• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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_DEFAULT_PARALLEL_AST_LIMIT 1024
68 
69 /**
70  * LDLM non-error return states
71  */
72 typedef enum {
73 	ELDLM_OK = 0,
74 
75 	ELDLM_LOCK_CHANGED = 300,
76 	ELDLM_LOCK_ABORTED = 301,
77 	ELDLM_LOCK_REPLACED = 302,
78 	ELDLM_NO_LOCK_DATA = 303,
79 	ELDLM_LOCK_WOULDBLOCK = 304,
80 
81 	ELDLM_NAMESPACE_EXISTS = 400,
82 	ELDLM_BAD_NAMESPACE    = 401
83 } ldlm_error_t;
84 
85 /**
86  * LDLM namespace type.
87  * The "client" type is actually an indication that this is a narrow local view
88  * into complete namespace on the server. Such namespaces cannot make any
89  * decisions about lack of conflicts or do any autonomous lock granting without
90  * first speaking to a server.
91  */
92 typedef enum {
93 	LDLM_NAMESPACE_SERVER = 1 << 0,
94 	LDLM_NAMESPACE_CLIENT = 1 << 1
95 } ldlm_side_t;
96 
97 /**
98  * The blocking callback is overloaded to perform two functions.  These flags
99  * indicate which operation should be performed.
100  */
101 #define LDLM_CB_BLOCKING    1
102 #define LDLM_CB_CANCELING   2
103 
104 /**
105  * \name Lock Compatibility Matrix.
106  *
107  * A lock has both a type (extent, flock, inode bits, or plain) and a mode.
108  * Lock types are described in their respective implementation files:
109  * ldlm_{extent,flock,inodebits,plain}.c.
110  *
111  * There are six lock modes along with a compatibility matrix to indicate if
112  * two locks are compatible.
113  *
114  * - EX: Exclusive mode. Before a new file is created, MDS requests EX lock
115  *   on the parent.
116  * - PW: Protective Write (normal write) mode. When a client requests a write
117  *   lock from an OST, a lock with PW mode will be issued.
118  * - PR: Protective Read (normal read) mode. When a client requests a read from
119  *   an OST, a lock with PR mode will be issued. Also, if the client opens a
120  *   file for execution, it is granted a lock with PR mode.
121  * - CW: Concurrent Write mode. The type of lock that the MDS grants if a client
122  *   requests a write lock during a file open operation.
123  * - CR Concurrent Read mode. When a client performs a path lookup, MDS grants
124  *   an inodebit lock with the CR mode on the intermediate path component.
125  * - NL Null mode.
126  *
127  * <PRE>
128  *       NL  CR  CW  PR  PW  EX
129  *  NL    1   1   1   1   1   1
130  *  CR    1   1   1   1   1   0
131  *  CW    1   1   1   0   0   0
132  *  PR    1   1   0   1   0   0
133  *  PW    1   1   0   0   0   0
134  *  EX    1   0   0   0   0   0
135  * </PRE>
136  */
137 /** @{ */
138 #define LCK_COMPAT_EX  LCK_NL
139 #define LCK_COMPAT_PW  (LCK_COMPAT_EX | LCK_CR)
140 #define LCK_COMPAT_PR  (LCK_COMPAT_PW | LCK_PR)
141 #define LCK_COMPAT_CW  (LCK_COMPAT_PW | LCK_CW)
142 #define LCK_COMPAT_CR  (LCK_COMPAT_CW | LCK_PR | LCK_PW)
143 #define LCK_COMPAT_NL  (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
144 #define LCK_COMPAT_GROUP  (LCK_GROUP | LCK_NL)
145 #define LCK_COMPAT_COS (LCK_COS)
146 /** @} Lock Compatibility Matrix */
147 
148 extern ldlm_mode_t lck_compat_array[];
149 
lockmode_verify(ldlm_mode_t mode)150 static inline void lockmode_verify(ldlm_mode_t mode)
151 {
152        LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
153 }
154 
lockmode_compat(ldlm_mode_t exist_mode,ldlm_mode_t new_mode)155 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
156 {
157        return (lck_compat_array[exist_mode] & new_mode);
158 }
159 
160 /*
161  *
162  * cluster name spaces
163  *
164  */
165 
166 #define DLM_OST_NAMESPACE 1
167 #define DLM_MDS_NAMESPACE 2
168 
169 /* XXX
170    - do we just separate this by security domains and use a prefix for
171      multiple namespaces in the same domain?
172    -
173 */
174 
175 /**
176  * Locking rules for LDLM:
177  *
178  * lr_lock
179  *
180  * lr_lock
181  *     waiting_locks_spinlock
182  *
183  * lr_lock
184  *     led_lock
185  *
186  * lr_lock
187  *     ns_lock
188  *
189  * lr_lvb_mutex
190  *     lr_lock
191  *
192  */
193 
194 struct ldlm_pool;
195 struct ldlm_lock;
196 struct ldlm_resource;
197 struct ldlm_namespace;
198 
199 /**
200  * Operations on LDLM pools.
201  * LDLM pool is a pool of locks in the namespace without any implicitly
202  * specified limits.
203  * Locks in the pool are organized in LRU.
204  * Local memory pressure or server instructions (e.g. mempressure on server)
205  * can trigger freeing of locks from the pool
206  */
207 struct ldlm_pool_ops {
208 	/** Recalculate pool \a pl usage */
209 	int (*po_recalc)(struct ldlm_pool *pl);
210 	/** Cancel at least \a nr locks from pool \a pl */
211 	int (*po_shrink)(struct ldlm_pool *pl, int nr,
212 			 gfp_t gfp_mask);
213 };
214 
215 /** One second for pools thread check interval. Each pool has own period. */
216 #define LDLM_POOLS_THREAD_PERIOD (1)
217 
218 /** ~6% margin for modest pools. See ldlm_pool.c for details. */
219 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
220 
221 /** Default recalc period for server side pools in sec. */
222 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
223 
224 /** Default recalc period for client side pools in sec. */
225 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
226 
227 /**
228  * LDLM pool structure to track granted locks.
229  * For purposes of determining when to release locks on e.g. memory pressure.
230  * This feature is commonly referred to as lru_resize.
231  */
232 struct ldlm_pool {
233 	/** Pool debugfs directory. */
234 	struct dentry		*pl_debugfs_entry;
235 	/** Pool name, must be long enough to hold compound proc entry name. */
236 	char			pl_name[100];
237 	/** Lock for protecting SLV/CLV updates. */
238 	spinlock_t		pl_lock;
239 	/** Number of allowed locks in in pool, both, client and server side. */
240 	atomic_t		pl_limit;
241 	/** Number of granted locks in */
242 	atomic_t		pl_granted;
243 	/** Grant rate per T. */
244 	atomic_t		pl_grant_rate;
245 	/** Cancel rate per T. */
246 	atomic_t		pl_cancel_rate;
247 	/** Server lock volume (SLV). Protected by pl_lock. */
248 	__u64			pl_server_lock_volume;
249 	/** Current biggest client lock volume. Protected by pl_lock. */
250 	__u64			pl_client_lock_volume;
251 	/** Lock volume factor. SLV on client is calculated as following:
252 	 *  server_slv * lock_volume_factor. */
253 	atomic_t		pl_lock_volume_factor;
254 	/** Time when last SLV from server was obtained. */
255 	time64_t		pl_recalc_time;
256 	/** Recalculation period for pool. */
257 	time64_t		pl_recalc_period;
258 	/** Recalculation and shrink operations. */
259 	const struct ldlm_pool_ops	*pl_ops;
260 	/** Number of planned locks for next period. */
261 	int			pl_grant_plan;
262 	/** Pool statistics. */
263 	struct lprocfs_stats	*pl_stats;
264 
265 	/* sysfs object */
266 	struct kobject		 pl_kobj;
267 	struct completion	 pl_kobj_unregister;
268 };
269 
270 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
271 
272 /**
273  * LVB operations.
274  * LVB is Lock Value Block. This is a special opaque (to LDLM) value that could
275  * be associated with an LDLM lock and transferred from client to server and
276  * back.
277  *
278  * Currently LVBs are used by:
279  *  - OSC-OST code to maintain current object size/times
280  *  - layout lock code to return the layout when the layout lock is granted
281  */
282 struct ldlm_valblock_ops {
283 	int (*lvbo_init)(struct ldlm_resource *res);
284 	int (*lvbo_update)(struct ldlm_resource *res,
285 			   struct ptlrpc_request *r,
286 			   int increase);
287 	int (*lvbo_free)(struct ldlm_resource *res);
288 	/* Return size of lvb data appropriate RPC size can be reserved */
289 	int (*lvbo_size)(struct ldlm_lock *lock);
290 	/* Called to fill in lvb data to RPC buffer @buf */
291 	int (*lvbo_fill)(struct ldlm_lock *lock, void *buf, int buflen);
292 };
293 
294 /**
295  * LDLM pools related, type of lock pool in the namespace.
296  * Greedy means release cached locks aggressively
297  */
298 typedef enum {
299 	LDLM_NAMESPACE_GREEDY = 1 << 0,
300 	LDLM_NAMESPACE_MODEST = 1 << 1
301 } ldlm_appetite_t;
302 
303 struct ldlm_ns_bucket {
304 	/** back pointer to namespace */
305 	struct ldlm_namespace      *nsb_namespace;
306 	/**
307 	 * Estimated lock callback time.  Used by adaptive timeout code to
308 	 * avoid spurious client evictions due to unresponsiveness when in
309 	 * fact the network or overall system load is at fault
310 	 */
311 	struct adaptive_timeout     nsb_at_estimate;
312 };
313 
314 enum {
315 	/** LDLM namespace lock stats */
316 	LDLM_NSS_LOCKS	  = 0,
317 	LDLM_NSS_LAST
318 };
319 
320 typedef enum {
321 	/** invalid type */
322 	LDLM_NS_TYPE_UNKNOWN    = 0,
323 	/** mdc namespace */
324 	LDLM_NS_TYPE_MDC,
325 	/** mds namespace */
326 	LDLM_NS_TYPE_MDT,
327 	/** osc namespace */
328 	LDLM_NS_TYPE_OSC,
329 	/** ost namespace */
330 	LDLM_NS_TYPE_OST,
331 	/** mgc namespace */
332 	LDLM_NS_TYPE_MGC,
333 	/** mgs namespace */
334 	LDLM_NS_TYPE_MGT,
335 } ldlm_ns_type_t;
336 
337 /**
338  * LDLM Namespace.
339  *
340  * Namespace serves to contain locks related to a particular service.
341  * There are two kinds of namespaces:
342  * - Server namespace has knowledge of all locks and is therefore authoritative
343  *   to make decisions like what locks could be granted and what conflicts
344  *   exist during new lock enqueue.
345  * - Client namespace only has limited knowledge about locks in the namespace,
346  *   only seeing locks held by the client.
347  *
348  * Every Lustre service has one server namespace present on the server serving
349  * that service. Every client connected to the service has a client namespace
350  * for it.
351  * Every lock obtained by client in that namespace is actually represented by
352  * two in-memory locks. One on the server and one on the client. The locks are
353  * linked by a special cookie by which one node can tell to the other which lock
354  * it actually means during communications. Such locks are called remote locks.
355  * The locks held by server only without any reference to a client are called
356  * local locks.
357  */
358 struct ldlm_namespace {
359 	/** Backward link to OBD, required for LDLM pool to store new SLV. */
360 	struct obd_device	*ns_obd;
361 
362 	/** Flag indicating if namespace is on client instead of server */
363 	ldlm_side_t		ns_client;
364 
365 	/** Resource hash table for namespace. */
366 	struct cfs_hash		*ns_rs_hash;
367 
368 	/** serialize */
369 	spinlock_t		ns_lock;
370 
371 	/** big refcount (by bucket) */
372 	atomic_t		ns_bref;
373 
374 	/**
375 	 * Namespace connect flags supported by server (may be changed via
376 	 * /proc, LRU resize may be disabled/enabled).
377 	 */
378 	__u64			ns_connect_flags;
379 
380 	/** Client side original connect flags supported by server. */
381 	__u64			ns_orig_connect_flags;
382 
383 	/* namespace debugfs dir entry */
384 	struct dentry		*ns_debugfs_entry;
385 
386 	/**
387 	 * Position in global namespace list linking all namespaces on
388 	 * the node.
389 	 */
390 	struct list_head		ns_list_chain;
391 
392 	/**
393 	 * List of unused locks for this namespace. This list is also called
394 	 * LRU lock list.
395 	 * Unused locks are locks with zero reader/writer reference counts.
396 	 * This list is only used on clients for lock caching purposes.
397 	 * When we want to release some locks voluntarily or if server wants
398 	 * us to release some locks due to e.g. memory pressure, we take locks
399 	 * to release from the head of this list.
400 	 * Locks are linked via l_lru field in \see struct ldlm_lock.
401 	 */
402 	struct list_head		ns_unused_list;
403 	/** Number of locks in the LRU list above */
404 	int			ns_nr_unused;
405 
406 	/**
407 	 * Maximum number of locks permitted in the LRU. If 0, means locks
408 	 * are managed by pools and there is no preset limit, rather it is all
409 	 * controlled by available memory on this client and on server.
410 	 */
411 	unsigned int		ns_max_unused;
412 	/** Maximum allowed age (last used time) for locks in the LRU */
413 	unsigned int		ns_max_age;
414 
415 	/**
416 	 * Used to rate-limit ldlm_namespace_dump calls.
417 	 * \see ldlm_namespace_dump. Increased by 10 seconds every time
418 	 * it is called.
419 	 */
420 	unsigned long		ns_next_dump;
421 
422 	/**
423 	 * LVB operations for this namespace.
424 	 * \see struct ldlm_valblock_ops
425 	 */
426 	struct ldlm_valblock_ops *ns_lvbo;
427 
428 	/**
429 	 * Used by filter code to store pointer to OBD of the service.
430 	 * Should be dropped in favor of \a ns_obd
431 	 */
432 	void			*ns_lvbp;
433 
434 	/**
435 	 * Wait queue used by __ldlm_namespace_free. Gets woken up every time
436 	 * a resource is removed.
437 	 */
438 	wait_queue_head_t		ns_waitq;
439 	/** LDLM pool structure for this namespace */
440 	struct ldlm_pool	ns_pool;
441 	/** Definition of how eagerly unused locks will be released from LRU */
442 	ldlm_appetite_t		ns_appetite;
443 
444 	/** Limit of parallel AST RPC count. */
445 	unsigned		ns_max_parallel_ast;
446 
447 	/** Callback to cancel locks before replaying it during recovery. */
448 	ldlm_cancel_for_recovery ns_cancel_for_recovery;
449 
450 	/** LDLM lock stats */
451 	struct lprocfs_stats	*ns_stats;
452 
453 	/**
454 	 * Flag to indicate namespace is being freed. Used to determine if
455 	 * recalculation of LDLM pool statistics should be skipped.
456 	 */
457 	unsigned		ns_stopping:1;
458 
459 	struct kobject		ns_kobj; /* sysfs object */
460 	struct completion	ns_kobj_unregister;
461 };
462 
463 /**
464  * Returns 1 if namespace \a ns supports early lock cancel (ELC).
465  */
ns_connect_cancelset(struct ldlm_namespace * ns)466 static inline int ns_connect_cancelset(struct ldlm_namespace *ns)
467 {
468 	LASSERT(ns != NULL);
469 	return !!(ns->ns_connect_flags & OBD_CONNECT_CANCELSET);
470 }
471 
472 /**
473  * Returns 1 if this namespace supports lru_resize.
474  */
ns_connect_lru_resize(struct ldlm_namespace * ns)475 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
476 {
477 	LASSERT(ns != NULL);
478 	return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
479 }
480 
ns_register_cancel(struct ldlm_namespace * ns,ldlm_cancel_for_recovery arg)481 static inline void ns_register_cancel(struct ldlm_namespace *ns,
482 				      ldlm_cancel_for_recovery arg)
483 {
484 	LASSERT(ns != NULL);
485 	ns->ns_cancel_for_recovery = arg;
486 }
487 
488 struct ldlm_lock;
489 
490 /** Type for blocking callback function of a lock. */
491 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
492 				      struct ldlm_lock_desc *new, void *data,
493 				      int flag);
494 /** Type for completion callback function of a lock. */
495 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, __u64 flags,
496 					void *data);
497 /** Type for glimpse callback function of a lock. */
498 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
499 
500 /** Work list for sending GL ASTs to multiple locks. */
501 struct ldlm_glimpse_work {
502 	struct ldlm_lock	*gl_lock; /* lock to glimpse */
503 	struct list_head		 gl_list; /* linkage to other gl work structs */
504 	__u32			 gl_flags;/* see LDLM_GL_WORK_* below */
505 	union ldlm_gl_desc	*gl_desc; /* glimpse descriptor to be packed in
506 					   * glimpse callback request */
507 };
508 
509 /** The ldlm_glimpse_work is allocated on the stack and should not be freed. */
510 #define LDLM_GL_WORK_NOFREE 0x1
511 
512 /** Interval node data for each LDLM_EXTENT lock. */
513 struct ldlm_interval {
514 	struct interval_node	li_node;  /* node for tree management */
515 	struct list_head		li_group; /* the locks which have the same
516 					   * policy - group of the policy */
517 };
518 
519 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
520 
521 /**
522  * Interval tree for extent locks.
523  * The interval tree must be accessed under the resource lock.
524  * Interval trees are used for granted extent locks to speed up conflicts
525  * lookup. See ldlm/interval_tree.c for more details.
526  */
527 struct ldlm_interval_tree {
528 	/** Tree size. */
529 	int			lit_size;
530 	ldlm_mode_t		lit_mode;  /* lock mode */
531 	struct interval_node	*lit_root; /* actual ldlm_interval */
532 };
533 
534 /** Whether to track references to exports by LDLM locks. */
535 #define LUSTRE_TRACKS_LOCK_EXP_REFS (0)
536 
537 /** Cancel flags. */
538 typedef enum {
539 	LCF_ASYNC      = 0x1, /* Cancel locks asynchronously. */
540 	LCF_LOCAL      = 0x2, /* Cancel locks locally, not notifing server */
541 	LCF_BL_AST     = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
542 			       * in the same RPC */
543 } ldlm_cancel_flags_t;
544 
545 struct ldlm_flock {
546 	__u64 start;
547 	__u64 end;
548 	__u64 owner;
549 	__u64 blocking_owner;
550 	struct obd_export *blocking_export;
551 	/* Protected by the hash lock */
552 	__u32 blocking_refs;
553 	__u32 pid;
554 };
555 
556 typedef union {
557 	struct ldlm_extent l_extent;
558 	struct ldlm_flock l_flock;
559 	struct ldlm_inodebits l_inodebits;
560 } ldlm_policy_data_t;
561 
562 void ldlm_convert_policy_to_local(struct obd_export *exp, ldlm_type_t type,
563 				  const ldlm_wire_policy_data_t *wpolicy,
564 				  ldlm_policy_data_t *lpolicy);
565 
566 enum lvb_type {
567 	LVB_T_NONE	= 0,
568 	LVB_T_OST	= 1,
569 	LVB_T_LQUOTA	= 2,
570 	LVB_T_LAYOUT	= 3,
571 };
572 
573 /**
574  * LDLM lock structure
575  *
576  * Represents a single LDLM lock and its state in memory. Each lock is
577  * associated with a single ldlm_resource, the object which is being
578  * locked. There may be multiple ldlm_locks on a single resource,
579  * depending on the lock type and whether the locks are conflicting or
580  * not.
581  */
582 struct ldlm_lock {
583 	/**
584 	 * Local lock handle.
585 	 * When remote side wants to tell us about a lock, they address
586 	 * it by this opaque handle.  The handle does not hold a
587 	 * reference on the ldlm_lock, so it can be safely passed to
588 	 * other threads or nodes. When the lock needs to be accessed
589 	 * from the handle, it is looked up again in the lock table, and
590 	 * may no longer exist.
591 	 *
592 	 * Must be first in the structure.
593 	 */
594 	struct portals_handle	l_handle;
595 	/**
596 	 * Lock reference count.
597 	 * This is how many users have pointers to actual structure, so that
598 	 * we do not accidentally free lock structure that is in use.
599 	 */
600 	atomic_t		l_refc;
601 	/**
602 	 * Internal spinlock protects l_resource.  We should hold this lock
603 	 * first before taking res_lock.
604 	 */
605 	spinlock_t		l_lock;
606 	/**
607 	 * Pointer to actual resource this lock is in.
608 	 * ldlm_lock_change_resource() can change this.
609 	 */
610 	struct ldlm_resource	*l_resource;
611 	/**
612 	 * List item for client side LRU list.
613 	 * Protected by ns_lock in struct ldlm_namespace.
614 	 */
615 	struct list_head		l_lru;
616 	/**
617 	 * Linkage to resource's lock queues according to current lock state.
618 	 * (could be granted, waiting or converting)
619 	 * Protected by lr_lock in struct ldlm_resource.
620 	 */
621 	struct list_head		l_res_link;
622 	/**
623 	 * Tree node for ldlm_extent.
624 	 */
625 	struct ldlm_interval	*l_tree_node;
626 	/**
627 	 * Per export hash of locks.
628 	 * Protected by per-bucket exp->exp_lock_hash locks.
629 	 */
630 	struct hlist_node	l_exp_hash;
631 	/**
632 	 * Per export hash of flock locks.
633 	 * Protected by per-bucket exp->exp_flock_hash locks.
634 	 */
635 	struct hlist_node	l_exp_flock_hash;
636 	/**
637 	 * Requested mode.
638 	 * Protected by lr_lock.
639 	 */
640 	ldlm_mode_t		l_req_mode;
641 	/**
642 	 * Granted mode, also protected by lr_lock.
643 	 */
644 	ldlm_mode_t		l_granted_mode;
645 	/** Lock completion handler pointer. Called when lock is granted. */
646 	ldlm_completion_callback l_completion_ast;
647 	/**
648 	 * Lock blocking AST handler pointer.
649 	 * It plays two roles:
650 	 * - as a notification of an attempt to queue a conflicting lock (once)
651 	 * - as a notification when the lock is being cancelled.
652 	 *
653 	 * As such it's typically called twice: once for the initial conflict
654 	 * and then once more when the last user went away and the lock is
655 	 * cancelled (could happen recursively).
656 	 */
657 	ldlm_blocking_callback	l_blocking_ast;
658 	/**
659 	 * Lock glimpse handler.
660 	 * Glimpse handler is used to obtain LVB updates from a client by
661 	 * server
662 	 */
663 	ldlm_glimpse_callback	l_glimpse_ast;
664 
665 	/**
666 	 * Lock export.
667 	 * This is a pointer to actual client export for locks that were granted
668 	 * to clients. Used server-side.
669 	 */
670 	struct obd_export	*l_export;
671 	/**
672 	 * Lock connection export.
673 	 * Pointer to server export on a client.
674 	 */
675 	struct obd_export	*l_conn_export;
676 
677 	/**
678 	 * Remote lock handle.
679 	 * If the lock is remote, this is the handle of the other side lock
680 	 * (l_handle)
681 	 */
682 	struct lustre_handle	l_remote_handle;
683 
684 	/**
685 	 * Representation of private data specific for a lock type.
686 	 * Examples are: extent range for extent lock or bitmask for ibits locks
687 	 */
688 	ldlm_policy_data_t	l_policy_data;
689 
690 	/**
691 	 * Lock state flags. Protected by lr_lock.
692 	 * \see lustre_dlm_flags.h where the bits are defined.
693 	 */
694 	__u64			l_flags;
695 
696 	/**
697 	 * Lock r/w usage counters.
698 	 * Protected by lr_lock.
699 	 */
700 	__u32			l_readers;
701 	__u32			l_writers;
702 	/**
703 	 * If the lock is granted, a process sleeps on this waitq to learn when
704 	 * it's no longer in use.  If the lock is not granted, a process sleeps
705 	 * on this waitq to learn when it becomes granted.
706 	 */
707 	wait_queue_head_t		l_waitq;
708 
709 	/**
710 	 * Seconds. It will be updated if there is any activity related to
711 	 * the lock, e.g. enqueue the lock or send blocking AST.
712 	 */
713 	time64_t		l_last_activity;
714 
715 	/**
716 	 * Time last used by e.g. being matched by lock match.
717 	 * Jiffies. Should be converted to time if needed.
718 	 */
719 	unsigned long		l_last_used;
720 
721 	/** Originally requested extent for the extent lock. */
722 	struct ldlm_extent	l_req_extent;
723 
724 	/*
725 	 * Client-side-only members.
726 	 */
727 
728 	enum lvb_type	      l_lvb_type;
729 
730 	/**
731 	 * Temporary storage for a LVB received during an enqueue operation.
732 	 */
733 	__u32			l_lvb_len;
734 	void			*l_lvb_data;
735 
736 	/** Private storage for lock user. Opaque to LDLM. */
737 	void			*l_ast_data;
738 
739 	/*
740 	 * Server-side-only members.
741 	 */
742 
743 	/**
744 	 * Connection cookie for the client originating the operation.
745 	 * Used by Commit on Share (COS) code. Currently only used for
746 	 * inodebits locks on MDS.
747 	 */
748 	__u64			l_client_cookie;
749 
750 	/**
751 	 * List item for locks waiting for cancellation from clients.
752 	 * The lists this could be linked into are:
753 	 * waiting_locks_list (protected by waiting_locks_spinlock),
754 	 * then if the lock timed out, it is moved to
755 	 * expired_lock_thread.elt_expired_locks for further processing.
756 	 * Protected by elt_lock.
757 	 */
758 	struct list_head		l_pending_chain;
759 
760 	/**
761 	 * Set when lock is sent a blocking AST. Time in seconds when timeout
762 	 * is reached and client holding this lock could be evicted.
763 	 * This timeout could be further extended by e.g. certain IO activity
764 	 * under this lock.
765 	 * \see ost_rw_prolong_locks
766 	 */
767 	unsigned long		l_callback_timeout;
768 
769 	/** Local PID of process which created this lock. */
770 	__u32			l_pid;
771 
772 	/**
773 	 * Number of times blocking AST was sent for this lock.
774 	 * This is for debugging. Valid values are 0 and 1, if there is an
775 	 * attempt to send blocking AST more than once, an assertion would be
776 	 * hit. \see ldlm_work_bl_ast_lock
777 	 */
778 	int			l_bl_ast_run;
779 	/** List item ldlm_add_ast_work_item() for case of blocking ASTs. */
780 	struct list_head		l_bl_ast;
781 	/** List item ldlm_add_ast_work_item() for case of completion ASTs. */
782 	struct list_head		l_cp_ast;
783 	/** For ldlm_add_ast_work_item() for "revoke" AST used in COS. */
784 	struct list_head		l_rk_ast;
785 
786 	/**
787 	 * Pointer to a conflicting lock that caused blocking AST to be sent
788 	 * for this lock
789 	 */
790 	struct ldlm_lock	*l_blocking_lock;
791 
792 	/**
793 	 * Protected by lr_lock, linkages to "skip lists".
794 	 * For more explanations of skip lists see ldlm/ldlm_inodebits.c
795 	 */
796 	struct list_head		l_sl_mode;
797 	struct list_head		l_sl_policy;
798 
799 	/** Reference tracking structure to debug leaked locks. */
800 	struct lu_ref		l_reference;
801 #if LUSTRE_TRACKS_LOCK_EXP_REFS
802 	/* Debugging stuff for bug 20498, for tracking export references. */
803 	/** number of export references taken */
804 	int			l_exp_refs_nr;
805 	/** link all locks referencing one export */
806 	struct list_head		l_exp_refs_link;
807 	/** referenced export object */
808 	struct obd_export	*l_exp_refs_target;
809 #endif
810 	/**
811 	 * export blocking dlm lock list, protected by
812 	 * l_export->exp_bl_list_lock.
813 	 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
814 	 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock.
815 	 */
816 	struct list_head		l_exp_list;
817 };
818 
819 /**
820  * LDLM resource description.
821  * Basically, resource is a representation for a single object.
822  * Object has a name which is currently 4 64-bit integers. LDLM user is
823  * responsible for creation of a mapping between objects it wants to be
824  * protected and resource names.
825  *
826  * A resource can only hold locks of a single lock type, though there may be
827  * multiple ldlm_locks on a single resource, depending on the lock type and
828  * whether the locks are conflicting or not.
829  */
830 struct ldlm_resource {
831 	struct ldlm_ns_bucket	*lr_ns_bucket;
832 
833 	/**
834 	 * List item for list in namespace hash.
835 	 * protected by ns_lock
836 	 */
837 	struct hlist_node	lr_hash;
838 
839 	/** Spinlock to protect locks under this resource. */
840 	spinlock_t		lr_lock;
841 
842 	/**
843 	 * protected by lr_lock
844 	 * @{ */
845 	/** List of locks in granted state */
846 	struct list_head		lr_granted;
847 	/**
848 	 * List of locks that could not be granted due to conflicts and
849 	 * that are waiting for conflicts to go away */
850 	struct list_head		lr_waiting;
851 	/** @} */
852 
853 	/* XXX No longer needed? Remove ASAP */
854 	ldlm_mode_t		lr_most_restr;
855 
856 	/** Type of locks this resource can hold. Only one type per resource. */
857 	ldlm_type_t		lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK,IBITS} */
858 
859 	/** Resource name */
860 	struct ldlm_res_id	lr_name;
861 	/** Reference count for this resource */
862 	atomic_t		lr_refcount;
863 
864 	/**
865 	 * Interval trees (only for extent locks) for all modes of this resource
866 	 */
867 	struct ldlm_interval_tree lr_itree[LCK_MODE_NUM];
868 
869 	/**
870 	 * Server-side-only lock value block elements.
871 	 * To serialize lvbo_init.
872 	 */
873 	struct mutex		lr_lvb_mutex;
874 	int			lr_lvb_len;
875 	/** protected by lr_lock */
876 	void			*lr_lvb_data;
877 
878 	/** When the resource was considered as contended. */
879 	unsigned long		lr_contention_time;
880 	/** List of references to this resource. For debugging. */
881 	struct lu_ref		lr_reference;
882 
883 	struct inode		*lr_lvb_inode;
884 };
885 
ldlm_has_layout(struct ldlm_lock * lock)886 static inline bool ldlm_has_layout(struct ldlm_lock *lock)
887 {
888 	return lock->l_resource->lr_type == LDLM_IBITS &&
889 		lock->l_policy_data.l_inodebits.bits & MDS_INODELOCK_LAYOUT;
890 }
891 
892 static inline char *
ldlm_ns_name(struct ldlm_namespace * ns)893 ldlm_ns_name(struct ldlm_namespace *ns)
894 {
895 	return ns->ns_rs_hash->hs_name;
896 }
897 
898 static inline struct ldlm_namespace *
ldlm_res_to_ns(struct ldlm_resource * res)899 ldlm_res_to_ns(struct ldlm_resource *res)
900 {
901 	return res->lr_ns_bucket->nsb_namespace;
902 }
903 
904 static inline struct ldlm_namespace *
ldlm_lock_to_ns(struct ldlm_lock * lock)905 ldlm_lock_to_ns(struct ldlm_lock *lock)
906 {
907 	return ldlm_res_to_ns(lock->l_resource);
908 }
909 
910 static inline char *
ldlm_lock_to_ns_name(struct ldlm_lock * lock)911 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
912 {
913 	return ldlm_ns_name(ldlm_lock_to_ns(lock));
914 }
915 
916 static inline struct adaptive_timeout *
ldlm_lock_to_ns_at(struct ldlm_lock * lock)917 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
918 {
919 	return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
920 }
921 
ldlm_lvbo_init(struct ldlm_resource * res)922 static inline int ldlm_lvbo_init(struct ldlm_resource *res)
923 {
924 	struct ldlm_namespace *ns = ldlm_res_to_ns(res);
925 
926 	if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_init != NULL)
927 		return ns->ns_lvbo->lvbo_init(res);
928 
929 	return 0;
930 }
931 
ldlm_lvbo_size(struct ldlm_lock * lock)932 static inline int ldlm_lvbo_size(struct ldlm_lock *lock)
933 {
934 	struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
935 
936 	if (ns->ns_lvbo != NULL && ns->ns_lvbo->lvbo_size != NULL)
937 		return ns->ns_lvbo->lvbo_size(lock);
938 
939 	return 0;
940 }
941 
ldlm_lvbo_fill(struct ldlm_lock * lock,void * buf,int len)942 static inline int ldlm_lvbo_fill(struct ldlm_lock *lock, void *buf, int len)
943 {
944 	struct ldlm_namespace *ns = ldlm_lock_to_ns(lock);
945 
946 	if (ns->ns_lvbo != NULL) {
947 		LASSERT(ns->ns_lvbo->lvbo_fill != NULL);
948 		return ns->ns_lvbo->lvbo_fill(lock, buf, len);
949 	}
950 	return 0;
951 }
952 
953 struct ldlm_ast_work {
954 	struct ldlm_lock      *w_lock;
955 	int		    w_blocking;
956 	struct ldlm_lock_desc  w_desc;
957 	struct list_head	     w_list;
958 	int		    w_flags;
959 	void		  *w_data;
960 	int		    w_datalen;
961 };
962 
963 /**
964  * Common ldlm_enqueue parameters
965  */
966 struct ldlm_enqueue_info {
967 	__u32 ei_type;   /** Type of the lock being enqueued. */
968 	__u32 ei_mode;   /** Mode of the lock being enqueued. */
969 	void *ei_cb_bl;  /** blocking lock callback */
970 	void *ei_cb_cp;  /** lock completion callback */
971 	void *ei_cb_gl;  /** lock glimpse callback */
972 	void *ei_cbdata; /** Data to be passed into callbacks. */
973 };
974 
975 extern struct obd_ops ldlm_obd_ops;
976 
977 extern char *ldlm_lockname[];
978 char *ldlm_it2str(int it);
979 
980 /**
981  * Just a fancy CDEBUG call with log level preset to LDLM_DEBUG.
982  * For the cases where we do not have actual lock to print along
983  * with a debugging message that is ldlm-related
984  */
985 #define LDLM_DEBUG_NOLOCK(format, a...)			\
986 	CDEBUG(D_DLMTRACE, "### " format "\n", ##a)
987 
988 /**
989  * Support function for lock information printing into debug logs.
990  * \see LDLM_DEBUG
991  */
992 #define ldlm_lock_debug(msgdata, mask, cdls, lock, fmt, a...) do {      \
993 	CFS_CHECK_STACK(msgdata, mask, cdls);			   \
994 									\
995 	if (((mask) & D_CANTMASK) != 0 ||			       \
996 	    ((libcfs_debug & (mask)) != 0 &&			    \
997 	     (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0))	  \
998 		_ldlm_lock_debug(lock, msgdata, fmt, ##a);	      \
999 } while (0)
1000 
1001 void _ldlm_lock_debug(struct ldlm_lock *lock,
1002 		      struct libcfs_debug_msg_data *data,
1003 		      const char *fmt, ...)
1004 	__printf(3, 4);
1005 
1006 /**
1007  * Rate-limited version of lock printing function.
1008  */
1009 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do {			 \
1010 	static struct cfs_debug_limit_state _ldlm_cdls;			   \
1011 	LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, mask, &_ldlm_cdls);	      \
1012 	ldlm_lock_debug(&msgdata, mask, &_ldlm_cdls, lock, "### " fmt, ##a);\
1013 } while (0)
1014 
1015 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
1016 #define LDLM_WARN(lock, fmt, a...)  LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
1017 
1018 /** Non-rate-limited lock printing function for debugging purposes. */
1019 #define LDLM_DEBUG(lock, fmt, a...)   do {				  \
1020 	if (likely(lock != NULL)) {					    \
1021 		LIBCFS_DEBUG_MSG_DATA_DECL(msgdata, D_DLMTRACE, NULL);      \
1022 		ldlm_lock_debug(&msgdata, D_DLMTRACE, NULL, lock,	    \
1023 				"### " fmt, ##a);			    \
1024 	} else {							    \
1025 		LDLM_DEBUG_NOLOCK("no dlm lock: " fmt, ##a);		    \
1026 	}								    \
1027 } while (0)
1028 
1029 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, __u64 *flags,
1030 				      int first_enq, ldlm_error_t *err,
1031 				      struct list_head *work_list);
1032 
1033 /**
1034  * Return values for lock iterators.
1035  * Also used during deciding of lock grants and cancellations.
1036  */
1037 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
1038 #define LDLM_ITER_STOP     2 /* stop iterating */
1039 
1040 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
1041 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
1042 
1043 /** \defgroup ldlm_iterator Lock iterators
1044  *
1045  * LDLM provides for a way to iterate through every lock on a resource or
1046  * namespace or every resource in a namespace.
1047  * @{ */
1048 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
1049 			  ldlm_iterator_t iter, void *data);
1050 /** @} ldlm_iterator */
1051 
1052 int ldlm_replay_locks(struct obd_import *imp);
1053 
1054 /* ldlm_flock.c */
1055 int ldlm_flock_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1056 
1057 /* ldlm_extent.c */
1058 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
1059 
1060 struct ldlm_callback_suite {
1061 	ldlm_completion_callback lcs_completion;
1062 	ldlm_blocking_callback   lcs_blocking;
1063 	ldlm_glimpse_callback    lcs_glimpse;
1064 };
1065 
1066 /* ldlm_lockd.c */
1067 int ldlm_get_ref(void);
1068 void ldlm_put_ref(void);
1069 struct ldlm_lock *ldlm_request_lock(struct ptlrpc_request *req);
1070 
1071 /* ldlm_lock.c */
1072 void ldlm_lock2handle(const struct ldlm_lock *lock,
1073 		      struct lustre_handle *lockh);
1074 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, __u64 flags);
1075 void ldlm_cancel_callback(struct ldlm_lock *);
1076 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
1077 int ldlm_lock_set_data(struct lustre_handle *, void *);
1078 
1079 /**
1080  * Obtain a lock reference by its handle.
1081  */
ldlm_handle2lock(const struct lustre_handle * h)1082 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
1083 {
1084 	return __ldlm_handle2lock(h, 0);
1085 }
1086 
1087 #define LDLM_LOCK_REF_DEL(lock) \
1088 	lu_ref_del(&lock->l_reference, "handle", current)
1089 
1090 static inline struct ldlm_lock *
ldlm_handle2lock_long(const struct lustre_handle * h,__u64 flags)1091 ldlm_handle2lock_long(const struct lustre_handle *h, __u64 flags)
1092 {
1093 	struct ldlm_lock *lock;
1094 
1095 	lock = __ldlm_handle2lock(h, flags);
1096 	if (lock != NULL)
1097 		LDLM_LOCK_REF_DEL(lock);
1098 	return lock;
1099 }
1100 
1101 /**
1102  * Update Lock Value Block Operations (LVBO) on a resource taking into account
1103  * data from request \a r
1104  */
ldlm_res_lvbo_update(struct ldlm_resource * res,struct ptlrpc_request * r,int increase)1105 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1106 				       struct ptlrpc_request *r, int increase)
1107 {
1108 	if (ldlm_res_to_ns(res)->ns_lvbo &&
1109 	    ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1110 		return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1111 								 increase);
1112 	}
1113 	return 0;
1114 }
1115 
1116 int ldlm_error2errno(ldlm_error_t error);
1117 
1118 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1119 void ldlm_dump_export_locks(struct obd_export *exp);
1120 #endif
1121 
1122 /**
1123  * Release a temporary lock reference obtained by ldlm_handle2lock() or
1124  * __ldlm_handle2lock().
1125  */
1126 #define LDLM_LOCK_PUT(lock)		     \
1127 do {					    \
1128 	LDLM_LOCK_REF_DEL(lock);		\
1129 	/*LDLM_DEBUG((lock), "put");*/	  \
1130 	ldlm_lock_put(lock);		    \
1131 } while (0)
1132 
1133 /**
1134  * Release a lock reference obtained by some other means (see
1135  * LDLM_LOCK_PUT()).
1136  */
1137 #define LDLM_LOCK_RELEASE(lock)		 \
1138 do {					    \
1139 	/*LDLM_DEBUG((lock), "put");*/	  \
1140 	ldlm_lock_put(lock);		    \
1141 } while (0)
1142 
1143 #define LDLM_LOCK_GET(lock)		     \
1144 ({					      \
1145 	ldlm_lock_get(lock);		    \
1146 	/*LDLM_DEBUG((lock), "get");*/	  \
1147 	lock;				   \
1148 })
1149 
1150 #define ldlm_lock_list_put(head, member, count)		     \
1151 ({								  \
1152 	struct ldlm_lock *_lock, *_next;			    \
1153 	int c = count;					      \
1154 	list_for_each_entry_safe(_lock, _next, head, member) {  \
1155 		if (c-- == 0)				       \
1156 			break;				      \
1157 		list_del_init(&_lock->member);		  \
1158 		LDLM_LOCK_RELEASE(_lock);			   \
1159 	}							   \
1160 	LASSERT(c <= 0);					    \
1161 })
1162 
1163 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1164 void ldlm_lock_put(struct ldlm_lock *lock);
1165 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1166 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1167 int  ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1168 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1169 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1170 void ldlm_lock_fail_match_locked(struct ldlm_lock *lock);
1171 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1172 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1173 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, __u64 flags,
1174 			    const struct ldlm_res_id *, ldlm_type_t type,
1175 			    ldlm_policy_data_t *, ldlm_mode_t mode,
1176 			    struct lustre_handle *, int unref);
1177 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1178 					__u64 *bits);
1179 void ldlm_lock_cancel(struct ldlm_lock *lock);
1180 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1181 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1182 
1183 /* resource.c */
1184 struct ldlm_namespace *
1185 ldlm_namespace_new(struct obd_device *obd, char *name,
1186 		   ldlm_side_t client, ldlm_appetite_t apt,
1187 		   ldlm_ns_type_t ns_type);
1188 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, __u64 flags);
1189 void ldlm_namespace_get(struct ldlm_namespace *ns);
1190 void ldlm_namespace_put(struct ldlm_namespace *ns);
1191 int ldlm_debugfs_setup(void);
1192 void ldlm_debugfs_cleanup(void);
1193 
1194 /* resource.c - internal */
1195 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1196 					struct ldlm_resource *parent,
1197 					const struct ldlm_res_id *,
1198 					ldlm_type_t type, int create);
1199 int ldlm_resource_putref(struct ldlm_resource *res);
1200 void ldlm_resource_add_lock(struct ldlm_resource *res,
1201 			    struct list_head *head,
1202 			    struct ldlm_lock *lock);
1203 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1204 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1205 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1206 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1207 void ldlm_resource_dump(int level, struct ldlm_resource *);
1208 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1209 			      const struct ldlm_res_id *);
1210 
1211 #define LDLM_RESOURCE_ADDREF(res) do {				  \
1212 	lu_ref_add_atomic(&(res)->lr_reference, __func__, current);  \
1213 } while (0)
1214 
1215 #define LDLM_RESOURCE_DELREF(res) do {				  \
1216 	lu_ref_del(&(res)->lr_reference, __func__, current);	  \
1217 } while (0)
1218 
1219 /* ldlm_request.c */
1220 /** \defgroup ldlm_local_ast Default AST handlers for local locks
1221  * These AST handlers are typically used for server-side local locks and are
1222  * also used by client-side lock handlers to perform minimum level base
1223  * processing.
1224  * @{ */
1225 int ldlm_completion_ast_async(struct ldlm_lock *lock, __u64 flags, void *data);
1226 int ldlm_completion_ast(struct ldlm_lock *lock, __u64 flags, void *data);
1227 /** @} ldlm_local_ast */
1228 
1229 /** \defgroup ldlm_cli_api API to operate on locks from actual LDLM users.
1230  * These are typically used by client and server (*_local versions)
1231  * to obtain and release locks.
1232  * @{ */
1233 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1234 		     struct ldlm_enqueue_info *einfo,
1235 		     const struct ldlm_res_id *res_id,
1236 		     ldlm_policy_data_t const *policy, __u64 *flags,
1237 		     void *lvb, __u32 lvb_len, enum lvb_type lvb_type,
1238 		     struct lustre_handle *lockh, int async);
1239 int ldlm_prep_enqueue_req(struct obd_export *exp,
1240 			  struct ptlrpc_request *req,
1241 			  struct list_head *cancels,
1242 			  int count);
1243 int ldlm_prep_elc_req(struct obd_export *exp,
1244 		      struct ptlrpc_request *req,
1245 		      int version, int opc, int canceloff,
1246 		      struct list_head *cancels, int count);
1247 
1248 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1249 			  ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1250 			  __u64 *flags, void *lvb, __u32 lvb_len,
1251 			  struct lustre_handle *lockh, int rc);
1252 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1253 int ldlm_cli_cancel(struct lustre_handle *lockh,
1254 		    ldlm_cancel_flags_t cancel_flags);
1255 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1256 			   ldlm_cancel_flags_t flags, void *opaque);
1257 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1258 				    const struct ldlm_res_id *res_id,
1259 				    ldlm_policy_data_t *policy,
1260 				    ldlm_mode_t mode,
1261 				    ldlm_cancel_flags_t flags,
1262 				    void *opaque);
1263 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1264 			       struct list_head *cancels,
1265 			       ldlm_policy_data_t *policy,
1266 			       ldlm_mode_t mode, __u64 lock_flags,
1267 			       ldlm_cancel_flags_t cancel_flags, void *opaque);
1268 int ldlm_cli_cancel_list_local(struct list_head *cancels, int count,
1269 			       ldlm_cancel_flags_t flags);
1270 int ldlm_cli_cancel_list(struct list_head *head, int count,
1271 			 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1272 /** @} ldlm_cli_api */
1273 
1274 /* mds/handler.c */
1275 /* This has to be here because recursive inclusion sucks. */
1276 int intent_disposition(struct ldlm_reply *rep, int flag);
1277 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1278 
1279 /* ioctls for trying requests */
1280 #define IOC_LDLM_TYPE		   'f'
1281 #define IOC_LDLM_MIN_NR		 40
1282 
1283 #define IOC_LDLM_TEST		   _IOWR('f', 40, long)
1284 #define IOC_LDLM_DUMP		   _IOWR('f', 41, long)
1285 #define IOC_LDLM_REGRESS_START	  _IOWR('f', 42, long)
1286 #define IOC_LDLM_REGRESS_STOP	   _IOWR('f', 43, long)
1287 #define IOC_LDLM_MAX_NR		 43
1288 
1289 /**
1290  * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1291  * than one lock_res is dead-lock safe.
1292  */
1293 enum lock_res_type {
1294 	LRT_NORMAL,
1295 	LRT_NEW
1296 };
1297 
1298 /** Lock resource. */
lock_res(struct ldlm_resource * res)1299 static inline void lock_res(struct ldlm_resource *res)
1300 {
1301 	spin_lock(&res->lr_lock);
1302 }
1303 
1304 /** Lock resource with a way to instruct lockdep code about nestedness-safe. */
lock_res_nested(struct ldlm_resource * res,enum lock_res_type mode)1305 static inline void lock_res_nested(struct ldlm_resource *res,
1306 				   enum lock_res_type mode)
1307 {
1308 	spin_lock_nested(&res->lr_lock, mode);
1309 }
1310 
1311 /** Unlock resource. */
unlock_res(struct ldlm_resource * res)1312 static inline void unlock_res(struct ldlm_resource *res)
1313 {
1314 	spin_unlock(&res->lr_lock);
1315 }
1316 
1317 /** Check if resource is already locked, assert if not. */
check_res_locked(struct ldlm_resource * res)1318 static inline void check_res_locked(struct ldlm_resource *res)
1319 {
1320 	assert_spin_locked(&res->lr_lock);
1321 }
1322 
1323 struct ldlm_resource *lock_res_and_lock(struct ldlm_lock *lock);
1324 void unlock_res_and_lock(struct ldlm_lock *lock);
1325 
1326 /* ldlm_pool.c */
1327 /** \defgroup ldlm_pools Various LDLM pool related functions
1328  * There are not used outside of ldlm.
1329  * @{
1330  */
1331 int ldlm_pools_init(void);
1332 void ldlm_pools_fini(void);
1333 
1334 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1335 		   int idx, ldlm_side_t client);
1336 void ldlm_pool_fini(struct ldlm_pool *pl);
1337 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1338 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);
1339 /** @} */
1340 
1341 #endif
1342 /** @} LDLM */
1343