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 #include <linux/lustre_dlm.h>
53 #include <lustre_lib.h>
54 #include <lustre_net.h>
55 #include <lustre_import.h>
56 #include <lustre_handles.h>
57 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
63 #define OBD_LDLM_DEVICENAME "ldlm"
65 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
66 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
67 #define LDLM_CTIME_AGE_LIMIT (10)
68 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
71 * LDLM non-error return states
76 ELDLM_LOCK_CHANGED = 300,
77 ELDLM_LOCK_ABORTED = 301,
78 ELDLM_LOCK_REPLACED = 302,
79 ELDLM_NO_LOCK_DATA = 303,
80 ELDLM_LOCK_WOULDBLOCK = 304,
82 ELDLM_NAMESPACE_EXISTS = 400,
83 ELDLM_BAD_NAMESPACE = 401
87 * LDLM namespace type.
88 * The "client" type is actually an indication that this is a narrow local view
89 * into complete namespace on the server. Such namespaces cannot make any
90 * decisions about lack of conflicts or do any autonomous lock granting without
91 * first speaking to a server.
94 LDLM_NAMESPACE_SERVER = 1 << 0,
95 LDLM_NAMESPACE_CLIENT = 1 << 1
99 * Declaration of flags sent through the wire.
101 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
104 * If the server returns one of these flags, then the lock was put on that list.
105 * If the client sends one of these flags (during recovery ONLY!), it wants the
106 * lock added to the specified list, no questions asked.
108 #define LDLM_FL_BLOCK_GRANTED 0x000002
109 #define LDLM_FL_BLOCK_CONV 0x000004
110 #define LDLM_FL_BLOCK_WAIT 0x000008
112 /* Used to be LDLM_FL_CBPENDING 0x000010 moved to non-wire flags */
114 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
115 * queued for sending. */
116 /* Used to be LDLM_FL_WAIT_NOREPROC 0x000040 moved to non-wire flags */
117 /* Used to be LDLM_FL_CANCEL 0x000080 moved to non-wire flags */
120 * Lock is being replayed. This could probably be implied by the fact that one
121 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous.
123 #define LDLM_FL_REPLAY 0x000100
125 #define LDLM_FL_INTENT_ONLY 0x000200 /* Don't grant lock, just do intent. */
127 /* Used to be LDLM_FL_LOCAL_ONLY 0x000400 moved to non-wire flags */
128 /* Used to be LDLM_FL_FAILED 0x000800 moved to non-wire flags */
130 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
132 /* Used to be LDLM_FL_CANCELING 0x002000 moved to non-wire flags */
133 /* Used to be LDLM_FL_LOCAL 0x004000 moved to non-wire flags */
135 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
137 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
140 /** file & record locking */
141 #define LDLM_FL_BLOCK_NOWAIT 0x040000 /* Server told not to wait if blocked.
142 * For AGL, OST will not send glimpse
144 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
146 /* Used to be LDLM_FL_LVB_READY 0x100000 moved to non-wire flags */
147 /* Used to be LDLM_FL_KMS_IGNORE 0x200000 moved to non-wire flags */
148 /* Used to be LDLM_FL_NO_LRU 0x400000 moved to non-wire flags */
150 /* Immediatelly cancel such locks when they block some other locks. Send
151 * cancel notification to original lock holder, but expect no reply. This is
152 * for clients (like liblustre) that cannot be expected to reliably response
153 * to blocking AST. */
154 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
156 /* Flags flags inherited from parent lock when doing intents. */
157 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
159 /* Used to be LDLM_FL_CP_REQD 0x1000000 moved to non-wire flags */
160 /* Used to be LDLM_FL_CLEANED 0x2000000 moved to non-wire flags */
161 /* Used to be LDLM_FL_ATOMIC_CB 0x4000000 moved to non-wire flags */
162 /* Used to be LDLM_FL_BL_AST 0x10000000 moved to non-wire flags */
163 /* Used to be LDLM_FL_BL_DONE 0x20000000 moved to non-wire flags */
165 /* measure lock contention and return -EUSERS if locking contention is high */
166 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
168 /* These are flags that are mapped into the flags and ASTs of blocking locks */
169 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
171 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
172 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
175 * --------------------------------------------------------------------------
176 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
177 * 0x80000000 will not be sent over the wire.
178 * --------------------------------------------------------------------------
182 * Declaration of flags not sent through the wire.
186 * Used for marking lock as a target for -EINTR while cp_ast sleep
187 * emulation + race with upcoming bl_ast.
189 #define LDLM_FL_FAIL_LOC 0x100000000ULL
192 * Used while processing the unused list to know that we have already
193 * handled this lock and decided to skip it.
195 #define LDLM_FL_SKIPPED 0x200000000ULL
196 /* this lock is being destroyed */
197 #define LDLM_FL_CBPENDING 0x400000000ULL
198 /* not a real flag, not saved in lock */
199 #define LDLM_FL_WAIT_NOREPROC 0x800000000ULL
200 /* cancellation callback already run */
201 #define LDLM_FL_CANCEL 0x1000000000ULL
202 #define LDLM_FL_LOCAL_ONLY 0x2000000000ULL
203 /* don't run the cancel callback under ldlm_cli_cancel_unused */
204 #define LDLM_FL_FAILED 0x4000000000ULL
205 /* lock cancel has already been sent */
206 #define LDLM_FL_CANCELING 0x8000000000ULL
207 /* local lock (ie, no srv/cli split) */
208 #define LDLM_FL_LOCAL 0x10000000000ULL
209 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
210 * the LVB filling happens _after_ the lock has been granted, so another thread
211 * can match it before the LVB has been updated. As a dirty hack, we set
212 * LDLM_FL_LVB_READY only after we've done the LVB poop.
213 * this is only needed on LOV/OSC now, where LVB is actually used and callers
214 * must set it in input flags.
216 * The proper fix is to do the granting inside of the completion AST, which can
217 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
218 * pretty high-risk, though, and would need a lot more testing. */
219 #define LDLM_FL_LVB_READY 0x20000000000ULL
220 /* A lock contributes to the known minimum size (KMS) calculation until it has
221 * finished the part of its cancelation that performs write back on its dirty
222 * pages. It can remain on the granted list during this whole time. Threads
223 * racing to update the KMS after performing their writeback need to know to
224 * exclude each other's locks from the calculation as they walk the granted
226 #define LDLM_FL_KMS_IGNORE 0x40000000000ULL
227 /* completion AST to be executed */
228 #define LDLM_FL_CP_REQD 0x80000000000ULL
229 /* cleanup_resource has already handled the lock */
230 #define LDLM_FL_CLEANED 0x100000000000ULL
231 /* optimization hint: LDLM can run blocking callback from current context
232 * w/o involving separate thread. in order to decrease cs rate */
233 #define LDLM_FL_ATOMIC_CB 0x200000000000ULL
235 /* It may happen that a client initiates two operations, e.g. unlink and
236 * mkdir, such that the server sends a blocking AST for conflicting
237 * locks to this client for the first operation, whereas the second
238 * operation has canceled this lock and is waiting for rpc_lock which is
239 * taken by the first operation. LDLM_FL_BL_AST is set by
240 * ldlm_callback_handler() in the lock to prevent the Early Lock Cancel
241 * (ELC) code from cancelling it.
243 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock
244 * cache is dropped to let ldlm_callback_handler() return EINVAL to the
245 * server. It is used when ELC RPC is already prepared and is waiting
246 * for rpc_lock, too late to send a separate CANCEL RPC. */
247 #define LDLM_FL_BL_AST 0x400000000000ULL
248 #define LDLM_FL_BL_DONE 0x800000000000ULL
249 /* Don't put lock into the LRU list, so that it is not canceled due to aging.
250 * Used by MGC locks, they are cancelled only at unmount or by callback. */
251 #define LDLM_FL_NO_LRU 0x1000000000000ULL
254 * The blocking callback is overloaded to perform two functions. These flags
255 * indicate which operation should be performed.
257 #define LDLM_CB_BLOCKING 1
258 #define LDLM_CB_CANCELING 2
261 * \name Lock Compatibility Matrix.
263 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
264 * Lock types are described in their respective implementation files:
265 * ldlm_{extent,flock,inodebits,plain}.c.
267 * There are six lock modes along with a compatibility matrix to indicate if
268 * two locks are compatible.
270 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
272 * - PW: Protective Write (normal write) mode. When a client requests a write
273 * lock from an OST, a lock with PW mode will be issued.
274 * - PR: Protective Read (normal read) mode. When a client requests a read from
275 * an OST, a lock with PR mode will be issued. Also, if the client opens a
276 * file for execution, it is granted a lock with PR mode.
277 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
278 * requests a write lock during a file open operation.
279 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
280 * an inodebit lock with the CR mode on the intermediate path component.
294 #define LCK_COMPAT_EX LCK_NL
295 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
296 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
297 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
298 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
299 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
300 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
301 #define LCK_COMPAT_COS (LCK_COS)
302 /** @} Lock Compatibility Matrix */
304 extern ldlm_mode_t lck_compat_array[];
306 static inline void lockmode_verify(ldlm_mode_t mode)
308 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
311 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
313 return (lck_compat_array[exist_mode] & new_mode);
318 * cluster name spaces
322 #define DLM_OST_NAMESPACE 1
323 #define DLM_MDS_NAMESPACE 2
326 - do we just separate this by security domains and use a prefix for
327 multiple namespaces in the same domain?
332 * Locking rules for LDLM:
337 * waiting_locks_spinlock
352 struct ldlm_resource;
353 struct ldlm_namespace;
356 * Operations on LDLM pools.
357 * LDLM pool is a pool of locks in the namespace without any implicitly
359 * Locks in the pool are organized in LRU.
360 * Local memory pressure or server instructions (e.g. mempressure on server)
361 * can trigger freeing of locks from the pool
363 struct ldlm_pool_ops {
364 /** Recalculate pool \a pl usage */
365 int (*po_recalc)(struct ldlm_pool *pl);
366 /** Cancel at least \a nr locks from pool \a pl */
367 int (*po_shrink)(struct ldlm_pool *pl, int nr,
368 unsigned int gfp_mask);
369 int (*po_setup)(struct ldlm_pool *pl, int limit);
372 /** One second for pools thread check interval. Each pool has own period. */
373 #define LDLM_POOLS_THREAD_PERIOD (1)
375 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
376 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
378 /** Default recalc period for server side pools in sec. */
379 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
381 /** Default recalc period for client side pools in sec. */
382 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
385 * LDLM pool structure to track granted locks.
386 * For purposes of determining when to release locks on e.g. memory pressure.
387 * This feature is commonly referred to as lru_resize.
390 /** Pool proc directory. */
391 proc_dir_entry_t *pl_proc_dir;
392 /** Pool name, must be long enough to hold compound proc entry name. */
394 /** Lock for protecting SLV/CLV updates. */
396 /** Number of allowed locks in in pool, both, client and server side. */
398 /** Number of granted locks in */
400 /** Grant rate per T. */
401 atomic_t pl_grant_rate;
402 /** Cancel rate per T. */
403 atomic_t pl_cancel_rate;
404 /** Server lock volume (SLV). Protected by pl_lock. */
405 __u64 pl_server_lock_volume;
406 /** Current biggest client lock volume. Protected by pl_lock. */
407 __u64 pl_client_lock_volume;
408 /** Lock volume factor. SLV on client is calculated as following:
409 * server_slv * lock_volume_factor. */
410 atomic_t pl_lock_volume_factor;
411 /** Time when last SLV from server was obtained. */
412 time_t pl_recalc_time;
413 /** Recalculation period for pool. */
414 time_t pl_recalc_period;
415 /** Recalculation and shrink operations. */
416 struct ldlm_pool_ops *pl_ops;
417 /** Number of planned locks for next period. */
419 /** Pool statistics. */
420 struct lprocfs_stats *pl_stats;
423 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
424 void *req_cookie, ldlm_mode_t mode, __u64 flags,
427 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
431 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
432 * be associated with an LDLM lock and transferred from client to server and
435 * Currently LVBs are used by:
436 * - OSC-OST code to maintain current object size/times
437 * - layout lock code to return the layout when the layout lock is granted
439 struct ldlm_valblock_ops {
440 int (*lvbo_init)(struct ldlm_resource *res);
441 int (*lvbo_update)(struct ldlm_resource *res,
442 struct ptlrpc_request *r,
444 int (*lvbo_free)(struct ldlm_resource *res);
445 /* Return size of lvb data appropriate RPC size can be reserved */
446 int (*lvbo_size)(struct ldlm_lock *lock);
447 /* Called to fill in lvb data to RPC buffer @buf */
448 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
452 * LDLM pools related, type of lock pool in the namespace.
453 * Greedy means release cached locks aggressively
456 LDLM_NAMESPACE_GREEDY = 1 << 0,
457 LDLM_NAMESPACE_MODEST = 1 << 1
461 * Default values for the "max_nolock_size", "contention_time" and
462 * "contended_locks" namespace tunables.
464 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
465 #define NS_DEFAULT_CONTENTION_SECONDS 2
466 #define NS_DEFAULT_CONTENDED_LOCKS 32
468 struct ldlm_ns_bucket {
469 /** back pointer to namespace */
470 struct ldlm_namespace *nsb_namespace;
472 * Estimated lock callback time. Used by adaptive timeout code to
473 * avoid spurious client evictions due to unresponsiveness when in
474 * fact the network or overall system load is at fault
476 struct adaptive_timeout nsb_at_estimate;
480 /** LDLM namespace lock stats */
487 LDLM_NS_TYPE_UNKNOWN = 0,
505 * Namespace serves to contain locks related to a particular service.
506 * There are two kinds of namespaces:
507 * - Server namespace has knowledge of all locks and is therefore authoritative
508 * to make decisions like what locks could be granted and what conflicts
509 * exist during new lock enqueue.
510 * - Client namespace only has limited knowledge about locks in the namespace,
511 * only seeing locks held by the client.
513 * Every Lustre service has one server namespace present on the server serving
514 * that service. Every client connected to the service has a client namespace
516 * Every lock obtained by client in that namespace is actually represented by
517 * two in-memory locks. One on the server and one on the client. The locks are
518 * linked by a special cookie by which one node can tell to the other which lock
519 * it actually means during communications. Such locks are called remote locks.
520 * The locks held by server only without any reference to a client are called
523 struct ldlm_namespace {
524 /** Backward link to OBD, required for LDLM pool to store new SLV. */
525 struct obd_device *ns_obd;
527 /** Flag indicating if namespace is on client instead of server */
528 ldlm_side_t ns_client;
530 /** Resource hash table for namespace. */
531 cfs_hash_t *ns_rs_hash;
536 /** big refcount (by bucket) */
540 * Namespace connect flags supported by server (may be changed via
541 * /proc, LRU resize may be disabled/enabled).
543 __u64 ns_connect_flags;
545 /** Client side original connect flags supported by server. */
546 __u64 ns_orig_connect_flags;
548 /* namespace proc dir entry */
549 struct proc_dir_entry *ns_proc_dir_entry;
552 * Position in global namespace list linking all namespaces on
555 struct list_head ns_list_chain;
558 * List of unused locks for this namespace. This list is also called
560 * Unused locks are locks with zero reader/writer reference counts.
561 * This list is only used on clients for lock caching purposes.
562 * When we want to release some locks voluntarily or if server wants
563 * us to release some locks due to e.g. memory pressure, we take locks
564 * to release from the head of this list.
565 * Locks are linked via l_lru field in \see struct ldlm_lock.
567 struct list_head ns_unused_list;
568 /** Number of locks in the LRU list above */
572 * Maximum number of locks permitted in the LRU. If 0, means locks
573 * are managed by pools and there is no preset limit, rather it is all
574 * controlled by available memory on this client and on server.
576 unsigned int ns_max_unused;
577 /** Maximum allowed age (last used time) for locks in the LRU */
578 unsigned int ns_max_age;
580 * Server only: number of times we evicted clients due to lack of reply
583 unsigned int ns_timeouts;
585 * Number of seconds since the file change time after which the
586 * MDT will return an UPDATE lock along with a LOOKUP lock.
587 * This allows the client to start caching negative dentries
588 * for a directory and may save an RPC for a later stat.
590 unsigned int ns_ctime_age_limit;
593 * Used to rate-limit ldlm_namespace_dump calls.
594 * \see ldlm_namespace_dump. Increased by 10 seconds every time
597 cfs_time_t ns_next_dump;
599 /** "policy" function that does actual lock conflict determination */
600 ldlm_res_policy ns_policy;
603 * LVB operations for this namespace.
604 * \see struct ldlm_valblock_ops
606 struct ldlm_valblock_ops *ns_lvbo;
609 * Used by filter code to store pointer to OBD of the service.
610 * Should be dropped in favor of \a ns_obd
615 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
616 * a resource is removed.
618 wait_queue_head_t ns_waitq;
619 /** LDLM pool structure for this namespace */
620 struct ldlm_pool ns_pool;
621 /** Definition of how eagerly unused locks will be released from LRU */
622 ldlm_appetite_t ns_appetite;
625 * If more than \a ns_contended_locks are found, the resource is
626 * considered to be contended. Lock enqueues might specify that no
627 * contended locks should be granted
629 unsigned ns_contended_locks;
632 * The resources in this namespace remember contended state during
633 * \a ns_contention_time, in seconds.
635 unsigned ns_contention_time;
638 * Limit size of contended extent locks, in bytes.
639 * If extended lock is requested for more then this many bytes and
640 * caller instructs us not to grant contended locks, we would disregard
643 unsigned ns_max_nolock_size;
645 /** Limit of parallel AST RPC count. */
646 unsigned ns_max_parallel_ast;
648 /** Callback to cancel locks before replaying it during recovery. */
649 ldlm_cancel_for_recovery ns_cancel_for_recovery;
651 /** LDLM lock stats */
652 struct lprocfs_stats *ns_stats;
655 * Flag to indicate namespace is being freed. Used to determine if
656 * recalculation of LDLM pool statistics should be skipped.
658 unsigned ns_stopping:1;
662 * Returns 1 if namespace \a ns is a client namespace.
664 static inline int ns_is_client(struct ldlm_namespace *ns)
667 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
668 LDLM_NAMESPACE_SERVER)));
669 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
670 ns->ns_client == LDLM_NAMESPACE_SERVER);
671 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
675 * Returns 1 if namespace \a ns is a server namespace.
677 static inline int ns_is_server(struct ldlm_namespace *ns)
680 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
681 LDLM_NAMESPACE_SERVER)));
682 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
683 ns->ns_client == LDLM_NAMESPACE_SERVER);
684 return ns->ns_client == LDLM_NAMESPACE_SERVER;
688 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
690 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
693 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
697 * Returns 1 if this namespace supports lru_resize.
699 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
702 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
705 static inline void ns_register_cancel(struct ldlm_namespace *ns,
706 ldlm_cancel_for_recovery arg)
709 ns->ns_cancel_for_recovery = arg;
714 /** Type for blocking callback function of a lock. */
715 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
716 struct ldlm_lock_desc *new, void *data,
718 /** Type for completion callback function of a lock. */
719 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
721 /** Type for glimpse callback function of a lock. */
722 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
723 /** Type for weight callback function of a lock. */
724 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
726 /** Work list for sending GL ASTs to multiple locks. */
727 struct ldlm_glimpse_work {
728 struct ldlm_lock *gl_lock; /* lock to glimpse */
729 struct list_head gl_list; /* linkage to other gl work structs */
730 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
731 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
732 * glimpse callback request */
735 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
736 #define LDLM_GL_WORK_NOFREE 0x1
738 /** Interval node data for each LDLM_EXTENT lock. */
739 struct ldlm_interval {
740 struct interval_node li_node; /* node for tree management */
741 struct list_head li_group; /* the locks which have the same
742 * policy - group of the policy */
744 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
747 * Interval tree for extent locks.
748 * The interval tree must be accessed under the resource lock.
749 * Interval trees are used for granted extent locks to speed up conflicts
750 * lookup. See ldlm/interval_tree.c for more details.
752 struct ldlm_interval_tree {
755 ldlm_mode_t lit_mode; /* lock mode */
756 struct interval_node *lit_root; /* actual ldlm_interval */
759 /** Whether to track references to exports by LDLM locks. */
760 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
764 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
765 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
766 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
768 } ldlm_cancel_flags_t;
774 __u64 blocking_owner;
775 struct obd_export *blocking_export;
776 /* Protected by the hash lock */
782 struct ldlm_extent l_extent;
783 struct ldlm_flock l_flock;
784 struct ldlm_inodebits l_inodebits;
785 } ldlm_policy_data_t;
787 void ldlm_convert_policy_to_wire(ldlm_type_t type,
788 const ldlm_policy_data_t *lpolicy,
789 ldlm_wire_policy_data_t *wpolicy);
790 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
791 const ldlm_wire_policy_data_t *wpolicy,
792 ldlm_policy_data_t *lpolicy);
802 * LDLM lock structure
804 * Represents a single LDLM lock and its state in memory. Each lock is
805 * associated with a single ldlm_resource, the object which is being
806 * locked. There may be multiple ldlm_locks on a single resource,
807 * depending on the lock type and whether the locks are conflicting or
813 * When remote side wants to tell us about a lock, they address
814 * it by this opaque handle. The handle does not hold a
815 * reference on the ldlm_lock, so it can be safely passed to
816 * other threads or nodes. When the lock needs to be accessed
817 * from the handle, it is looked up again in the lock table, and
818 * may no longer exist.
820 * Must be first in the structure.
822 struct portals_handle l_handle;
824 * Lock reference count.
825 * This is how many users have pointers to actual structure, so that
826 * we do not accidentally free lock structure that is in use.
830 * Internal spinlock protects l_resource. We should hold this lock
831 * first before taking res_lock.
835 * Pointer to actual resource this lock is in.
836 * ldlm_lock_change_resource() can change this.
838 struct ldlm_resource *l_resource;
840 * List item for client side LRU list.
841 * Protected by ns_lock in struct ldlm_namespace.
843 struct list_head l_lru;
845 * Linkage to resource's lock queues according to current lock state.
846 * (could be granted, waiting or converting)
847 * Protected by lr_lock in struct ldlm_resource.
849 struct list_head l_res_link;
851 * Tree node for ldlm_extent.
853 struct ldlm_interval *l_tree_node;
855 * Per export hash of locks.
856 * Protected by per-bucket exp->exp_lock_hash locks.
858 struct hlist_node l_exp_hash;
860 * Per export hash of flock locks.
861 * Protected by per-bucket exp->exp_flock_hash locks.
863 struct hlist_node l_exp_flock_hash;
866 * Protected by lr_lock.
868 ldlm_mode_t l_req_mode;
870 * Granted mode, also protected by lr_lock.
872 ldlm_mode_t l_granted_mode;
873 /** Lock completion handler pointer. Called when lock is granted. */
874 ldlm_completion_callback l_completion_ast;
876 * Lock blocking AST handler pointer.
877 * It plays two roles:
878 * - as a notification of an attempt to queue a conflicting lock (once)
879 * - as a notification when the lock is being cancelled.
881 * As such it's typically called twice: once for the initial conflict
882 * and then once more when the last user went away and the lock is
883 * cancelled (could happen recursively).
885 ldlm_blocking_callback l_blocking_ast;
887 * Lock glimpse handler.
888 * Glimpse handler is used to obtain LVB updates from a client by
891 ldlm_glimpse_callback l_glimpse_ast;
893 /** XXX apparently unused "weight" handler. To be removed? */
894 ldlm_weigh_callback l_weigh_ast;
898 * This is a pointer to actual client export for locks that were granted
899 * to clients. Used server-side.
901 struct obd_export *l_export;
903 * Lock connection export.
904 * Pointer to server export on a client.
906 struct obd_export *l_conn_export;
909 * Remote lock handle.
910 * If the lock is remote, this is the handle of the other side lock
913 struct lustre_handle l_remote_handle;
916 * Representation of private data specific for a lock type.
917 * Examples are: extent range for extent lock or bitmask for ibits locks
919 ldlm_policy_data_t l_policy_data;
923 * Like whenever we receive any blocking requests for this lock, etc.
924 * Protected by lr_lock.
928 * Lock r/w usage counters.
929 * Protected by lr_lock.
934 * If the lock is granted, a process sleeps on this waitq to learn when
935 * it's no longer in use. If the lock is not granted, a process sleeps
936 * on this waitq to learn when it becomes granted.
938 wait_queue_head_t l_waitq;
941 * Seconds. It will be updated if there is any activity related to
942 * the lock, e.g. enqueue the lock or send blocking AST.
944 cfs_time_t l_last_activity;
947 * Time last used by e.g. being matched by lock match.
948 * Jiffies. Should be converted to time if needed.
950 cfs_time_t l_last_used;
952 /** Originally requested extent for the extent lock. */
953 struct ldlm_extent l_req_extent;
955 unsigned int l_failed:1,
957 * Set for locks that were removed from class hash table and will be
958 * destroyed when last reference to them is released. Set by
959 * ldlm_lock_destroy_internal().
961 * Protected by lock and resource locks.
965 * it's set in lock_res_and_lock() and unset in unlock_res_and_lock().
967 * NB: compared with check_res_locked(), checking this bit is cheaper.
968 * Also, spin_is_locked() is deprecated for kernel code; one reason is
969 * because it works only for SMP so user needs to add extra macros like
970 * LASSERT_SPIN_LOCKED for uniprocessor kernels.
974 * It's set once we call ldlm_add_waiting_lock_res_locked()
975 * to start the lock-timeout timer and it will never be reset.
977 * Protected by lock_res_and_lock().
980 /** Flag whether this is a server namespace lock. */
984 * Client-side-only members.
987 enum lvb_type l_lvb_type;
990 * Temporary storage for a LVB received during an enqueue operation.
995 /** Private storage for lock user. Opaque to LDLM. */
999 * Server-side-only members.
1003 * Connection cookie for the client originating the operation.
1004 * Used by Commit on Share (COS) code. Currently only used for
1005 * inodebits locks on MDS.
1007 __u64 l_client_cookie;
1010 * List item for locks waiting for cancellation from clients.
1011 * The lists this could be linked into are:
1012 * waiting_locks_list (protected by waiting_locks_spinlock),
1013 * then if the lock timed out, it is moved to
1014 * expired_lock_thread.elt_expired_locks for further processing.
1015 * Protected by elt_lock.
1017 struct list_head l_pending_chain;
1020 * Set when lock is sent a blocking AST. Time in seconds when timeout
1021 * is reached and client holding this lock could be evicted.
1022 * This timeout could be further extended by e.g. certain IO activity
1024 * \see ost_rw_prolong_locks
1026 cfs_time_t l_callback_timeout;
1028 /** Local PID of process which created this lock. */
1032 * Number of times blocking AST was sent for this lock.
1033 * This is for debugging. Valid values are 0 and 1, if there is an
1034 * attempt to send blocking AST more than once, an assertion would be
1035 * hit. \see ldlm_work_bl_ast_lock
1038 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
1039 struct list_head l_bl_ast;
1040 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
1041 struct list_head l_cp_ast;
1042 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
1043 struct list_head l_rk_ast;
1046 * Pointer to a conflicting lock that caused blocking AST to be sent
1049 struct ldlm_lock *l_blocking_lock;
1052 * Protected by lr_lock, linkages to "skip lists".
1053 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
1055 struct list_head l_sl_mode;
1056 struct list_head l_sl_policy;
1058 /** Reference tracking structure to debug leaked locks. */
1059 struct lu_ref l_reference;
1060 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1061 /* Debugging stuff for bug 20498, for tracking export references. */
1062 /** number of export references taken */
1064 /** link all locks referencing one export */
1065 struct list_head l_exp_refs_link;
1066 /** referenced export object */
1067 struct obd_export *l_exp_refs_target;
1070 * export blocking dlm lock list, protected by
1071 * l_export->exp_bl_list_lock.
1072 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
1073 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
1075 struct list_head l_exp_list;
1079 * LDLM resource description.
1080 * Basically, resource is a representation for a single object.
1081 * Object has a name which is currently 4 64-bit integers. LDLM user is
1082 * responsible for creation of a mapping between objects it wants to be
1083 * protected and resource names.
1085 * A resource can only hold locks of a single lock type, though there may be
1086 * multiple ldlm_locks on a single resource, depending on the lock type and
1087 * whether the locks are conflicting or not.
1089 struct ldlm_resource {
1090 struct ldlm_ns_bucket *lr_ns_bucket;
1093 * List item for list in namespace hash.
1094 * protected by ns_lock
1096 struct hlist_node lr_hash;
1098 /** Spinlock to protect locks under this resource. */
1102 * protected by lr_lock
1104 /** List of locks in granted state */
1105 struct list_head lr_granted;
1106 /** List of locks waiting to change their granted mode (converted) */
1107 struct list_head lr_converting;
1109 * List of locks that could not be granted due to conflicts and
1110 * that are waiting for conflicts to go away */
1111 struct list_head lr_waiting;
1114 /* XXX No longer needed? Remove ASAP */
1115 ldlm_mode_t lr_most_restr;
1117 /** Type of locks this resource can hold. Only one type per resource. */
1118 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
1120 /** Resource name */
1121 struct ldlm_res_id lr_name;
1122 /** Reference count for this resource */
1123 atomic_t lr_refcount;
1126 * Interval trees (only for extent locks) for all modes of this resource
1128 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM];
1131 * Server-side-only lock value block elements.
1132 * To serialize lvbo_init.
1134 struct mutex lr_lvb_mutex;
1136 /** protected by lr_lock */
1139 /** When the resource was considered as contended. */
1140 cfs_time_t lr_contention_time;
1141 /** List of references to this resource. For debugging. */
1142 struct lu_ref lr_reference;
1144 struct inode *lr_lvb_inode;
1147 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
1149 return lock->l_resource->lr_type == LDLM_IBITS &&
1150 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
1153 static inline char *
1154 ldlm_ns_name(struct ldlm_namespace *ns)
1156 return ns->ns_rs_hash->hs_name;
1159 static inline struct ldlm_namespace *
1160 ldlm_res_to_ns(struct ldlm_resource *res)
1162 return res->lr_ns_bucket->nsb_namespace;
1165 static inline struct ldlm_namespace *
1166 ldlm_lock_to_ns(struct ldlm_lock *lock)
1168 return ldlm_res_to_ns(lock->l_resource);
1171 static inline char *
1172 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
1174 return ldlm_ns_name(ldlm_lock_to_ns(lock));
1177 static inline struct adaptive_timeout *
1178 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
1180 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
1183 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
1185 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
1187 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_init != NULL)
1188 return ns->ns_lvbo->lvbo_init(res);
1193 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1195 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1197 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1198 return ns->ns_lvbo->lvbo_size(lock);
1203 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1205 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1207 if (ns->ns_lvbo != NULL) {
1208 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1209 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1214 struct ldlm_ast_work {
1215 struct ldlm_lock *w_lock;
1217 struct ldlm_lock_desc w_desc;
1218 struct list_head w_list;
1225 * Common ldlm_enqueue parameters
1227 struct ldlm_enqueue_info {
1228 __u32 ei_type; /** Type of the lock being enqueued. */
1229 __u32 ei_mode; /** Mode of the lock being enqueued. */
1230 void *ei_cb_bl; /** blocking lock callback */
1231 void *ei_cb_cp; /** lock completion callback */
1232 void *ei_cb_gl; /** lock glimpse callback */
1233 void *ei_cb_wg; /** lock weigh callback */
1234 void *ei_cbdata; /** Data to be passed into callbacks. */
1237 extern struct obd_ops ldlm_obd_ops;
1239 extern char *ldlm_lockname[];
1240 extern char *ldlm_typename[];
1241 extern char *ldlm_it2str(int it);
1244 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1245 * For the cases where we do not have actual lock to print along
1246 * with a debugging message that is ldlm-related
1248 #define LDLM_DEBUG_NOLOCK(format, a...) \
1249 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1252 * Support function for lock information printing into debug logs.
1255 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1256 CFS_CHECK_STACK(msgdata, mask, cdls); \
1258 if (((mask) & D_CANTMASK) != 0 || \
1259 ((libcfs_debug & (mask)) != 0 && \
1260 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1261 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1264 void _ldlm_lock_debug(struct ldlm_lock *lock,
1265 struct libcfs_debug_msg_data *data,
1266 const char *fmt, ...)
1267 __attribute__ ((format (printf, 3, 4)));
1270 * Rate-limited version of lock printing function.
1272 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1273 static cfs_debug_limit_state_t _ldlm_cdls; \
1274 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1275 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1278 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1279 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1281 /** Non-rate-limited lock printing function for debugging purposes. */
1282 #define LDLM_DEBUG(lock, fmt, a...) do { \
1283 if (likely(lock != NULL)) { \
1284 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1285 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1286 "### " fmt , ##a); \
1288 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1292 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1293 int first_enq, ldlm_error_t *err,
1294 struct list_head *work_list);
1297 * Return values for lock iterators.
1298 * Also used during deciding of lock grants and cancellations.
1300 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1301 #define LDLM_ITER_STOP 2 /* stop iterating */
1303 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1304 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1306 /** \defgroup ldlm_iterator Lock iterators
1308 * LDLM provides for a way to iterate through every lock on a resource or
1309 * namespace or every resource in a namespace.
1311 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1313 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1315 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1316 ldlm_iterator_t iter, void *data);
1317 /** @} ldlm_iterator */
1319 int ldlm_replay_locks(struct obd_import *imp);
1322 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1325 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1327 struct ldlm_callback_suite {
1328 ldlm_completion_callback lcs_completion;
1329 ldlm_blocking_callback lcs_blocking;
1330 ldlm_glimpse_callback lcs_glimpse;
1331 ldlm_weigh_callback lcs_weigh;
1335 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1336 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
1337 int ldlm_get_ref(void);
1338 void ldlm_put_ref(void);
1339 int ldlm_init_export(struct obd_export *exp);
1340 void ldlm_destroy_export(struct obd_export *exp);
1341 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1344 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1345 void ldlm_lock2handle(const struct ldlm_lock *lock,
1346 struct lustre_handle *lockh);
1347 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1348 void ldlm_cancel_callback(struct ldlm_lock *);
1349 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1350 int ldlm_lock_set_data(struct lustre_handle *, void *);
1353 * Obtain a lock reference by its handle.
1355 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1357 return __ldlm_handle2lock(h, 0);
1360 #define LDLM_LOCK_REF_DEL(lock) \
1361 lu_ref_del(&lock->l_reference, "handle", current)
1363 static inline struct ldlm_lock *
1364 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1366 struct ldlm_lock *lock;
1368 lock = __ldlm_handle2lock(h, flags);
1370 LDLM_LOCK_REF_DEL(lock);
1375 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1376 * data from reqest \a r
1378 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1379 struct ptlrpc_request *r, int increase)
1381 if (ldlm_res_to_ns(res)->ns_lvbo &&
1382 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1383 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1389 int ldlm_error2errno(ldlm_error_t error);
1390 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
1391 * confuses user-space. */
1392 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1393 void ldlm_dump_export_locks(struct obd_export *exp);
1397 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1398 * __ldlm_handle2lock().
1400 #define LDLM_LOCK_PUT(lock) \
1402 LDLM_LOCK_REF_DEL(lock); \
1403 /*LDLM_DEBUG((lock), "put");*/ \
1404 ldlm_lock_put(lock); \
1408 * Release a lock reference obtained by some other means (see
1411 #define LDLM_LOCK_RELEASE(lock) \
1413 /*LDLM_DEBUG((lock), "put");*/ \
1414 ldlm_lock_put(lock); \
1417 #define LDLM_LOCK_GET(lock) \
1419 ldlm_lock_get(lock); \
1420 /*LDLM_DEBUG((lock), "get");*/ \
1424 #define ldlm_lock_list_put(head, member, count) \
1426 struct ldlm_lock *_lock, *_next; \
1428 list_for_each_entry_safe(_lock, _next, head, member) { \
1431 list_del_init(&_lock->member); \
1432 LDLM_LOCK_RELEASE(_lock); \
1437 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1438 void ldlm_lock_put(struct ldlm_lock *lock);
1439 void ldlm_lock_destroy(struct ldlm_lock *lock);
1440 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1441 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1442 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1443 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1444 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1445 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1446 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1447 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1448 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1449 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1450 const struct ldlm_res_id *, ldlm_type_t type,
1451 ldlm_policy_data_t *, ldlm_mode_t mode,
1452 struct lustre_handle *, int unref);
1453 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1455 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1457 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
1458 void ldlm_lock_cancel(struct ldlm_lock *lock);
1459 void ldlm_reprocess_all(struct ldlm_resource *res);
1460 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1461 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1462 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1465 struct ldlm_namespace *
1466 ldlm_namespace_new(struct obd_device *obd, char *name,
1467 ldlm_side_t client, ldlm_appetite_t apt,
1468 ldlm_ns_type_t ns_type);
1469 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1470 void ldlm_namespace_free(struct ldlm_namespace *ns,
1471 struct obd_import *imp, int force);
1472 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1473 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1474 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1475 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1476 void ldlm_namespace_get(struct ldlm_namespace *ns);
1477 void ldlm_namespace_put(struct ldlm_namespace *ns);
1478 int ldlm_proc_setup(void);
1480 void ldlm_proc_cleanup(void);
1482 static inline void ldlm_proc_cleanup(void) {}
1485 /* resource.c - internal */
1486 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1487 struct ldlm_resource *parent,
1488 const struct ldlm_res_id *,
1489 ldlm_type_t type, int create);
1490 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1491 int ldlm_resource_putref(struct ldlm_resource *res);
1492 void ldlm_resource_add_lock(struct ldlm_resource *res,
1493 struct list_head *head,
1494 struct ldlm_lock *lock);
1495 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1496 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1497 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1498 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1499 void ldlm_resource_dump(int level, struct ldlm_resource *);
1500 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1501 const struct ldlm_res_id *);
1503 #define LDLM_RESOURCE_ADDREF(res) do { \
1504 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, current); \
1507 #define LDLM_RESOURCE_DELREF(res) do { \
1508 lu_ref_del(&(res)->lr_reference, __FUNCTION__, current); \
1511 /* ldlm_request.c */
1512 int ldlm_expired_completion_wait(void *data);
1513 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1514 * These AST handlers are typically used for server-side local locks and are
1515 * also used by client-side lock handlers to perform minimum level base
1518 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1519 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1520 void *data, int flag);
1521 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1522 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1523 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1524 /** @} ldlm_local_ast */
1526 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1527 * These are typically used by client and server (*_local versions)
1528 * to obtain and release locks.
1530 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1531 struct ldlm_enqueue_info *einfo,
1532 const struct ldlm_res_id *res_id,
1533 ldlm_policy_data_t const *policy, __u64 *flags,
1534 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1535 struct lustre_handle *lockh, int async);
1536 int ldlm_prep_enqueue_req(struct obd_export *exp,
1537 struct ptlrpc_request *req,
1538 struct list_head *cancels,
1540 int ldlm_prep_elc_req(struct obd_export *exp,
1541 struct ptlrpc_request *req,
1542 int version, int opc, int canceloff,
1543 struct list_head *cancels, int count);
1545 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1546 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1547 const struct ldlm_request *dlm_req,
1548 const struct ldlm_callback_suite *cbs);
1549 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1550 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1551 __u64 *flags, void *lvb, __u32 lvb_len,
1552 struct lustre_handle *lockh, int rc);
1553 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1554 const struct ldlm_res_id *res_id,
1555 ldlm_type_t type, ldlm_policy_data_t *policy,
1556 ldlm_mode_t mode, __u64 *flags,
1557 ldlm_blocking_callback blocking,
1558 ldlm_completion_callback completion,
1559 ldlm_glimpse_callback glimpse,
1560 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1561 const __u64 *client_cookie,
1562 struct lustre_handle *lockh);
1563 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1564 void *data, __u32 data_len);
1565 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1566 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1567 int ldlm_cli_cancel(struct lustre_handle *lockh,
1568 ldlm_cancel_flags_t cancel_flags);
1569 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1570 ldlm_cancel_flags_t flags, void *opaque);
1571 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1572 const struct ldlm_res_id *res_id,
1573 ldlm_policy_data_t *policy,
1575 ldlm_cancel_flags_t flags,
1577 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1578 int count, ldlm_cancel_flags_t flags);
1579 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1580 struct list_head *cancels,
1581 ldlm_policy_data_t *policy,
1582 ldlm_mode_t mode, int lock_flags,
1583 ldlm_cancel_flags_t cancel_flags, void *opaque);
1584 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1585 ldlm_cancel_flags_t flags);
1586 int ldlm_cli_cancel_list(struct list_head *head, int count,
1587 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1588 /** @} ldlm_cli_api */
1591 /* This has to be here because recursive inclusion sucks. */
1592 int intent_disposition(struct ldlm_reply *rep, int flag);
1593 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1596 /* ioctls for trying requests */
1597 #define IOC_LDLM_TYPE 'f'
1598 #define IOC_LDLM_MIN_NR 40
1600 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1601 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1602 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1603 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1604 #define IOC_LDLM_MAX_NR 43
1607 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1608 * than one lock_res is dead-lock safe.
1610 enum lock_res_type {
1615 /** Lock resource. */
1616 static inline void lock_res(struct ldlm_resource *res)
1618 spin_lock(&res->lr_lock);
1621 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
1622 static inline void lock_res_nested(struct ldlm_resource *res,
1623 enum lock_res_type mode)
1625 spin_lock_nested(&res->lr_lock, mode);
1628 /** Unlock resource. */
1629 static inline void unlock_res(struct ldlm_resource *res)
1631 spin_unlock(&res->lr_lock);
1634 /** Check if resource is already locked, assert if not. */
1635 static inline void check_res_locked(struct ldlm_resource *res)
1637 LASSERT(spin_is_locked(&res->lr_lock));
1640 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1641 void unlock_res_and_lock(struct ldlm_lock *lock);
1644 /** \defgroup ldlm_pools Various LDLM pool related functions
1645 * There are not used outside of ldlm.
1648 void ldlm_pools_recalc(ldlm_side_t client);
1649 int ldlm_pools_init(void);
1650 void ldlm_pools_fini(void);
1652 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1653 int idx, ldlm_side_t client);
1654 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1655 unsigned int gfp_mask);
1656 void ldlm_pool_fini(struct ldlm_pool *pl);
1657 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1658 int ldlm_pool_recalc(struct ldlm_pool *pl);
1659 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1660 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1661 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1662 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1663 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1664 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1665 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1666 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1667 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);