ns_contention_time;
542 * Limit size of contended extent locks, in bytes.
543 * If extended lock is requested for more then this many bytes and
544 * caller instructs us not to grant contended locks, we would disregard
547 unsigned ns_max_nolock_size;
549 /** Limit of parallel AST RPC count. */
550 unsigned ns_max_parallel_ast;
553 * Callback to check if a lock is good to be canceled by ELC or
556 ldlm_cancel_cbt ns_cancel;
558 /** LDLM lock stats */
559 struct lprocfs_stats *ns_stats;
562 * Flag to indicate namespace is being freed. Used to determine if
563 * recalculation of LDLM pool statistics should be skipped.
565 unsigned ns_stopping:1,
568 * Flag to indicate the LRU recalc on RPC reply is in progress.
569 * Used to limit the process by 1 thread only.
574 * Which bucket should we start with the lock reclaim.
576 int ns_reclaim_start;
578 struct kobject ns_kobj; /* sysfs object */
579 struct completion ns_kobj_unregister;
582 * To avoid another ns_lock usage, a separate bitops field.
584 unsigned long ns_flags;
588 * Returns 1 if namespace \a ns is a client namespace.
590 static inline int ns_is_client(struct ldlm_namespace *ns)
593 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
594 ns->ns_client == LDLM_NAMESPACE_SERVER);
595 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
599 * Returns 1 if namespace \a ns is a server namespace.
601 static inline int ns_is_server(struct ldlm_namespace *ns)
604 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
605 ns->ns_client == LDLM_NAMESPACE_SERVER);
606 return ns->ns_client == LDLM_NAMESPACE_SERVER;
610 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
612 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
615 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
619 * Returns 1 if this namespace supports lru_resize.
621 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
624 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
627 static inline void ns_register_cancel(struct ldlm_namespace *ns,
636 /** Type for blocking callback function of a lock. */
637 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
638 struct ldlm_lock_desc *new, void *data,
640 /** Type for completion callback function of a lock. */
641 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
643 /** Type for glimpse callback function of a lock. */
644 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
646 /** Type for created callback function of a lock. */
647 typedef void (*ldlm_created_callback)(struct ldlm_lock *lock);
649 /** Work list for sending GL ASTs to multiple locks. */
650 struct ldlm_glimpse_work {
651 struct ldlm_lock *gl_lock; /* lock to glimpse */
652 struct list_head gl_list; /* linkage to other gl work structs */
653 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
654 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
655 * glimpse callback request */
656 ptlrpc_interpterer_t gl_interpret_reply;
657 void *gl_interpret_data;
660 struct ldlm_bl_desc {
661 unsigned int bl_same_client:1,
665 struct ldlm_cb_set_arg {
666 struct ptlrpc_request_set *set;
667 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
669 struct list_head *list;
670 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
671 ptlrpc_interpterer_t gl_interpret_reply;
672 void *gl_interpret_data;
673 struct ldlm_bl_desc *bl_desc;
676 struct ldlm_cb_async_args {
677 struct ldlm_cb_set_arg *ca_set_arg;
678 struct ldlm_lock *ca_lock;
681 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
682 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
684 /** Interval node data for each LDLM_EXTENT lock. */
685 struct ldlm_interval {
686 struct interval_node li_node; /* node for tree management */
687 struct list_head li_group; /* the locks which have the same
688 * policy - group of the policy */
690 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
693 * Interval tree for extent locks.
694 * The interval tree must be accessed under the resource lock.
695 * Interval trees are used for granted extent locks to speed up conflicts
696 * lookup. See ldlm/interval_tree.c for more details.
698 struct ldlm_interval_tree {
701 enum ldlm_mode lit_mode; /* lock mode */
702 struct interval_node *lit_root; /* actual ldlm_interval */
706 * Lists of waiting locks for each inodebit type.
707 * A lock can be in several liq_waiting lists and it remains in lr_waiting.
709 struct ldlm_ibits_queues {
710 struct list_head liq_waiting[MDS_INODELOCK_NUMBITS];
713 struct ldlm_ibits_node {
714 struct list_head lin_link[MDS_INODELOCK_NUMBITS];
715 struct ldlm_lock *lock;
718 /** Whether to track references to exports by LDLM locks. */
719 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
722 enum ldlm_cancel_flags {
723 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
724 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
725 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
732 __u64 blocking_owner;
733 struct obd_export *blocking_export;
734 atomic_t blocking_refs;
738 union ldlm_policy_data {
739 struct ldlm_extent l_extent;
740 struct ldlm_flock l_flock;
741 struct ldlm_inodebits l_inodebits;
744 void ldlm_convert_policy_to_wire(enum ldlm_type type,
745 const union ldlm_policy_data *lpolicy,
746 union ldlm_wire_policy_data *wpolicy);
747 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
748 const union ldlm_wire_policy_data *wpolicy,
749 union ldlm_policy_data *lpolicy);
759 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
761 #define LDLM_GID_ANY ((__u64)-1)
764 * LDLM lock structure
766 * Represents a single LDLM lock and its state in memory. Each lock is
767 * associated with a single ldlm_resource, the object which is being
768 * locked. There may be multiple ldlm_locks on a single resource,
769 * depending on the lock type and whether the locks are conflicting or
775 * When remote side wants to tell us about a lock, they address
776 * it by this opaque handle. The handle does not hold a
777 * reference on the ldlm_lock, so it can be safely passed to
778 * other threads or nodes. When the lock needs to be accessed
779 * from the handle, it is looked up again in the lock table, and
780 * may no longer exist.
782 * Must be first in the structure.
784 struct portals_handle l_handle;
786 * Internal spinlock protects l_resource. We should hold this lock
787 * first before taking res_lock.
791 * Pointer to actual resource this lock is in.
792 * ldlm_lock_change_resource() can change this.
794 struct ldlm_resource *l_resource;
796 * List item for client side LRU list.
797 * Protected by ns_lock in struct ldlm_namespace.
799 struct list_head l_lru;
801 * Linkage to resource's lock queues according to current lock state.
802 * (could be granted or waiting)
803 * Protected by lr_lock in struct ldlm_resource.
805 struct list_head l_res_link;
807 * Internal structures per lock type..
810 struct ldlm_interval *l_tree_node;
811 struct ldlm_ibits_node *l_ibits_node;
814 * Per export hash of locks.
815 * Protected by per-bucket exp->exp_lock_hash locks.
817 struct hlist_node l_exp_hash;
819 * Per export hash of flock locks.
820 * Protected by per-bucket exp->exp_flock_hash locks.
822 struct hlist_node l_exp_flock_hash;
825 * Protected by lr_lock.
827 enum ldlm_mode l_req_mode;
829 * Granted mode, also protected by lr_lock.
831 enum ldlm_mode l_granted_mode;
832 /** Lock completion handler pointer. Called when lock is granted. */
833 ldlm_completion_callback l_completion_ast;
835 * Lock blocking AST handler pointer.
836 * It plays two roles:
837 * - as a notification of an attempt to queue a conflicting lock (once)
838 * - as a notification when the lock is being cancelled.
840 * As such it's typically called twice: once for the initial conflict
841 * and then once more when the last user went away and the lock is
842 * cancelled (could happen recursively).
844 ldlm_blocking_callback l_blocking_ast;
846 * Lock glimpse handler.
847 * Glimpse handler is used to obtain LVB updates from a client by
850 ldlm_glimpse_callback l_glimpse_ast;
854 * This is a pointer to actual client export for locks that were granted
855 * to clients. Used server-side.
857 struct obd_export *l_export;
859 * Lock connection export.
860 * Pointer to server export on a client.
862 struct obd_export *l_conn_export;
865 * Remote lock handle.
866 * If the lock is remote, this is the handle of the other side lock
869 struct lustre_handle l_remote_handle;
872 * Representation of private data specific for a lock type.
873 * Examples are: extent range for extent lock or bitmask for ibits locks
875 union ldlm_policy_data l_policy_data;
878 * Lock state flags. Protected by lr_lock.
879 * \see lustre_dlm_flags.h where the bits are defined.
884 * Lock r/w usage counters.
885 * Protected by lr_lock.
890 * If the lock is granted, a process sleeps on this waitq to learn when
891 * it's no longer in use. If the lock is not granted, a process sleeps
892 * on this waitq to learn when it becomes granted.
894 wait_queue_head_t l_waitq;
897 * Time, in nanoseconds, last used by e.g. being matched by lock match.
901 /** Originally requested extent for the extent lock. */
902 struct ldlm_extent l_req_extent;
905 * Client-side-only members.
908 enum lvb_type l_lvb_type;
911 * Temporary storage for a LVB received during an enqueue operation.
912 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
917 /** Private storage for lock user. Opaque to LDLM. */
922 * Seconds. It will be updated if there is any activity related to
923 * the lock at client, e.g. enqueue the lock. For server it is the
924 * time when blocking ast was sent.
927 time64_t l_blast_sent;
930 /* separate ost_lvb used mostly by Data-on-MDT for now.
931 * It is introduced to don't mix with layout lock data. */
932 struct ost_lvb l_ost_lvb;
934 * Server-side-only members.
938 * Connection cookie for the client originating the operation.
939 * Used by Commit on Share (COS) code. Currently only used for
940 * inodebits locks on MDS.
942 __u64 l_client_cookie;
945 * List item for locks waiting for cancellation from clients.
946 * The lists this could be linked into are:
947 * waiting_locks_list (protected by waiting_locks_spinlock),
948 * then if the lock timed out, it is moved to
949 * expired_lock_list for further processing.
951 struct list_head l_pending_chain;
954 * Set when lock is sent a blocking AST. Time in seconds when timeout
955 * is reached and client holding this lock could be evicted.
956 * This timeout could be further extended by e.g. certain IO activity
958 * \see ost_rw_prolong_locks
960 time64_t l_callback_timestamp;
962 /** Local PID of process which created this lock. */
966 * Number of times blocking AST was sent for this lock.
967 * This is for debugging. Valid values are 0 and 1, if there is an
968 * attempt to send blocking AST more than once, an assertion would be
969 * hit. \see ldlm_work_bl_ast_lock
972 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
973 struct list_head l_bl_ast;
974 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
975 struct list_head l_cp_ast;
976 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
977 struct list_head l_rk_ast;
980 * Pointer to a conflicting lock that caused blocking AST to be sent
983 struct ldlm_lock *l_blocking_lock;
986 * Protected by lr_lock, linkages to "skip lists".
987 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
989 struct list_head l_sl_mode;
990 struct list_head l_sl_policy;
992 /** Reference tracking structure to debug leaked locks. */
993 struct lu_ref l_reference;
994 #if LUSTRE_TRACKS_LOCK_EXP_REFS
995 /* Debugging stuff for bug 20498, for tracking export references. */
996 /** number of export references taken */
998 /** link all locks referencing one export */
999 struct list_head l_exp_refs_link;
1000 /** referenced export object */
1001 struct obd_export *l_exp_refs_target;
1004 * export blocking dlm lock list, protected by
1005 * l_export->exp_bl_list_lock.
1006 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
1007 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
1009 struct list_head l_exp_list;
1012 enum ldlm_match_flags {
1013 LDLM_MATCH_UNREF = BIT(0),
1014 LDLM_MATCH_AST = BIT(1),
1015 LDLM_MATCH_AST_ANY = BIT(2),
1016 LDLM_MATCH_RIGHT = BIT(3),
1020 * Describe the overlap between two locks. itree_overlap_cb data.
1022 struct ldlm_match_data {
1023 struct ldlm_lock *lmd_old;
1024 struct ldlm_lock *lmd_lock;
1025 enum ldlm_mode *lmd_mode;
1026 union ldlm_policy_data *lmd_policy;
1028 __u64 lmd_skip_flags;
1029 enum ldlm_match_flags lmd_match;
1032 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
1033 #define l_transno l_client_cookie
1035 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
1036 * which is for server. */
1037 #define l_slc_link l_rk_ast
1039 #define HANDLE_MAP_SIZE ((LMV_MAX_STRIPE_COUNT + 7) >> 3)
1041 struct lustre_handle_array {
1042 unsigned int ha_count;
1043 /* ha_map is used as bit flag to indicate handle is remote or local */
1044 char ha_map[HANDLE_MAP_SIZE];
1045 struct lustre_handle ha_handles[0];
1049 * LDLM resource description.
1050 * Basically, resource is a representation for a single object.
1051 * Object has a name which is currently 4 64-bit integers. LDLM user is
1052 * responsible for creation of a mapping between objects it wants to be
1053 * protected and resource names.
1055 * A resource can only hold locks of a single lock type, though there may be
1056 * multiple ldlm_locks on a single resource, depending on the lock type and
1057 * whether the locks are conflicting or not.
1059 struct ldlm_resource {
1060 struct ldlm_ns_bucket *lr_ns_bucket;
1063 * List item for list in namespace hash.
1064 * protected by ns_lock
1066 struct hlist_node lr_hash;
1068 /** Reference count for this resource */
1069 atomic_t lr_refcount;
1071 /** Spinlock to protect locks under this resource. */
1075 * protected by lr_lock
1077 /** List of locks in granted state */
1078 struct list_head lr_granted;
1080 * List of locks that could not be granted due to conflicts and
1081 * that are waiting for conflicts to go away */
1082 struct list_head lr_waiting;
1085 /** Resource name */
1086 struct ldlm_res_id lr_name;
1090 * Interval trees (only for extent locks) for all modes of
1093 struct ldlm_interval_tree *lr_itree;
1094 struct ldlm_ibits_queues *lr_ibits_queues;
1099 * When the resource was considered as contended,
1100 * used only on server side.
1102 time64_t lr_contention_time;
1104 * Associated inode, used only on client side.
1106 struct inode *lr_lvb_inode;
1109 /** Type of locks this resource can hold. Only one type per resource. */
1110 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1113 * Server-side-only lock value block elements.
1114 * To serialize lvbo_init.
1117 struct mutex lr_lvb_mutex;
1118 /** protected by lr_lock */
1120 /** is lvb initialized ? */
1121 bool lr_lvb_initialized;
1123 /** List of references to this resource. For debugging. */
1124 struct lu_ref lr_reference;
1127 static inline int ldlm_is_granted(struct ldlm_lock *lock)
1129 return lock->l_req_mode == lock->l_granted_mode;
1132 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1134 return lock->l_resource->lr_type == LDLM_IBITS &&
1135 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1138 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1140 return lock->l_resource->lr_type == LDLM_IBITS &&
1141 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1144 static inline char *
1145 ldlm_ns_name(struct ldlm_namespace *ns)
1150 static inline struct ldlm_namespace *
1151 ldlm_res_to_ns(struct ldlm_resource *res)
1153 return res->lr_ns_bucket->nsb_namespace;
1156 static inline struct ldlm_namespace *
1157 ldlm_lock_to_ns(struct ldlm_lock *lock)
1159 return ldlm_res_to_ns(lock->l_resource);
1162 static inline char *
1163 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1165 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1168 static inline struct adaptive_timeout *
1169 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1171 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1174 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1176 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1179 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1180 res->lr_lvb_initialized)
1183 mutex_lock(&res->lr_lvb_mutex);
1184 /* Did we lose the race? */
1185 if (res->lr_lvb_initialized) {
1186 mutex_unlock(&res->lr_lvb_mutex);
1189 rc = ns->ns_lvbo->lvbo_init(res);
1191 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1193 if (res->lr_lvb_data != NULL) {
1194 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1195 res->lr_lvb_data = NULL;
1197 res->lr_lvb_len = rc;
1199 res->lr_lvb_initialized = true;
1201 mutex_unlock(&res->lr_lvb_mutex);
1205 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1207 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1209 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1210 return ns->ns_lvbo->lvbo_size(lock);
1215 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int *len)
1217 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1220 if (ns->ns_lvbo != NULL) {
1221 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1222 /* init lvb now if not already */
1223 rc = ldlm_lvbo_init(lock->l_resource);
1225 CERROR("lock %p: delayed lvb init failed (rc %d)",
1229 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1234 struct ldlm_ast_work {
1235 struct ldlm_lock *w_lock;
1237 struct ldlm_lock_desc w_desc;
1238 struct list_head w_list;
1245 * Common ldlm_enqueue parameters
1247 struct ldlm_enqueue_info {
1248 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1249 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1250 void *ei_cb_bl; /** blocking lock callback */
1251 void *ei_cb_local_bl; /** blocking local lock callback */
1252 void *ei_cb_cp; /** lock completion callback */
1253 void *ei_cb_gl; /** lock glimpse callback */
1254 ldlm_created_callback ei_cb_created; /** lock created callback */
1255 void *ei_cbdata; /** Data to be passed into callbacks. */
1256 void *ei_namespace; /** lock namespace **/
1257 u64 ei_inodebits; /** lock inode bits **/
1258 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1259 unsigned int ei_enq_slot:1; /** whether acquire rpc slot */
1262 #define ei_res_id ei_cb_gl
1264 extern char *ldlm_lockname[];
1265 extern char *ldlm_typename[];
1266 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1269 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1270 * For the cases where we do not have actual lock to print along
1271 * with a debugging message that is ldlm-related
1273 #define LDLM_DEBUG_NOLOCK(format, a...) \
1274 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1277 * Support function for lock information printing into debug logs.
1281 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1282 CFS_CHECK_STACK(msgdata, mask, cdls); \
1284 if (((mask) & D_CANTMASK) != 0 || \
1285 ((libcfs_debug & (mask)) != 0 && \
1286 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1287 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1290 void _ldlm_lock_debug(struct ldlm_lock *lock,
1291 struct libcfs_debug_msg_data *data,
1292 const char *fmt, ...)
1293 __attribute__ ((format (printf, 3, 4)));
1296 * Rate-limited version of lock printing function.
1298 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1299 static struct cfs_debug_limit_state _ldlm_cdls; \
1300 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1301 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1304 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1305 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1307 /** Non-rate-limited lock printing function for debugging purposes. */
1308 #define LDLM_DEBUG(lock, fmt, a...) do { \
1309 if (likely(lock != NULL)) { \
1310 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1311 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1312 "### " fmt , ##a); \
1314 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1317 #else /* !LIBCFS_DEBUG */
1318 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1319 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1320 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1324 * Three intentions can be used for the policy functions in
1325 * ldlm_processing_policy.
1327 * LDLM_PROCESS_RESCAN:
1329 * It's used when policy functions are called from ldlm_reprocess_queue() to
1330 * reprocess the wait list and try to grant locks, blocking ASTs
1331 * have already been sent in this situation, completion ASTs need be sent for
1332 * the locks being granted.
1334 * LDLM_PROCESS_ENQUEUE:
1336 * It's used when policy functions are called from ldlm_lock_enqueue() to
1337 * process the wait list for handling an enqueue request, blocking
1338 * ASTs have not been sent yet, so list of conflicting locks would be
1339 * collected and ASTs sent.
1341 * LDLM_PROCESS_RECOVERY:
1343 * It's used when policy functions are called from ldlm_reprocess_queue() to
1344 * reprocess the wait list when recovery done. In case of blocking
1345 * ASTs are lost before recovery, it needs not only to grant locks if
1346 * available, but also send blocking ASTs to the locks doesn't have AST sent
1347 * flag. Completion ASTs need be sent for the locks being granted.
1349 enum ldlm_process_intention {
1350 LDLM_PROCESS_RESCAN = 0,
1351 LDLM_PROCESS_ENQUEUE = 1,
1352 LDLM_PROCESS_RECOVERY = 2,
1355 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1356 enum ldlm_process_intention intention,
1357 enum ldlm_error *err,
1358 struct list_head *work_list);
1360 typedef int (*ldlm_reprocessing_policy)(struct ldlm_resource *res,
1361 struct list_head *queue,
1362 struct list_head *work_list,
1363 enum ldlm_process_intention intention,
1364 struct ldlm_lock *hint);
1367 * Return values for lock iterators.
1368 * Also used during deciding of lock grants and cancellations.
1370 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1371 #define LDLM_ITER_STOP 2 /* stop iterating */
1373 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1374 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1376 /** \defgroup ldlm_iterator Lock iterators
1378 * LDLM provides for a way to iterate through every lock on a resource or
1379 * namespace or every resource in a namespace.
1381 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1383 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1385 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1386 ldlm_iterator_t iter, void *data);
1387 /** @} ldlm_iterator */
1389 int ldlm_replay_locks(struct obd_import *imp);
1392 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1395 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1397 struct ldlm_prolong_args {
1398 struct obd_export *lpa_export;
1399 struct ldlm_res_id lpa_resid;
1400 struct ldlm_extent lpa_extent;
1401 enum ldlm_mode lpa_mode;
1402 timeout_t lpa_timeout;
1406 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1407 struct ldlm_prolong_args *arg);
1408 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1410 struct ldlm_callback_suite {
1411 ldlm_completion_callback lcs_completion;
1412 ldlm_blocking_callback lcs_blocking;
1413 ldlm_glimpse_callback lcs_glimpse;
1417 #ifdef HAVE_SERVER_SUPPORT
1418 /** \defgroup ldlm_srv_ast Server AST handlers
1419 * These are AST handlers used by server code.
1420 * Their property is that they are just preparing RPCs to be sent to clients.
1423 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1424 void *data, int flag);
1425 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1426 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1427 int ldlm_glimpse_locks(struct ldlm_resource *res,
1428 struct list_head *gl_work_list);
1429 /** @} ldlm_srv_ast */
1431 /** \defgroup ldlm_handlers Server LDLM handlers
1432 * These are handler functions that should be called by "frontends" such as
1433 * MDT or OST to pass through LDLM requests to LDLM for handling
1436 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1437 const struct ldlm_request *dlm_req,
1438 const struct ldlm_callback_suite *cbs);
1439 int ldlm_handle_convert0(struct ptlrpc_request *req,
1440 const struct ldlm_request *dlm_req);
1441 int ldlm_handle_cancel(struct ptlrpc_request *req);
1442 int ldlm_request_cancel(struct ptlrpc_request *req,
1443 const struct ldlm_request *dlm_req,
1444 int first, enum lustre_at_flags flags);
1445 /** @} ldlm_handlers */
1447 void ldlm_revoke_export_locks(struct obd_export *exp);
1448 timeout_t ldlm_bl_timeout(struct ldlm_lock *lock);
1450 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1451 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, timeout_t timeout);
1452 int ldlm_get_ref(void);
1453 void ldlm_put_ref(void);
1454 int ldlm_init_export(struct obd_export *exp);
1455 void ldlm_destroy_export(struct obd_export *exp);
1456 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1459 #ifdef HAVE_SERVER_SUPPORT
1460 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1461 ldlm_reprocessing_policy
1462 ldlm_get_reprocessing_policy(struct ldlm_resource *res);
1464 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1465 void ldlm_lock2handle(const struct ldlm_lock *lock,
1466 struct lustre_handle *lockh);
1467 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1468 void ldlm_cancel_callback(struct ldlm_lock *);
1469 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1470 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1473 * Obtain a lock reference by its handle.
1475 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1477 return __ldlm_handle2lock(h, 0);
1480 #define LDLM_LOCK_REF_DEL(lock) \
1481 lu_ref_del(&lock->l_reference, "handle", lock)
1483 static inline struct ldlm_lock *
1484 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1486 struct ldlm_lock *lock;
1488 lock = __ldlm_handle2lock(h, flags);
1490 LDLM_LOCK_REF_DEL(lock);
1495 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1496 * data from request \a r
1498 static inline int ldlm_lvbo_update(struct ldlm_resource *res,
1499 struct ldlm_lock *lock,
1500 struct ptlrpc_request *req, int increase)
1502 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1505 /* delayed lvb init may be required */
1506 rc = ldlm_lvbo_init(res);
1508 CERROR("delayed lvb init failed (rc %d)\n", rc);
1512 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1513 return ns->ns_lvbo->lvbo_update(res, lock, req, increase);
1518 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1519 struct ptlrpc_request *req,
1522 return ldlm_lvbo_update(res, NULL, req, increase);
1525 int is_granted_or_cancelled_nolock(struct ldlm_lock *lock);
1527 int ldlm_error2errno(enum ldlm_error error);
1528 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1529 * confuses user-space. */
1530 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1531 void ldlm_dump_export_locks(struct obd_export *exp);
1535 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1536 * __ldlm_handle2lock().
1538 #define LDLM_LOCK_PUT(lock) \
1540 LDLM_LOCK_REF_DEL(lock); \
1541 /*LDLM_DEBUG((lock), "put");*/ \
1542 ldlm_lock_put(lock); \
1546 * Release a lock reference obtained by some other means (see
1549 #define LDLM_LOCK_RELEASE(lock) \
1551 /*LDLM_DEBUG((lock), "put");*/ \
1552 ldlm_lock_put(lock); \
1555 #define LDLM_LOCK_GET(lock) \
1557 ldlm_lock_get(lock); \
1558 /*LDLM_DEBUG((lock), "get");*/ \
1562 #define ldlm_lock_list_put(head, member, count) \
1564 struct ldlm_lock *_lock, *_next; \
1566 list_for_each_entry_safe(_lock, _next, head, member) { \
1569 list_del_init(&_lock->member); \
1570 LDLM_LOCK_RELEASE(_lock); \
1575 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1576 void ldlm_lock_put(struct ldlm_lock *lock);
1577 void ldlm_lock_destroy(struct ldlm_lock *lock);
1578 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1579 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1580 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1581 enum ldlm_mode mode);
1582 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1583 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1584 enum ldlm_mode mode);
1585 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1586 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1587 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1588 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1590 enum ldlm_mode ldlm_lock_match_with_skip(struct ldlm_namespace *ns,
1591 __u64 flags, __u64 skip_flags,
1592 const struct ldlm_res_id *res_id,
1593 enum ldlm_type type,
1594 union ldlm_policy_data *policy,
1595 enum ldlm_mode mode,
1596 struct lustre_handle *lh,
1597 enum ldlm_match_flags match_flags);
1598 static inline enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns,
1600 const struct ldlm_res_id *res_id,
1601 enum ldlm_type type,
1602 union ldlm_policy_data *policy,
1603 enum ldlm_mode mode,
1604 struct lustre_handle *lh)
1606 return ldlm_lock_match_with_skip(ns, flags, 0, res_id, type, policy,
1609 struct ldlm_lock *search_itree(struct ldlm_resource *res,
1610 struct ldlm_match_data *data);
1611 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1613 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1614 void ldlm_lock_cancel(struct ldlm_lock *lock);
1615 void ldlm_reprocess_all(struct ldlm_resource *res, struct ldlm_lock *hint);
1616 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1617 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1618 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1621 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1622 enum ldlm_side client,
1623 enum ldlm_appetite apt,
1624 enum ldlm_ns_type ns_type);
1625 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1626 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1627 struct obd_import *imp,
1629 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1630 void ldlm_namespace_free(struct ldlm_namespace *ns,
1631 struct obd_import *imp, int force);
1632 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1633 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1634 enum ldlm_side client);
1635 void ldlm_namespace_get(struct ldlm_namespace *ns);
1636 void ldlm_namespace_put(struct ldlm_namespace *ns);
1638 int ldlm_debugfs_setup(void);
1639 void ldlm_debugfs_cleanup(void);
1641 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1642 struct lprocfs_stats *srv_stats)
1644 int lock_type = 0, op = 0;
1646 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1648 switch (lock_type) {
1650 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1653 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1656 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1659 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1667 lprocfs_counter_incr(srv_stats, op);
1670 /* resource.c - internal */
1671 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1672 struct ldlm_resource *parent,
1673 const struct ldlm_res_id *,
1674 enum ldlm_type type, int create);
1675 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1676 int ldlm_resource_putref(struct ldlm_resource *res);
1677 void ldlm_resource_add_lock(struct ldlm_resource *res,
1678 struct list_head *head,
1679 struct ldlm_lock *lock);
1680 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1681 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1682 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1683 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1684 void ldlm_resource_dump(int level, struct ldlm_resource *);
1685 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1686 const struct ldlm_res_id *);
1688 #define LDLM_RESOURCE_ADDREF(res) do { \
1689 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1692 #define LDLM_RESOURCE_DELREF(res) do { \
1693 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1696 /* ldlm_request.c */
1697 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1698 * These AST handlers are typically used for server-side local locks and are
1699 * also used by client-side lock handlers to perform minimum level base
1702 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1703 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1704 void *data, int flag);
1705 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1706 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1707 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1708 /** @} ldlm_local_ast */
1710 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1711 * These are typically used by client and server (*_local versions)
1712 * to obtain and release locks.
1714 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1715 struct ldlm_enqueue_info *einfo,
1716 const struct ldlm_res_id *res_id,
1717 union ldlm_policy_data const *policy, __u64 *flags,
1718 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1719 struct lustre_handle *lockh, int async);
1720 int ldlm_prep_enqueue_req(struct obd_export *exp,
1721 struct ptlrpc_request *req,
1722 struct list_head *cancels,
1724 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1725 int version, int opc, int canceloff,
1726 struct list_head *cancels, int count);
1728 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1729 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1730 const struct ldlm_request *dlm_req,
1731 const struct ldlm_callback_suite *cbs);
1732 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1733 enum ldlm_type type, __u8 with_policy,
1734 enum ldlm_mode mode, __u64 *flags, void *lvb,
1736 const struct lustre_handle *lockh, int rc);
1737 int ldlm_cli_enqueue_local(const struct lu_env *env,
1738 struct ldlm_namespace *ns,
1739 const struct ldlm_res_id *res_id,
1740 enum ldlm_type type, union ldlm_policy_data *policy,
1741 enum ldlm_mode mode, __u64 *flags,
1742 ldlm_blocking_callback blocking,
1743 ldlm_completion_callback completion,
1744 ldlm_glimpse_callback glimpse,
1745 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1746 const __u64 *client_cookie,
1747 struct lustre_handle *lockh);
1748 int ldlm_cli_convert_req(struct ldlm_lock *lock, __u32 *flags, __u64 new_bits);
1749 int ldlm_cli_convert(struct ldlm_lock *lock,
1750 enum ldlm_cancel_flags cancel_flags);
1751 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1752 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1753 enum ldlm_cancel_flags cancel_flags);
1754 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1755 enum ldlm_cancel_flags flags, void *opaque);
1756 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1757 const struct ldlm_res_id *res_id,
1758 union ldlm_policy_data *policy,
1759 enum ldlm_mode mode,
1760 enum ldlm_cancel_flags flags, void *opaque);
1761 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1762 int count, enum ldlm_cancel_flags flags);
1763 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1764 struct list_head *cancels,
1765 union ldlm_policy_data *policy,
1766 enum ldlm_mode mode, __u64 lock_flags,
1767 enum ldlm_cancel_flags cancel_flags,
1769 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1770 enum ldlm_cancel_flags flags);
1771 int ldlm_cli_cancel_list(struct list_head *head, int count,
1772 struct ptlrpc_request *req,
1773 enum ldlm_cancel_flags flags);
1775 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);
1776 int ldlm_cli_inodebits_convert(struct ldlm_lock *lock,
1777 enum ldlm_cancel_flags cancel_flags);
1779 /** @} ldlm_cli_api */
1781 extern unsigned int ldlm_enqueue_min;
1784 /* This has to be here because recursive inclusion sucks. */
1785 int intent_disposition(struct ldlm_reply *rep, int flag);
1786 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1789 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1790 * than one lock_res is dead-lock safe.
1792 enum lock_res_type {
1797 /** Lock resource. */
1798 static inline void lock_res(struct ldlm_resource *res)
1800 spin_lock(&res->lr_lock);
1803 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1804 static inline void lock_res_nested(struct ldlm_resource *res,
1805 enum lock_res_type mode)
1807 spin_lock_nested(&res->lr_lock, mode);
1810 /** Unlock resource. */
1811 static inline void unlock_res(struct ldlm_resource *res)
1813 spin_unlock(&res->lr_lock);
1816 /** Check if resource is already locked, assert if not. */
1817 static inline void check_res_locked(struct ldlm_resource *res)
1819 assert_spin_locked(&res->lr_lock);
1822 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1823 void unlock_res_and_lock(struct ldlm_lock *lock);
1826 /** \defgroup ldlm_pools Various LDLM pool related functions
1827 * There are not used outside of ldlm.
1830 int ldlm_pools_init(void);
1831 void ldlm_pools_fini(void);
1833 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1834 int idx, enum ldlm_side client);
1835 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1836 void ldlm_pool_fini(struct ldlm_pool *pl);
1837 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1838 time64_t ldlm_pool_recalc(struct ldlm_pool *pl, bool force);
1839 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1840 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1841 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1842 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1843 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1844 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1845 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1846 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1847 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1850 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1851 const struct ldlm_extent *ex2)
1853 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1856 /* check if @ex1 contains @ex2 */
1857 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1858 const struct ldlm_extent *ex2)
1860 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1863 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);