1 /*
2 * GPL HEADER START
3 *
4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
5 *
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.
9 *
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).
15 *
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
19 *
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
22 * have any questions.
23 *
24 * GPL HEADER END
25 */
26 /*
27 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
29 *
30 * Copyright (c) 2010, 2012, Intel Corporation.
31 */
32 /*
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
35 */
36
37 /** \defgroup LDLM Lustre Distributed Lock Manager
38 *
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.
44 *
45 * @{
46 */
47
48 #ifndef _LUSTRE_DLM_H__
49 #define _LUSTRE_DLM_H__
50
51 #include "lustre_lib.h"
52 #include "lustre_net.h"
53 #include "lustre_import.h"
54 #include "lustre_handles.h"
55 #include "interval_tree.h" /* for interval_node{}, ldlm_extent */
56 #include "lu_ref.h"
57
58 #include "lustre_dlm_flags.h"
59
60 struct obd_ops;
61 struct obd_device;
62
63 #define OBD_LDLM_DEVICENAME "ldlm"
64
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
69
70 /**
71 * LDLM non-error return states
72 */
73 typedef enum {
74 ELDLM_OK = 0,
75
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,
81
82 ELDLM_NAMESPACE_EXISTS = 400,
83 ELDLM_BAD_NAMESPACE = 401
84 } ldlm_error_t;
85
86 /**
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.
92 */
93 typedef enum {
94 LDLM_NAMESPACE_SERVER = 1 << 0,
95 LDLM_NAMESPACE_CLIENT = 1 << 1
96 } ldlm_side_t;
97
98 /**
99 * The blocking callback is overloaded to perform two functions. These flags
100 * indicate which operation should be performed.
101 */
102 #define LDLM_CB_BLOCKING 1
103 #define LDLM_CB_CANCELING 2
104
105 /**
106 * \name Lock Compatibility Matrix.
107 *
108 * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
109 * Lock types are described in their respective implementation files:
110 * ldlm_{extent,flock,inodebits,plain}.c.
111 *
112 * There are six lock modes along with a compatibility matrix to indicate if
113 * two locks are compatible.
114 *
115 * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
116 * on the parent.
117 * - PW: Protective Write (normal write) mode. When a client requests a write
118 * lock from an OST, a lock with PW mode will be issued.
119 * - PR: Protective Read (normal read) mode. When a client requests a read from
120 * an OST, a lock with PR mode will be issued. Also, if the client opens a
121 * file for execution, it is granted a lock with PR mode.
122 * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
123 * requests a write lock during a file open operation.
124 * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
125 * an inodebit lock with the CR mode on the intermediate path component.
126 * - NL Null mode.
127 *
128 * <PRE>
129 * NL CR CW PR PW EX
130 * NL 1 1 1 1 1 1
131 * CR 1 1 1 1 1 0
132 * CW 1 1 1 0 0 0
133 * PR 1 1 0 1 0 0
134 * PW 1 1 0 0 0 0
135 * EX 1 0 0 0 0 0
136 * </PRE>
137 */
138 /** @{ */
139 #define LCK_COMPAT_EX LCK_NL
140 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
141 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
142 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
143 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
144 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
145 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
146 #define LCK_COMPAT_COS (LCK_COS)
147 /** @} Lock Compatibility Matrix */
148
149 extern ldlm_mode_t lck_compat_array[];
150
lockmode_verify(ldlm_mode_t mode)151 static inline void lockmode_verify(ldlm_mode_t mode)
152 {
153 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
154 }
155
lockmode_compat(ldlm_mode_t exist_mode,ldlm_mode_t new_mode)156 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
157 {
158 return (lck_compat_array[exist_mode] & new_mode);
159 }
160
161 /*
162 *
163 * cluster name spaces
164 *
165 */
166
167 #define DLM_OST_NAMESPACE 1
168 #define DLM_MDS_NAMESPACE 2
169
170 /* XXX
171 - do we just separate this by security domains and use a prefix for
172 multiple namespaces in the same domain?
173 -
174 */
175
176 /**
177 * Locking rules for LDLM:
178 *
179 * lr_lock
180 *
181 * lr_lock
182 * waiting_locks_spinlock
183 *
184 * lr_lock
185 * led_lock
186 *
187 * lr_lock
188 * ns_lock
189 *
190 * lr_lvb_mutex
191 * lr_lock
192 *
193 */
194
195 struct ldlm_pool;
196 struct ldlm_lock;
197 struct ldlm_resource;
198 struct ldlm_namespace;
199
200 /**
201 * Operations on LDLM pools.
202 * LDLM pool is a pool of locks in the namespace without any implicitly
203 * specified limits.
204 * Locks in the pool are organized in LRU.
205 * Local memory pressure or server instructions (e.g. mempressure on server)
206 * can trigger freeing of locks from the pool
207 */
208 struct ldlm_pool_ops {
209 /** Recalculate pool \a pl usage */
210 int (*po_recalc)(struct ldlm_pool *pl);
211 /** Cancel at least \a nr locks from pool \a pl */
212 int (*po_shrink)(struct ldlm_pool *pl, int nr,
213 gfp_t gfp_mask);
214 int (*po_setup)(struct ldlm_pool *pl, int limit);
215 };
216
217 /** One second for pools thread check interval. Each pool has own period. */
218 #define LDLM_POOLS_THREAD_PERIOD (1)
219
220 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
221 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
222
223 /** Default recalc period for server side pools in sec. */
224 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
225
226 /** Default recalc period for client side pools in sec. */
227 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
228
229 /**
230 * LDLM pool structure to track granted locks.
231 * For purposes of determining when to release locks on e.g. memory pressure.
232 * This feature is commonly referred to as lru_resize.
233 */
234 struct ldlm_pool {
235 /** Pool proc directory. */
236 struct proc_dir_entry *pl_proc_dir;
237 /** Pool name, must be long enough to hold compound proc entry name. */
238 char pl_name[100];
239 /** Lock for protecting SLV/CLV updates. */
240 spinlock_t pl_lock;
241 /** Number of allowed locks in in pool, both, client and server side. */
242 atomic_t pl_limit;
243 /** Number of granted locks in */
244 atomic_t pl_granted;
245 /** Grant rate per T. */
246 atomic_t pl_grant_rate;
247 /** Cancel rate per T. */
248 atomic_t pl_cancel_rate;
249 /** Server lock volume (SLV). Protected by pl_lock. */
250 __u64 pl_server_lock_volume;
251 /** Current biggest client lock volume. Protected by pl_lock. */
252 __u64 pl_client_lock_volume;
253 /** Lock volume factor. SLV on client is calculated as following:
254 * server_slv * lock_volume_factor. */
255 atomic_t pl_lock_volume_factor;
256 /** Time when last SLV from server was obtained. */
257 time_t pl_recalc_time;
258 /** Recalculation period for pool. */
259 time_t pl_recalc_period;
260 /** Recalculation and shrink operations. */
261 const struct ldlm_pool_ops *pl_ops;
262 /** Number of planned locks for next period. */
263 int pl_grant_plan;
264 /** Pool statistics. */
265 struct lprocfs_stats *pl_stats;
266 };
267
268 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
269 void *req_cookie, ldlm_mode_t mode, __u64 flags,
270 void *data);
271
272 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
273
274 /**
275 * LVB operations.
276 * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
277 * be associated with an LDLM lock and transferred from client to server and
278 * back.
279 *
280 * Currently LVBs are used by:
281 * - OSC-OST code to maintain current object size/times
282 * - layout lock code to return the layout when the layout lock is granted
283 */
284 struct ldlm_valblock_ops {
285 int (*lvbo_init)(struct ldlm_resource *res);
286 int (*lvbo_update)(struct ldlm_resource *res,
287 struct ptlrpc_request *r,
288 int increase);
289 int (*lvbo_free)(struct ldlm_resource *res);
290 /* Return size of lvb data appropriate RPC size can be reserved */
291 int (*lvbo_size)(struct ldlm_lock *lock);
292 /* Called to fill in lvb data to RPC buffer @buf */
293 int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
294 };
295
296 /**
297 * LDLM pools related, type of lock pool in the namespace.
298 * Greedy means release cached locks aggressively
299 */
300 typedef enum {
301 LDLM_NAMESPACE_GREEDY = 1 << 0,
302 LDLM_NAMESPACE_MODEST = 1 << 1
303 } ldlm_appetite_t;
304
305 /**
306 * Default values for the "max_nolock_size", "contention_time" and
307 * "contended_locks" namespace tunables.
308 */
309 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
310 #define NS_DEFAULT_CONTENTION_SECONDS 2
311 #define NS_DEFAULT_CONTENDED_LOCKS 32
312
313 struct ldlm_ns_bucket {
314 /** back pointer to namespace */
315 struct ldlm_namespace *nsb_namespace;
316 /**
317 * Estimated lock callback time. Used by adaptive timeout code to
318 * avoid spurious client evictions due to unresponsiveness when in
319 * fact the network or overall system load is at fault
320 */
321 struct adaptive_timeout nsb_at_estimate;
322 };
323
324 enum {
325 /** LDLM namespace lock stats */
326 LDLM_NSS_LOCKS = 0,
327 LDLM_NSS_LAST
328 };
329
330 typedef enum {
331 /** invalid type */
332 LDLM_NS_TYPE_UNKNOWN = 0,
333 /** mdc namespace */
334 LDLM_NS_TYPE_MDC,
335 /** mds namespace */
336 LDLM_NS_TYPE_MDT,
337 /** osc namespace */
338 LDLM_NS_TYPE_OSC,
339 /** ost namespace */
340 LDLM_NS_TYPE_OST,
341 /** mgc namespace */
342 LDLM_NS_TYPE_MGC,
343 /** mgs namespace */
344 LDLM_NS_TYPE_MGT,
345 } ldlm_ns_type_t;
346
347 /**
348 * LDLM Namespace.
349 *
350 * Namespace serves to contain locks related to a particular service.
351 * There are two kinds of namespaces:
352 * - Server namespace has knowledge of all locks and is therefore authoritative
353 * to make decisions like what locks could be granted and what conflicts
354 * exist during new lock enqueue.
355 * - Client namespace only has limited knowledge about locks in the namespace,
356 * only seeing locks held by the client.
357 *
358 * Every Lustre service has one server namespace present on the server serving
359 * that service. Every client connected to the service has a client namespace
360 * for it.
361 * Every lock obtained by client in that namespace is actually represented by
362 * two in-memory locks. One on the server and one on the client. The locks are
363 * linked by a special cookie by which one node can tell to the other which lock
364 * it actually means during communications. Such locks are called remote locks.
365 * The locks held by server only without any reference to a client are called
366 * local locks.
367 */
368 struct ldlm_namespace {
369 /** Backward link to OBD, required for LDLM pool to store new SLV. */
370 struct obd_device *ns_obd;
371
372 /** Flag indicating if namespace is on client instead of server */
373 ldlm_side_t ns_client;
374
375 /** Resource hash table for namespace. */
376 struct cfs_hash *ns_rs_hash;
377
378 /** serialize */
379 spinlock_t ns_lock;
380
381 /** big refcount (by bucket) */
382 atomic_t ns_bref;
383
384 /**
385 * Namespace connect flags supported by server (may be changed via
386 * /proc, LRU resize may be disabled/enabled).
387 */
388 __u64 ns_connect_flags;
389
390 /** Client side original connect flags supported by server. */
391 __u64 ns_orig_connect_flags;
392
393 /* namespace proc dir entry */
394 struct proc_dir_entry *ns_proc_dir_entry;
395
396 /**
397 * Position in global namespace list linking all namespaces on
398 * the node.
399 */
400 struct list_head ns_list_chain;
401
402 /**
403 * List of unused locks for this namespace. This list is also called
404 * LRU lock list.
405 * Unused locks are locks with zero reader/writer reference counts.
406 * This list is only used on clients for lock caching purposes.
407 * When we want to release some locks voluntarily or if server wants
408 * us to release some locks due to e.g. memory pressure, we take locks
409 * to release from the head of this list.
410 * Locks are linked via l_lru field in \see struct ldlm_lock.
411 */
412 struct list_head ns_unused_list;
413 /** Number of locks in the LRU list above */
414 int ns_nr_unused;
415
416 /**
417 * Maximum number of locks permitted in the LRU. If 0, means locks
418 * are managed by pools and there is no preset limit, rather it is all
419 * controlled by available memory on this client and on server.
420 */
421 unsigned int ns_max_unused;
422 /** Maximum allowed age (last used time) for locks in the LRU */
423 unsigned int ns_max_age;
424 /**
425 * Server only: number of times we evicted clients due to lack of reply
426 * to ASTs.
427 */
428 unsigned int ns_timeouts;
429 /**
430 * Number of seconds since the file change time after which the
431 * MDT will return an UPDATE lock along with a LOOKUP lock.
432 * This allows the client to start caching negative dentries
433 * for a directory and may save an RPC for a later stat.
434 */
435 unsigned int ns_ctime_age_limit;
436
437 /**
438 * Used to rate-limit ldlm_namespace_dump calls.
439 * \see ldlm_namespace_dump. Increased by 10 seconds every time
440 * it is called.
441 */
442 unsigned long ns_next_dump;
443
444 /** "policy" function that does actual lock conflict determination */
445 ldlm_res_policy ns_policy;
446
447 /**
448 * LVB operations for this namespace.
449 * \see struct ldlm_valblock_ops
450 */
451 struct ldlm_valblock_ops *ns_lvbo;
452
453 /**
454 * Used by filter code to store pointer to OBD of the service.
455 * Should be dropped in favor of \a ns_obd
456 */
457 void *ns_lvbp;
458
459 /**
460 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
461 * a resource is removed.
462 */
463 wait_queue_head_t ns_waitq;
464 /** LDLM pool structure for this namespace */
465 struct ldlm_pool ns_pool;
466 /** Definition of how eagerly unused locks will be released from LRU */
467 ldlm_appetite_t ns_appetite;
468
469 /**
470 * If more than \a ns_contended_locks are found, the resource is
471 * considered to be contended. Lock enqueues might specify that no
472 * contended locks should be granted
473 */
474 unsigned ns_contended_locks;
475
476 /**
477 * The resources in this namespace remember contended state during
478 * \a ns_contention_time, in seconds.
479 */
480 unsigned ns_contention_time;
481
482 /**
483 * Limit size of contended extent locks, in bytes.
484 * If extended lock is requested for more then this many bytes and
485 * caller instructs us not to grant contended locks, we would disregard
486 * such a request.
487 */
488 unsigned ns_max_nolock_size;
489
490 /** Limit of parallel AST RPC count. */
491 unsigned ns_max_parallel_ast;
492
493 /** Callback to cancel locks before replaying it during recovery. */
494 ldlm_cancel_for_recovery ns_cancel_for_recovery;
495
496 /** LDLM lock stats */
497 struct lprocfs_stats *ns_stats;
498
499 /**
500 * Flag to indicate namespace is being freed. Used to determine if
501 * recalculation of LDLM pool statistics should be skipped.
502 */
503 unsigned ns_stopping:1;
504 };
505
506 /**
507 * Returns 1 if namespace \a ns is a client namespace.
508 */
ns_is_client(struct ldlm_namespace * ns)509 static inline int ns_is_client(struct ldlm_namespace *ns)
510 {
511 LASSERT(ns != NULL);
512 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
513 LDLM_NAMESPACE_SERVER)));
514 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
515 ns->ns_client == LDLM_NAMESPACE_SERVER);
516 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
517 }
518
519 /**
520 * Returns 1 if namespace \a ns is a server namespace.
521 */
ns_is_server(struct ldlm_namespace * ns)522 static inline int ns_is_server(struct ldlm_namespace *ns)
523 {
524 LASSERT(ns != NULL);
525 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
526 LDLM_NAMESPACE_SERVER)));
527 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
528 ns->ns_client == LDLM_NAMESPACE_SERVER);
529 return ns->ns_client == LDLM_NAMESPACE_SERVER;
530 }
531
532 /**
533 * Returns 1 if namespace \a ns supports early lock cancel (ELC).
534 */
ns_connect_cancelset(struct ldlm_namespace * ns)535 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
536 {
537 LASSERT(ns != NULL);
538 return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
539 }
540
541 /**
542 * Returns 1 if this namespace supports lru_resize.
543 */
ns_connect_lru_resize(struct ldlm_namespace * ns)544 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
545 {
546 LASSERT(ns != NULL);
547 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
548 }
549
ns_register_cancel(struct ldlm_namespace * ns,ldlm_cancel_for_recovery arg)550 static inline void ns_register_cancel(struct ldlm_namespace *ns,
551 ldlm_cancel_for_recovery arg)
552 {
553 LASSERT(ns != NULL);
554 ns->ns_cancel_for_recovery = arg;
555 }
556
557 struct ldlm_lock;
558
559 /** Type for blocking callback function of a lock. */
560 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
561 struct ldlm_lock_desc *new, void *data,
562 int flag);
563 /** Type for completion callback function of a lock. */
564 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
565 void *data);
566 /** Type for glimpse callback function of a lock. */
567 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
568
569 /** Work list for sending GL ASTs to multiple locks. */
570 struct ldlm_glimpse_work {
571 struct ldlm_lock *gl_lock; /* lock to glimpse */
572 struct list_head gl_list; /* linkage to other gl work structs */
573 __u32 gl_flags;/* see LDLM_GL_WORK_* below */
574 union ldlm_gl_desc *gl_desc; /* glimpse descriptor to be packed in
575 * glimpse callback request */
576 };
577
578 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
579 #define LDLM_GL_WORK_NOFREE 0x1
580
581 /** Interval node data for each LDLM_EXTENT lock. */
582 struct ldlm_interval {
583 struct interval_node li_node; /* node for tree management */
584 struct list_head li_group; /* the locks which have the same
585 * policy - group of the policy */
586 };
587 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
588
589 /**
590 * Interval tree for extent locks.
591 * The interval tree must be accessed under the resource lock.
592 * Interval trees are used for granted extent locks to speed up conflicts
593 * lookup. See ldlm/interval_tree.c for more details.
594 */
595 struct ldlm_interval_tree {
596 /** Tree size. */
597 int lit_size;
598 ldlm_mode_t lit_mode; /* lock mode */
599 struct interval_node *lit_root; /* actual ldlm_interval */
600 };
601
602 /** Whether to track references to exports by LDLM locks. */
603 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
604
605 /** Cancel flags. */
606 typedef enum {
607 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
608 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
609 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
610 * in the same RPC */
611 } ldlm_cancel_flags_t;
612
613 struct ldlm_flock {
614 __u64 start;
615 __u64 end;
616 __u64 owner;
617 __u64 blocking_owner;
618 struct obd_export *blocking_export;
619 /* Protected by the hash lock */
620 __u32 blocking_refs;
621 __u32 pid;
622 };
623
624 typedef union {
625 struct ldlm_extent l_extent;
626 struct ldlm_flock l_flock;
627 struct ldlm_inodebits l_inodebits;
628 } ldlm_policy_data_t;
629
630 void ldlm_convert_policy_to_wire(ldlm_type_t type,
631 const ldlm_policy_data_t *lpolicy,
632 ldlm_wire_policy_data_t *wpolicy);
633 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
634 const ldlm_wire_policy_data_t *wpolicy,
635 ldlm_policy_data_t *lpolicy);
636
637 enum lvb_type {
638 LVB_T_NONE = 0,
639 LVB_T_OST = 1,
640 LVB_T_LQUOTA = 2,
641 LVB_T_LAYOUT = 3,
642 };
643
644 /**
645 * LDLM lock structure
646 *
647 * Represents a single LDLM lock and its state in memory. Each lock is
648 * associated with a single ldlm_resource, the object which is being
649 * locked. There may be multiple ldlm_locks on a single resource,
650 * depending on the lock type and whether the locks are conflicting or
651 * not.
652 */
653 struct ldlm_lock {
654 /**
655 * Local lock handle.
656 * When remote side wants to tell us about a lock, they address
657 * it by this opaque handle. The handle does not hold a
658 * reference on the ldlm_lock, so it can be safely passed to
659 * other threads or nodes. When the lock needs to be accessed
660 * from the handle, it is looked up again in the lock table, and
661 * may no longer exist.
662 *
663 * Must be first in the structure.
664 */
665 struct portals_handle l_handle;
666 /**
667 * Lock reference count.
668 * This is how many users have pointers to actual structure, so that
669 * we do not accidentally free lock structure that is in use.
670 */
671 atomic_t l_refc;
672 /**
673 * Internal spinlock protects l_resource. We should hold this lock
674 * first before taking res_lock.
675 */
676 spinlock_t l_lock;
677 /**
678 * Pointer to actual resource this lock is in.
679 * ldlm_lock_change_resource() can change this.
680 */
681 struct ldlm_resource *l_resource;
682 /**
683 * List item for client side LRU list.
684 * Protected by ns_lock in struct ldlm_namespace.
685 */
686 struct list_head l_lru;
687 /**
688 * Linkage to resource's lock queues according to current lock state.
689 * (could be granted, waiting or converting)
690 * Protected by lr_lock in struct ldlm_resource.
691 */
692 struct list_head l_res_link;
693 /**
694 * Tree node for ldlm_extent.
695 */
696 struct ldlm_interval *l_tree_node;
697 /**
698 * Per export hash of locks.
699 * Protected by per-bucket exp->exp_lock_hash locks.
700 */
701 struct hlist_node l_exp_hash;
702 /**
703 * Per export hash of flock locks.
704 * Protected by per-bucket exp->exp_flock_hash locks.
705 */
706 struct hlist_node l_exp_flock_hash;
707 /**
708 * Requested mode.
709 * Protected by lr_lock.
710 */
711 ldlm_mode_t l_req_mode;
712 /**
713 * Granted mode, also protected by lr_lock.
714 */
715 ldlm_mode_t l_granted_mode;
716 /** Lock completion handler pointer. Called when lock is granted. */
717 ldlm_completion_callback l_completion_ast;
718 /**
719 * Lock blocking AST handler pointer.
720 * It plays two roles:
721 * - as a notification of an attempt to queue a conflicting lock (once)
722 * - as a notification when the lock is being cancelled.
723 *
724 * As such it's typically called twice: once for the initial conflict
725 * and then once more when the last user went away and the lock is
726 * cancelled (could happen recursively).
727 */
728 ldlm_blocking_callback l_blocking_ast;
729 /**
730 * Lock glimpse handler.
731 * Glimpse handler is used to obtain LVB updates from a client by
732 * server
733 */
734 ldlm_glimpse_callback l_glimpse_ast;
735
736 /**
737 * Lock export.
738 * This is a pointer to actual client export for locks that were granted
739 * to clients. Used server-side.
740 */
741 struct obd_export *l_export;
742 /**
743 * Lock connection export.
744 * Pointer to server export on a client.
745 */
746 struct obd_export *l_conn_export;
747
748 /**
749 * Remote lock handle.
750 * If the lock is remote, this is the handle of the other side lock
751 * (l_handle)
752 */
753 struct lustre_handle l_remote_handle;
754
755 /**
756 * Representation of private data specific for a lock type.
757 * Examples are: extent range for extent lock or bitmask for ibits locks
758 */
759 ldlm_policy_data_t l_policy_data;
760
761 /**
762 * Lock state flags. Protected by lr_lock.
763 * \see lustre_dlm_flags.h where the bits are defined.
764 */
765 __u64 l_flags;
766
767 /**
768 * Lock r/w usage counters.
769 * Protected by lr_lock.
770 */
771 __u32 l_readers;
772 __u32 l_writers;
773 /**
774 * If the lock is granted, a process sleeps on this waitq to learn when
775 * it's no longer in use. If the lock is not granted, a process sleeps
776 * on this waitq to learn when it becomes granted.
777 */
778 wait_queue_head_t l_waitq;
779
780 /**
781 * Seconds. It will be updated if there is any activity related to
782 * the lock, e.g. enqueue the lock or send blocking AST.
783 */
784 unsigned long l_last_activity;
785
786 /**
787 * Time last used by e.g. being matched by lock match.
788 * Jiffies. Should be converted to time if needed.
789 */
790 unsigned long l_last_used;
791
792 /** Originally requested extent for the extent lock. */
793 struct ldlm_extent l_req_extent;
794
795 /*
796 * Client-side-only members.
797 */
798
799 enum lvb_type l_lvb_type;
800
801 /**
802 * Temporary storage for a LVB received during an enqueue operation.
803 */
804 __u32 l_lvb_len;
805 void *l_lvb_data;
806
807 /** Private storage for lock user. Opaque to LDLM. */
808 void *l_ast_data;
809
810 /*
811 * Server-side-only members.
812 */
813
814 /**
815 * Connection cookie for the client originating the operation.
816 * Used by Commit on Share (COS) code. Currently only used for
817 * inodebits locks on MDS.
818 */
819 __u64 l_client_cookie;
820
821 /**
822 * List item for locks waiting for cancellation from clients.
823 * The lists this could be linked into are:
824 * waiting_locks_list (protected by waiting_locks_spinlock),
825 * then if the lock timed out, it is moved to
826 * expired_lock_thread.elt_expired_locks for further processing.
827 * Protected by elt_lock.
828 */
829 struct list_head l_pending_chain;
830
831 /**
832 * Set when lock is sent a blocking AST. Time in seconds when timeout
833 * is reached and client holding this lock could be evicted.
834 * This timeout could be further extended by e.g. certain IO activity
835 * under this lock.
836 * \see ost_rw_prolong_locks
837 */
838 unsigned long l_callback_timeout;
839
840 /** Local PID of process which created this lock. */
841 __u32 l_pid;
842
843 /**
844 * Number of times blocking AST was sent for this lock.
845 * This is for debugging. Valid values are 0 and 1, if there is an
846 * attempt to send blocking AST more than once, an assertion would be
847 * hit. \see ldlm_work_bl_ast_lock
848 */
849 int l_bl_ast_run;
850 /** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
851 struct list_head l_bl_ast;
852 /** List item ldlm_add_ast_work_item() for case of completion ASTs. */
853 struct list_head l_cp_ast;
854 /** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
855 struct list_head l_rk_ast;
856
857 /**
858 * Pointer to a conflicting lock that caused blocking AST to be sent
859 * for this lock
860 */
861 struct ldlm_lock *l_blocking_lock;
862
863 /**
864 * Protected by lr_lock, linkages to "skip lists".
865 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
866 */
867 struct list_head l_sl_mode;
868 struct list_head l_sl_policy;
869
870 /** Reference tracking structure to debug leaked locks. */
871 struct lu_ref l_reference;
872 #if LUSTRE_TRACKS_LOCK_EXP_REFS
873 /* Debugging stuff for bug 20498, for tracking export references. */
874 /** number of export references taken */
875 int l_exp_refs_nr;
876 /** link all locks referencing one export */
877 struct list_head l_exp_refs_link;
878 /** referenced export object */
879 struct obd_export *l_exp_refs_target;
880 #endif
881 /**
882 * export blocking dlm lock list, protected by
883 * l_export->exp_bl_list_lock.
884 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
885 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
886 */
887 struct list_head l_exp_list;
888 };
889
890 /**
891 * LDLM resource description.
892 * Basically, resource is a representation for a single object.
893 * Object has a name which is currently 4 64-bit integers. LDLM user is
894 * responsible for creation of a mapping between objects it wants to be
895 * protected and resource names.
896 *
897 * A resource can only hold locks of a single lock type, though there may be
898 * multiple ldlm_locks on a single resource, depending on the lock type and
899 * whether the locks are conflicting or not.
900 */
901 struct ldlm_resource {
902 struct ldlm_ns_bucket *lr_ns_bucket;
903
904 /**
905 * List item for list in namespace hash.
906 * protected by ns_lock
907 */
908 struct hlist_node lr_hash;
909
910 /** Spinlock to protect locks under this resource. */
911 spinlock_t lr_lock;
912
913 /**
914 * protected by lr_lock
915 * @{ */
916 /** List of locks in granted state */
917 struct list_head lr_granted;
918 /** List of locks waiting to change their granted mode (converted) */
919 struct list_head lr_converting;
920 /**
921 * List of locks that could not be granted due to conflicts and
922 * that are waiting for conflicts to go away */
923 struct list_head lr_waiting;
924 /** @} */
925
926 /* XXX No longer needed? Remove ASAP */
927 ldlm_mode_t lr_most_restr;
928
929 /** Type of locks this resource can hold. Only one type per resource. */
930 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
931
932 /** Resource name */
933 struct ldlm_res_id lr_name;
934 /** Reference count for this resource */
935 atomic_t lr_refcount;
936
937 /**
938 * Interval trees (only for extent locks) for all modes of this resource
939 */
940 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM];
941
942 /**
943 * Server-side-only lock value block elements.
944 * To serialize lvbo_init.
945 */
946 struct mutex lr_lvb_mutex;
947 int lr_lvb_len;
948 /** protected by lr_lock */
949 void *lr_lvb_data;
950
951 /** When the resource was considered as contended. */
952 unsigned long lr_contention_time;
953 /** List of references to this resource. For debugging. */
954 struct lu_ref lr_reference;
955
956 struct inode *lr_lvb_inode;
957 };
958
ldlm_has_layout(struct ldlm_lock * lock)959 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
960 {
961 return lock->l_resource->lr_type == LDLM_IBITS &&
962 lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
963 }
964
965 static inline char *
ldlm_ns_name(struct ldlm_namespace * ns)966 ldlm_ns_name(struct ldlm_namespace *ns)
967 {
968 return ns->ns_rs_hash->hs_name;
969 }
970
971 static inline struct ldlm_namespace *
ldlm_res_to_ns(struct ldlm_resource * res)972 ldlm_res_to_ns(struct ldlm_resource *res)
973 {
974 return res->lr_ns_bucket->nsb_namespace;
975 }
976
977 static inline struct ldlm_namespace *
ldlm_lock_to_ns(struct ldlm_lock * lock)978 ldlm_lock_to_ns(struct ldlm_lock *lock)
979 {
980 return ldlm_res_to_ns(lock->l_resource);
981 }
982
983 static inline char *
ldlm_lock_to_ns_name(struct ldlm_lock * lock)984 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
985 {
986 return ldlm_ns_name(ldlm_lock_to_ns(lock));
987 }
988
989 static inline struct adaptive_timeout *
ldlm_lock_to_ns_at(struct ldlm_lock * lock)990 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
991 {
992 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
993 }
994
ldlm_lvbo_init(struct ldlm_resource * res)995 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
996 {
997 struct ldlm_namespace *ns = ldlm_res_to_ns(res);
998
999 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_init != NULL)
1000 return ns->ns_lvbo->lvbo_init(res);
1001
1002 return 0;
1003 }
1004
ldlm_lvbo_size(struct ldlm_lock * lock)1005 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
1006 {
1007 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1008
1009 if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
1010 return ns->ns_lvbo->lvbo_size(lock);
1011
1012 return 0;
1013 }
1014
ldlm_lvbo_fill(struct ldlm_lock * lock,void * buf,int len)1015 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
1016 {
1017 struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
1018
1019 if (ns->ns_lvbo != NULL) {
1020 LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
1021 return ns->ns_lvbo->lvbo_fill(lock, buf, len);
1022 }
1023 return 0;
1024 }
1025
1026 struct ldlm_ast_work {
1027 struct ldlm_lock *w_lock;
1028 int w_blocking;
1029 struct ldlm_lock_desc w_desc;
1030 struct list_head w_list;
1031 int w_flags;
1032 void *w_data;
1033 int w_datalen;
1034 };
1035
1036 /**
1037 * Common ldlm_enqueue parameters
1038 */
1039 struct ldlm_enqueue_info {
1040 __u32 ei_type; /** Type of the lock being enqueued. */
1041 __u32 ei_mode; /** Mode of the lock being enqueued. */
1042 void *ei_cb_bl; /** blocking lock callback */
1043 void *ei_cb_cp; /** lock completion callback */
1044 void *ei_cb_gl; /** lock glimpse callback */
1045 void *ei_cbdata; /** Data to be passed into callbacks. */
1046 };
1047
1048 extern struct obd_ops ldlm_obd_ops;
1049
1050 extern char *ldlm_lockname[];
1051 extern char *ldlm_typename[];
1052 extern char *ldlm_it2str(int it);
1053
1054 /**
1055 * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
1056 * For the cases where we do not have actual lock to print along
1057 * with a debugging message that is ldlm-related
1058 */
1059 #define LDLM_DEBUG_NOLOCK(format, a...) \
1060 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
1061
1062 /**
1063 * Support function for lock information printing into debug logs.
1064 * \see LDLM_DEBUG
1065 */
1066 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do { \
1067 CFS_CHECK_STACK(msgdata, mask, cdls); \
1068 \
1069 if (((mask) & D_CANTMASK) != 0 || \
1070 ((libcfs_debug & (mask)) != 0 && \
1071 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) \
1072 _ldlm_lock_debug(lock, msgdata, fmt, ##a); \
1073 } while (0)
1074
1075 void _ldlm_lock_debug(struct ldlm_lock *lock,
1076 struct libcfs_debug_msg_data *data,
1077 const char *fmt, ...)
1078 __printf(3, 4);
1079
1080 /**
1081 * Rate-limited version of lock printing function.
1082 */
1083 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
1084 static struct cfs_debug_limit_state _ldlm_cdls; \
1085 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls); \
1086 ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt , ##a);\
1087 } while (0)
1088
1089 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1090 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1091
1092 /** Non-rate-limited lock printing function for debugging purposes. */
1093 #define LDLM_DEBUG(lock, fmt, a...) do { \
1094 if (likely(lock != NULL)) { \
1095 LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL); \
1096 ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock, \
1097 "### " fmt , ##a); \
1098 } else { \
1099 LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a); \
1100 } \
1101 } while (0)
1102
1103 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1104 int first_enq, ldlm_error_t *err,
1105 struct list_head *work_list);
1106
1107 /**
1108 * Return values for lock iterators.
1109 * Also used during deciding of lock grants and cancellations.
1110 */
1111 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1112 #define LDLM_ITER_STOP 2 /* stop iterating */
1113
1114 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1115 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1116
1117 /** \defgroup ldlm_iterator Lock iterators
1118 *
1119 * LDLM provides for a way to iterate through every lock on a resource or
1120 * namespace or every resource in a namespace.
1121 * @{ */
1122 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
1123 void *closure);
1124 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
1125 void *closure);
1126 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1127 ldlm_iterator_t iter, void *data);
1128 /** @} ldlm_iterator */
1129
1130 int ldlm_replay_locks(struct obd_import *imp);
1131
1132 /* ldlm_flock.c */
1133 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1134
1135 /* ldlm_extent.c */
1136 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1137
1138 struct ldlm_callback_suite {
1139 ldlm_completion_callback lcs_completion;
1140 ldlm_blocking_callback lcs_blocking;
1141 ldlm_glimpse_callback lcs_glimpse;
1142 };
1143
1144 /* ldlm_lockd.c */
1145 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
1146 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
1147 int ldlm_get_ref(void);
1148 void ldlm_put_ref(void);
1149 int ldlm_init_export(struct obd_export *exp);
1150 void ldlm_destroy_export(struct obd_export *exp);
1151 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1152
1153 /* ldlm_lock.c */
1154 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
1155 void ldlm_lock2handle(const struct ldlm_lock *lock,
1156 struct lustre_handle *lockh);
1157 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1158 void ldlm_cancel_callback(struct ldlm_lock *);
1159 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1160 int ldlm_lock_set_data(struct lustre_handle *, void *);
1161
1162 /**
1163 * Obtain a lock reference by its handle.
1164 */
ldlm_handle2lock(const struct lustre_handle * h)1165 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1166 {
1167 return __ldlm_handle2lock(h, 0);
1168 }
1169
1170 #define LDLM_LOCK_REF_DEL(lock) \
1171 lu_ref_del(&lock->l_reference, "handle", current)
1172
1173 static inline struct ldlm_lock *
ldlm_handle2lock_long(const struct lustre_handle * h,__u64 flags)1174 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1175 {
1176 struct ldlm_lock *lock;
1177
1178 lock = __ldlm_handle2lock(h, flags);
1179 if (lock != NULL)
1180 LDLM_LOCK_REF_DEL(lock);
1181 return lock;
1182 }
1183
1184 /**
1185 * Update Lock Value Block Operations (LVBO) on a resource taking into account
1186 * data from request \a r
1187 */
ldlm_res_lvbo_update(struct ldlm_resource * res,struct ptlrpc_request * r,int increase)1188 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1189 struct ptlrpc_request *r, int increase)
1190 {
1191 if (ldlm_res_to_ns(res)->ns_lvbo &&
1192 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1193 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1194 increase);
1195 }
1196 return 0;
1197 }
1198
1199 int ldlm_error2errno(ldlm_error_t error);
1200 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
1201 * confuses user-space. */
1202 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1203 void ldlm_dump_export_locks(struct obd_export *exp);
1204 #endif
1205
1206 /**
1207 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1208 * __ldlm_handle2lock().
1209 */
1210 #define LDLM_LOCK_PUT(lock) \
1211 do { \
1212 LDLM_LOCK_REF_DEL(lock); \
1213 /*LDLM_DEBUG((lock), "put");*/ \
1214 ldlm_lock_put(lock); \
1215 } while (0)
1216
1217 /**
1218 * Release a lock reference obtained by some other means (see
1219 * LDLM_LOCK_PUT()).
1220 */
1221 #define LDLM_LOCK_RELEASE(lock) \
1222 do { \
1223 /*LDLM_DEBUG((lock), "put");*/ \
1224 ldlm_lock_put(lock); \
1225 } while (0)
1226
1227 #define LDLM_LOCK_GET(lock) \
1228 ({ \
1229 ldlm_lock_get(lock); \
1230 /*LDLM_DEBUG((lock), "get");*/ \
1231 lock; \
1232 })
1233
1234 #define ldlm_lock_list_put(head, member, count) \
1235 ({ \
1236 struct ldlm_lock *_lock, *_next; \
1237 int c = count; \
1238 list_for_each_entry_safe(_lock, _next, head, member) { \
1239 if (c-- == 0) \
1240 break; \
1241 list_del_init(&_lock->member); \
1242 LDLM_LOCK_RELEASE(_lock); \
1243 } \
1244 LASSERT(c <= 0); \
1245 })
1246
1247 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1248 void ldlm_lock_put(struct ldlm_lock *lock);
1249 void ldlm_lock_destroy(struct ldlm_lock *lock);
1250 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1251 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1252 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1253 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1254 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1255 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1256 void ldlm_lock_fail_match(struct ldlm_lock *lock);
1257 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1258 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1259 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1260 const struct ldlm_res_id *, ldlm_type_t type,
1261 ldlm_policy_data_t *, ldlm_mode_t mode,
1262 struct lustre_handle *, int unref);
1263 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1264 __u64 *bits);
1265 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1266 __u32 *flags);
1267 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
1268 void ldlm_lock_cancel(struct ldlm_lock *lock);
1269 void ldlm_reprocess_all(struct ldlm_resource *res);
1270 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1271 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1272 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1273
1274 /* resource.c */
1275 struct ldlm_namespace *
1276 ldlm_namespace_new(struct obd_device *obd, char *name,
1277 ldlm_side_t client, ldlm_appetite_t apt,
1278 ldlm_ns_type_t ns_type);
1279 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1280 void ldlm_namespace_free(struct ldlm_namespace *ns,
1281 struct obd_import *imp, int force);
1282 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1283 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1284 void ldlm_namespace_get(struct ldlm_namespace *ns);
1285 void ldlm_namespace_put(struct ldlm_namespace *ns);
1286 #if defined (CONFIG_PROC_FS)
1287 int ldlm_proc_setup(void);
1288 void ldlm_proc_cleanup(void);
1289 #else
ldlm_proc_setup(void)1290 static inline int ldlm_proc_setup(void) { return 0; }
ldlm_proc_cleanup(void)1291 static inline void ldlm_proc_cleanup(void) {}
1292 #endif
1293
1294 /* resource.c - internal */
1295 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1296 struct ldlm_resource *parent,
1297 const struct ldlm_res_id *,
1298 ldlm_type_t type, int create);
1299 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1300 int ldlm_resource_putref(struct ldlm_resource *res);
1301 void ldlm_resource_add_lock(struct ldlm_resource *res,
1302 struct list_head *head,
1303 struct ldlm_lock *lock);
1304 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1305 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1306 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1307 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1308 void ldlm_resource_dump(int level, struct ldlm_resource *);
1309 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1310 const struct ldlm_res_id *);
1311
1312 #define LDLM_RESOURCE_ADDREF(res) do { \
1313 lu_ref_add_atomic(&(res)->lr_reference, __func__, current); \
1314 } while (0)
1315
1316 #define LDLM_RESOURCE_DELREF(res) do { \
1317 lu_ref_del(&(res)->lr_reference, __func__, current); \
1318 } while (0)
1319
1320 /* ldlm_request.c */
1321 int ldlm_expired_completion_wait(void *data);
1322 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1323 * These AST handlers are typically used for server-side local locks and are
1324 * also used by client-side lock handlers to perform minimum level base
1325 * processing.
1326 * @{ */
1327 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1328 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1329 void *data, int flag);
1330 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1331 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1332 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1333 /** @} ldlm_local_ast */
1334
1335 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1336 * These are typically used by client and server (*_local versions)
1337 * to obtain and release locks.
1338 * @{ */
1339 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1340 struct ldlm_enqueue_info *einfo,
1341 const struct ldlm_res_id *res_id,
1342 ldlm_policy_data_t const *policy, __u64 *flags,
1343 void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1344 struct lustre_handle *lockh, int async);
1345 int ldlm_prep_enqueue_req(struct obd_export *exp,
1346 struct ptlrpc_request *req,
1347 struct list_head *cancels,
1348 int count);
1349 int ldlm_prep_elc_req(struct obd_export *exp,
1350 struct ptlrpc_request *req,
1351 int version, int opc, int canceloff,
1352 struct list_head *cancels, int count);
1353
1354 struct ptlrpc_request *ldlm_enqueue_pack(struct obd_export *exp, int lvb_len);
1355 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1356 const struct ldlm_request *dlm_req,
1357 const struct ldlm_callback_suite *cbs);
1358 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1359 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1360 __u64 *flags, void *lvb, __u32 lvb_len,
1361 struct lustre_handle *lockh, int rc);
1362 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1363 const struct ldlm_res_id *res_id,
1364 ldlm_type_t type, ldlm_policy_data_t *policy,
1365 ldlm_mode_t mode, __u64 *flags,
1366 ldlm_blocking_callback blocking,
1367 ldlm_completion_callback completion,
1368 ldlm_glimpse_callback glimpse,
1369 void *data, __u32 lvb_len, enum lvb_type lvb_type,
1370 const __u64 *client_cookie,
1371 struct lustre_handle *lockh);
1372 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1373 void *data, __u32 data_len);
1374 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1375 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1376 int ldlm_cli_cancel(struct lustre_handle *lockh,
1377 ldlm_cancel_flags_t cancel_flags);
1378 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1379 ldlm_cancel_flags_t flags, void *opaque);
1380 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1381 const struct ldlm_res_id *res_id,
1382 ldlm_policy_data_t *policy,
1383 ldlm_mode_t mode,
1384 ldlm_cancel_flags_t flags,
1385 void *opaque);
1386 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1387 int count, ldlm_cancel_flags_t flags);
1388 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1389 struct list_head *cancels,
1390 ldlm_policy_data_t *policy,
1391 ldlm_mode_t mode, __u64 lock_flags,
1392 ldlm_cancel_flags_t cancel_flags, void *opaque);
1393 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1394 ldlm_cancel_flags_t flags);
1395 int ldlm_cli_cancel_list(struct list_head *head, int count,
1396 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1397 /** @} ldlm_cli_api */
1398
1399 /* mds/handler.c */
1400 /* This has to be here because recursive inclusion sucks. */
1401 int intent_disposition(struct ldlm_reply *rep, int flag);
1402 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1403
1404
1405 /* ioctls for trying requests */
1406 #define IOC_LDLM_TYPE 'f'
1407 #define IOC_LDLM_MIN_NR 40
1408
1409 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1410 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1411 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1412 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1413 #define IOC_LDLM_MAX_NR 43
1414
1415 /**
1416 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1417 * than one lock_res is dead-lock safe.
1418 */
1419 enum lock_res_type {
1420 LRT_NORMAL,
1421 LRT_NEW
1422 };
1423
1424 /** Lock resource. */
lock_res(struct ldlm_resource * res)1425 static inline void lock_res(struct ldlm_resource *res)
1426 {
1427 spin_lock(&res->lr_lock);
1428 }
1429
1430 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
lock_res_nested(struct ldlm_resource * res,enum lock_res_type mode)1431 static inline void lock_res_nested(struct ldlm_resource *res,
1432 enum lock_res_type mode)
1433 {
1434 spin_lock_nested(&res->lr_lock, mode);
1435 }
1436
1437 /** Unlock resource. */
unlock_res(struct ldlm_resource * res)1438 static inline void unlock_res(struct ldlm_resource *res)
1439 {
1440 spin_unlock(&res->lr_lock);
1441 }
1442
1443 /** Check if resource is already locked, assert if not. */
check_res_locked(struct ldlm_resource * res)1444 static inline void check_res_locked(struct ldlm_resource *res)
1445 {
1446 assert_spin_locked(&res->lr_lock);
1447 }
1448
1449 struct ldlm_resource *lock_res_and_lock(struct ldlm_lock *lock);
1450 void unlock_res_and_lock(struct ldlm_lock *lock);
1451
1452 /* ldlm_pool.c */
1453 /** \defgroup ldlm_pools Various LDLM pool related functions
1454 * There are not used outside of ldlm.
1455 * @{
1456 */
1457 int ldlm_pools_recalc(ldlm_side_t client);
1458 int ldlm_pools_init(void);
1459 void ldlm_pools_fini(void);
1460
1461 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1462 int idx, ldlm_side_t client);
1463 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1464 gfp_t gfp_mask);
1465 void ldlm_pool_fini(struct ldlm_pool *pl);
1466 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1467 int ldlm_pool_recalc(struct ldlm_pool *pl);
1468 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1469 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1470 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1471 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1472 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1473 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1474 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1475 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1476 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1477 /** @} */
1478
1479 #endif
1480 /** @} LDLM */
1481