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
73 * LDLM non-error return states
77 ELDLM_LOCK_MATCHED = 1,
79 ELDLM_LOCK_CHANGED = 300,
80 ELDLM_LOCK_ABORTED = 301,
81 ELDLM_LOCK_REPLACED = 302,
82 ELDLM_NO_LOCK_DATA = 303,
83 ELDLM_LOCK_WOULDBLOCK = 304,
85 ELDLM_NAMESPACE_EXISTS = 400,
86 ELDLM_BAD_NAMESPACE = 401,
90 * LDLM namespace type.
91 * The "client" type is actually an indication that this is a narrow local view
92 * into complete namespace on the server. Such namespaces cannot make any
93 * decisions about lack of conflicts or do any autonomous lock granting without
94 * first speaking to a server.
97 LDLM_NAMESPACE_SERVER = 0x01,
98 LDLM_NAMESPACE_CLIENT = 0x02
102 * The blocking callback is overloaded to perform two functions. These flags
103 * indicate which operation should be performed.
105 #define LDLM_CB_BLOCKING 1
106 #define LDLM_CB_CANCELING 2
109 * \name Lock Compatibility Matrix.
111 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
112 * Lock types are described in their respective implementation files:
113 * ldlm_{extent,flock,inodebits,plain}.c.
115 * There are six lock modes along with a compatibility matrix to indicate if
116 * two locks are compatible.
118 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
120 * - PW: Protective Write (normal write) mode. When a client requests a write
121 * lock from an OST, a lock with PW mode will be issued.
122 * - PR: Protective Read (normal read) mode. When a client requests a read from
123 * an OST, a lock with PR mode will be issued. Also, if the client opens a
124 * file for execution, it is granted a lock with PR mode.
125 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
126 * requests a write lock during a file open operation.
127 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
128 * an inodebit lock with the CR mode on the intermediate path component.
142 #define LCK_COMPAT_EX LCK_NL
143 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
144 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
145 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
146 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
147 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
148 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
149 #define LCK_COMPAT_COS (LCK_COS)
150 /** @} Lock Compatibility Matrix */
152 extern enum ldlm_mode lck_compat_array[];
154 static inline void lockmode_verify(enum ldlm_mode mode)
156 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
159 static inline int lockmode_compat(enum ldlm_mode exist_mode,
160 enum ldlm_mode new_mode)
162 return lck_compat_array[exist_mode] & new_mode;
167 * cluster name spaces
171 #define DLM_OST_NAMESPACE 1
172 #define DLM_MDS_NAMESPACE 2
175 - do we just separate this by security domains and use a prefix for
176 multiple namespaces in the same domain?
181 * Locking rules for LDLM:
186 * waiting_locks_spinlock
201 struct ldlm_resource;
202 struct ldlm_namespace;
205 * Operations on LDLM pools.
206 * LDLM pool is a pool of locks in the namespace without any implicitly
208 * Locks in the pool are organized in LRU.
209 * Local memory pressure or server instructions (e.g. mempressure on server)
210 * can trigger freeing of locks from the pool
212 struct ldlm_pool_ops {
213 /** Recalculate pool \a pl usage */
214 int (*po_recalc)(struct ldlm_pool *pl);
215 /** Cancel at least \a nr locks from pool \a pl */
216 int (*po_shrink)(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
217 int (*po_setup)(struct ldlm_pool *pl, int limit);
220 /** One second for pools thread check interval. Each pool has own period. */
221 #define LDLM_POOLS_THREAD_PERIOD (1)
223 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
224 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
226 /** Default recalc period for server side pools in sec. */
227 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
229 /** Default recalc period for client side pools in sec. */
230 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
233 * LDLM pool structure to track granted locks.
234 * For purposes of determining when to release locks on e.g. memory pressure.
235 * This feature is commonly referred to as lru_resize.
238 /** Pool debugfs directory. */
239 struct dentry *pl_debugfs_entry;
240 /** Pool name, must be long enough to hold compound proc entry name. */
242 /** Lock for protecting SLV/CLV updates. */
244 /** Number of allowed locks in in pool, both, client and server side. */
246 /** Number of granted locks in */
248 /** Grant rate per T. */
249 atomic_t pl_grant_rate;
250 /** Cancel rate per T. */
251 atomic_t pl_cancel_rate;
252 /** Server lock volume (SLV). Protected by pl_lock. */
253 __u64 pl_server_lock_volume;
254 /** Current biggest client lock volume. Protected by pl_lock. */
255 __u64 pl_client_lock_volume;
256 /** Lock volume factor. SLV on client is calculated as following:
257 * server_slv * lock_volume_factor. */
258 atomic_t pl_lock_volume_factor;
259 /** Time when last SLV from server was obtained. */
260 time64_t pl_recalc_time;
261 /** Recalculation period for pool. */
262 time64_t pl_recalc_period;
263 /** Recalculation and shrink operations. */
264 struct ldlm_pool_ops *pl_ops;
265 /** Number of planned locks for next period. */
267 /** Pool statistics. */
268 struct lprocfs_stats *pl_stats;
271 struct kobject pl_kobj;
272 struct completion pl_kobj_unregister;
275 typedef int (*ldlm_res_policy)(const struct lu_env *env,
276 struct ldlm_namespace *,
277 struct ldlm_lock **, void *req_cookie,
278 enum ldlm_mode mode, __u64 flags, void *data);
280 typedef int (*ldlm_cancel_cbt)(struct ldlm_lock *lock);
284 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
285 * be associated with an LDLM lock and transferred from client to server and
288 * Currently LVBs are used by:
289 * - OSC-OST code to maintain current object size/times
290 * - layout lock code to return the layout when the layout lock is granted
292 * To ensure delayed LVB initialization, it is highly recommended to use the set
293 * of ldlm_[res_]lvbo_[init,update,fill]() functions.
295 struct ldlm_valblock_ops {
296 int (*lvbo_init)(const struct lu_env *env, struct ldlm_resource *res);
297 int (*lvbo_update)(const struct lu_env *env, struct ldlm_resource *res,
298 struct ldlm_lock *lock, struct ptlrpc_request *r,
300 int (*lvbo_free)(struct ldlm_resource *res);
301 /* Return size of lvb data appropriate RPC size can be reserved */
302 int (*lvbo_size)(struct ldlm_lock *lock);
303 /* Called to fill in lvb data to RPC buffer @buf */
304 int (*lvbo_fill)(const struct lu_env *env, struct ldlm_lock *lock,
305 void *buf, int *buflen);
309 * LDLM pools related, type of lock pool in the namespace.
310 * Greedy means release cached locks aggressively
313 LDLM_NAMESPACE_GREEDY = 1 << 0,
314 LDLM_NAMESPACE_MODEST = 1 << 1
318 * Default values for the "max_nolock_size", "contention_time" and
319 * "contended_locks" namespace tunables.
321 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
322 #define NS_DEFAULT_CONTENTION_SECONDS 2
323 #define NS_DEFAULT_CONTENDED_LOCKS 32
325 struct ldlm_ns_bucket {
326 /** back pointer to namespace */
327 struct ldlm_namespace *nsb_namespace;
329 * Estimated lock callback time. Used by adaptive timeout code to
330 * avoid spurious client evictions due to unresponsiveness when in
331 * fact the network or overall system load is at fault
333 struct adaptive_timeout nsb_at_estimate;
335 * Which res in the bucket should we start with the reclaim.
337 int nsb_reclaim_start;
341 /** LDLM namespace lock stats */
347 LDLM_NS_TYPE_UNKNOWN = 0, /**< invalid type */
348 LDLM_NS_TYPE_MDC, /**< MDC namespace */
349 LDLM_NS_TYPE_MDT, /**< MDT namespace */
350 LDLM_NS_TYPE_OSC, /**< OSC namespace */
351 LDLM_NS_TYPE_OST, /**< OST namespace */
352 LDLM_NS_TYPE_MGC, /**< MGC namespace */
353 LDLM_NS_TYPE_MGT, /**< MGT namespace */
359 * Namespace serves to contain locks related to a particular service.
360 * There are two kinds of namespaces:
361 * - Server namespace has knowledge of all locks and is therefore authoritative
362 * to make decisions like what locks could be granted and what conflicts
363 * exist during new lock enqueue.
364 * - Client namespace only has limited knowledge about locks in the namespace,
365 * only seeing locks held by the client.
367 * Every Lustre service has one server namespace present on the server serving
368 * that service. Every client connected to the service has a client namespace
370 * Every lock obtained by client in that namespace is actually represented by
371 * two in-memory locks. One on the server and one on the client. The locks are
372 * linked by a special cookie by which one node can tell to the other which lock
373 * it actually means during communications. Such locks are called remote locks.
374 * The locks held by server only without any reference to a client are called
377 struct ldlm_namespace {
378 /** Backward link to OBD, required for LDLM pool to store new SLV. */
379 struct obd_device *ns_obd;
381 /** Flag indicating if namespace is on client instead of server */
382 enum ldlm_side ns_client;
384 /** name of this namespace */
387 /** Resource hash table for namespace. */
388 struct cfs_hash *ns_rs_hash;
393 /** big refcount (by bucket) */
397 * Namespace connect flags supported by server (may be changed via
398 * /proc, LRU resize may be disabled/enabled).
400 __u64 ns_connect_flags;
402 /** Client side original connect flags supported by server. */
403 __u64 ns_orig_connect_flags;
405 /* namespace debugfs dir entry */
406 struct dentry *ns_debugfs_entry;
409 * Position in global namespace list linking all namespaces on
412 struct list_head ns_list_chain;
415 * List of unused locks for this namespace. This list is also called
417 * Unused locks are locks with zero reader/writer reference counts.
418 * This list is only used on clients for lock caching purposes.
419 * When we want to release some locks voluntarily or if server wants
420 * us to release some locks due to e.g. memory pressure, we take locks
421 * to release from the head of this list.
422 * Locks are linked via l_lru field in \see struct ldlm_lock.
424 struct list_head ns_unused_list;
425 /** Number of locks in the LRU list above */
427 struct list_head *ns_last_pos;
430 * Maximum number of locks permitted in the LRU. If 0, means locks
431 * are managed by pools and there is no preset limit, rather it is all
432 * controlled by available memory on this client and on server.
434 unsigned int ns_max_unused;
436 /** Maximum allowed age (last used time) for locks in the LRU */
440 * Server only: number of times we evicted clients due to lack of reply
443 unsigned int ns_timeouts;
445 * Number of seconds since the file change time after which the
446 * MDT will return an UPDATE lock along with a LOOKUP lock.
447 * This allows the client to start caching negative dentries
448 * for a directory and may save an RPC for a later stat.
450 time64_t ns_ctime_age_limit;
452 * Number of seconds since the lock was last used. The client may
453 * cancel the lock limited by this age and flush related data if
454 * any other client shows interest in it doing glimpse request.
455 * This allows to cache stat data locally for such files early.
457 time64_t ns_dirty_age_limit;
459 * Used to rate-limit ldlm_namespace_dump calls.
460 * \see ldlm_namespace_dump. Increased by 10 seconds every time
463 time64_t ns_next_dump;
465 /** "policy" function that does actual lock conflict determination */
466 ldlm_res_policy ns_policy;
469 * LVB operations for this namespace.
470 * \see struct ldlm_valblock_ops
472 struct ldlm_valblock_ops *ns_lvbo;
475 * Used by filter code to store pointer to OBD of the service.
476 * Should be dropped in favor of \a ns_obd
481 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
482 * a resource is removed.
484 wait_queue_head_t ns_waitq;
485 /** LDLM pool structure for this namespace */
486 struct ldlm_pool ns_pool;
487 /** Definition of how eagerly unused locks will be released from LRU */
488 enum ldlm_appetite ns_appetite;
491 * If more than \a ns_contended_locks are found, the resource is
492 * considered to be contended. Lock enqueues might specify that no
493 * contended locks should be granted
495 unsigned ns_contended_locks;
498 * The resources in this namespace remember contended state during
499 * \a ns_contention_time, in seconds.
501 time64_t ns_contention_time;
504 * Limit size of contended extent locks, in bytes.
505 * If extended lock is requested for more then this many bytes and
506 * caller instructs us not to grant contended locks, we would disregard
509 unsigned ns_max_nolock_size;
511 /** Limit of parallel AST RPC count. */
512 unsigned ns_max_parallel_ast;
515 * Callback to check if a lock is good to be canceled by ELC or
518 ldlm_cancel_cbt ns_cancel;
520 /** LDLM lock stats */
521 struct lprocfs_stats *ns_stats;
524 * Flag to indicate namespace is being freed. Used to determine if
525 * recalculation of LDLM pool statistics should be skipped.
527 unsigned ns_stopping:1;
530 * Which bucket should we start with the lock reclaim.
532 int ns_reclaim_start;
534 struct kobject ns_kobj; /* sysfs object */
535 struct completion ns_kobj_unregister;
539 * Returns 1 if namespace \a ns is a client namespace.
541 static inline int ns_is_client(struct ldlm_namespace *ns)
544 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
545 LDLM_NAMESPACE_SERVER)));
546 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
547 ns->ns_client == LDLM_NAMESPACE_SERVER);
548 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
552 * Returns 1 if namespace \a ns is a server namespace.
554 static inline int ns_is_server(struct ldlm_namespace *ns)
557 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
558 LDLM_NAMESPACE_SERVER)));
559 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
560 ns->ns_client == LDLM_NAMESPACE_SERVER);
561 return ns->ns_client == LDLM_NAMESPACE_SERVER;
565 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
567 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
570 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
574 * Returns 1 if this namespace supports lru_resize.
576 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
579 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
582 static inline void ns_register_cancel(struct ldlm_namespace *ns,
591 /** Type for blocking callback function of a lock. */
592 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
593 struct ldlm_lock_desc *new, void *data,
595 /** Type for completion callback function of a lock. */
596 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
598 /** Type for glimpse callback function of a lock. */
599 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
601 /** Work list for sending GL ASTs to multiple locks. */
602 struct ldlm_glimpse_work {
603 struct ldlm_lock *gl_lock; /* lock to glimpse */
604 struct list_head gl_list; /* linkage to other gl work structs */
605 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
606 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
607 * glimpse callback request */
608 ptlrpc_interpterer_t gl_interpret_reply;
609 void *gl_interpret_data;
612 struct ldlm_bl_desc {
613 unsigned int bl_same_client:1,
617 struct ldlm_cb_set_arg {
618 struct ptlrpc_request_set *set;
619 int type; /* LDLM_{CP,BL,GL}_CALLBACK */
621 struct list_head *list;
622 union ldlm_gl_desc *gl_desc; /* glimpse AST descriptor */
623 ptlrpc_interpterer_t gl_interpret_reply;
624 void *gl_interpret_data;
625 struct ldlm_bl_desc *bl_desc;
628 struct ldlm_cb_async_args {
629 struct ldlm_cb_set_arg *ca_set_arg;
630 struct ldlm_lock *ca_lock;
633 /** The ldlm_glimpse_work was slab allocated & must be freed accordingly.*/
634 #define LDLM_GL_WORK_SLAB_ALLOCATED 0x1
636 /** Interval node data for each LDLM_EXTENT lock. */
637 struct ldlm_interval {
638 struct interval_node li_node; /* node for tree management */
639 struct list_head li_group; /* the locks which have the same
640 * policy - group of the policy */
642 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
645 * Interval tree for extent locks.
646 * The interval tree must be accessed under the resource lock.
647 * Interval trees are used for granted extent locks to speed up conflicts
648 * lookup. See ldlm/interval_tree.c for more details.
650 struct ldlm_interval_tree {
653 enum ldlm_mode lit_mode; /* lock mode */
654 struct interval_node *lit_root; /* actual ldlm_interval */
657 /** Whether to track references to exports by LDLM locks. */
658 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
661 enum ldlm_cancel_flags {
662 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
663 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
664 LCF_BL_AST = 0x4, /* Cancel LDLM_FL_BL_AST locks in the same RPC */
665 LCF_CONVERT = 0x8, /* Try to convert IBITS lock before cancel */
672 __u64 blocking_owner;
673 struct obd_export *blocking_export;
674 atomic_t blocking_refs;
678 union ldlm_policy_data {
679 struct ldlm_extent l_extent;
680 struct ldlm_flock l_flock;
681 struct ldlm_inodebits l_inodebits;
684 void ldlm_convert_policy_to_wire(enum ldlm_type type,
685 const union ldlm_policy_data *lpolicy,
686 union ldlm_wire_policy_data *wpolicy);
687 void ldlm_convert_policy_to_local(struct obd_export *exp, enum ldlm_type type,
688 const union ldlm_wire_policy_data *wpolicy,
689 union ldlm_policy_data *lpolicy);
699 * LDLM_GID_ANY is used to match any group id in ldlm_lock_match().
701 #define LDLM_GID_ANY ((__u64)-1)
704 * LDLM lock structure
706 * Represents a single LDLM lock and its state in memory. Each lock is
707 * associated with a single ldlm_resource, the object which is being
708 * locked. There may be multiple ldlm_locks on a single resource,
709 * depending on the lock type and whether the locks are conflicting or
715 * When remote side wants to tell us about a lock, they address
716 * it by this opaque handle. The handle does not hold a
717 * reference on the ldlm_lock, so it can be safely passed to
718 * other threads or nodes. When the lock needs to be accessed
719 * from the handle, it is looked up again in the lock table, and
720 * may no longer exist.
722 * Must be first in the structure.
724 struct portals_handle l_handle;
726 * Lock reference count.
727 * This is how many users have pointers to actual structure, so that
728 * we do not accidentally free lock structure that is in use.
732 * Internal spinlock protects l_resource. We should hold this lock
733 * first before taking res_lock.
737 * Pointer to actual resource this lock is in.
738 * ldlm_lock_change_resource() can change this.
740 struct ldlm_resource *l_resource;
742 * List item for client side LRU list.
743 * Protected by ns_lock in struct ldlm_namespace.
745 struct list_head l_lru;
747 * Linkage to resource's lock queues according to current lock state.
748 * (could be granted or waiting)
749 * Protected by lr_lock in struct ldlm_resource.
751 struct list_head l_res_link;
753 * Tree node for ldlm_extent.
755 struct ldlm_interval *l_tree_node;
757 * Per export hash of locks.
758 * Protected by per-bucket exp->exp_lock_hash locks.
760 struct hlist_node l_exp_hash;
762 * Per export hash of flock locks.
763 * Protected by per-bucket exp->exp_flock_hash locks.
765 struct hlist_node l_exp_flock_hash;
768 * Protected by lr_lock.
770 enum ldlm_mode l_req_mode;
772 * Granted mode, also protected by lr_lock.
774 enum ldlm_mode l_granted_mode;
775 /** Lock completion handler pointer. Called when lock is granted. */
776 ldlm_completion_callback l_completion_ast;
778 * Lock blocking AST handler pointer.
779 * It plays two roles:
780 * - as a notification of an attempt to queue a conflicting lock (once)
781 * - as a notification when the lock is being cancelled.
783 * As such it's typically called twice: once for the initial conflict
784 * and then once more when the last user went away and the lock is
785 * cancelled (could happen recursively).
787 ldlm_blocking_callback l_blocking_ast;
789 * Lock glimpse handler.
790 * Glimpse handler is used to obtain LVB updates from a client by
793 ldlm_glimpse_callback l_glimpse_ast;
797 * This is a pointer to actual client export for locks that were granted
798 * to clients. Used server-side.
800 struct obd_export *l_export;
802 * Lock connection export.
803 * Pointer to server export on a client.
805 struct obd_export *l_conn_export;
808 * Remote lock handle.
809 * If the lock is remote, this is the handle of the other side lock
812 struct lustre_handle l_remote_handle;
815 * Representation of private data specific for a lock type.
816 * Examples are: extent range for extent lock or bitmask for ibits locks
818 union ldlm_policy_data l_policy_data;
821 * Lock state flags. Protected by lr_lock.
822 * \see lustre_dlm_flags.h where the bits are defined.
827 * Lock r/w usage counters.
828 * Protected by lr_lock.
833 * If the lock is granted, a process sleeps on this waitq to learn when
834 * it's no longer in use. If the lock is not granted, a process sleeps
835 * on this waitq to learn when it becomes granted.
837 wait_queue_head_t l_waitq;
840 * Time, in nanoseconds, last used by e.g. being matched by lock match.
844 /** Originally requested extent for the extent lock. */
845 struct ldlm_extent l_req_extent;
848 * Client-side-only members.
851 enum lvb_type l_lvb_type;
854 * Temporary storage for a LVB received during an enqueue operation.
855 * May be vmalloc'd, so needs to be freed with OBD_FREE_LARGE().
860 /** Private storage for lock user. Opaque to LDLM. */
865 * Seconds. It will be updated if there is any activity related to
866 * the lock at client, e.g. enqueue the lock. For server it is the
867 * time when blocking ast was sent.
870 time64_t l_blast_sent;
873 /* separate ost_lvb used mostly by Data-on-MDT for now.
874 * It is introduced to don't mix with layout lock data. */
875 struct ost_lvb l_ost_lvb;
877 * Server-side-only members.
881 * Connection cookie for the client originating the operation.
882 * Used by Commit on Share (COS) code. Currently only used for
883 * inodebits locks on MDS.
885 __u64 l_client_cookie;
888 * List item for locks waiting for cancellation from clients.
889 * The lists this could be linked into are:
890 * waiting_locks_list (protected by waiting_locks_spinlock),
891 * then if the lock timed out, it is moved to
892 * expired_lock_list for further processing.
894 struct list_head l_pending_chain;
897 * Set when lock is sent a blocking AST. Time in seconds when timeout
898 * is reached and client holding this lock could be evicted.
899 * This timeout could be further extended by e.g. certain IO activity
901 * \see ost_rw_prolong_locks
903 time64_t l_callback_timeout;
905 /** Local PID of process which created this lock. */
909 * Number of times blocking AST was sent for this lock.
910 * This is for debugging. Valid values are 0 and 1, if there is an
911 * attempt to send blocking AST more than once, an assertion would be
912 * hit. \see ldlm_work_bl_ast_lock
915 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
916 struct list_head l_bl_ast;
917 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
918 struct list_head l_cp_ast;
919 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
920 struct list_head l_rk_ast;
923 * Pointer to a conflicting lock that caused blocking AST to be sent
926 struct ldlm_lock *l_blocking_lock;
929 * Protected by lr_lock, linkages to "skip lists".
930 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
932 struct list_head l_sl_mode;
933 struct list_head l_sl_policy;
935 /** Reference tracking structure to debug leaked locks. */
936 struct lu_ref l_reference;
937 #if LUSTRE_TRACKS_LOCK_EXP_REFS
938 /* Debugging stuff for bug 20498, for tracking export references. */
939 /** number of export references taken */
941 /** link all locks referencing one export */
942 struct list_head l_exp_refs_link;
943 /** referenced export object */
944 struct obd_export *l_exp_refs_target;
947 * export blocking dlm lock list, protected by
948 * l_export->exp_bl_list_lock.
949 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
950 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
952 struct list_head l_exp_list;
955 /** For uncommitted cross-MDT lock, store transno this lock belongs to */
956 #define l_transno l_client_cookie
958 /** For uncommitted cross-MDT lock, which is client lock, share with l_rk_ast
959 * which is for server. */
960 #define l_slc_link l_rk_ast
962 #define HANDLE_MAP_SIZE ((LMV_MAX_STRIPE_COUNT + 7) >> 3)
964 struct lustre_handle_array {
965 unsigned int ha_count;
966 /* ha_map is used as bit flag to indicate handle is remote or local */
967 char ha_map[HANDLE_MAP_SIZE];
968 struct lustre_handle ha_handles[0];
972 * LDLM resource description.
973 * Basically, resource is a representation for a single object.
974 * Object has a name which is currently 4 64-bit integers. LDLM user is
975 * responsible for creation of a mapping between objects it wants to be
976 * protected and resource names.
978 * A resource can only hold locks of a single lock type, though there may be
979 * multiple ldlm_locks on a single resource, depending on the lock type and
980 * whether the locks are conflicting or not.
982 struct ldlm_resource {
983 struct ldlm_ns_bucket *lr_ns_bucket;
986 * List item for list in namespace hash.
987 * protected by ns_lock
989 struct hlist_node lr_hash;
991 /** Reference count for this resource */
992 atomic_t lr_refcount;
994 /** Spinlock to protect locks under this resource. */
998 * protected by lr_lock
1000 /** List of locks in granted state */
1001 struct list_head lr_granted;
1003 * List of locks that could not be granted due to conflicts and
1004 * that are waiting for conflicts to go away */
1005 struct list_head lr_waiting;
1008 /** Resource name */
1009 struct ldlm_res_id lr_name;
1012 * Interval trees (only for extent locks) for all modes of this resource
1014 struct ldlm_interval_tree *lr_itree;
1018 * When the resource was considered as contended,
1019 * used only on server side.
1021 time64_t lr_contention_time;
1023 * Associated inode, used only on client side.
1025 struct inode *lr_lvb_inode;
1028 /** Type of locks this resource can hold. Only one type per resource. */
1029 enum ldlm_type lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1032 * Server-side-only lock value block elements.
1033 * To serialize lvbo_init.
1036 struct mutex lr_lvb_mutex;
1037 /** protected by lr_lock */
1039 /** is lvb initialized ? */
1040 bool lr_lvb_initialized;
1042 /** List of references to this resource. For debugging. */
1043 struct lu_ref lr_reference;
1046 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1048 return lock->l_resource->lr_type == LDLM_IBITS &&
1049 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1052 static inline bool ldlm_has_dom(struct ldlm_lock *lock)
1054 return lock->l_resource->lr_type == LDLM_IBITS &&
1055 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_DOM;
1058 static inline char *
1059 ldlm_ns_name(struct ldlm_namespace *ns)
1064 static inline struct ldlm_namespace *
1065 ldlm_res_to_ns(struct ldlm_resource *res)
1067 return res->lr_ns_bucket->nsb_namespace;
1070 static inline struct ldlm_namespace *
1071 ldlm_lock_to_ns(struct ldlm_lock *lock)
1073 return ldlm_res_to_ns(lock->l_resource);
1076 static inline char *
1077 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1079 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1082 static inline struct adaptive_timeout *
1083 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1085 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1088 static inline int ldlm_lvbo_init(const struct lu_env *env,
1089 struct ldlm_resource *res)
1091 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1094 if (ns->ns_lvbo == NULL || ns->ns_lvbo->lvbo_init == NULL ||
1095 res->lr_lvb_initialized)
1098 mutex_lock(&res->lr_lvb_mutex);
1099 /* Did we lose the race? */
1100 if (res->lr_lvb_initialized) {
1101 mutex_unlock(&res->lr_lvb_mutex);
1104 rc = ns->ns_lvbo->lvbo_init(env, res);
1106 CDEBUG(D_DLMTRACE, "lvbo_init failed for resource : rc = %d\n",
1108 if (res->lr_lvb_data != NULL) {
1109 OBD_FREE(res->lr_lvb_data, res->lr_lvb_len);
1110 res->lr_lvb_data = NULL;
1112 res->lr_lvb_len = rc;
1114 res->lr_lvb_initialized = true;
1116 mutex_unlock(&res->lr_lvb_mutex);
1120 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1122 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1124 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1125 return ns->ns_lvbo->lvbo_size(lock);
1130 static inline int ldlm_lvbo_fill(const struct lu_env *env,
1131 struct ldlm_lock *lock, void *buf, int *len)
1133 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1136 if (ns->ns_lvbo != NULL) {
1137 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1138 /* init lvb now if not already */
1139 rc = ldlm_lvbo_init(env, lock->l_resource);
1141 CERROR("lock %p: delayed lvb init failed (rc %d)",
1145 return ns->ns_lvbo->lvbo_fill(env, lock, buf, len);
1150 struct ldlm_ast_work {
1151 struct ldlm_lock *w_lock;
1153 struct ldlm_lock_desc w_desc;
1154 struct list_head w_list;
1161 * Common ldlm_enqueue parameters
1163 struct ldlm_enqueue_info {
1164 enum ldlm_type ei_type; /** Type of the lock being enqueued. */
1165 enum ldlm_mode ei_mode; /** Mode of the lock being enqueued. */
1166 void *ei_cb_bl; /** blocking lock callback */
1167 void *ei_cb_local_bl; /** blocking local lock callback */
1168 void *ei_cb_cp; /** lock completion callback */
1169 void *ei_cb_gl; /** lock glimpse callback */
1170 void *ei_cbdata; /** Data to be passed into callbacks. */
1171 void *ei_namespace; /** lock namespace **/
1172 u64 ei_inodebits; /** lock inode bits **/
1173 unsigned int ei_enq_slave:1; /** whether enqueue slave stripes */
1176 #define ei_res_id ei_cb_gl
1178 extern struct obd_ops ldlm_obd_ops;
1180 extern char *ldlm_lockname[];
1181 extern char *ldlm_typename[];
1182 extern const char *ldlm_it2str(enum ldlm_intent_flags it);
1185 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1186 * For the cases where we do not have actual lock to print along
1187 * with a debugging message that is ldlm-related
1189 #define LDLM_DEBUG_NOLOCK(format, a...) \
1190 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1193 * Support function for lock information printing into debug logs.
1197 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1198 CFS_CHECK_STACK(msgdata, mask, cdls); \
1200 if (((mask) & D_CANTMASK) != 0 || \
1201 ((libcfs_debug & (mask)) != 0 && \
1202 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1203 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1206 void _ldlm_lock_debug(struct ldlm_lock *lock,
1207 struct libcfs_debug_msg_data *data,
1208 const char *fmt, ...)
1209 __attribute__ ((format (printf, 3, 4)));
1212 * Rate-limited version of lock printing function.
1214 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1215 static struct cfs_debug_limit_state _ldlm_cdls; \
1216 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1217 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1220 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1221 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1223 /** Non-rate-limited lock printing function for debugging purposes. */
1224 #define LDLM_DEBUG(lock, fmt, a...) do { \
1225 if (likely(lock != NULL)) { \
1226 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1227 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1228 "### " fmt , ##a); \
1230 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1233 #else /* !LIBCFS_DEBUG */
1234 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1235 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1236 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1240 * Three intentions can be used for the policy functions in
1241 * ldlm_processing_policy.
1243 * LDLM_PROCESS_RESCAN:
1245 * It's used when policy functions are called from ldlm_reprocess_queue() to
1246 * reprocess the wait list and try to grant locks, blocking ASTs
1247 * have already been sent in this situation, completion ASTs need be sent for
1248 * the locks being granted.
1250 * LDLM_PROCESS_ENQUEUE:
1252 * It's used when policy functions are called from ldlm_lock_enqueue() to
1253 * process the wait list for handling an enqueue request, blocking
1254 * ASTs have not been sent yet, so list of conflicting locks would be
1255 * collected and ASTs sent.
1257 * LDLM_PROCESS_RECOVERY:
1259 * It's used when policy functions are called from ldlm_reprocess_queue() to
1260 * reprocess the wait list when recovery done. In case of blocking
1261 * ASTs are lost before recovery, it needs not only to grant locks if
1262 * available, but also send blocking ASTs to the locks doesn't have AST sent
1263 * flag. Completion ASTs need be sent for the locks being granted.
1265 enum ldlm_process_intention {
1266 LDLM_PROCESS_RESCAN = 0,
1267 LDLM_PROCESS_ENQUEUE = 1,
1268 LDLM_PROCESS_RECOVERY = 2,
1271 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1272 enum ldlm_process_intention intention,
1273 enum ldlm_error *err,
1274 struct list_head *work_list);
1277 * Return values for lock iterators.
1278 * Also used during deciding of lock grants and cancellations.
1280 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1281 #define LDLM_ITER_STOP 2 /* stop iterating */
1283 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1284 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1286 /** \defgroup ldlm_iterator Lock iterators
1288 * LDLM provides for a way to iterate through every lock on a resource or
1289 * namespace or every resource in a namespace.
1291 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1293 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1295 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1296 ldlm_iterator_t iter, void *data);
1297 /** @} ldlm_iterator */
1299 int ldlm_replay_locks(struct obd_import *imp);
1302 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1305 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1307 struct ldlm_prolong_args {
1308 struct obd_export *lpa_export;
1309 struct ldlm_res_id lpa_resid;
1310 struct ldlm_extent lpa_extent;
1311 enum ldlm_mode lpa_mode;
1312 time64_t lpa_timeout;
1316 void ldlm_lock_prolong_one(struct ldlm_lock *lock,
1317 struct ldlm_prolong_args *arg);
1318 void ldlm_resource_prolong(struct ldlm_prolong_args *arg);
1320 struct ldlm_callback_suite {
1321 ldlm_completion_callback lcs_completion;
1322 ldlm_blocking_callback lcs_blocking;
1323 ldlm_glimpse_callback lcs_glimpse;
1327 #ifdef HAVE_SERVER_SUPPORT
1328 /** \defgroup ldlm_srv_ast Server AST handlers
1329 * These are AST handlers used by server code.
1330 * Their property is that they are just preparing RPCs to be sent to clients.
1333 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1334 void *data, int flag);
1335 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1336 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1337 int ldlm_glimpse_locks(struct ldlm_resource *res,
1338 struct list_head *gl_work_list);
1339 /** @} ldlm_srv_ast */
1341 /** \defgroup ldlm_handlers Server LDLM handlers
1342 * These are handler functions that should be called by "frontends" such as
1343 * MDT or OST to pass through LDLM requests to LDLM for handling
1346 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1347 const struct ldlm_request *dlm_req,
1348 const struct ldlm_callback_suite *cbs);
1349 int ldlm_handle_convert0(struct ptlrpc_request *req,
1350 const struct ldlm_request *dlm_req);
1351 int ldlm_handle_cancel(struct ptlrpc_request *req);
1352 int ldlm_request_cancel(struct ptlrpc_request *req,
1353 const struct ldlm_request *dlm_req,
1354 int first, enum lustre_at_flags flags);
1355 /** @} ldlm_handlers */
1357 void ldlm_revoke_export_locks(struct obd_export *exp);
1358 time64_t ldlm_bl_timeout(struct ldlm_lock *lock);
1360 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1361 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, time64_t timeout);
1362 int ldlm_get_ref(void);
1363 void ldlm_put_ref(void);
1364 int ldlm_init_export(struct obd_export *exp);
1365 void ldlm_destroy_export(struct obd_export *exp);
1366 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1369 #ifdef HAVE_SERVER_SUPPORT
1370 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1372 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1373 void ldlm_lock2handle(const struct ldlm_lock *lock,
1374 struct lustre_handle *lockh);
1375 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1376 void ldlm_cancel_callback(struct ldlm_lock *);
1377 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1378 int ldlm_lock_set_data(const struct lustre_handle *lockh, void *data);
1381 * Obtain a lock reference by its handle.
1383 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1385 return __ldlm_handle2lock(h, 0);
1388 #define LDLM_LOCK_REF_DEL(lock) \
1389 lu_ref_del(&lock->l_reference, "handle", current)
1391 static inline struct ldlm_lock *
1392 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1394 struct ldlm_lock *lock;
1396 lock = __ldlm_handle2lock(h, flags);
1398 LDLM_LOCK_REF_DEL(lock);
1403 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1404 * data from request \a r
1406 static inline int ldlm_lvbo_update(const struct lu_env *env,
1407 struct ldlm_resource *res,
1408 struct ldlm_lock *lock,
1409 struct ptlrpc_request *req, int increase)
1411 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1414 /* delayed lvb init may be required */
1415 rc = ldlm_lvbo_init(env, res);
1417 CERROR("delayed lvb init failed (rc %d)\n", rc);
1421 if (ns->ns_lvbo && ns->ns_lvbo->lvbo_update)
1422 return ns->ns_lvbo->lvbo_update(env, res, lock, req, increase);
1427 static inline int ldlm_res_lvbo_update(const struct lu_env *env,
1428 struct ldlm_resource *res,
1429 struct ptlrpc_request *req,
1432 return ldlm_lvbo_update(env, res, NULL, req, increase);
1435 int ldlm_error2errno(enum ldlm_error error);
1436 enum ldlm_error ldlm_errno2error(int err_no); /* don't call it `errno': this
1437 * confuses user-space. */
1438 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1439 void ldlm_dump_export_locks(struct obd_export *exp);
1443 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1444 * __ldlm_handle2lock().
1446 #define LDLM_LOCK_PUT(lock) \
1448 LDLM_LOCK_REF_DEL(lock); \
1449 /*LDLM_DEBUG((lock), "put");*/ \
1450 ldlm_lock_put(lock); \
1454 * Release a lock reference obtained by some other means (see
1457 #define LDLM_LOCK_RELEASE(lock) \
1459 /*LDLM_DEBUG((lock), "put");*/ \
1460 ldlm_lock_put(lock); \
1463 #define LDLM_LOCK_GET(lock) \
1465 ldlm_lock_get(lock); \
1466 /*LDLM_DEBUG((lock), "get");*/ \
1470 #define ldlm_lock_list_put(head, member, count) \
1472 struct ldlm_lock *_lock, *_next; \
1474 list_for_each_entry_safe(_lock, _next, head, member) { \
1477 list_del_init(&_lock->member); \
1478 LDLM_LOCK_RELEASE(_lock); \
1483 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1484 void ldlm_lock_put(struct ldlm_lock *lock);
1485 void ldlm_lock_destroy(struct ldlm_lock *lock);
1486 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1487 void ldlm_lock_addref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1488 int ldlm_lock_addref_try(const struct lustre_handle *lockh,
1489 enum ldlm_mode mode);
1490 void ldlm_lock_decref(const struct lustre_handle *lockh, enum ldlm_mode mode);
1491 void ldlm_lock_decref_and_cancel(const struct lustre_handle *lockh,
1492 enum ldlm_mode mode);
1493 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1494 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1495 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1496 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1497 enum ldlm_mode ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1498 const struct ldlm_res_id *, enum ldlm_type type,
1499 union ldlm_policy_data *, enum ldlm_mode mode,
1500 struct lustre_handle *, int unref);
1501 enum ldlm_mode ldlm_revalidate_lock_handle(const struct lustre_handle *lockh,
1503 void ldlm_lock_mode_downgrade(struct ldlm_lock *lock, enum ldlm_mode new_mode);
1504 void ldlm_lock_cancel(struct ldlm_lock *lock);
1505 void ldlm_reprocess_all(struct ldlm_resource *res);
1506 void ldlm_reprocess_recovery_done(struct ldlm_namespace *ns);
1507 void ldlm_lock_dump_handle(int level, const struct lustre_handle *lockh);
1508 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1511 struct ldlm_namespace *ldlm_namespace_new(struct obd_device *obd, char *name,
1512 enum ldlm_side client,
1513 enum ldlm_appetite apt,
1514 enum ldlm_ns_type ns_type);
1515 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1516 void ldlm_namespace_free_prior(struct ldlm_namespace *ns,
1517 struct obd_import *imp,
1519 void ldlm_namespace_free_post(struct ldlm_namespace *ns);
1520 void ldlm_namespace_free(struct ldlm_namespace *ns,
1521 struct obd_import *imp, int force);
1522 void ldlm_namespace_register(struct ldlm_namespace *ns, enum ldlm_side client);
1523 void ldlm_namespace_unregister(struct ldlm_namespace *ns,
1524 enum ldlm_side client);
1525 void ldlm_namespace_get(struct ldlm_namespace *ns);
1526 void ldlm_namespace_put(struct ldlm_namespace *ns);
1528 int ldlm_debugfs_setup(void);
1529 void ldlm_debugfs_cleanup(void);
1531 static inline void ldlm_svc_get_eopc(const struct ldlm_request *dlm_req,
1532 struct lprocfs_stats *srv_stats)
1534 int lock_type = 0, op = 0;
1536 lock_type = dlm_req->lock_desc.l_resource.lr_type;
1538 switch (lock_type) {
1540 op = PTLRPC_LAST_CNTR + LDLM_PLAIN_ENQUEUE;
1543 op = PTLRPC_LAST_CNTR + LDLM_EXTENT_ENQUEUE;
1546 op = PTLRPC_LAST_CNTR + LDLM_FLOCK_ENQUEUE;
1549 op = PTLRPC_LAST_CNTR + LDLM_IBITS_ENQUEUE;
1557 lprocfs_counter_incr(srv_stats, op);
1562 /* resource.c - internal */
1563 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1564 struct ldlm_resource *parent,
1565 const struct ldlm_res_id *,
1566 enum ldlm_type type, int create);
1567 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1568 int ldlm_resource_putref(struct ldlm_resource *res);
1569 void ldlm_resource_add_lock(struct ldlm_resource *res,
1570 struct list_head *head,
1571 struct ldlm_lock *lock);
1572 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1573 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1574 void ldlm_dump_all_namespaces(enum ldlm_side client, int level);
1575 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1576 void ldlm_resource_dump(int level, struct ldlm_resource *);
1577 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1578 const struct ldlm_res_id *);
1580 #define LDLM_RESOURCE_ADDREF(res) do { \
1581 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1584 #define LDLM_RESOURCE_DELREF(res) do { \
1585 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1588 /* ldlm_request.c */
1589 int ldlm_expired_completion_wait(void *data);
1590 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1591 * These AST handlers are typically used for server-side local locks and are
1592 * also used by client-side lock handlers to perform minimum level base
1595 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1596 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1597 void *data, int flag);
1598 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1599 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1600 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1601 /** @} ldlm_local_ast */
1603 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1604 * These are typically used by client and server (*_local versions)
1605 * to obtain and release locks.
1607 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1608 struct ldlm_enqueue_info *einfo,
1609 const struct ldlm_res_id *res_id,
1610 union ldlm_policy_data const *policy, __u64 *flags,
1611 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1612 struct lustre_handle *lockh, int async);
1613 int ldlm_prep_enqueue_req(struct obd_export *exp,
1614 struct ptlrpc_request *req,
1615 struct list_head *cancels,
1617 int ldlm_prep_elc_req(struct obd_export *exp, struct ptlrpc_request *req,
1618 int version, int opc, int canceloff,
1619 struct list_head *cancels, int count);
1621 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1622 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1623 const struct ldlm_request *dlm_req,
1624 const struct ldlm_callback_suite *cbs);
1625 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1626 enum ldlm_type type, __u8 with_policy,
1627 enum ldlm_mode mode, __u64 *flags, void *lvb,
1629 const struct lustre_handle *lockh, int rc);
1630 int ldlm_cli_enqueue_local(const struct lu_env *env,
1631 struct ldlm_namespace *ns,
1632 const struct ldlm_res_id *res_id,
1633 enum ldlm_type type, union ldlm_policy_data *policy,
1634 enum ldlm_mode mode, __u64 *flags,
1635 ldlm_blocking_callback blocking,
1636 ldlm_completion_callback completion,
1637 ldlm_glimpse_callback glimpse,
1638 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1639 const __u64 *client_cookie,
1640 struct lustre_handle *lockh);
1641 int ldlm_cli_convert(struct ldlm_lock *lock, __u32 *flags);
1642 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1643 int ldlm_cli_cancel(const struct lustre_handle *lockh,
1644 enum ldlm_cancel_flags cancel_flags);
1645 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1646 enum ldlm_cancel_flags flags, void *opaque);
1647 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1648 const struct ldlm_res_id *res_id,
1649 union ldlm_policy_data *policy,
1650 enum ldlm_mode mode,
1651 enum ldlm_cancel_flags flags, void *opaque);
1652 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1653 int count, enum ldlm_cancel_flags flags);
1654 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1655 struct list_head *cancels,
1656 union ldlm_policy_data *policy,
1657 enum ldlm_mode mode, __u64 lock_flags,
1658 enum ldlm_cancel_flags cancel_flags,
1660 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1661 enum ldlm_cancel_flags flags);
1662 int ldlm_cli_cancel_list(struct list_head *head, int count,
1663 struct ptlrpc_request *req,
1664 enum ldlm_cancel_flags flags);
1666 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);
1667 int ldlm_cli_dropbits(struct ldlm_lock *lock, __u64 drop_bits);
1668 int ldlm_cli_dropbits_list(struct list_head *converts, __u64 drop_bits);
1670 /** @} ldlm_cli_api */
1673 /* This has to be here because recursive inclusion sucks. */
1674 int intent_disposition(struct ldlm_reply *rep, int flag);
1675 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1678 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1679 * than one lock_res is dead-lock safe.
1681 enum lock_res_type {
1686 /** Lock resource. */
1687 static inline void lock_res(struct ldlm_resource *res)
1689 spin_lock(&res->lr_lock);
1692 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1693 static inline void lock_res_nested(struct ldlm_resource *res,
1694 enum lock_res_type mode)
1696 spin_lock_nested(&res->lr_lock, mode);
1699 /** Unlock resource. */
1700 static inline void unlock_res(struct ldlm_resource *res)
1702 spin_unlock(&res->lr_lock);
1705 /** Check if resource is already locked, assert if not. */
1706 static inline void check_res_locked(struct ldlm_resource *res)
1708 assert_spin_locked(&res->lr_lock);
1711 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1712 void unlock_res_and_lock(struct ldlm_lock *lock);
1715 /** \defgroup ldlm_pools Various LDLM pool related functions
1716 * There are not used outside of ldlm.
1719 int ldlm_pools_init(void);
1720 void ldlm_pools_fini(void);
1722 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1723 int idx, enum ldlm_side client);
1724 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr, gfp_t gfp_mask);
1725 void ldlm_pool_fini(struct ldlm_pool *pl);
1726 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1727 time64_t ldlm_pool_recalc(struct ldlm_pool *pl);
1728 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1729 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1730 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1731 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1732 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1733 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1734 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1735 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1736 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1739 static inline int ldlm_extent_overlap(const struct ldlm_extent *ex1,
1740 const struct ldlm_extent *ex2)
1742 return ex1->start <= ex2->end && ex2->start <= ex1->end;
1745 /* check if @ex1 contains @ex2 */
1746 static inline int ldlm_extent_contain(const struct ldlm_extent *ex1,
1747 const struct ldlm_extent *ex2)
1749 return ex1->start <= ex2->start && ex1->end >= ex2->end;
1752 int ldlm_inodebits_drop(struct ldlm_lock *lock, __u64 to_drop);