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.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2010, 2012, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 /** \defgroup LDLM Lustre Distributed Lock Manager
39 * Lustre DLM is based on VAX DLM.
40 * Its two main roles are:
41 * - To provide locking assuring consistency of data on all Lustre nodes.
42 * - To allow clients to cache state protected by a lock by holding the
43 * lock until a conflicting lock is requested or it is expired by the LRU.
48 #ifndef _LUSTRE_DLM_H__
49 #define _LUSTRE_DLM_H__
51 #if defined(__linux__)
52 #include <linux/lustre_dlm.h>
53 #elif defined(__APPLE__)
54 #include <darwin/lustre_dlm.h>
55 #elif defined(__WINNT__)
56 #include <winnt/lustre_dlm.h>
58 #error Unsupported operating system.
61 #include <lustre_lib.h>
62 #include <lustre_net.h>
63 #include <lustre_import.h>
64 #include <lustre_handles.h>
65 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
71 #define OBD_LDLM_DEVICENAME "ldlm"
73 #define LDLM_DEFAULT_LRU_SIZE (100 * cfs_num_online_cpus())
74 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
75 #define LDLM_CTIME_AGE_LIMIT (10)
76 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
79 * LDLM non-error return states
84 ELDLM_LOCK_CHANGED = 300,
85 ELDLM_LOCK_ABORTED = 301,
86 ELDLM_LOCK_REPLACED = 302,
87 ELDLM_NO_LOCK_DATA = 303,
88 ELDLM_LOCK_WOULDBLOCK = 304,
90 ELDLM_NAMESPACE_EXISTS = 400,
91 ELDLM_BAD_NAMESPACE = 401
95 * LDLM namespace type.
96 * The "client" type is actually an indication that this is a narrow local view
97 * into complete namespace on the server. Such namespaces cannot make any
98 * decisions about lack of conflicts or do any autonomous lock granting without
99 * first speaking to a server.
102 LDLM_NAMESPACE_SERVER = 1 << 0,
103 LDLM_NAMESPACE_CLIENT = 1 << 1
107 * Declaration of flags sent through the wire.
109 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
112 * If the server returns one of these flags, then the lock was put on that list.
113 * If the client sends one of these flags (during recovery ONLY!), it wants the
114 * lock added to the specified list, no questions asked.
116 #define LDLM_FL_BLOCK_GRANTED 0x000002
117 #define LDLM_FL_BLOCK_CONV 0x000004
118 #define LDLM_FL_BLOCK_WAIT 0x000008
120 /* Used to be LDLM_FL_CBPENDING 0x000010 moved to non-wire flags */
122 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
123 * queued for sending. */
124 /* Used to be LDLM_FL_WAIT_NOREPROC 0x000040 moved to non-wire flags */
125 /* Used to be LDLM_FL_CANCEL 0x000080 moved to non-wire flags */
128 * Lock is being replayed. This could probably be implied by the fact that one
129 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous.
131 #define LDLM_FL_REPLAY 0x000100
133 #define LDLM_FL_INTENT_ONLY 0x000200 /* Don't grant lock, just do intent. */
135 /* Used to be LDLM_FL_LOCAL_ONLY 0x000400 moved to non-wire flags */
136 /* Used to be LDLM_FL_FAILED 0x000800 moved to non-wire flags */
138 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
140 /* Used to be LDLM_FL_CANCELING 0x002000 moved to non-wire flags */
141 /* Used to be LDLM_FL_LOCAL 0x004000 moved to non-wire flags */
143 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
145 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
148 /** file & record locking */
149 #define LDLM_FL_BLOCK_NOWAIT 0x040000 /* Server told not to wait if blocked.
150 * For AGL, OST will not send glimpse
152 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
154 /* Used to be LDLM_FL_LVB_READY 0x100000 moved to non-wire flags */
155 /* Used to be LDLM_FL_KMS_IGNORE 0x200000 moved to non-wire flags */
156 /* Used to be LDLM_FL_NO_LRU 0x400000 moved to non-wire flags */
158 /* Immediatelly cancel such locks when they block some other locks. Send
159 * cancel notification to original lock holder, but expect no reply. This is
160 * for clients (like liblustre) that cannot be expected to reliably response
161 * to blocking AST. */
162 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
164 /* Flags flags inherited from parent lock when doing intents. */
165 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
167 /* Used to be LDLM_FL_CP_REQD 0x1000000 moved to non-wire flags */
168 /* Used to be LDLM_FL_CLEANED 0x2000000 moved to non-wire flags */
169 /* Used to be LDLM_FL_ATOMIC_CB 0x4000000 moved to non-wire flags */
170 /* Used to be LDLM_FL_BL_AST 0x10000000 moved to non-wire flags */
171 /* Used to be LDLM_FL_BL_DONE 0x20000000 moved to non-wire flags */
173 /* measure lock contention and return -EUSERS if locking contention is high */
174 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
176 /* These are flags that are mapped into the flags and ASTs of blocking locks */
177 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
179 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
180 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
183 * --------------------------------------------------------------------------
184 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
185 * 0x80000000 will not be sent over the wire.
186 * --------------------------------------------------------------------------
190 * Declaration of flags not sent through the wire.
194 * Used for marking lock as a target for -EINTR while cp_ast sleep
195 * emulation + race with upcoming bl_ast.
197 #define LDLM_FL_FAIL_LOC 0x100000000ULL
200 * Used while processing the unused list to know that we have already
201 * handled this lock and decided to skip it.
203 #define LDLM_FL_SKIPPED 0x200000000ULL
204 /* this lock is being destroyed */
205 #define LDLM_FL_CBPENDING 0x400000000ULL
206 /* not a real flag, not saved in lock */
207 #define LDLM_FL_WAIT_NOREPROC 0x800000000ULL
208 /* cancellation callback already run */
209 #define LDLM_FL_CANCEL 0x1000000000ULL
210 #define LDLM_FL_LOCAL_ONLY 0x2000000000ULL
211 /* don't run the cancel callback under ldlm_cli_cancel_unused */
212 #define LDLM_FL_FAILED 0x4000000000ULL
213 /* lock cancel has already been sent */
214 #define LDLM_FL_CANCELING 0x8000000000ULL
215 /* local lock (ie, no srv/cli split) */
216 #define LDLM_FL_LOCAL 0x10000000000ULL
217 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
218 * the LVB filling happens _after_ the lock has been granted, so another thread
219 * can match it before the LVB has been updated. As a dirty hack, we set
220 * LDLM_FL_LVB_READY only after we've done the LVB poop.
221 * this is only needed on LOV/OSC now, where LVB is actually used and callers
222 * must set it in input flags.
224 * The proper fix is to do the granting inside of the completion AST, which can
225 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
226 * pretty high-risk, though, and would need a lot more testing. */
227 #define LDLM_FL_LVB_READY 0x20000000000ULL
228 /* A lock contributes to the known minimum size (KMS) calculation until it has
229 * finished the part of its cancelation that performs write back on its dirty
230 * pages. It can remain on the granted list during this whole time. Threads
231 * racing to update the KMS after performing their writeback need to know to
232 * exclude each other's locks from the calculation as they walk the granted
234 #define LDLM_FL_KMS_IGNORE 0x40000000000ULL
235 /* completion AST to be executed */
236 #define LDLM_FL_CP_REQD 0x80000000000ULL
237 /* cleanup_resource has already handled the lock */
238 #define LDLM_FL_CLEANED 0x100000000000ULL
239 /* optimization hint: LDLM can run blocking callback from current context
240 * w/o involving separate thread. in order to decrease cs rate */
241 #define LDLM_FL_ATOMIC_CB 0x200000000000ULL
243 /* It may happen that a client initiates two operations, e.g. unlink and
244 * mkdir, such that the server sends a blocking AST for conflicting
245 * locks to this client for the first operation, whereas the second
246 * operation has canceled this lock and is waiting for rpc_lock which is
247 * taken by the first operation. LDLM_FL_BL_AST is set by
248 * ldlm_callback_handler() in the lock to prevent the Early Lock Cancel
249 * (ELC) code from cancelling it.
251 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock
252 * cache is dropped to let ldlm_callback_handler() return EINVAL to the
253 * server. It is used when ELC RPC is already prepared and is waiting
254 * for rpc_lock, too late to send a separate CANCEL RPC. */
255 #define LDLM_FL_BL_AST 0x400000000000ULL
256 #define LDLM_FL_BL_DONE 0x800000000000ULL
257 /* Don't put lock into the LRU list, so that it is not canceled due to aging.
258 * Used by MGC locks, they are cancelled only at unmount or by callback. */
259 #define LDLM_FL_NO_LRU 0x1000000000000ULL
262 * The blocking callback is overloaded to perform two functions. These flags
263 * indicate which operation should be performed.
265 #define LDLM_CB_BLOCKING 1
266 #define LDLM_CB_CANCELING 2
269 * \name Lock Compatibility Matrix.
271 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
272 * Lock types are described in their respective implementation files:
273 * ldlm_{extent,flock,inodebits,plain}.c.
275 * There are six lock modes along with a compatibility matrix to indicate if
276 * two locks are compatible.
278 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
280 * - PW: Protective Write (normal write) mode. When a client requests a write
281 * lock from an OST, a lock with PW mode will be issued.
282 * - PR: Protective Read (normal read) mode. When a client requests a read from
283 * an OST, a lock with PR mode will be issued. Also, if the client opens a
284 * file for execution, it is granted a lock with PR mode.
285 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
286 * requests a write lock during a file open operation.
287 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
288 * an inodebit lock with the CR mode on the intermediate path component.
302 #define LCK_COMPAT_EX LCK_NL
303 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
304 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
305 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
306 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
307 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
308 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
309 #define LCK_COMPAT_COS (LCK_COS)
310 /** @} Lock Compatibility Matrix */
312 extern ldlm_mode_t lck_compat_array[];
314 static inline void lockmode_verify(ldlm_mode_t mode)
316 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
319 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
321 return (lck_compat_array[exist_mode] & new_mode);
326 * cluster name spaces
330 #define DLM_OST_NAMESPACE 1
331 #define DLM_MDS_NAMESPACE 2
334 - do we just separate this by security domains and use a prefix for
335 multiple namespaces in the same domain?
340 * Locking rules for LDLM:
345 * waiting_locks_spinlock
360 struct ldlm_resource;
361 struct ldlm_namespace;
364 * Operations on LDLM pools.
365 * LDLM pool is a pool of locks in the namespace without any implicitly
367 * Locks in the pool are organized in LRU.
368 * Local memory pressure or server instructions (e.g. mempressure on server)
369 * can trigger freeing of locks from the pool
371 struct ldlm_pool_ops {
372 /** Recalculate pool \a pl usage */
373 int (*po_recalc)(struct ldlm_pool *pl);
374 /** Cancel at least \a nr locks from pool \a pl */
375 int (*po_shrink)(struct ldlm_pool *pl, int nr,
376 unsigned int gfp_mask);
377 int (*po_setup)(struct ldlm_pool *pl, int limit);
380 /** One second for pools thread check interval. Each pool has own period. */
381 #define LDLM_POOLS_THREAD_PERIOD (1)
383 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
384 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
386 /** Default recalc period for server side pools in sec. */
387 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
389 /** Default recalc period for client side pools in sec. */
390 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
393 * LDLM pool structure to track granted locks.
394 * For purposes of determining when to release locks on e.g. memory pressure.
395 * This feature is commonly referred to as lru_resize.
398 /** Pool proc directory. */
399 cfs_proc_dir_entry_t *pl_proc_dir;
400 /** Pool name, must be long enough to hold compound proc entry name. */
402 /** Lock for protecting SLV/CLV updates. */
404 /** Number of allowed locks in in pool, both, client and server side. */
405 cfs_atomic_t pl_limit;
406 /** Number of granted locks in */
407 cfs_atomic_t pl_granted;
408 /** Grant rate per T. */
409 cfs_atomic_t pl_grant_rate;
410 /** Cancel rate per T. */
411 cfs_atomic_t pl_cancel_rate;
412 /** Server lock volume (SLV). Protected by pl_lock. */
413 __u64 pl_server_lock_volume;
414 /** Current biggest client lock volume. Protected by pl_lock. */
415 __u64 pl_client_lock_volume;
416 /** Lock volume factor. SLV on client is calculated as following:
417 * server_slv * lock_volume_factor. */
418 cfs_atomic_t pl_lock_volume_factor;
419 /** Time when last SLV from server was obtained. */
420 time_t pl_recalc_time;
421 /** Recalculation period for pool. */
422 time_t pl_recalc_period;
423 /** Recalculation and shrink operations. */
424 struct ldlm_pool_ops *pl_ops;
425 /** Number of planned locks for next period. */
427 /** Pool statistics. */
428 struct lprocfs_stats *pl_stats;
431 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
432 void *req_cookie, ldlm_mode_t mode, __u64 flags,
435 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
439 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
440 * be associated with an LDLM lock and transferred from client to server and
443 * Currently LVBs are used by:
444 * - OSC-OST code to maintain current object size/times
445 * - layout lock code to return the layout when the layout lock is granted
447 struct ldlm_valblock_ops {
448 int (*lvbo_init)(struct ldlm_resource *res);
449 int (*lvbo_update)(struct ldlm_resource *res,
450 struct ptlrpc_request *r,
452 int (*lvbo_free)(struct ldlm_resource *res);
453 /* Return size of lvb data appropriate RPC size can be reserved */
454 int (*lvbo_size)(struct ldlm_lock *lock);
455 /* Called to fill in lvb data to RPC buffer @buf */
456 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
460 * LDLM pools related, type of lock pool in the namespace.
461 * Greedy means release cached locks aggressively
464 LDLM_NAMESPACE_GREEDY = 1 << 0,
465 LDLM_NAMESPACE_MODEST = 1 << 1
469 * Default values for the "max_nolock_size", "contention_time" and
470 * "contended_locks" namespace tunables.
472 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
473 #define NS_DEFAULT_CONTENTION_SECONDS 2
474 #define NS_DEFAULT_CONTENDED_LOCKS 32
476 struct ldlm_ns_bucket {
477 /** back pointer to namespace */
478 struct ldlm_namespace *nsb_namespace;
480 * Estimated lock callback time. Used by adaptive timeout code to
481 * avoid spurious client evictions due to unresponsiveness when in
482 * fact the network or overall system load is at fault
484 struct adaptive_timeout nsb_at_estimate;
488 /** LDLM namespace lock stats */
495 LDLM_NS_TYPE_UNKNOWN = 0,
513 * Namespace serves to contain locks related to a particular service.
514 * There are two kinds of namespaces:
515 * - Server namespace has knowledge of all locks and is therefore authoritative
516 * to make decisions like what locks could be granted and what conflicts
517 * exist during new lock enqueue.
518 * - Client namespace only has limited knowledge about locks in the namespace,
519 * only seeing locks held by the client.
521 * Every Lustre service has one server namespace present on the server serving
522 * that service. Every client connected to the service has a client namespace
524 * Every lock obtained by client in that namespace is actually represented by
525 * two in-memory locks. One on the server and one on the client. The locks are
526 * linked by a special cookie by which one node can tell to the other which lock
527 * it actually means during communications. Such locks are called remote locks.
528 * The locks held by server only without any reference to a client are called
531 struct ldlm_namespace {
532 /** Backward link to OBD, required for LDLM pool to store new SLV. */
533 struct obd_device *ns_obd;
535 /** Flag indicating if namespace is on client instead of server */
536 ldlm_side_t ns_client;
538 /** Resource hash table for namespace. */
539 cfs_hash_t *ns_rs_hash;
544 /** big refcount (by bucket) */
545 cfs_atomic_t ns_bref;
548 * Namespace connect flags supported by server (may be changed via
549 * /proc, LRU resize may be disabled/enabled).
551 __u64 ns_connect_flags;
553 /** Client side original connect flags supported by server. */
554 __u64 ns_orig_connect_flags;
557 * Position in global namespace list linking all namespaces on
560 cfs_list_t ns_list_chain;
563 * List of unused locks for this namespace. This list is also called
565 * Unused locks are locks with zero reader/writer reference counts.
566 * This list is only used on clients for lock caching purposes.
567 * When we want to release some locks voluntarily or if server wants
568 * us to release some locks due to e.g. memory pressure, we take locks
569 * to release from the head of this list.
570 * Locks are linked via l_lru field in \see struct ldlm_lock.
572 cfs_list_t ns_unused_list;
573 /** Number of locks in the LRU list above */
577 * Maximum number of locks permitted in the LRU. If 0, means locks
578 * are managed by pools and there is no preset limit, rather it is all
579 * controlled by available memory on this client and on server.
581 unsigned int ns_max_unused;
582 /** Maximum allowed age (last used time) for locks in the LRU */
583 unsigned int ns_max_age;
585 * Server only: number of times we evicted clients due to lack of reply
588 unsigned int ns_timeouts;
590 * Number of seconds since the file change time after which the
591 * MDT will return an UPDATE lock along with a LOOKUP lock.
592 * This allows the client to start caching negative dentries
593 * for a directory and may save an RPC for a later stat.
595 unsigned int ns_ctime_age_limit;
598 * Used to rate-limit ldlm_namespace_dump calls.
599 * \see ldlm_namespace_dump. Increased by 10 seconds every time
602 cfs_time_t ns_next_dump;
604 /** "policy" function that does actual lock conflict determination */
605 ldlm_res_policy ns_policy;
608 * LVB operations for this namespace.
609 * \see struct ldlm_valblock_ops
611 struct ldlm_valblock_ops *ns_lvbo;
614 * Used by filter code to store pointer to OBD of the service.
615 * Should be dropped in favor of \a ns_obd
620 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
621 * a resource is removed.
623 cfs_waitq_t ns_waitq;
624 /** LDLM pool structure for this namespace */
625 struct ldlm_pool ns_pool;
626 /** Definition of how eagerly unused locks will be released from LRU */
627 ldlm_appetite_t ns_appetite;
630 * If more than \a ns_contended_locks are found, the resource is
631 * considered to be contended. Lock enqueues might specify that no
632 * contended locks should be granted
634 unsigned ns_contended_locks;
637 * The resources in this namespace remember contended state during
638 * \a ns_contention_time, in seconds.
640 unsigned ns_contention_time;
643 * Limit size of contended extent locks, in bytes.
644 * If extended lock is requested for more then this many bytes and
645 * caller instructs us not to grant contended locks, we would disregard
648 unsigned ns_max_nolock_size;
650 /** Limit of parallel AST RPC count. */
651 unsigned ns_max_parallel_ast;
653 /** Callback to cancel locks before replaying it during recovery. */
654 ldlm_cancel_for_recovery ns_cancel_for_recovery;
656 /** LDLM lock stats */
657 struct lprocfs_stats *ns_stats;
660 * Flag to indicate namespace is being freed. Used to determine if
661 * recalculation of LDLM pool statistics should be skipped.
663 unsigned ns_stopping:1;
667 * Returns 1 if namespace \a ns is a client namespace.
669 static inline int ns_is_client(struct ldlm_namespace *ns)
672 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
673 LDLM_NAMESPACE_SERVER)));
674 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
675 ns->ns_client == LDLM_NAMESPACE_SERVER);
676 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
680 * Returns 1 if namespace \a ns is a server namespace.
682 static inline int ns_is_server(struct ldlm_namespace *ns)
685 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
686 LDLM_NAMESPACE_SERVER)));
687 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
688 ns->ns_client == LDLM_NAMESPACE_SERVER);
689 return ns->ns_client == LDLM_NAMESPACE_SERVER;
693 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
695 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
698 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
702 * Returns 1 if this namespace supports lru_resize.
704 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
707 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
710 static inline void ns_register_cancel(struct ldlm_namespace *ns,
711 ldlm_cancel_for_recovery arg)
714 ns->ns_cancel_for_recovery = arg;
719 /** Type for blocking callback function of a lock. */
720 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
721 struct ldlm_lock_desc *new, void *data,
723 /** Type for completion callback function of a lock. */
724 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
726 /** Type for glimpse callback function of a lock. */
727 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
728 /** Type for weight callback function of a lock. */
729 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
731 /** Work list for sending GL ASTs to multiple locks. */
732 struct ldlm_glimpse_work {
733 struct ldlm_lock *gl_lock; /* lock to glimpse */
734 cfs_list_t gl_list; /* linkage to other gl work structs */
735 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
736 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
737 * glimpse callback request */
740 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
741 #define LDLM_GL_WORK_NOFREE 0x1
743 /** Interval node data for each LDLM_EXTENT lock. */
744 struct ldlm_interval {
745 struct interval_node li_node; /* node for tree management */
746 cfs_list_t li_group; /* the locks which have the same
747 * policy - group of the policy */
749 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
752 * Interval tree for extent locks.
753 * The interval tree must be accessed under the resource lock.
754 * Interval trees are used for granted extent locks to speed up conflicts
755 * lookup. See ldlm/interval_tree.c for more details.
757 struct ldlm_interval_tree {
760 ldlm_mode_t lit_mode; /* lock mode */
761 struct interval_node *lit_root; /* actual ldlm_interval */
764 /** Whether to track references to exports by LDLM locks. */
765 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
769 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
770 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
771 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
773 } ldlm_cancel_flags_t;
779 __u64 blocking_owner;
780 struct obd_export *blocking_export;
781 /* Protected by the hash lock */
787 struct ldlm_extent l_extent;
788 struct ldlm_flock l_flock;
789 struct ldlm_inodebits l_inodebits;
790 } ldlm_policy_data_t;
792 void ldlm_convert_policy_to_wire(ldlm_type_t type,
793 const ldlm_policy_data_t *lpolicy,
794 ldlm_wire_policy_data_t *wpolicy);
795 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
796 const ldlm_wire_policy_data_t *wpolicy,
797 ldlm_policy_data_t *lpolicy);
807 * LDLM lock structure
809 * Represents a single LDLM lock and its state in memory. Each lock is
810 * associated with a single ldlm_resource, the object which is being
811 * locked. There may be multiple ldlm_locks on a single resource,
812 * depending on the lock type and whether the locks are conflicting or
818 * When remote side wants to tell us about a lock, they address
819 * it by this opaque handle. The handle does not hold a
820 * reference on the ldlm_lock, so it can be safely passed to
821 * other threads or nodes. When the lock needs to be accessed
822 * from the handle, it is looked up again in the lock table, and
823 * may no longer exist.
825 * Must be first in the structure.
827 struct portals_handle l_handle;
829 * Lock reference count.
830 * This is how many users have pointers to actual structure, so that
831 * we do not accidentally free lock structure that is in use.
835 * Internal spinlock protects l_resource. We should hold this lock
836 * first before taking res_lock.
840 * Pointer to actual resource this lock is in.
841 * ldlm_lock_change_resource() can change this.
843 struct ldlm_resource *l_resource;
845 * List item for client side LRU list.
846 * Protected by ns_lock in struct ldlm_namespace.
850 * Linkage to resource's lock queues according to current lock state.
851 * (could be granted, waiting or converting)
852 * Protected by lr_lock in struct ldlm_resource.
854 cfs_list_t l_res_link;
856 * Tree node for ldlm_extent.
858 struct ldlm_interval *l_tree_node;
860 * Per export hash of locks.
861 * Protected by per-bucket exp->exp_lock_hash locks.
863 cfs_hlist_node_t l_exp_hash;
865 * Per export hash of flock locks.
866 * Protected by per-bucket exp->exp_flock_hash locks.
868 cfs_hlist_node_t l_exp_flock_hash;
871 * Protected by lr_lock.
873 ldlm_mode_t l_req_mode;
875 * Granted mode, also protected by lr_lock.
877 ldlm_mode_t l_granted_mode;
878 /** Lock completion handler pointer. Called when lock is granted. */
879 ldlm_completion_callback l_completion_ast;
881 * Lock blocking AST handler pointer.
882 * It plays two roles:
883 * - as a notification of an attempt to queue a conflicting lock (once)
884 * - as a notification when the lock is being cancelled.
886 * As such it's typically called twice: once for the initial conflict
887 * and then once more when the last user went away and the lock is
888 * cancelled (could happen recursively).
890 ldlm_blocking_callback l_blocking_ast;
892 * Lock glimpse handler.
893 * Glimpse handler is used to obtain LVB updates from a client by
896 ldlm_glimpse_callback l_glimpse_ast;
898 /** XXX apparently unused "weight" handler. To be removed? */
899 ldlm_weigh_callback l_weigh_ast;
903 * This is a pointer to actual client export for locks that were granted
904 * to clients. Used server-side.
906 struct obd_export *l_export;
908 * Lock connection export.
909 * Pointer to server export on a client.
911 struct obd_export *l_conn_export;
914 * Remote lock handle.
915 * If the lock is remote, this is the handle of the other side lock
918 struct lustre_handle l_remote_handle;
921 * Representation of private data specific for a lock type.
922 * Examples are: extent range for extent lock or bitmask for ibits locks
924 ldlm_policy_data_t l_policy_data;
928 * Like whenever we receive any blocking requests for this lock, etc.
929 * Protected by lr_lock.
933 * Lock r/w usage counters.
934 * Protected by lr_lock.
939 * If the lock is granted, a process sleeps on this waitq to learn when
940 * it's no longer in use. If the lock is not granted, a process sleeps
941 * on this waitq to learn when it becomes granted.
946 * Seconds. It will be updated if there is any activity related to
947 * the lock, e.g. enqueue the lock or send blocking AST.
949 cfs_time_t l_last_activity;
952 * Time last used by e.g. being matched by lock match.
953 * Jiffies. Should be converted to time if needed.
955 cfs_time_t l_last_used;
957 /** Originally requested extent for the extent lock. */
958 struct ldlm_extent l_req_extent;
960 unsigned int l_failed:1,
962 * Set for locks that were removed from class hash table and will be
963 * destroyed when last reference to them is released. Set by
964 * ldlm_lock_destroy_internal().
966 * Protected by lock and resource locks.
970 * it's set in lock_res_and_lock() and unset in unlock_res_and_lock().
972 * NB: compared with check_res_locked(), checking this bit is cheaper.
973 * Also, spin_is_locked() is deprecated for kernel code; one reason is
974 * because it works only for SMP so user needs to add extra macros like
975 * LASSERT_SPIN_LOCKED for uniprocessor kernels.
979 * It's set once we call ldlm_add_waiting_lock_res_locked()
980 * to start the lock-timeout timer and it will never be reset.
982 * Protected by lock_res_and_lock().
985 /** Flag whether this is a server namespace lock. */
989 * Client-side-only members.
992 enum lvb_type l_lvb_type;
995 * Temporary storage for a LVB received during an enqueue operation.
1000 /** Private storage for lock user. Opaque to LDLM. */
1004 * Server-side-only members.
1008 * Connection cookie for the client originating the operation.
1009 * Used by Commit on Share (COS) code. Currently only used for
1010 * inodebits locks on MDS.
1012 __u64 l_client_cookie;
1015 * List item for locks waiting for cancellation from clients.
1016 * The lists this could be linked into are:
1017 * waiting_locks_list (protected by waiting_locks_spinlock),
1018 * then if the lock timed out, it is moved to
1019 * expired_lock_thread.elt_expired_locks for further processing.
1020 * Protected by elt_lock.
1022 cfs_list_t l_pending_chain;
1025 * Set when lock is sent a blocking AST. Time in seconds when timeout
1026 * is reached and client holding this lock could be evicted.
1027 * This timeout could be further extended by e.g. certain IO activity
1029 * \see ost_rw_prolong_locks
1031 cfs_time_t l_callback_timeout;
1033 /** Local PID of process which created this lock. */
1037 * Number of times blocking AST was sent for this lock.
1038 * This is for debugging. Valid values are 0 and 1, if there is an
1039 * attempt to send blocking AST more than once, an assertion would be
1040 * hit. \see ldlm_work_bl_ast_lock
1043 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
1044 cfs_list_t l_bl_ast;
1045 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
1046 cfs_list_t l_cp_ast;
1047 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
1048 cfs_list_t l_rk_ast;
1051 * Pointer to a conflicting lock that caused blocking AST to be sent
1054 struct ldlm_lock *l_blocking_lock;
1057 * Protected by lr_lock, linkages to "skip lists".
1058 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
1060 cfs_list_t l_sl_mode;
1061 cfs_list_t l_sl_policy;
1063 /** Reference tracking structure to debug leaked locks. */
1064 struct lu_ref l_reference;
1065 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1066 /* Debugging stuff for bug 20498, for tracking export references. */
1067 /** number of export references taken */
1069 /** link all locks referencing one export */
1070 cfs_list_t l_exp_refs_link;
1071 /** referenced export object */
1072 struct obd_export *l_exp_refs_target;
1075 * export blocking dlm lock list, protected by
1076 * l_export->exp_bl_list_lock.
1077 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
1078 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
1080 cfs_list_t l_exp_list;
1084 * LDLM resource description.
1085 * Basically, resource is a representation for a single object.
1086 * Object has a name which is currently 4 64-bit integers. LDLM user is
1087 * responsible for creation of a mapping between objects it wants to be
1088 * protected and resource names.
1090 * A resource can only hold locks of a single lock type, though there may be
1091 * multiple ldlm_locks on a single resource, depending on the lock type and
1092 * whether the locks are conflicting or not.
1094 struct ldlm_resource {
1095 struct ldlm_ns_bucket *lr_ns_bucket;
1098 * List item for list in namespace hash.
1099 * protected by ns_lock
1101 cfs_hlist_node_t lr_hash;
1103 /** Spinlock to protect locks under this resource. */
1107 * protected by lr_lock
1109 /** List of locks in granted state */
1110 cfs_list_t lr_granted;
1111 /** List of locks waiting to change their granted mode (converted) */
1112 cfs_list_t lr_converting;
1114 * List of locks that could not be granted due to conflicts and
1115 * that are waiting for conflicts to go away */
1116 cfs_list_t lr_waiting;
1119 /* XXX No longer needed? Remove ASAP */
1120 ldlm_mode_t lr_most_restr;
1122 /** Type of locks this resource can hold. Only one type per resource. */
1123 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1125 /** Resource name */
1126 struct ldlm_res_id lr_name;
1127 /** Reference count for this resource */
1128 cfs_atomic_t lr_refcount;
1131 * Interval trees (only for extent locks) for all modes of this resource
1133 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM];
1136 * Server-side-only lock value block elements.
1137 * To serialize lvbo_init.
1139 struct mutex lr_lvb_mutex;
1141 /** protected by lr_lock */
1144 /** When the resource was considered as contended. */
1145 cfs_time_t lr_contention_time;
1146 /** List of references to this resource. For debugging. */
1147 struct lu_ref lr_reference;
1149 struct inode *lr_lvb_inode;
1152 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1154 return lock->l_resource->lr_type == LDLM_IBITS &&
1155 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1158 static inline char *
1159 ldlm_ns_name(struct ldlm_namespace *ns)
1161 return ns->ns_rs_hash->hs_name;
1164 static inline struct ldlm_namespace *
1165 ldlm_res_to_ns(struct ldlm_resource *res)
1167 return res->lr_ns_bucket->nsb_namespace;
1170 static inline struct ldlm_namespace *
1171 ldlm_lock_to_ns(struct ldlm_lock *lock)
1173 return ldlm_res_to_ns(lock->l_resource);
1176 static inline char *
1177 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1179 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1182 static inline struct adaptive_timeout *
1183 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1185 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1188 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1190 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1192 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_init != NULL)
1193 return ns->ns_lvbo->lvbo_init(res);
1198 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1200 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1202 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1203 return ns->ns_lvbo->lvbo_size(lock);
1208 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1210 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1212 if (ns->ns_lvbo != NULL) {
1213 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1214 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1219 struct ldlm_ast_work {
1220 struct ldlm_lock *w_lock;
1222 struct ldlm_lock_desc w_desc;
1230 * Common ldlm_enqueue parameters
1232 struct ldlm_enqueue_info {
1233 __u32 ei_type; /** Type of the lock being enqueued. */
1234 __u32 ei_mode; /** Mode of the lock being enqueued. */
1235 void *ei_cb_bl; /** blocking lock callback */
1236 void *ei_cb_cp; /** lock completion callback */
1237 void *ei_cb_gl; /** lock glimpse callback */
1238 void *ei_cb_wg; /** lock weigh callback */
1239 void *ei_cbdata; /** Data to be passed into callbacks. */
1242 extern struct obd_ops ldlm_obd_ops;
1244 extern char *ldlm_lockname[];
1245 extern char *ldlm_typename[];
1246 extern char *ldlm_it2str(int it);
1249 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1250 * For the cases where we do not have actual lock to print along
1251 * with a debugging message that is ldlm-related
1253 #define LDLM_DEBUG_NOLOCK(format, a...) \
1254 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1257 * Support function for lock information printing into debug logs.
1261 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1262 CFS_CHECK_STACK(msgdata, mask, cdls); \
1264 if (((mask) & D_CANTMASK) != 0 || \
1265 ((libcfs_debug & (mask)) != 0 && \
1266 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1267 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1270 void _ldlm_lock_debug(struct ldlm_lock *lock,
1271 struct libcfs_debug_msg_data *data,
1272 const char *fmt, ...)
1273 __attribute__ ((format (printf, 3, 4)));
1276 * Rate-limited version of lock printing function.
1278 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1279 static cfs_debug_limit_state_t _ldlm_cdls; \
1280 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1281 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1284 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1285 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1287 /** Non-rate-limited lock printing function for debugging purposes. */
1288 #define LDLM_DEBUG(lock, fmt, a...) do { \
1289 if (likely(lock != NULL)) { \
1290 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1291 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1292 "### " fmt , ##a); \
1294 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1297 #else /* !LIBCFS_DEBUG */
1298 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1299 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1300 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1303 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1304 int first_enq, ldlm_error_t *err,
1305 cfs_list_t *work_list);
1308 * Return values for lock iterators.
1309 * Also used during deciding of lock grants and cancellations.
1311 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1312 #define LDLM_ITER_STOP 2 /* stop iterating */
1314 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1315 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1317 /** \defgroup ldlm_iterator Lock iterators
1319 * LDLM provides for a way to iterate through every lock on a resource or
1320 * namespace or every resource in a namespace.
1322 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1324 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1326 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1327 ldlm_iterator_t iter, void *data);
1328 /** @} ldlm_iterator */
1330 int ldlm_replay_locks(struct obd_import *imp);
1333 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1336 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1338 struct ldlm_callback_suite {
1339 ldlm_completion_callback lcs_completion;
1340 ldlm_blocking_callback lcs_blocking;
1341 ldlm_glimpse_callback lcs_glimpse;
1342 ldlm_weigh_callback lcs_weigh;
1346 #ifdef HAVE_SERVER_SUPPORT
1347 /** \defgroup ldlm_srv_ast Server AST handlers
1348 * These are AST handlers used by server code.
1349 * Their property is that they are just preparing RPCs to be sent to clients.
1352 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1353 void *data, int flag);
1354 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1355 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1356 int ldlm_glimpse_locks(struct ldlm_resource *res, cfs_list_t *gl_work_list);
1357 /** @} ldlm_srv_ast */
1359 /** \defgroup ldlm_handlers Server LDLM handlers
1360 * These are handler functions that should be called by "frontends" such as
1361 * MDT or OST to pass through LDLM requests to LDLM for handling
1364 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
1365 ldlm_blocking_callback, ldlm_glimpse_callback);
1366 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1367 const struct ldlm_request *dlm_req,
1368 const struct ldlm_callback_suite *cbs);
1369 int ldlm_handle_convert(struct ptlrpc_request *req);
1370 int ldlm_handle_convert0(struct ptlrpc_request *req,
1371 const struct ldlm_request *dlm_req);
1372 int ldlm_handle_cancel(struct ptlrpc_request *req);
1373 int ldlm_request_cancel(struct ptlrpc_request *req,
1374 const struct ldlm_request *dlm_req, int first);
1375 /** @} ldlm_handlers */
1377 void ldlm_revoke_export_locks(struct obd_export *exp);
1379 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1380 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
1381 int ldlm_get_ref(void);
1382 void ldlm_put_ref(void);
1383 int ldlm_init_export(struct obd_export *exp);
1384 void ldlm_destroy_export(struct obd_export *exp);
1385 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1388 #ifdef HAVE_SERVER_SUPPORT
1389 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1391 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1392 void ldlm_lock2handle(const struct ldlm_lock *lock,
1393 struct lustre_handle *lockh);
1394 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1395 void ldlm_cancel_callback(struct ldlm_lock *);
1396 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1397 int ldlm_lock_set_data(struct lustre_handle *, void *);
1400 * Obtain a lock reference by its handle.
1402 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1404 return __ldlm_handle2lock(h, 0);
1407 #define LDLM_LOCK_REF_DEL(lock) \
1408 lu_ref_del(&lock->l_reference, "handle", cfs_current())
1410 static inline struct ldlm_lock *
1411 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1413 struct ldlm_lock *lock;
1415 lock = __ldlm_handle2lock(h, flags);
1417 LDLM_LOCK_REF_DEL(lock);
1422 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1423 * data from reqest \a r
1425 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1426 struct ptlrpc_request *r, int increase)
1428 if (ldlm_res_to_ns(res)->ns_lvbo &&
1429 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1430 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1436 int ldlm_error2errno(ldlm_error_t error);
1437 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
1438 * confuses user-space. */
1439 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1440 void ldlm_dump_export_locks(struct obd_export *exp);
1444 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1445 * __ldlm_handle2lock().
1447 #define LDLM_LOCK_PUT(lock) \
1449 LDLM_LOCK_REF_DEL(lock); \
1450 /*LDLM_DEBUG((lock), "put");*/ \
1451 ldlm_lock_put(lock); \
1455 * Release a lock reference obtained by some other means (see
1458 #define LDLM_LOCK_RELEASE(lock) \
1460 /*LDLM_DEBUG((lock), "put");*/ \
1461 ldlm_lock_put(lock); \
1464 #define LDLM_LOCK_GET(lock) \
1466 ldlm_lock_get(lock); \
1467 /*LDLM_DEBUG((lock), "get");*/ \
1471 #define ldlm_lock_list_put(head, member, count) \
1473 struct ldlm_lock *_lock, *_next; \
1475 cfs_list_for_each_entry_safe(_lock, _next, head, member) { \
1478 cfs_list_del_init(&_lock->member); \
1479 LDLM_LOCK_RELEASE(_lock); \
1484 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1485 void ldlm_lock_put(struct ldlm_lock *lock);
1486 void ldlm_lock_destroy(struct ldlm_lock *lock);
1487 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1488 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1489 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1490 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1491 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1492 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1493 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1494 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1495 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1496 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1497 const struct ldlm_res_id *, ldlm_type_t type,
1498 ldlm_policy_data_t *, ldlm_mode_t mode,
1499 struct lustre_handle *, int unref);
1500 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1502 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1504 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
1505 void ldlm_lock_cancel(struct ldlm_lock *lock);
1506 void ldlm_reprocess_all(struct ldlm_resource *res);
1507 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1508 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1509 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1512 struct ldlm_namespace *
1513 ldlm_namespace_new(struct obd_device *obd, char *name,
1514 ldlm_side_t client, ldlm_appetite_t apt,
1515 ldlm_ns_type_t ns_type);
1516 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1517 void ldlm_namespace_free(struct ldlm_namespace *ns,
1518 struct obd_import *imp, int force);
1519 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1520 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1521 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1522 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1523 void ldlm_namespace_get(struct ldlm_namespace *ns);
1524 void ldlm_namespace_put(struct ldlm_namespace *ns);
1525 int ldlm_proc_setup(void);
1527 void ldlm_proc_cleanup(void);
1529 static inline void ldlm_proc_cleanup(void) {}
1532 /* resource.c - internal */
1533 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1534 struct ldlm_resource *parent,
1535 const struct ldlm_res_id *,
1536 ldlm_type_t type, int create);
1537 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1538 int ldlm_resource_putref(struct ldlm_resource *res);
1539 void ldlm_resource_add_lock(struct ldlm_resource *res,
1541 struct ldlm_lock *lock);
1542 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1543 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1544 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1545 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1546 void ldlm_resource_dump(int level, struct ldlm_resource *);
1547 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1548 const struct ldlm_res_id *);
1550 #define LDLM_RESOURCE_ADDREF(res) do { \
1551 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1554 #define LDLM_RESOURCE_DELREF(res) do { \
1555 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1558 /* ldlm_request.c */
1559 int ldlm_expired_completion_wait(void *data);
1560 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1561 * These AST handlers are typically used for server-side local locks and are
1562 * also used by client-side lock handlers to perform minimum level base
1565 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1566 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1567 void *data, int flag);
1568 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1569 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1570 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1571 /** @} ldlm_local_ast */
1573 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1574 * These are typically used by client and server (*_local versions)
1575 * to obtain and release locks.
1577 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1578 struct ldlm_enqueue_info *einfo,
1579 const struct ldlm_res_id *res_id,
1580 ldlm_policy_data_t const *policy, __u64 *flags,
1581 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1582 struct lustre_handle *lockh, int async);
1583 int ldlm_prep_enqueue_req(struct obd_export *exp,
1584 struct ptlrpc_request *req,
1585 cfs_list_t *cancels,
1587 int ldlm_prep_elc_req(struct obd_export *exp,
1588 struct ptlrpc_request *req,
1589 int version, int opc, int canceloff,
1590 cfs_list_t *cancels, int count);
1592 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1593 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1594 const struct ldlm_request *dlm_req,
1595 const struct ldlm_callback_suite *cbs);
1596 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1597 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1598 __u64 *flags, void *lvb, __u32 lvb_len,
1599 struct lustre_handle *lockh, int rc);
1600 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1601 const struct ldlm_res_id *res_id,
1602 ldlm_type_t type, ldlm_policy_data_t *policy,
1603 ldlm_mode_t mode, __u64 *flags,
1604 ldlm_blocking_callback blocking,
1605 ldlm_completion_callback completion,
1606 ldlm_glimpse_callback glimpse,
1607 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1608 const __u64 *client_cookie,
1609 struct lustre_handle *lockh);
1610 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1611 void *data, __u32 data_len);
1612 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1613 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1614 int ldlm_cli_cancel(struct lustre_handle *lockh,
1615 ldlm_cancel_flags_t cancel_flags);
1616 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1617 ldlm_cancel_flags_t flags, void *opaque);
1618 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1619 const struct ldlm_res_id *res_id,
1620 ldlm_policy_data_t *policy,
1622 ldlm_cancel_flags_t flags,
1624 int ldlm_cli_cancel_req(struct obd_export *exp, cfs_list_t *head,
1625 int count, ldlm_cancel_flags_t flags);
1626 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1627 cfs_list_t *cancels,
1628 ldlm_policy_data_t *policy,
1629 ldlm_mode_t mode, int lock_flags,
1630 ldlm_cancel_flags_t cancel_flags, void *opaque);
1631 int ldlm_cli_cancel_list_local(cfs_list_t *cancels, int count,
1632 ldlm_cancel_flags_t flags);
1633 int ldlm_cli_cancel_list(cfs_list_t *head, int count,
1634 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1635 /** @} ldlm_cli_api */
1638 /* This has to be here because recursive inclusion sucks. */
1639 int intent_disposition(struct ldlm_reply *rep, int flag);
1640 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1643 /* ioctls for trying requests */
1644 #define IOC_LDLM_TYPE 'f'
1645 #define IOC_LDLM_MIN_NR 40
1647 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1648 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1649 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1650 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1651 #define IOC_LDLM_MAX_NR 43
1654 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1655 * than one lock_res is dead-lock safe.
1657 enum lock_res_type {
1662 /** Lock resource. */
1663 static inline void lock_res(struct ldlm_resource *res)
1665 spin_lock(&res->lr_lock);
1668 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1669 static inline void lock_res_nested(struct ldlm_resource *res,
1670 enum lock_res_type mode)
1672 spin_lock_nested(&res->lr_lock, mode);
1675 /** Unlock resource. */
1676 static inline void unlock_res(struct ldlm_resource *res)
1678 spin_unlock(&res->lr_lock);
1681 /** Check if resource is already locked, assert if not. */
1682 static inline void check_res_locked(struct ldlm_resource *res)
1684 LASSERT_SPIN_LOCKED(&res->lr_lock);
1687 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1688 void unlock_res_and_lock(struct ldlm_lock *lock);
1691 /** \defgroup ldlm_pools Various LDLM pool related functions
1692 * There are not used outside of ldlm.
1695 void ldlm_pools_recalc(ldlm_side_t client);
1696 int ldlm_pools_init(void);
1697 void ldlm_pools_fini(void);
1699 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1700 int idx, ldlm_side_t client);
1701 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1702 unsigned int gfp_mask);
1703 void ldlm_pool_fini(struct ldlm_pool *pl);
1704 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1705 int ldlm_pool_recalc(struct ldlm_pool *pl);
1706 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1707 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1708 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1709 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1710 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1711 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1712 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1713 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1714 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);