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 #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,
93 ELDLM_LOCK_WOULDBLOCK = 304,
95 ELDLM_NAMESPACE_EXISTS = 400,
96 ELDLM_BAD_NAMESPACE = 401
100 * LDLM namespace type.
101 * The "client" type is actually an indication that this is a narrow local view
102 * into complete namespace on the server. Such namespaces cannot make any
103 * decisions about lack of conflicts or do any autonomous lock granting without
104 * first speaking to a server.
107 LDLM_NAMESPACE_SERVER = 1 << 0,
108 LDLM_NAMESPACE_CLIENT = 1 << 1
112 * Declaration of flags sent through the wire.
114 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
117 * If the server returns one of these flags, then the lock was put on that list.
118 * If the client sends one of these flags (during recovery ONLY!), it wants the
119 * lock added to the specified list, no questions asked.
121 #define LDLM_FL_BLOCK_GRANTED 0x000002
122 #define LDLM_FL_BLOCK_CONV 0x000004
123 #define LDLM_FL_BLOCK_WAIT 0x000008
125 /* Used to be LDLM_FL_CBPENDING 0x000010 moved to non-wire flags */
127 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
128 * queued for sending. */
129 /* Used to be LDLM_FL_WAIT_NOREPROC 0x000040 moved to non-wire flags */
130 /* Used to be LDLM_FL_CANCEL 0x000080 moved to non-wire flags */
133 * Lock is being replayed. This could probably be implied by the fact that one
134 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous.
136 #define LDLM_FL_REPLAY 0x000100
138 #define LDLM_FL_INTENT_ONLY 0x000200 /* Don't grant lock, just do intent. */
140 /* Used to be LDLM_FL_LOCAL_ONLY 0x000400 moved to non-wire flags */
141 /* Used to be LDLM_FL_FAILED 0x000800 moved to non-wire flags */
143 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
145 /* Used to be LDLM_FL_CANCELING 0x002000 moved to non-wire flags */
146 /* Used to be LDLM_FL_LOCAL 0x004000 moved to non-wire flags */
148 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
150 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
153 /** file & record locking */
154 #define LDLM_FL_BLOCK_NOWAIT 0x040000 /* Server told not to wait if blocked.
155 * For AGL, OST will not send glimpse
157 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
159 /* Used to be LDLM_FL_LVB_READY 0x100000 moved to non-wire flags */
160 /* Used to be LDLM_FL_KMS_IGNORE 0x200000 moved to non-wire flags */
161 /* Used to be LDLM_FL_NO_LRU 0x400000 moved to non-wire flags */
163 /* Immediatelly cancel such locks when they block some other locks. Send
164 * cancel notification to original lock holder, but expect no reply. This is
165 * for clients (like liblustre) that cannot be expected to reliably response
166 * to blocking AST. */
167 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
169 /* Flags flags inherited from parent lock when doing intents. */
170 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
172 /* Used to be LDLM_FL_CP_REQD 0x1000000 moved to non-wire flags */
173 /* Used to be LDLM_FL_CLEANED 0x2000000 moved to non-wire flags */
174 /* Used to be LDLM_FL_ATOMIC_CB 0x4000000 moved to non-wire flags */
175 /* Used to be LDLM_FL_BL_AST 0x10000000 moved to non-wire flags */
176 /* Used to be LDLM_FL_BL_DONE 0x20000000 moved to non-wire flags */
178 /* measure lock contention and return -EUSERS if locking contention is high */
179 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
181 /* These are flags that are mapped into the flags and ASTs of blocking locks */
182 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
184 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
185 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
188 * --------------------------------------------------------------------------
189 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
190 * 0x80000000 will not be sent over the wire.
191 * --------------------------------------------------------------------------
195 * Declaration of flags not sent through the wire.
199 * Used for marking lock as a target for -EINTR while cp_ast sleep
200 * emulation + race with upcoming bl_ast.
202 #define LDLM_FL_FAIL_LOC 0x100000000ULL
205 * Used while processing the unused list to know that we have already
206 * handled this lock and decided to skip it.
208 #define LDLM_FL_SKIPPED 0x200000000ULL
209 /* this lock is being destroyed */
210 #define LDLM_FL_CBPENDING 0x400000000ULL
211 /* not a real flag, not saved in lock */
212 #define LDLM_FL_WAIT_NOREPROC 0x800000000ULL
213 /* cancellation callback already run */
214 #define LDLM_FL_CANCEL 0x1000000000ULL
215 #define LDLM_FL_LOCAL_ONLY 0x2000000000ULL
216 /* don't run the cancel callback under ldlm_cli_cancel_unused */
217 #define LDLM_FL_FAILED 0x4000000000ULL
218 /* lock cancel has already been sent */
219 #define LDLM_FL_CANCELING 0x8000000000ULL
220 /* local lock (ie, no srv/cli split) */
221 #define LDLM_FL_LOCAL 0x10000000000ULL
222 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
223 * the LVB filling happens _after_ the lock has been granted, so another thread
224 * can match it before the LVB has been updated. As a dirty hack, we set
225 * LDLM_FL_LVB_READY only after we've done the LVB poop.
226 * this is only needed on LOV/OSC now, where LVB is actually used and callers
227 * must set it in input flags.
229 * The proper fix is to do the granting inside of the completion AST, which can
230 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
231 * pretty high-risk, though, and would need a lot more testing. */
232 #define LDLM_FL_LVB_READY 0x20000000000ULL
233 /* A lock contributes to the known minimum size (KMS) calculation until it has
234 * finished the part of its cancelation that performs write back on its dirty
235 * pages. It can remain on the granted list during this whole time. Threads
236 * racing to update the KMS after performing their writeback need to know to
237 * exclude each other's locks from the calculation as they walk the granted
239 #define LDLM_FL_KMS_IGNORE 0x40000000000ULL
240 /* completion AST to be executed */
241 #define LDLM_FL_CP_REQD 0x80000000000ULL
242 /* cleanup_resource has already handled the lock */
243 #define LDLM_FL_CLEANED 0x100000000000ULL
244 /* optimization hint: LDLM can run blocking callback from current context
245 * w/o involving separate thread. in order to decrease cs rate */
246 #define LDLM_FL_ATOMIC_CB 0x200000000000ULL
248 /* It may happen that a client initiates two operations, e.g. unlink and
249 * mkdir, such that the server sends a blocking AST for conflicting
250 * locks to this client for the first operation, whereas the second
251 * operation has canceled this lock and is waiting for rpc_lock which is
252 * taken by the first operation. LDLM_FL_BL_AST is set by
253 * ldlm_callback_handler() in the lock to prevent the Early Lock Cancel
254 * (ELC) code from cancelling it.
256 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock
257 * cache is dropped to let ldlm_callback_handler() return EINVAL to the
258 * server. It is used when ELC RPC is already prepared and is waiting
259 * for rpc_lock, too late to send a separate CANCEL RPC. */
260 #define LDLM_FL_BL_AST 0x400000000000ULL
261 #define LDLM_FL_BL_DONE 0x800000000000ULL
262 /* Don't put lock into the LRU list, so that it is not canceled due to aging.
263 * Used by MGC locks, they are cancelled only at unmount or by callback. */
264 #define LDLM_FL_NO_LRU 0x1000000000000ULL
267 * The blocking callback is overloaded to perform two functions. These flags
268 * indicate which operation should be performed.
270 #define LDLM_CB_BLOCKING 1
271 #define LDLM_CB_CANCELING 2
274 * \name Lock Compatibility Matrix.
276 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
277 * Lock types are described in their respective implementation files:
278 * ldlm_{extent,flock,inodebits,plain}.c.
280 * There are six lock modes along with a compatibility matrix to indicate if
281 * two locks are compatible.
283 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
285 * - PW: Protective Write (normal write) mode. When a client requests a write
286 * lock from an OST, a lock with PW mode will be issued.
287 * - PR: Protective Read (normal read) mode. When a client requests a read from
288 * an OST, a lock with PR mode will be issued. Also, if the client opens a
289 * file for execution, it is granted a lock with PR mode.
290 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
291 * requests a write lock during a file open operation.
292 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
293 * an inodebit lock with the CR mode on the intermediate path component.
307 #define LCK_COMPAT_EX LCK_NL
308 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
309 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
310 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
311 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
312 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
313 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
314 #define LCK_COMPAT_COS (LCK_COS)
315 /** @} Lock Compatibility Matrix */
317 extern ldlm_mode_t lck_compat_array[];
319 static inline void lockmode_verify(ldlm_mode_t mode)
321 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
324 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
326 return (lck_compat_array[exist_mode] & new_mode);
331 * cluster name spaces
335 #define DLM_OST_NAMESPACE 1
336 #define DLM_MDS_NAMESPACE 2
339 - do we just separate this by security domains and use a prefix for
340 multiple namespaces in the same domain?
345 * Locking rules for LDLM:
350 * waiting_locks_spinlock
365 struct ldlm_resource;
366 struct ldlm_namespace;
369 * Operations on LDLM pools.
370 * LDLM pool is a pool of locks in the namespace without any implicitly
372 * Locks in the pool are organized in LRU.
373 * Local memory pressure or server instructions (e.g. mempressure on server)
374 * can trigger freeing of locks from the pool
376 struct ldlm_pool_ops {
377 /** Recalculate pool \a pl usage */
378 int (*po_recalc)(struct ldlm_pool *pl);
379 /** Cancel at least \a nr locks from pool \a pl */
380 int (*po_shrink)(struct ldlm_pool *pl, int nr,
381 unsigned int gfp_mask);
382 int (*po_setup)(struct ldlm_pool *pl, int limit);
385 /** One second for pools thread check interval. Each pool has own period. */
386 #define LDLM_POOLS_THREAD_PERIOD (1)
388 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
389 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
391 /** Default recalc period for server side pools in sec. */
392 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
394 /** Default recalc period for client side pools in sec. */
395 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
398 * LDLM pool structure to track granted locks.
399 * For purposes of determining when to release locks on e.g. memory pressure.
400 * This feature is commonly referred to as lru_resize.
403 /** Pool proc directory. */
404 cfs_proc_dir_entry_t *pl_proc_dir;
405 /** Pool name, must be long enough to hold compound proc entry name. */
407 /** Lock for protecting SLV/CLV updates. */
409 /** Number of allowed locks in in pool, both, client and server side. */
410 cfs_atomic_t pl_limit;
411 /** Number of granted locks in */
412 cfs_atomic_t pl_granted;
413 /** Grant rate per T. */
414 cfs_atomic_t pl_grant_rate;
415 /** Cancel rate per T. */
416 cfs_atomic_t pl_cancel_rate;
417 /** Server lock volume (SLV). Protected by pl_lock. */
418 __u64 pl_server_lock_volume;
419 /** Current biggest client lock volume. Protected by pl_lock. */
420 __u64 pl_client_lock_volume;
421 /** Lock volume factor. SLV on client is calculated as following:
422 * server_slv * lock_volume_factor. */
423 cfs_atomic_t pl_lock_volume_factor;
424 /** Time when last SLV from server was obtained. */
425 time_t pl_recalc_time;
426 /** Recalculation period for pool. */
427 time_t pl_recalc_period;
428 /** Recalculation and shrink operations. */
429 struct ldlm_pool_ops *pl_ops;
430 /** Number of planned locks for next period. */
432 /** Pool statistics. */
433 struct lprocfs_stats *pl_stats;
436 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
437 void *req_cookie, ldlm_mode_t mode, __u64 flags,
440 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
444 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
445 * be associated with an LDLM lock and transferred from client to server and
448 * Currently LVBs are used by:
449 * - OSC-OST code to maintain current object size/times
450 * - layout lock code to return the layout when the layout lock is granted
452 struct ldlm_valblock_ops {
453 int (*lvbo_init)(struct ldlm_resource *res);
454 int (*lvbo_update)(struct ldlm_resource *res,
455 struct ptlrpc_request *r,
457 int (*lvbo_free)(struct ldlm_resource *res);
458 /* Return size of lvb data appropriate RPC size can be reserved */
459 int (*lvbo_size)(struct ldlm_lock *lock);
460 /* Called to fill in lvb data to RPC buffer @buf */
461 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
465 * LDLM pools related, type of lock pool in the namespace.
466 * Greedy means release cached locks aggressively
469 LDLM_NAMESPACE_GREEDY = 1 << 0,
470 LDLM_NAMESPACE_MODEST = 1 << 1
474 * Default values for the "max_nolock_size", "contention_time" and
475 * "contended_locks" namespace tunables.
477 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
478 #define NS_DEFAULT_CONTENTION_SECONDS 2
479 #define NS_DEFAULT_CONTENDED_LOCKS 32
481 struct ldlm_ns_bucket {
482 /** back pointer to namespace */
483 struct ldlm_namespace *nsb_namespace;
485 * Estimated lock callback time. Used by adaptive timeout code to
486 * avoid spurious client evictions due to unresponsiveness when in
487 * fact the network or overall system load is at fault
489 struct adaptive_timeout nsb_at_estimate;
493 /** LDLM namespace lock stats */
500 LDLM_NS_TYPE_UNKNOWN = 0,
518 * Namespace serves to contain locks related to a particular service.
519 * There are two kinds of namespaces:
520 * - Server namespace has knowledge of all locks and is therefore authoritative
521 * to make decisions like what locks could be granted and what conflicts
522 * exist during new lock enqueue.
523 * - Client namespace only has limited knowledge about locks in the namespace,
524 * only seeing locks held by the client.
526 * Every Lustre service has one server namespace present on the server serving
527 * that service. Every client connected to the service has a client namespace
529 * Every lock obtained by client in that namespace is actually represented by
530 * two in-memory locks. One on the server and one on the client. The locks are
531 * linked by a special cookie by which one node can tell to the other which lock
532 * it actually means during communications. Such locks are called remote locks.
533 * The locks held by server only without any reference to a client are called
536 struct ldlm_namespace {
537 /** Backward link to OBD, required for LDLM pool to store new SLV. */
538 struct obd_device *ns_obd;
540 /** Flag indicating if namespace is on client instead of server */
541 ldlm_side_t ns_client;
543 /** Resource hash table for namespace. */
544 cfs_hash_t *ns_rs_hash;
549 /** big refcount (by bucket) */
550 cfs_atomic_t ns_bref;
553 * Namespace connect flags supported by server (may be changed via
554 * /proc, LRU resize may be disabled/enabled).
556 __u64 ns_connect_flags;
558 /** Client side original connect flags supported by server. */
559 __u64 ns_orig_connect_flags;
562 * Position in global namespace list linking all namespaces on
565 cfs_list_t ns_list_chain;
568 * List of unused locks for this namespace. This list is also called
570 * Unused locks are locks with zero reader/writer reference counts.
571 * This list is only used on clients for lock caching purposes.
572 * When we want to release some locks voluntarily or if server wants
573 * us to release some locks due to e.g. memory pressure, we take locks
574 * to release from the head of this list.
575 * Locks are linked via l_lru field in \see struct ldlm_lock.
577 cfs_list_t ns_unused_list;
578 /** Number of locks in the LRU list above */
582 * Maximum number of locks permitted in the LRU. If 0, means locks
583 * are managed by pools and there is no preset limit, rather it is all
584 * controlled by available memory on this client and on server.
586 unsigned int ns_max_unused;
587 /** Maximum allowed age (last used time) for locks in the LRU */
588 unsigned int ns_max_age;
590 * Server only: number of times we evicted clients due to lack of reply
593 unsigned int ns_timeouts;
595 * Number of seconds since the file change time after which the
596 * MDT will return an UPDATE lock along with a LOOKUP lock.
597 * This allows the client to start caching negative dentries
598 * for a directory and may save an RPC for a later stat.
600 unsigned int ns_ctime_age_limit;
603 * Used to rate-limit ldlm_namespace_dump calls.
604 * \see ldlm_namespace_dump. Increased by 10 seconds every time
607 cfs_time_t ns_next_dump;
609 /** "policy" function that does actual lock conflict determination */
610 ldlm_res_policy ns_policy;
613 * LVB operations for this namespace.
614 * \see struct ldlm_valblock_ops
616 struct ldlm_valblock_ops *ns_lvbo;
619 * Used by filter code to store pointer to OBD of the service.
620 * Should be dropped in favor of \a ns_obd
625 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
626 * a resource is removed.
628 cfs_waitq_t ns_waitq;
629 /** LDLM pool structure for this namespace */
630 struct ldlm_pool ns_pool;
631 /** Definition of how eagerly unused locks will be released from LRU */
632 ldlm_appetite_t ns_appetite;
635 * If more than \a ns_contended_locks are found, the resource is
636 * considered to be contended. Lock enqueues might specify that no
637 * contended locks should be granted
639 unsigned ns_contended_locks;
642 * The resources in this namespace remember contended state during
643 * \a ns_contention_time, in seconds.
645 unsigned ns_contention_time;
648 * Limit size of contended extent locks, in bytes.
649 * If extended lock is requested for more then this many bytes and
650 * caller instructs us not to grant contended locks, we would disregard
653 unsigned ns_max_nolock_size;
655 /** Limit of parallel AST RPC count. */
656 unsigned ns_max_parallel_ast;
658 /** Callback to cancel locks before replaying it during recovery. */
659 ldlm_cancel_for_recovery ns_cancel_for_recovery;
661 /** LDLM lock stats */
662 struct lprocfs_stats *ns_stats;
665 * Flag to indicate namespace is being freed. Used to determine if
666 * recalculation of LDLM pool statistics should be skipped.
668 unsigned ns_stopping:1;
672 * Returns 1 if namespace \a ns is a client namespace.
674 static inline int ns_is_client(struct ldlm_namespace *ns)
677 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
678 LDLM_NAMESPACE_SERVER)));
679 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
680 ns->ns_client == LDLM_NAMESPACE_SERVER);
681 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
685 * Returns 1 if namespace \a ns is a server namespace.
687 static inline int ns_is_server(struct ldlm_namespace *ns)
690 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
691 LDLM_NAMESPACE_SERVER)));
692 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
693 ns->ns_client == LDLM_NAMESPACE_SERVER);
694 return ns->ns_client == LDLM_NAMESPACE_SERVER;
698 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
700 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
703 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
707 * Returns 1 if this namespace supports lru_resize.
709 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
712 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
715 static inline void ns_register_cancel(struct ldlm_namespace *ns,
716 ldlm_cancel_for_recovery arg)
719 ns->ns_cancel_for_recovery = arg;
724 /** Type for blocking callback function of a lock. */
725 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
726 struct ldlm_lock_desc *new, void *data,
728 /** Type for completion callback function of a lock. */
729 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
731 /** Type for glimpse callback function of a lock. */
732 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
733 /** Type for weight callback function of a lock. */
734 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
736 /** Work list for sending GL ASTs to multiple locks. */
737 struct ldlm_glimpse_work {
738 struct ldlm_lock *gl_lock; /* lock to glimpse */
739 cfs_list_t gl_list; /* linkage to other gl work structs */
740 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
741 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
742 * glimpse callback request */
745 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
746 #define LDLM_GL_WORK_NOFREE 0x1
748 /** Interval node data for each LDLM_EXTENT lock. */
749 struct ldlm_interval {
750 struct interval_node li_node; /* node for tree management */
751 cfs_list_t li_group; /* the locks which have the same
752 * policy - group of the policy */
754 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
757 * Interval tree for extent locks.
758 * The interval tree must be accessed under the resource lock.
759 * Interval trees are used for granted extent locks to speed up conflicts
760 * lookup. See ldlm/interval_tree.c for more details.
762 struct ldlm_interval_tree {
765 ldlm_mode_t lit_mode; /* lock mode */
766 struct interval_node *lit_root; /* actual ldlm_interval */
769 /** Whether to track references to exports by LDLM locks. */
770 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
774 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
775 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
776 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
778 } ldlm_cancel_flags_t;
784 __u64 blocking_owner;
785 struct obd_export *blocking_export;
786 /* Protected by the hash lock */
792 struct ldlm_extent l_extent;
793 struct ldlm_flock l_flock;
794 struct ldlm_inodebits l_inodebits;
795 } ldlm_policy_data_t;
797 void ldlm_convert_policy_to_wire(ldlm_type_t type,
798 const ldlm_policy_data_t *lpolicy,
799 ldlm_wire_policy_data_t *wpolicy);
800 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
801 const ldlm_wire_policy_data_t *wpolicy,
802 ldlm_policy_data_t *lpolicy);
812 * LDLM lock structure
814 * Represents a single LDLM lock and its state in memory. Each lock is
815 * associated with a single ldlm_resource, the object which is being
816 * locked. There may be multiple ldlm_locks on a single resource,
817 * depending on the lock type and whether the locks are conflicting or
823 * When remote side wants to tell us about a lock, they address
824 * it by this opaque handle. The handle does not hold a
825 * reference on the ldlm_lock, so it can be safely passed to
826 * other threads or nodes. When the lock needs to be accessed
827 * from the handle, it is looked up again in the lock table, and
828 * may no longer exist.
830 * Must be first in the structure.
832 struct portals_handle l_handle;
834 * Lock reference count.
835 * This is how many users have pointers to actual structure, so that
836 * we do not accidentally free lock structure that is in use.
840 * Internal spinlock protects l_resource. We should hold this lock
841 * first before taking res_lock.
845 * Pointer to actual resource this lock is in.
846 * ldlm_lock_change_resource() can change this.
848 struct ldlm_resource *l_resource;
850 * List item for client side LRU list.
851 * Protected by ns_lock in struct ldlm_namespace.
855 * Linkage to resource's lock queues according to current lock state.
856 * (could be granted, waiting or converting)
857 * Protected by lr_lock in struct ldlm_resource.
859 cfs_list_t l_res_link;
861 * Tree node for ldlm_extent.
863 struct ldlm_interval *l_tree_node;
865 * Per export hash of locks.
866 * Protected by per-bucket exp->exp_lock_hash locks.
868 cfs_hlist_node_t l_exp_hash;
870 * Per export hash of flock locks.
871 * Protected by per-bucket exp->exp_flock_hash locks.
873 cfs_hlist_node_t l_exp_flock_hash;
876 * Protected by lr_lock.
878 ldlm_mode_t l_req_mode;
880 * Granted mode, also protected by lr_lock.
882 ldlm_mode_t l_granted_mode;
883 /** Lock completion handler pointer. Called when lock is granted. */
884 ldlm_completion_callback l_completion_ast;
886 * Lock blocking AST handler pointer.
887 * It plays two roles:
888 * - as a notification of an attempt to queue a conflicting lock (once)
889 * - as a notification when the lock is being cancelled.
891 * As such it's typically called twice: once for the initial conflict
892 * and then once more when the last user went away and the lock is
893 * cancelled (could happen recursively).
895 ldlm_blocking_callback l_blocking_ast;
897 * Lock glimpse handler.
898 * Glimpse handler is used to obtain LVB updates from a client by
901 ldlm_glimpse_callback l_glimpse_ast;
903 /** XXX apparently unused "weight" handler. To be removed? */
904 ldlm_weigh_callback l_weigh_ast;
908 * This is a pointer to actual client export for locks that were granted
909 * to clients. Used server-side.
911 struct obd_export *l_export;
913 * Lock connection export.
914 * Pointer to server export on a client.
916 struct obd_export *l_conn_export;
919 * Remote lock handle.
920 * If the lock is remote, this is the handle of the other side lock
923 struct lustre_handle l_remote_handle;
926 * Representation of private data specific for a lock type.
927 * Examples are: extent range for extent lock or bitmask for ibits locks
929 ldlm_policy_data_t l_policy_data;
933 * Like whenever we receive any blocking requests for this lock, etc.
934 * Protected by lr_lock.
938 * Lock r/w usage counters.
939 * Protected by lr_lock.
944 * If the lock is granted, a process sleeps on this waitq to learn when
945 * it's no longer in use. If the lock is not granted, a process sleeps
946 * on this waitq to learn when it becomes granted.
951 * Seconds. It will be updated if there is any activity related to
952 * the lock, e.g. enqueue the lock or send blocking AST.
954 cfs_time_t l_last_activity;
957 * Time last used by e.g. being matched by lock match.
958 * Jiffies. Should be converted to time if needed.
960 cfs_time_t l_last_used;
962 /** Originally requested extent for the extent lock. */
963 struct ldlm_extent l_req_extent;
965 unsigned int l_failed:1,
967 * Set for locks that were removed from class hash table and will be
968 * destroyed when last reference to them is released. Set by
969 * ldlm_lock_destroy_internal().
971 * Protected by lock and resource locks.
975 * it's set in lock_res_and_lock() and unset in unlock_res_and_lock().
977 * NB: compared with check_res_locked(), checking this bit is cheaper.
978 * Also, spin_is_locked() is deprecated for kernel code; one reason is
979 * because it works only for SMP so user needs to add extra macros like
980 * LASSERT_SPIN_LOCKED for uniprocessor kernels.
984 * It's set once we call ldlm_add_waiting_lock_res_locked()
985 * to start the lock-timeout timer and it will never be reset.
987 * Protected by lock_res_and_lock().
990 /** Flag whether this is a server namespace lock. */
994 * Client-side-only members.
997 enum lvb_type l_lvb_type;
1000 * Temporary storage for a LVB received during an enqueue operation.
1005 /** Private storage for lock user. Opaque to LDLM. */
1009 * Server-side-only members.
1013 * Connection cookie for the client originating the operation.
1014 * Used by Commit on Share (COS) code. Currently only used for
1015 * inodebits locks on MDS.
1017 __u64 l_client_cookie;
1020 * List item for locks waiting for cancellation from clients.
1021 * The lists this could be linked into are:
1022 * waiting_locks_list (protected by waiting_locks_spinlock),
1023 * then if the lock timed out, it is moved to
1024 * expired_lock_thread.elt_expired_locks for further processing.
1025 * Protected by elt_lock.
1027 cfs_list_t l_pending_chain;
1030 * Set when lock is sent a blocking AST. Time in seconds when timeout
1031 * is reached and client holding this lock could be evicted.
1032 * This timeout could be further extended by e.g. certain IO activity
1034 * \see ost_rw_prolong_locks
1036 cfs_time_t l_callback_timeout;
1038 /** Local PID of process which created this lock. */
1042 * Number of times blocking AST was sent for this lock.
1043 * This is for debugging. Valid values are 0 and 1, if there is an
1044 * attempt to send blocking AST more than once, an assertion would be
1045 * hit. \see ldlm_work_bl_ast_lock
1048 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
1049 cfs_list_t l_bl_ast;
1050 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
1051 cfs_list_t l_cp_ast;
1052 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
1053 cfs_list_t l_rk_ast;
1056 * Pointer to a conflicting lock that caused blocking AST to be sent
1059 struct ldlm_lock *l_blocking_lock;
1062 * Protected by lr_lock, linkages to "skip lists".
1063 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
1065 cfs_list_t l_sl_mode;
1066 cfs_list_t l_sl_policy;
1068 /** Reference tracking structure to debug leaked locks. */
1069 struct lu_ref l_reference;
1070 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1071 /* Debugging stuff for bug 20498, for tracking export references. */
1072 /** number of export references taken */
1074 /** link all locks referencing one export */
1075 cfs_list_t l_exp_refs_link;
1076 /** referenced export object */
1077 struct obd_export *l_exp_refs_target;
1080 * export blocking dlm lock list, protected by
1081 * l_export->exp_bl_list_lock.
1082 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
1083 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
1085 cfs_list_t l_exp_list;
1089 * LDLM resource description.
1090 * Basically, resource is a representation for a single object.
1091 * Object has a name which is currently 4 64-bit integers. LDLM user is
1092 * responsible for creation of a mapping between objects it wants to be
1093 * protected and resource names.
1095 * A resource can only hold locks of a single lock type, though there may be
1096 * multiple ldlm_locks on a single resource, depending on the lock type and
1097 * whether the locks are conflicting or not.
1099 struct ldlm_resource {
1100 struct ldlm_ns_bucket *lr_ns_bucket;
1103 * List item for list in namespace hash.
1104 * protected by ns_lock
1106 cfs_hlist_node_t lr_hash;
1108 /** Spinlock to protect locks under this resource. */
1112 * protected by lr_lock
1114 /** List of locks in granted state */
1115 cfs_list_t lr_granted;
1116 /** List of locks waiting to change their granted mode (converted) */
1117 cfs_list_t lr_converting;
1119 * List of locks that could not be granted due to conflicts and
1120 * that are waiting for conflicts to go away */
1121 cfs_list_t lr_waiting;
1124 /* XXX No longer needed? Remove ASAP */
1125 ldlm_mode_t lr_most_restr;
1127 /** Type of locks this resource can hold. Only one type per resource. */
1128 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1130 /** Resource name */
1131 struct ldlm_res_id lr_name;
1132 /** Reference count for this resource */
1133 cfs_atomic_t lr_refcount;
1136 * Interval trees (only for extent locks) for all modes of this resource
1138 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM];
1141 * Server-side-only lock value block elements.
1142 * To serialize lvbo_init.
1144 struct mutex lr_lvb_mutex;
1146 /** protected by lr_lock */
1149 /** When the resource was considered as contended. */
1150 cfs_time_t lr_contention_time;
1151 /** List of references to this resource. For debugging. */
1152 struct lu_ref lr_reference;
1154 struct inode *lr_lvb_inode;
1157 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1159 return lock->l_resource->lr_type == LDLM_IBITS &&
1160 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1163 static inline char *
1164 ldlm_ns_name(struct ldlm_namespace *ns)
1166 return ns->ns_rs_hash->hs_name;
1169 static inline struct ldlm_namespace *
1170 ldlm_res_to_ns(struct ldlm_resource *res)
1172 return res->lr_ns_bucket->nsb_namespace;
1175 static inline struct ldlm_namespace *
1176 ldlm_lock_to_ns(struct ldlm_lock *lock)
1178 return ldlm_res_to_ns(lock->l_resource);
1181 static inline char *
1182 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1184 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1187 static inline struct adaptive_timeout *
1188 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1190 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1193 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1195 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1197 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_init != NULL)
1198 return ns->ns_lvbo->lvbo_init(res);
1203 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1205 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1207 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1208 return ns->ns_lvbo->lvbo_size(lock);
1213 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1215 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1217 if (ns->ns_lvbo != NULL) {
1218 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1219 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1224 struct ldlm_ast_work {
1225 struct ldlm_lock *w_lock;
1227 struct ldlm_lock_desc w_desc;
1235 * Common ldlm_enqueue parameters
1237 struct ldlm_enqueue_info {
1238 __u32 ei_type; /** Type of the lock being enqueued. */
1239 __u32 ei_mode; /** Mode of the lock being enqueued. */
1240 void *ei_cb_bl; /** blocking lock callback */
1241 void *ei_cb_cp; /** lock completion callback */
1242 void *ei_cb_gl; /** lock glimpse callback */
1243 void *ei_cb_wg; /** lock weigh callback */
1244 void *ei_cbdata; /** Data to be passed into callbacks. */
1247 extern struct obd_ops ldlm_obd_ops;
1249 extern char *ldlm_lockname[];
1250 extern char *ldlm_typename[];
1251 extern char *ldlm_it2str(int it);
1254 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1255 * For the cases where we do not have actual lock to print along
1256 * with a debugging message that is ldlm-related
1258 #define LDLM_DEBUG_NOLOCK(format, a...) \
1259 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1262 * Support function for lock information printing into debug logs.
1266 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1267 CFS_CHECK_STACK(msgdata, mask, cdls); \
1269 if (((mask) & D_CANTMASK) != 0 || \
1270 ((libcfs_debug & (mask)) != 0 && \
1271 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1272 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1275 void _ldlm_lock_debug(struct ldlm_lock *lock,
1276 struct libcfs_debug_msg_data *data,
1277 const char *fmt, ...)
1278 __attribute__ ((format (printf, 3, 4)));
1281 * Rate-limited version of lock printing function.
1283 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1284 static cfs_debug_limit_state_t _ldlm_cdls; \
1285 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1286 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1289 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1290 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1292 /** Non-rate-limited lock printing function for debugging purposes. */
1293 #define LDLM_DEBUG(lock, fmt, a...) do { \
1294 if (likely(lock != NULL)) { \
1295 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1296 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1297 "### " fmt , ##a); \
1299 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1302 #else /* !LIBCFS_DEBUG */
1303 # define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) ((void)0)
1304 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
1305 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
1308 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1309 int first_enq, ldlm_error_t *err,
1310 cfs_list_t *work_list);
1313 * Return values for lock iterators.
1314 * Also used during deciding of lock grants and cancellations.
1316 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1317 #define LDLM_ITER_STOP 2 /* stop iterating */
1319 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1320 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1322 /** \defgroup ldlm_iterator Lock iterators
1324 * LDLM provides for a way to iterate through every lock on a resource or
1325 * namespace or every resource in a namespace.
1327 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1329 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1331 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1332 ldlm_iterator_t iter, void *data);
1333 /** @} ldlm_iterator */
1335 int ldlm_replay_locks(struct obd_import *imp);
1338 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1341 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1343 struct ldlm_callback_suite {
1344 ldlm_completion_callback lcs_completion;
1345 ldlm_blocking_callback lcs_blocking;
1346 ldlm_glimpse_callback lcs_glimpse;
1347 ldlm_weigh_callback lcs_weigh;
1351 #ifdef HAVE_SERVER_SUPPORT
1352 /** \defgroup ldlm_srv_ast Server AST handlers
1353 * These are AST handlers used by server code.
1354 * Their property is that they are just preparing RPCs to be sent to clients.
1357 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
1358 void *data, int flag);
1359 int ldlm_server_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1360 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
1361 int ldlm_glimpse_locks(struct ldlm_resource *res, cfs_list_t *gl_work_list);
1362 /** @} ldlm_srv_ast */
1364 /** \defgroup ldlm_handlers Server LDLM handlers
1365 * These are handler functions that should be called by "frontends" such as
1366 * MDT or OST to pass through LDLM requests to LDLM for handling
1369 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
1370 ldlm_blocking_callback, ldlm_glimpse_callback);
1371 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1372 const struct ldlm_request *dlm_req,
1373 const struct ldlm_callback_suite *cbs);
1374 int ldlm_handle_convert(struct ptlrpc_request *req);
1375 int ldlm_handle_convert0(struct ptlrpc_request *req,
1376 const struct ldlm_request *dlm_req);
1377 int ldlm_handle_cancel(struct ptlrpc_request *req);
1378 int ldlm_request_cancel(struct ptlrpc_request *req,
1379 const struct ldlm_request *dlm_req, int first);
1380 /** @} ldlm_handlers */
1382 void ldlm_revoke_export_locks(struct obd_export *exp);
1384 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1385 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
1386 int ldlm_get_ref(void);
1387 void ldlm_put_ref(void);
1388 int ldlm_init_export(struct obd_export *exp);
1389 void ldlm_destroy_export(struct obd_export *exp);
1390 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1393 #ifdef HAVE_SERVER_SUPPORT
1394 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
1396 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1397 void ldlm_lock2handle(const struct ldlm_lock *lock,
1398 struct lustre_handle *lockh);
1399 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1400 void ldlm_cancel_callback(struct ldlm_lock *);
1401 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1402 int ldlm_lock_set_data(struct lustre_handle *, void *);
1405 * Obtain a lock reference by its handle.
1407 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1409 return __ldlm_handle2lock(h, 0);
1412 #define LDLM_LOCK_REF_DEL(lock) \
1413 lu_ref_del(&lock->l_reference, "handle", cfs_current())
1415 static inline struct ldlm_lock *
1416 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1418 struct ldlm_lock *lock;
1420 lock = __ldlm_handle2lock(h, flags);
1422 LDLM_LOCK_REF_DEL(lock);
1427 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1428 * data from reqest \a r
1430 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1431 struct ptlrpc_request *r, int increase)
1433 if (ldlm_res_to_ns(res)->ns_lvbo &&
1434 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1435 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1441 int ldlm_error2errno(ldlm_error_t error);
1442 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
1443 * confuses user-space. */
1444 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1445 void ldlm_dump_export_locks(struct obd_export *exp);
1449 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1450 * __ldlm_handle2lock().
1452 #define LDLM_LOCK_PUT(lock) \
1454 LDLM_LOCK_REF_DEL(lock); \
1455 /*LDLM_DEBUG((lock), "put");*/ \
1456 ldlm_lock_put(lock); \
1460 * Release a lock reference obtained by some other means (see
1463 #define LDLM_LOCK_RELEASE(lock) \
1465 /*LDLM_DEBUG((lock), "put");*/ \
1466 ldlm_lock_put(lock); \
1469 #define LDLM_LOCK_GET(lock) \
1471 ldlm_lock_get(lock); \
1472 /*LDLM_DEBUG((lock), "get");*/ \
1476 #define ldlm_lock_list_put(head, member, count) \
1478 struct ldlm_lock *_lock, *_next; \
1480 cfs_list_for_each_entry_safe(_lock, _next, head, member) { \
1483 cfs_list_del_init(&_lock->member); \
1484 LDLM_LOCK_RELEASE(_lock); \
1489 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1490 void ldlm_lock_put(struct ldlm_lock *lock);
1491 void ldlm_lock_destroy(struct ldlm_lock *lock);
1492 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1493 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1494 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1495 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1496 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1497 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1498 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1499 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1500 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1501 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1502 const struct ldlm_res_id *, ldlm_type_t type,
1503 ldlm_policy_data_t *, ldlm_mode_t mode,
1504 struct lustre_handle *, int unref);
1505 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1507 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1509 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
1510 void ldlm_lock_cancel(struct ldlm_lock *lock);
1511 void ldlm_reprocess_all(struct ldlm_resource *res);
1512 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1513 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1514 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1517 struct ldlm_namespace *
1518 ldlm_namespace_new(struct obd_device *obd, char *name,
1519 ldlm_side_t client, ldlm_appetite_t apt,
1520 ldlm_ns_type_t ns_type);
1521 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1522 void ldlm_namespace_free(struct ldlm_namespace *ns,
1523 struct obd_import *imp, int force);
1524 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1525 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1526 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1527 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1528 void ldlm_namespace_get(struct ldlm_namespace *ns);
1529 void ldlm_namespace_put(struct ldlm_namespace *ns);
1530 int ldlm_proc_setup(void);
1532 void ldlm_proc_cleanup(void);
1534 static inline void ldlm_proc_cleanup(void) {}
1537 /* resource.c - internal */
1538 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1539 struct ldlm_resource *parent,
1540 const struct ldlm_res_id *,
1541 ldlm_type_t type, int create);
1542 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1543 int ldlm_resource_putref(struct ldlm_resource *res);
1544 void ldlm_resource_add_lock(struct ldlm_resource *res,
1546 struct ldlm_lock *lock);
1547 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1548 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1549 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1550 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1551 void ldlm_resource_dump(int level, struct ldlm_resource *);
1552 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1553 const struct ldlm_res_id *);
1555 #define LDLM_RESOURCE_ADDREF(res) do { \
1556 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1559 #define LDLM_RESOURCE_DELREF(res) do { \
1560 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1563 /* ldlm_request.c */
1564 int ldlm_expired_completion_wait(void *data);
1565 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1566 * These AST handlers are typically used for server-side local locks and are
1567 * also used by client-side lock handlers to perform minimum level base
1570 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1571 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1572 void *data, int flag);
1573 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1574 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1575 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1576 /** @} ldlm_local_ast */
1578 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1579 * These are typically used by client and server (*_local versions)
1580 * to obtain and release locks.
1582 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1583 struct ldlm_enqueue_info *einfo,
1584 const struct ldlm_res_id *res_id,
1585 ldlm_policy_data_t const *policy, __u64 *flags,
1586 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1587 struct lustre_handle *lockh, int async);
1588 int ldlm_prep_enqueue_req(struct obd_export *exp,
1589 struct ptlrpc_request *req,
1590 cfs_list_t *cancels,
1592 int ldlm_prep_elc_req(struct obd_export *exp,
1593 struct ptlrpc_request *req,
1594 int version, int opc, int canceloff,
1595 cfs_list_t *cancels, int count);
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 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1616 ldlm_cancel_flags_t flags, void *opaque);
1617 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1618 const struct ldlm_res_id *res_id,
1619 ldlm_policy_data_t *policy,
1621 ldlm_cancel_flags_t flags,
1623 int ldlm_cli_cancel_req(struct obd_export *exp, cfs_list_t *head,
1624 int count, ldlm_cancel_flags_t flags);
1625 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1626 cfs_list_t *cancels,
1627 ldlm_policy_data_t *policy,
1628 ldlm_mode_t mode, int lock_flags,
1629 ldlm_cancel_flags_t cancel_flags, void *opaque);
1630 int ldlm_cli_cancel_list_local(cfs_list_t *cancels, int count,
1631 ldlm_cancel_flags_t flags);
1632 int ldlm_cli_cancel_list(cfs_list_t *head, int count,
1633 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1634 /** @} ldlm_cli_api */
1637 /* This has to be here because recursive inclusion sucks. */
1638 int intent_disposition(struct ldlm_reply *rep, int flag);
1639 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1642 /* ioctls for trying requests */
1643 #define IOC_LDLM_TYPE 'f'
1644 #define IOC_LDLM_MIN_NR 40
1646 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1647 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1648 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1649 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1650 #define IOC_LDLM_MAX_NR 43
1653 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1654 * than one lock_res is dead-lock safe.
1656 enum lock_res_type {
1661 /** Lock resource. */
1662 static inline void lock_res(struct ldlm_resource *res)
1664 spin_lock(&res->lr_lock);
1667 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1668 static inline void lock_res_nested(struct ldlm_resource *res,
1669 enum lock_res_type mode)
1671 spin_lock_nested(&res->lr_lock, mode);
1674 /** Unlock resource. */
1675 static inline void unlock_res(struct ldlm_resource *res)
1677 spin_unlock(&res->lr_lock);
1680 /** Check if resource is already locked, assert if not. */
1681 static inline void check_res_locked(struct ldlm_resource *res)
1683 LASSERT_SPIN_LOCKED(&res->lr_lock);
1686 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1687 void unlock_res_and_lock(struct ldlm_lock *lock);
1690 /** \defgroup ldlm_pools Various LDLM pool related functions
1691 * There are not used outside of ldlm.
1694 void ldlm_pools_recalc(ldlm_side_t client);
1695 int ldlm_pools_init(void);
1696 void ldlm_pools_fini(void);
1698 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1699 int idx, ldlm_side_t client);
1700 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1701 unsigned int gfp_mask);
1702 void ldlm_pool_fini(struct ldlm_pool *pl);
1703 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1704 int ldlm_pool_recalc(struct ldlm_pool *pl);
1705 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1706 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1707 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1708 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1709 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1710 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1711 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1712 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1713 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);