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, Whamcloud, Inc.
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 #ifdef HAVE_BGL_SUPPORT
74 /* 1.5 times the maximum 128 tasks available in VN mode */
75 #define LDLM_DEFAULT_LRU_SIZE 196
77 #define LDLM_DEFAULT_LRU_SIZE (100 * cfs_num_online_cpus())
79 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
80 #define LDLM_CTIME_AGE_LIMIT (10)
81 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
84 * LDLM non-error return states
89 ELDLM_LOCK_CHANGED = 300,
90 ELDLM_LOCK_ABORTED = 301,
91 ELDLM_LOCK_REPLACED = 302,
92 ELDLM_NO_LOCK_DATA = 303,
94 ELDLM_NAMESPACE_EXISTS = 400,
95 ELDLM_BAD_NAMESPACE = 401
99 * LDLM namespace type.
100 * The "client" type is actually an indication that this is a narrow local view
101 * into complete namespace on the server. Such namespaces cannot make any
102 * decisions about lack of conflicts or do any autonomous lock granting without
103 * first speaking to a server.
106 LDLM_NAMESPACE_SERVER = 1 << 0,
107 LDLM_NAMESPACE_CLIENT = 1 << 1
111 * Declaration of flags sent through the wire.
113 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
116 * If the server returns one of these flags, then the lock was put on that list.
117 * If the client sends one of these flags (during recovery ONLY!), it wants the
118 * lock added to the specified list, no questions asked.
120 #define LDLM_FL_BLOCK_GRANTED 0x000002
121 #define LDLM_FL_BLOCK_CONV 0x000004
122 #define LDLM_FL_BLOCK_WAIT 0x000008
124 /* Used to be LDLM_FL_CBPENDING 0x000010 moved to non-wire flags */
126 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
127 * queued for sending. */
128 /* Used to be LDLM_FL_WAIT_NOREPROC 0x000040 moved to non-wire flags */
129 /* Used to be LDLM_FL_CANCEL 0x000080 moved to non-wire flags */
132 * Lock is being replayed. This could probably be implied by the fact that one
133 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous.
135 #define LDLM_FL_REPLAY 0x000100
137 #define LDLM_FL_INTENT_ONLY 0x000200 /* Don't grant lock, just do intent. */
139 /* Used to be LDLM_FL_LOCAL_ONLY 0x000400 moved to non-wire flags */
140 /* Used to be LDLM_FL_FAILED 0x000800 moved to non-wire flags */
142 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
144 /* Used to be LDLM_FL_CANCELING 0x002000 moved to non-wire flags */
145 /* Used to be LDLM_FL_LOCAL 0x004000 moved to non-wire flags */
147 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
149 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
152 /** file & record locking */
153 #define LDLM_FL_BLOCK_NOWAIT 0x040000 /* Server told not to wait if blocked.
154 * For AGL, OST will not send glimpse
156 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
158 /* Used to be LDLM_FL_LVB_READY 0x100000 moved to non-wire flags */
159 /* Used to be LDLM_FL_KMS_IGNORE 0x200000 moved to non-wire flags */
160 /* Used to be LDLM_FL_NO_LRU 0x400000 moved to non-wire flags */
162 /* Immediatelly cancel such locks when they block some other locks. Send
163 * cancel notification to original lock holder, but expect no reply. This is
164 * for clients (like liblustre) that cannot be expected to reliably response
165 * to blocking AST. */
166 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
168 /* Flags flags inherited from parent lock when doing intents. */
169 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
171 /* Used to be LDLM_FL_CP_REQD 0x1000000 moved to non-wire flags */
172 /* Used to be LDLM_FL_CLEANED 0x2000000 moved to non-wire flags */
173 /* Used to be LDLM_FL_ATOMIC_CB 0x4000000 moved to non-wire flags */
174 /* Used to be LDLM_FL_BL_AST 0x10000000 moved to non-wire flags */
175 /* Used to be LDLM_FL_BL_DONE 0x20000000 moved to non-wire flags */
177 /* measure lock contention and return -EUSERS if locking contention is high */
178 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
180 /* These are flags that are mapped into the flags and ASTs of blocking locks */
181 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
183 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
184 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
187 * --------------------------------------------------------------------------
188 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
189 * 0x80000000 will not be sent over the wire.
190 * --------------------------------------------------------------------------
194 * Declaration of flags not sent through the wire.
198 * Used for marking lock as a target for -EINTR while cp_ast sleep
199 * emulation + race with upcoming bl_ast.
201 #define LDLM_FL_FAIL_LOC 0x100000000ULL
204 * Used while processing the unused list to know that we have already
205 * handled this lock and decided to skip it.
207 #define LDLM_FL_SKIPPED 0x200000000ULL
208 /* this lock is being destroyed */
209 #define LDLM_FL_CBPENDING 0x400000000ULL
210 /* not a real flag, not saved in lock */
211 #define LDLM_FL_WAIT_NOREPROC 0x800000000ULL
212 /* cancellation callback already run */
213 #define LDLM_FL_CANCEL 0x1000000000ULL
214 #define LDLM_FL_LOCAL_ONLY 0x2000000000ULL
215 /* don't run the cancel callback under ldlm_cli_cancel_unused */
216 #define LDLM_FL_FAILED 0x4000000000ULL
217 /* lock cancel has already been sent */
218 #define LDLM_FL_CANCELING 0x8000000000ULL
219 /* local lock (ie, no srv/cli split) */
220 #define LDLM_FL_LOCAL 0x10000000000ULL
221 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
222 * the LVB filling happens _after_ the lock has been granted, so another thread
223 * can match it before the LVB has been updated. As a dirty hack, we set
224 * LDLM_FL_LVB_READY only after we've done the LVB poop.
225 * this is only needed on LOV/OSC now, where LVB is actually used and callers
226 * must set it in input flags.
228 * The proper fix is to do the granting inside of the completion AST, which can
229 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
230 * pretty high-risk, though, and would need a lot more testing. */
231 #define LDLM_FL_LVB_READY 0x20000000000ULL
232 /* A lock contributes to the known minimum size (KMS) calculation until it has
233 * finished the part of its cancelation that performs write back on its dirty
234 * pages. It can remain on the granted list during this whole time. Threads
235 * racing to update the KMS after performing their writeback need to know to
236 * exclude each other's locks from the calculation as they walk the granted
238 #define LDLM_FL_KMS_IGNORE 0x40000000000ULL
239 /* completion AST to be executed */
240 #define LDLM_FL_CP_REQD 0x80000000000ULL
241 /* cleanup_resource has already handled the lock */
242 #define LDLM_FL_CLEANED 0x100000000000ULL
243 /* optimization hint: LDLM can run blocking callback from current context
244 * w/o involving separate thread. in order to decrease cs rate */
245 #define LDLM_FL_ATOMIC_CB 0x200000000000ULL
247 /* It may happen that a client initiates two operations, e.g. unlink and
248 * mkdir, such that the server sends a blocking AST for conflicting
249 * locks to this client for the first operation, whereas the second
250 * operation has canceled this lock and is waiting for rpc_lock which is
251 * taken by the first operation. LDLM_FL_BL_AST is set by
252 * ldlm_callback_handler() in the lock to prevent the Early Lock Cancel
253 * (ELC) code from cancelling it.
255 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock
256 * cache is dropped to let ldlm_callback_handler() return EINVAL to the
257 * server. It is used when ELC RPC is already prepared and is waiting
258 * for rpc_lock, too late to send a separate CANCEL RPC. */
259 #define LDLM_FL_BL_AST 0x400000000000ULL
260 #define LDLM_FL_BL_DONE 0x800000000000ULL
261 /* Don't put lock into the LRU list, so that it is not canceled due to aging.
262 * Used by MGC locks, they are cancelled only at unmount or by callback. */
263 #define LDLM_FL_NO_LRU 0x1000000000000ULL
266 * The blocking callback is overloaded to perform two functions. These flags
267 * indicate which operation should be performed.
269 #define LDLM_CB_BLOCKING 1
270 #define LDLM_CB_CANCELING 2
273 * \name Lock Compatibility Matrix.
275 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
276 * Lock types are described in their respective implementation files:
277 * ldlm_{extent,flock,inodebits,plain}.c.
279 * There are six lock modes along with a compatibility matrix to indicate if
280 * two locks are compatible.
282 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
284 * - PW: Protective Write (normal write) mode. When a client requests a write
285 * lock from an OST, a lock with PW mode will be issued.
286 * - PR: Protective Read (normal read) mode. When a client requests a read from
287 * an OST, a lock with PR mode will be issued. Also, if the client opens a
288 * file for execution, it is granted a lock with PR mode.
289 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
290 * requests a write lock during a file open operation.
291 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
292 * an inodebit lock with the CR mode on the intermediate path component.
306 #define LCK_COMPAT_EX LCK_NL
307 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
308 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
309 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
310 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
311 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
312 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
313 #define LCK_COMPAT_COS (LCK_COS)
314 /** @} Lock Compatibility Matrix */
316 extern ldlm_mode_t lck_compat_array[];
318 static inline void lockmode_verify(ldlm_mode_t mode)
320 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
323 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
325 return (lck_compat_array[exist_mode] & new_mode);
330 * cluster name spaces
334 #define DLM_OST_NAMESPACE 1
335 #define DLM_MDS_NAMESPACE 2
338 - do we just separate this by security domains and use a prefix for
339 multiple namespaces in the same domain?
344 * Locking rules for LDLM:
349 * waiting_locks_spinlock
364 struct ldlm_resource;
365 struct ldlm_namespace;
368 * Operations on LDLM pools.
369 * LDLM pool is a pool of locks in the namespace without any implicitly
371 * Locks in the pool are organized in LRU.
372 * Local memory pressure or server instructions (e.g. mempressure on server)
373 * can trigger freeing of locks from the pool
375 struct ldlm_pool_ops {
376 /** Recalculate pool \a pl usage */
377 int (*po_recalc)(struct ldlm_pool *pl);
378 /** Cancel at least \a nr locks from pool \a pl */
379 int (*po_shrink)(struct ldlm_pool *pl, int nr,
380 unsigned int gfp_mask);
381 int (*po_setup)(struct ldlm_pool *pl, int limit);
384 /** One second for pools thread check interval. Each pool has own period. */
385 #define LDLM_POOLS_THREAD_PERIOD (1)
387 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
388 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
390 /** Default recalc period for server side pools in sec. */
391 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
393 /** Default recalc period for client side pools in sec. */
394 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
397 * LDLM pool structure to track granted locks.
398 * For purposes of determining when to release locks on e.g. memory pressure.
399 * This feature is commonly referred to as lru_resize.
402 /** Pool proc directory. */
403 cfs_proc_dir_entry_t *pl_proc_dir;
404 /** Pool name, must be long enough to hold compound proc entry name. */
406 /** Lock for protecting SLV/CLV updates. */
408 /** Number of allowed locks in in pool, both, client and server side. */
409 cfs_atomic_t pl_limit;
410 /** Number of granted locks in */
411 cfs_atomic_t pl_granted;
412 /** Grant rate per T. */
413 cfs_atomic_t pl_grant_rate;
414 /** Cancel rate per T. */
415 cfs_atomic_t pl_cancel_rate;
416 /** Server lock volume (SLV). Protected by pl_lock. */
417 __u64 pl_server_lock_volume;
418 /** Current biggest client lock volume. Protected by pl_lock. */
419 __u64 pl_client_lock_volume;
420 /** Lock volume factor. SLV on client is calculated as following:
421 * server_slv * lock_volume_factor. */
422 cfs_atomic_t pl_lock_volume_factor;
423 /** Time when last SLV from server was obtained. */
424 time_t pl_recalc_time;
425 /** Recalculation period for pool. */
426 time_t pl_recalc_period;
427 /** Recalculation and shrink operations. */
428 struct ldlm_pool_ops *pl_ops;
429 /** Number of planned locks for next period. */
431 /** Pool statistics. */
432 struct lprocfs_stats *pl_stats;
435 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
436 void *req_cookie, ldlm_mode_t mode, __u64 flags,
439 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
443 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
444 * be associated with an LDLM lock and transferred from client to server and
447 * Currently LVBs are used by:
448 * - OSC-OST code to maintain current object size/times
449 * - layout lock code to return the layout when the layout lock is granted
451 struct ldlm_valblock_ops {
452 int (*lvbo_init)(struct ldlm_resource *res);
453 int (*lvbo_update)(struct ldlm_resource *res,
454 struct ptlrpc_request *r,
456 int (*lvbo_free)(struct ldlm_resource *res);
457 /* Return size of lvb data appropriate RPC size can be reserved */
458 int (*lvbo_size)(struct ldlm_lock *lock);
459 /* Called to fill in lvb data to RPC buffer @buf */
460 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
464 * LDLM pools related, type of lock pool in the namespace.
465 * Greedy means release cached locks aggressively
468 LDLM_NAMESPACE_GREEDY = 1 << 0,
469 LDLM_NAMESPACE_MODEST = 1 << 1
473 * Default values for the "max_nolock_size", "contention_time" and
474 * "contended_locks" namespace tunables.
476 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
477 #define NS_DEFAULT_CONTENTION_SECONDS 2
478 #define NS_DEFAULT_CONTENDED_LOCKS 32
480 struct ldlm_ns_bucket {
481 /** back pointer to namespace */
482 struct ldlm_namespace *nsb_namespace;
484 * Estimated lock callback time. Used by adaptive timeout code to
485 * avoid spurious client evictions due to unresponsiveness when in
486 * fact the network or overall system load is at fault
488 struct adaptive_timeout nsb_at_estimate;
492 /** LDLM namespace lock stats */
499 LDLM_NS_TYPE_UNKNOWN = 0,
517 * Namespace serves to contain locks related to a particular service.
518 * There are two kinds of namespaces:
519 * - Server namespace has knowledge of all locks and is therefore authoritative
520 * to make decisions like what locks could be granted and what conflicts
521 * exist during new lock enqueue.
522 * - Client namespace only has limited knowledge about locks in the namespace,
523 * only seeing locks held by the client.
525 * Every Lustre service has one server namespace present on the server serving
526 * that service. Every client connected to the service has a client namespace
528 * Every lock obtained by client in that namespace is actually represented by
529 * two in-memory locks. One on the server and one on the client. The locks are
530 * linked by a special cookie by which one node can tell to the other which lock
531 * it actually means during communications. Such locks are called remote locks.
532 * The locks held by server only without any reference to a client are called
535 struct ldlm_namespace {
536 /** Backward link to OBD, required for LDLM pool to store new SLV. */
537 struct obd_device *ns_obd;
539 /** Flag indicating if namespace is on client instead of server */
540 ldlm_side_t ns_client;
542 /** Resource hash table for namespace. */
543 cfs_hash_t *ns_rs_hash;
548 /** big refcount (by bucket) */
549 cfs_atomic_t ns_bref;
552 * Namespace connect flags supported by server (may be changed via
553 * /proc, LRU resize may be disabled/enabled).
555 __u64 ns_connect_flags;
557 /** Client side original connect flags supported by server. */
558 __u64 ns_orig_connect_flags;
561 * Position in global namespace list linking all namespaces on
564 cfs_list_t ns_list_chain;
567 * List of unused locks for this namespace. This list is also called
569 * Unused locks are locks with zero reader/writer reference counts.
570 * This list is only used on clients for lock caching purposes.
571 * When we want to release some locks voluntarily or if server wants
572 * us to release some locks due to e.g. memory pressure, we take locks
573 * to release from the head of this list.
574 * Locks are linked via l_lru field in \see struct ldlm_lock.
576 cfs_list_t ns_unused_list;
577 /** Number of locks in the LRU list above */
581 * Maximum number of locks permitted in the LRU. If 0, means locks
582 * are managed by pools and there is no preset limit, rather it is all
583 * controlled by available memory on this client and on server.
585 unsigned int ns_max_unused;
586 /** Maximum allowed age (last used time) for locks in the LRU */
587 unsigned int ns_max_age;
589 * Server only: number of times we evicted clients due to lack of reply
592 unsigned int ns_timeouts;
594 * Number of seconds since the file change time after which the
595 * MDT will return an UPDATE lock along with a LOOKUP lock.
596 * This allows the client to start caching negative dentries
597 * for a directory and may save an RPC for a later stat.
599 unsigned int ns_ctime_age_limit;
602 * Used to rate-limit ldlm_namespace_dump calls.
603 * \see ldlm_namespace_dump. Increased by 10 seconds every time
606 cfs_time_t ns_next_dump;
608 /** "policy" function that does actual lock conflict determination */
609 ldlm_res_policy ns_policy;
612 * LVB operations for this namespace.
613 * \see struct ldlm_valblock_ops
615 struct ldlm_valblock_ops *ns_lvbo;
618 * Used by filter code to store pointer to OBD of the service.
619 * Should be dropped in favor of \a ns_obd
624 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
625 * a resource is removed.
627 cfs_waitq_t ns_waitq;
628 /** LDLM pool structure for this namespace */
629 struct ldlm_pool ns_pool;
630 /** Definition of how eagerly unused locks will be released from LRU */
631 ldlm_appetite_t ns_appetite;
634 * If more than \a ns_contended_locks are found, the resource is
635 * considered to be contended. Lock enqueues might specify that no
636 * contended locks should be granted
638 unsigned ns_contended_locks;
641 * The resources in this namespace remember contended state during
642 * \a ns_contention_time, in seconds.
644 unsigned ns_contention_time;
647 * Limit size of contended extent locks, in bytes.
648 * If extended lock is requested for more then this many bytes and
649 * caller instructs us not to grant contended locks, we would disregard
652 unsigned ns_max_nolock_size;
654 /** Limit of parallel AST RPC count. */
655 unsigned ns_max_parallel_ast;
657 /** Callback to cancel locks before replaying it during recovery. */
658 ldlm_cancel_for_recovery ns_cancel_for_recovery;
660 /** LDLM lock stats */
661 struct lprocfs_stats *ns_stats;
664 * Flag to indicate namespace is being freed. Used to determine if
665 * recalculation of LDLM pool statistics should be skipped.
667 unsigned ns_stopping:1;
671 * Returns 1 if namespace \a ns is a client namespace.
673 static inline int ns_is_client(struct ldlm_namespace *ns)
676 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
677 LDLM_NAMESPACE_SERVER)));
678 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
679 ns->ns_client == LDLM_NAMESPACE_SERVER);
680 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
684 * Returns 1 if namespace \a ns is a server namespace.
686 static inline int ns_is_server(struct ldlm_namespace *ns)
689 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
690 LDLM_NAMESPACE_SERVER)));
691 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
692 ns->ns_client == LDLM_NAMESPACE_SERVER);
693 return ns->ns_client == LDLM_NAMESPACE_SERVER;
697 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
699 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
702 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
706 * Returns 1 if this namespace supports lru_resize.
708 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
711 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
714 static inline void ns_register_cancel(struct ldlm_namespace *ns,
715 ldlm_cancel_for_recovery arg)
718 ns->ns_cancel_for_recovery = arg;
723 /** Type for blocking callback function of a lock. */
724 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
725 struct ldlm_lock_desc *new, void *data,
727 /** Type for completion callback function of a lock. */
728 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
730 /** Type for glimpse callback function of a lock. */
731 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
732 /** Type for weight callback function of a lock. */
733 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
735 /** Work list for sending GL ASTs to multiple locks. */
736 struct ldlm_glimpse_work {
737 struct ldlm_lock *gl_lock; /* lock to glimpse */
738 cfs_list_t gl_list; /* linkage to other gl work structs */
739 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
740 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
741 * glimpse callback request */
744 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
745 #define LDLM_GL_WORK_NOFREE 0x1
747 /** Interval node data for each LDLM_EXTENT lock. */
748 struct ldlm_interval {
749 struct interval_node li_node; /* node for tree management */
750 cfs_list_t li_group; /* the locks which have the same
751 * policy - group of the policy */
753 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
756 * Interval tree for extent locks.
757 * The interval tree must be accessed under the resource lock.
758 * Interval trees are used for granted extent locks to speed up conflicts
759 * lookup. See ldlm/interval_tree.c for more details.
761 struct ldlm_interval_tree {
764 ldlm_mode_t lit_mode; /* lock mode */
765 struct interval_node *lit_root; /* actual ldlm_interval */
768 /** Whether to track references to exports by LDLM locks. */
769 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
773 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
774 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
775 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
777 } ldlm_cancel_flags_t;
783 __u64 blocking_owner;
784 struct obd_export *blocking_export;
785 /* Protected by the hash lock */
791 struct ldlm_extent l_extent;
792 struct ldlm_flock l_flock;
793 struct ldlm_inodebits l_inodebits;
794 } ldlm_policy_data_t;
796 void ldlm_convert_policy_to_wire(ldlm_type_t type,
797 const ldlm_policy_data_t *lpolicy,
798 ldlm_wire_policy_data_t *wpolicy);
799 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
800 const ldlm_wire_policy_data_t *wpolicy,
801 ldlm_policy_data_t *lpolicy);
811 * LDLM lock structure
813 * Represents a single LDLM lock and its state in memory. Each lock is
814 * associated with a single ldlm_resource, the object which is being
815 * locked. There may be multiple ldlm_locks on a single resource,
816 * depending on the lock type and whether the locks are conflicting or
822 * When remote side wants to tell us about a lock, they address
823 * it by this opaque handle. The handle does not hold a
824 * reference on the ldlm_lock, so it can be safely passed to
825 * other threads or nodes. When the lock needs to be accessed
826 * from the handle, it is looked up again in the lock table, and
827 * may no longer exist.
829 * Must be first in the structure.
831 struct portals_handle l_handle;
833 * Lock reference count.
834 * This is how many users have pointers to actual structure, so that
835 * we do not accidentally free lock structure that is in use.
839 * Internal spinlock protects l_resource. We should hold this lock
840 * first before taking res_lock.
844 * Pointer to actual resource this lock is in.
845 * ldlm_lock_change_resource() can change this.
847 struct ldlm_resource *l_resource;
849 * List item for client side LRU list.
850 * Protected by ns_lock in struct ldlm_namespace.
854 * Linkage to resource's lock queues according to current lock state.
855 * (could be granted, waiting or converting)
856 * Protected by lr_lock in struct ldlm_resource.
858 cfs_list_t l_res_link;
860 * Tree node for ldlm_extent.
862 struct ldlm_interval *l_tree_node;
864 * Per export hash of locks.
865 * Protected by per-bucket exp->exp_lock_hash locks.
867 cfs_hlist_node_t l_exp_hash;
869 * Per export hash of flock locks.
870 * Protected by per-bucket exp->exp_flock_hash locks.
872 cfs_hlist_node_t l_exp_flock_hash;
875 * Protected by lr_lock.
877 ldlm_mode_t l_req_mode;
879 * Granted mode, also protected by lr_lock.
881 ldlm_mode_t l_granted_mode;
882 /** Lock completion handler pointer. Called when lock is granted. */
883 ldlm_completion_callback l_completion_ast;
885 * Lock blocking AST handler pointer.
886 * It plays two roles:
887 * - as a notification of an attempt to queue a conflicting lock (once)
888 * - as a notification when the lock is being cancelled.
890 * As such it's typically called twice: once for the initial conflict
891 * and then once more when the last user went away and the lock is
892 * cancelled (could happen recursively).
894 ldlm_blocking_callback l_blocking_ast;
896 * Lock glimpse handler.
897 * Glimpse handler is used to obtain LVB updates from a client by
900 ldlm_glimpse_callback l_glimpse_ast;
902 /** XXX apparently unused "weight" handler. To be removed? */
903 ldlm_weigh_callback l_weigh_ast;
907 * This is a pointer to actual client export for locks that were granted
908 * to clients. Used server-side.
910 struct obd_export *l_export;
912 * Lock connection export.
913 * Pointer to server export on a client.
915 struct obd_export *l_conn_export;
918 * Remote lock handle.
919 * If the lock is remote, this is the handle of the other side lock
922 struct lustre_handle l_remote_handle;
925 * Representation of private data specific for a lock type.
926 * Examples are: extent range for extent lock or bitmask for ibits locks
928 ldlm_policy_data_t l_policy_data;
932 * Like whenever we receive any blocking requests for this lock, etc.
933 * Protected by lr_lock.
937 * Lock r/w usage counters.
938 * Protected by lr_lock.
943 * If the lock is granted, a process sleeps on this waitq to learn when
944 * it's no longer in use. If the lock is not granted, a process sleeps
945 * on this waitq to learn when it becomes granted.
950 * Seconds. It will be updated if there is any activity related to
951 * the lock, e.g. enqueue the lock or send blocking AST.
953 cfs_time_t l_last_activity;
956 * Time last used by e.g. being matched by lock match.
957 * Jiffies. Should be converted to time if needed.
959 cfs_time_t l_last_used;
961 /** Originally requested extent for the extent lock. */
962 struct ldlm_extent l_req_extent;
964 unsigned int l_failed:1,
966 * Set for locks that were removed from class hash table and will be
967 * destroyed when last reference to them is released. Set by
968 * ldlm_lock_destroy_internal().
970 * Protected by lock and resource locks.
974 * it's set in lock_res_and_lock() and unset in unlock_res_and_lock().
976 * NB: compared with check_res_locked(), checking this bit is cheaper.
977 * Also, spin_is_locked() is deprecated for kernel code; one reason is
978 * because it works only for SMP so user needs to add extra macros like
979 * LASSERT_SPIN_LOCKED for uniprocessor kernels.
983 * It's set once we call ldlm_add_waiting_lock_res_locked()
984 * to start the lock-timeout timer and it will never be reset.
986 * Protected by lock_res_and_lock().
989 /** Flag whether this is a server namespace lock. */
993 * Client-side-only members.
996 enum lvb_type l_lvb_type;
999 * Temporary storage for a LVB received during an enqueue operation.
1004 /** Private storage for lock user. Opaque to LDLM. */
1008 * Server-side-only members.
1012 * Connection cookie for the client originating the operation.
1013 * Used by Commit on Share (COS) code. Currently only used for
1014 * inodebits locks on MDS.
1016 __u64 l_client_cookie;
1019 * List item for locks waiting for cancellation from clients.
1020 * The lists this could be linked into are:
1021 * waiting_locks_list (protected by waiting_locks_spinlock),
1022 * then if the lock timed out, it is moved to
1023 * expired_lock_thread.elt_expired_locks for further processing.
1024 * Protected by elt_lock.
1026 cfs_list_t l_pending_chain;
1029 * Set when lock is sent a blocking AST. Time in seconds when timeout
1030 * is reached and client holding this lock could be evicted.
1031 * This timeout could be further extended by e.g. certain IO activity
1033 * \see ost_rw_prolong_locks
1035 cfs_time_t l_callback_timeout;
1037 /** Local PID of process which created this lock. */
1041 * Number of times blocking AST was sent for this lock.
1042 * This is for debugging. Valid values are 0 and 1, if there is an
1043 * attempt to send blocking AST more than once, an assertion would be
1044 * hit. \see ldlm_work_bl_ast_lock
1047 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
1048 cfs_list_t l_bl_ast;
1049 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
1050 cfs_list_t l_cp_ast;
1051 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
1052 cfs_list_t l_rk_ast;
1055 * Pointer to a conflicting lock that caused blocking AST to be sent
1058 struct ldlm_lock *l_blocking_lock;
1061 * Protected by lr_lock, linkages to "skip lists".
1062 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
1064 cfs_list_t l_sl_mode;
1065 cfs_list_t l_sl_policy;
1067 /** Reference tracking structure to debug leaked locks. */
1068 struct lu_ref l_reference;
1069 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1070 /* Debugging stuff for bug 20498, for tracking export references. */
1071 /** number of export references taken */
1073 /** link all locks referencing one export */
1074 cfs_list_t l_exp_refs_link;
1075 /** referenced export object */
1076 struct obd_export *l_exp_refs_target;
1079 * export blocking dlm lock list, protected by
1080 * l_export->exp_bl_list_lock.
1081 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
1082 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
1084 cfs_list_t l_exp_list;
1088 * LDLM resource description.
1089 * Basically, resource is a representation for a single object.
1090 * Object has a name which is currently 4 64-bit integers. LDLM user is
1091 * responsible for creation of a mapping between objects it wants to be
1092 * protected and resource names.
1094 * A resource can only hold locks of a single lock type, though there may be
1095 * multiple ldlm_locks on a single resource, depending on the lock type and
1096 * whether the locks are conflicting or not.
1098 struct ldlm_resource {
1099 struct ldlm_ns_bucket *lr_ns_bucket;
1102 * List item for list in namespace hash.
1103 * protected by ns_lock
1105 cfs_hlist_node_t lr_hash;
1107 /** Spinlock to protect locks under this resource. */
1111 * protected by lr_lock
1113 /** List of locks in granted state */
1114 cfs_list_t lr_granted;
1115 /** List of locks waiting to change their granted mode (converted) */
1116 cfs_list_t lr_converting;
1118 * List of locks that could not be granted due to conflicts and
1119 * that are waiting for conflicts to go away */
1120 cfs_list_t lr_waiting;
1123 /* XXX No longer needed? Remove ASAP */
1124 ldlm_mode_t lr_most_restr;
1126 /** Type of locks this resource can hold. Only one type per resource. */
1127 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1129 /** Resource name */
1130 struct ldlm_res_id lr_name;
1131 /** Reference count for this resource */
1132 cfs_atomic_t lr_refcount;
1135 * Interval trees (only for extent locks) for all modes of this resource
1137 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM];
1140 * Server-side-only lock value block elements.
1141 * To serialize lvbo_init.
1143 struct mutex lr_lvb_mutex;
1145 /** protected by lr_lock */
1148 /** When the resource was considered as contended. */
1149 cfs_time_t lr_contention_time;
1150 /** List of references to this resource. For debugging. */
1151 struct lu_ref lr_reference;
1153 struct inode *lr_lvb_inode;
1156 static inline char *
1157 ldlm_ns_name(struct ldlm_namespace *ns)
1159 return ns->ns_rs_hash->hs_name;
1162 static inline struct ldlm_namespace *
1163 ldlm_res_to_ns(struct ldlm_resource *res)
1165 return res->lr_ns_bucket->nsb_namespace;
1168 static inline struct ldlm_namespace *
1169 ldlm_lock_to_ns(struct ldlm_lock *lock)
1171 return ldlm_res_to_ns(lock->l_resource);
1174 static inline char *
1175 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1177 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1180 static inline struct adaptive_timeout *
1181 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1183 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1186 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1188 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1190 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_init != NULL)
1191 return ns->ns_lvbo->lvbo_init(res);
1196 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1198 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1200 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1201 return ns->ns_lvbo->lvbo_size(lock);
1206 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1208 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1210 if (ns->ns_lvbo != NULL) {
1211 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1212 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1217 struct ldlm_ast_work {
1218 struct ldlm_lock *w_lock;
1220 struct ldlm_lock_desc w_desc;
1228 * Common ldlm_enqueue parameters
1230 struct ldlm_enqueue_info {
1231 __u32 ei_type; /** Type of the lock being enqueued. */
1232 __u32 ei_mode; /** Mode of the lock being enqueued. */
1233 void *ei_cb_bl; /** blocking lock callback */
1234 void *ei_cb_cp; /** lock completion callback */
1235 void *ei_cb_gl; /** lock glimpse callback */
1236 void *ei_cb_wg; /** lock weigh callback */
1237 void *ei_cbdata; /** Data to be passed into callbacks. */
1240 extern struct obd_ops ldlm_obd_ops;
1242 extern char *ldlm_lockname[];
1243 extern char *ldlm_typename[];
1244 extern char *ldlm_it2str(int it);
1247 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1248 * For the cases where we do not have actual lock to print along
1249 * with a debugging message that is ldlm-related
1251 #define LDLM_DEBUG_NOLOCK(format, a...) \
1252 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1255 * Support function for lock information printing into debug logs.
1259 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1260 CFS_CHECK_STACK(msgdata, mask, cdls); \
1262 if (((mask) & D_CANTMASK) != 0 || \
1263 ((libcfs_debug & (mask)) != 0 && \
1264 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1265 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1268 void _ldlm_lock_debug(struct ldlm_lock *lock,
1269 struct libcfs_debug_msg_data *data,
1270 const char *fmt, ...)
1271 __attribute__ ((format (printf, 3, 4)));
1274 * Rate-limited version of lock printing function.
1276 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1277 static cfs_debug_limit_state_t _ldlm_cdls; \
1278 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1279 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1282 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1283 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1285 /** Non-rate-limited lock printing function for debugging purposes. */
1286 #define LDLM_DEBUG(lock, fmt, a...) do { \
1287 if (likely(lock != NULL)) { \
1288 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1289 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1290 "### " fmt , ##a); \
1292 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1295 #else /* !LIBCFS_DEBUG */
1296 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1297 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1298 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1301 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1302 int first_enq, ldlm_error_t *err,
1303 cfs_list_t *work_list);
1306 * Return values for lock iterators.
1307 * Also used during deciding of lock grants and cancellations.
1309 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1310 #define LDLM_ITER_STOP 2 /* stop iterating */
1312 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1313 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1315 /** \defgroup ldlm_iterator Lock iterators
1317 * LDLM provides for a way to iterate through every lock on a resource or
1318 * namespace or every resource in a namespace.
1320 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1322 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1324 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1325 ldlm_iterator_t iter, void *data);
1326 /** @} ldlm_iterator */
1328 int ldlm_replay_locks(struct obd_import *imp);
1331 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1334 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1336 struct ldlm_callback_suite {
1337 ldlm_completion_callback lcs_completion;
1338 ldlm_blocking_callback lcs_blocking;
1339 ldlm_glimpse_callback lcs_glimpse;
1340 ldlm_weigh_callback lcs_weigh;
1344 #ifdef HAVE_SERVER_SUPPORT
1345 /** \defgroup ldlm_srv_ast Server AST handlers
1346 * These are AST handlers used by server code.
1347 * Their property is that they are just preparing RPCs to be sent to clients.
1350 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1351 void *data, int flag);
1352 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1353 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1354 int ldlm_glimpse_locks(struct ldlm_resource *res, cfs_list_t *gl_work_list);
1355 /** @} ldlm_srv_ast */
1357 /** \defgroup ldlm_handlers Server LDLM handlers
1358 * These are handler functions that should be called by "frontends" such as
1359 * MDT or OST to pass through LDLM requests to LDLM for handling
1362 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
1363 ldlm_blocking_callback, ldlm_glimpse_callback);
1364 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1365 const struct ldlm_request *dlm_req,
1366 const struct ldlm_callback_suite *cbs);
1367 int ldlm_handle_convert(struct ptlrpc_request *req);
1368 int ldlm_handle_convert0(struct ptlrpc_request *req,
1369 const struct ldlm_request *dlm_req);
1370 int ldlm_handle_cancel(struct ptlrpc_request *req);
1371 int ldlm_request_cancel(struct ptlrpc_request *req,
1372 const struct ldlm_request *dlm_req, int first);
1373 /** @} ldlm_handlers */
1375 void ldlm_revoke_export_locks(struct obd_export *exp);
1377 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1378 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
1379 int ldlm_get_ref(void);
1380 void ldlm_put_ref(void);
1381 int ldlm_init_export(struct obd_export *exp);
1382 void ldlm_destroy_export(struct obd_export *exp);
1383 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1386 #ifdef HAVE_SERVER_SUPPORT
1387 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1389 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1390 void ldlm_lock2handle(const struct ldlm_lock *lock,
1391 struct lustre_handle *lockh);
1392 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1393 void ldlm_cancel_callback(struct ldlm_lock *);
1394 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1395 int ldlm_lock_set_data(struct lustre_handle *, void *);
1398 * Obtain a lock reference by its handle.
1400 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1402 return __ldlm_handle2lock(h, 0);
1405 #define LDLM_LOCK_REF_DEL(lock) \
1406 lu_ref_del(&lock->l_reference, "handle", cfs_current())
1408 static inline struct ldlm_lock *
1409 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1411 struct ldlm_lock *lock;
1413 lock = __ldlm_handle2lock(h, flags);
1415 LDLM_LOCK_REF_DEL(lock);
1420 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1421 * data from reqest \a r
1423 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1424 struct ptlrpc_request *r, int increase)
1426 if (ldlm_res_to_ns(res)->ns_lvbo &&
1427 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1428 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1434 int ldlm_error2errno(ldlm_error_t error);
1435 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
1436 * confuses user-space. */
1437 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1438 void ldlm_dump_export_locks(struct obd_export *exp);
1442 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1443 * __ldlm_handle2lock().
1445 #define LDLM_LOCK_PUT(lock) \
1447 LDLM_LOCK_REF_DEL(lock); \
1448 /*LDLM_DEBUG((lock), "put");*/ \
1449 ldlm_lock_put(lock); \
1453 * Release a lock reference obtained by some other means (see
1456 #define LDLM_LOCK_RELEASE(lock) \
1458 /*LDLM_DEBUG((lock), "put");*/ \
1459 ldlm_lock_put(lock); \
1462 #define LDLM_LOCK_GET(lock) \
1464 ldlm_lock_get(lock); \
1465 /*LDLM_DEBUG((lock), "get");*/ \
1469 #define ldlm_lock_list_put(head, member, count) \
1471 struct ldlm_lock *_lock, *_next; \
1473 cfs_list_for_each_entry_safe(_lock, _next, head, member) { \
1476 cfs_list_del_init(&_lock->member); \
1477 LDLM_LOCK_RELEASE(_lock); \
1482 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1483 void ldlm_lock_put(struct ldlm_lock *lock);
1484 void ldlm_lock_destroy(struct ldlm_lock *lock);
1485 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1486 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1487 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1488 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1489 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1490 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1491 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1492 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1493 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1494 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1495 const struct ldlm_res_id *, ldlm_type_t type,
1496 ldlm_policy_data_t *, ldlm_mode_t mode,
1497 struct lustre_handle *, int unref);
1498 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1500 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1502 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
1503 void ldlm_lock_cancel(struct ldlm_lock *lock);
1504 void ldlm_reprocess_all(struct ldlm_resource *res);
1505 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1506 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1507 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1510 struct ldlm_namespace *
1511 ldlm_namespace_new(struct obd_device *obd, char *name,
1512 ldlm_side_t client, ldlm_appetite_t apt,
1513 ldlm_ns_type_t ns_type);
1514 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1515 void ldlm_namespace_free(struct ldlm_namespace *ns,
1516 struct obd_import *imp, int force);
1517 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1518 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1519 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1520 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1521 void ldlm_namespace_get(struct ldlm_namespace *ns);
1522 void ldlm_namespace_put(struct ldlm_namespace *ns);
1523 int ldlm_proc_setup(void);
1525 void ldlm_proc_cleanup(void);
1527 static inline void ldlm_proc_cleanup(void) {}
1530 /* resource.c - internal */
1531 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1532 struct ldlm_resource *parent,
1533 const struct ldlm_res_id *,
1534 ldlm_type_t type, int create);
1535 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1536 int ldlm_resource_putref(struct ldlm_resource *res);
1537 void ldlm_resource_add_lock(struct ldlm_resource *res,
1539 struct ldlm_lock *lock);
1540 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1541 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1542 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1543 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1544 void ldlm_resource_dump(int level, struct ldlm_resource *);
1545 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1546 const struct ldlm_res_id *);
1548 #define LDLM_RESOURCE_ADDREF(res) do { \
1549 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1552 #define LDLM_RESOURCE_DELREF(res) do { \
1553 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1556 /* ldlm_request.c */
1557 int ldlm_expired_completion_wait(void *data);
1558 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1559 * These AST handlers are typically used for server-side local locks and are
1560 * also used by client-side lock handlers to perform minimum level base
1563 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1564 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1565 void *data, int flag);
1566 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1567 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1568 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1569 /** @} ldlm_local_ast */
1571 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1572 * These are typically used by client and server (*_local versions)
1573 * to obtain and release locks.
1575 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1576 struct ldlm_enqueue_info *einfo,
1577 const struct ldlm_res_id *res_id,
1578 ldlm_policy_data_t const *policy, __u64 *flags,
1579 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1580 struct lustre_handle *lockh, int async);
1581 int ldlm_prep_enqueue_req(struct obd_export *exp,
1582 struct ptlrpc_request *req,
1583 cfs_list_t *cancels,
1585 int ldlm_prep_elc_req(struct obd_export *exp,
1586 struct ptlrpc_request *req,
1587 int version, int opc, int canceloff,
1588 cfs_list_t *cancels, int count);
1589 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1590 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1591 __u64 *flags, void *lvb, __u32 lvb_len,
1592 struct lustre_handle *lockh, int rc);
1593 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1594 const struct ldlm_res_id *res_id,
1595 ldlm_type_t type, ldlm_policy_data_t *policy,
1596 ldlm_mode_t mode, __u64 *flags,
1597 ldlm_blocking_callback blocking,
1598 ldlm_completion_callback completion,
1599 ldlm_glimpse_callback glimpse,
1600 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1601 const __u64 *client_cookie,
1602 struct lustre_handle *lockh);
1603 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1604 void *data, __u32 data_len);
1605 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1606 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1607 int ldlm_cli_cancel(struct lustre_handle *lockh);
1608 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1609 ldlm_cancel_flags_t flags, void *opaque);
1610 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1611 const struct ldlm_res_id *res_id,
1612 ldlm_policy_data_t *policy,
1614 ldlm_cancel_flags_t flags,
1616 int ldlm_cli_cancel_req(struct obd_export *exp, cfs_list_t *head,
1617 int count, ldlm_cancel_flags_t flags);
1618 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1619 cfs_list_t *cancels,
1620 ldlm_policy_data_t *policy,
1621 ldlm_mode_t mode, int lock_flags,
1622 ldlm_cancel_flags_t cancel_flags, void *opaque);
1623 int ldlm_cli_cancel_list_local(cfs_list_t *cancels, int count,
1624 ldlm_cancel_flags_t flags);
1625 int ldlm_cli_cancel_list(cfs_list_t *head, int count,
1626 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1627 /** @} ldlm_cli_api */
1630 /* This has to be here because recursive inclusion sucks. */
1631 int intent_disposition(struct ldlm_reply *rep, int flag);
1632 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1635 /* ioctls for trying requests */
1636 #define IOC_LDLM_TYPE 'f'
1637 #define IOC_LDLM_MIN_NR 40
1639 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1640 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1641 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1642 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1643 #define IOC_LDLM_MAX_NR 43
1646 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1647 * than one lock_res is dead-lock safe.
1649 enum lock_res_type {
1654 /** Lock resource. */
1655 static inline void lock_res(struct ldlm_resource *res)
1657 spin_lock(&res->lr_lock);
1660 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1661 static inline void lock_res_nested(struct ldlm_resource *res,
1662 enum lock_res_type mode)
1664 spin_lock_nested(&res->lr_lock, mode);
1667 /** Unlock resource. */
1668 static inline void unlock_res(struct ldlm_resource *res)
1670 spin_unlock(&res->lr_lock);
1673 /** Check if resource is already locked, assert if not. */
1674 static inline void check_res_locked(struct ldlm_resource *res)
1676 LASSERT_SPIN_LOCKED(&res->lr_lock);
1679 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1680 void unlock_res_and_lock(struct ldlm_lock *lock);
1683 /** \defgroup ldlm_pools Various LDLM pool related functions
1684 * There are not used outside of ldlm.
1687 void ldlm_pools_recalc(ldlm_side_t client);
1688 int ldlm_pools_init(void);
1689 void ldlm_pools_fini(void);
1691 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1692 int idx, ldlm_side_t client);
1693 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1694 unsigned int gfp_mask);
1695 void ldlm_pool_fini(struct ldlm_pool *pl);
1696 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1697 int ldlm_pool_recalc(struct ldlm_pool *pl);
1698 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1699 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1700 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1701 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1702 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1703 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1704 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1705 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1706 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);