1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see [sun.com URL with a
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
37 #ifndef _LUSTRE_DLM_H__
38 #define _LUSTRE_DLM_H__
40 #if defined(__linux__)
41 #include <linux/lustre_dlm.h>
42 #elif defined(__APPLE__)
43 #include <darwin/lustre_dlm.h>
44 #elif defined(__WINNT__)
45 #include <winnt/lustre_dlm.h>
47 #error Unsupported operating system.
50 #include <lustre_lib.h>
51 #include <lustre_net.h>
52 #include <lustre_import.h>
53 #include <lustre_handles.h>
54 #include <lustre_export.h> /* for obd_export, for LDLM_DEBUG */
55 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
60 #define OBD_LDLM_DEVICENAME "ldlm"
62 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
63 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
64 #define LDLM_CTIME_AGE_LIMIT (10)
69 ELDLM_LOCK_CHANGED = 300,
70 ELDLM_LOCK_ABORTED = 301,
71 ELDLM_LOCK_REPLACED = 302,
72 ELDLM_NO_LOCK_DATA = 303,
74 ELDLM_NAMESPACE_EXISTS = 400,
75 ELDLM_BAD_NAMESPACE = 401
79 LDLM_NAMESPACE_SERVER = 1 << 0,
80 LDLM_NAMESPACE_CLIENT = 1 << 1
83 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
85 /* If the server returns one of these flags, then the lock was put on that list.
86 * If the client sends one of these flags (during recovery ONLY!), it wants the
87 * lock added to the specified list, no questions asked. -p */
88 #define LDLM_FL_BLOCK_GRANTED 0x000002
89 #define LDLM_FL_BLOCK_CONV 0x000004
90 #define LDLM_FL_BLOCK_WAIT 0x000008
92 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
93 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was sent */
94 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
95 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
97 /* Lock is being replayed. This could probably be implied by the fact that one
98 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
99 #define LDLM_FL_REPLAY 0x000100
101 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
102 #define LDLM_FL_LOCAL_ONLY 0x000400 /* see ldlm_cli_cancel_unused */
104 /* don't run the cancel callback under ldlm_cli_cancel_unused */
105 #define LDLM_FL_FAILED 0x000800
107 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
108 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
109 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
110 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
111 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
113 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
116 /* file & record locking */
117 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
118 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
120 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
121 * the LVB filling happens _after_ the lock has been granted, so another thread
122 * can match`t before the LVB has been updated. As a dirty hack, we set
123 * LDLM_FL_LVB_READY only after we've done the LVB poop.
124 * this is only needed on lov/osc now, where lvb is actually used and callers
125 * must set it in input flags.
127 * The proper fix is to do the granting inside of the completion AST, which can
128 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
129 * pretty high-risk, though, and would need a lot more testing. */
131 #define LDLM_FL_LVB_READY 0x100000
133 /* A lock contributes to the kms calculation until it has finished the part
134 * of it's cancelation that performs write back on its dirty pages. It
135 * can remain on the granted list during this whole time. Threads racing
136 * to update the kms after performing their writeback need to know to
137 * exclude each others locks from the calculation as they walk the granted
139 #define LDLM_FL_KMS_IGNORE 0x200000
141 /* Don't drop lock covering mmapped file in LRU */
142 #define LDLM_FL_NO_LRU 0x400000
144 /* Immediatelly cancel such locks when they block some other locks. Send
145 cancel notification to original lock holder, but expect no reply. */
146 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
148 /* Flags flags inherited from parent lock when doing intents. */
149 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
151 /* These are flags that are mapped into the flags and ASTs of blocking locks */
152 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
153 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
154 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
156 /* completion ast to be executed */
157 #define LDLM_FL_CP_REQD 0x1000000
159 /* cleanup_resource has already handled the lock */
160 #define LDLM_FL_CLEANED 0x2000000
162 /* optimization hint: LDLM can run blocking callback from current context
163 * w/o involving separate thread. in order to decrease cs rate */
164 #define LDLM_FL_ATOMIC_CB 0x4000000
166 /* Cancel lock asynchronously. See ldlm_cli_cancel_unused_resource. */
167 #define LDLM_FL_ASYNC 0x8000000
169 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
170 * such that server send blocking ast for conflict locks to this client for
171 * the 1st operation, whereas the 2nd operation has canceled this lock and
172 * is waiting for rpc_lock which is taken by the 1st operation.
173 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
174 * ELC code to cancel it.
175 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
176 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
177 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
178 * to send a separate CANCEL rpc. */
179 #define LDLM_FL_BL_AST 0x10000000
180 #define LDLM_FL_BL_DONE 0x20000000
182 /* measure lock contention and return -EUSERS if locking contention is high */
183 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
185 /* The blocking callback is overloaded to perform two functions. These flags
186 * indicate which operation should be performed. */
187 #define LDLM_CB_BLOCKING 1
188 #define LDLM_CB_CANCELING 2
190 /* compatibility matrix */
191 #define LCK_COMPAT_EX LCK_NL
192 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
193 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
194 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
195 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
196 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
197 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
199 extern ldlm_mode_t lck_compat_array[];
201 static inline void lockmode_verify(ldlm_mode_t mode)
203 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
206 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
208 return (lck_compat_array[exist_mode] & new_mode);
213 * cluster name spaces
217 #define DLM_OST_NAMESPACE 1
218 #define DLM_MDS_NAMESPACE 2
221 - do we just separate this by security domains and use a prefix for
222 multiple namespaces in the same domain?
232 * waiting_locks_spinlock
247 struct ldlm_resource;
248 struct ldlm_namespace;
250 struct ldlm_pool_ops {
251 int (*po_recalc)(struct ldlm_pool *pl);
252 int (*po_shrink)(struct ldlm_pool *pl, int nr,
253 unsigned int gfp_mask);
254 int (*po_setup)(struct ldlm_pool *pl, int limit);
257 /* One second for pools thread check interval. */
258 #define LDLM_POOLS_THREAD_PERIOD (1)
260 /* 5% margin for modest pools. See ldlm_pool.c for details. */
261 #define LDLM_POOLS_MODEST_MARGIN (5)
263 /* A change to SLV in % after which we want to wake up pools thread asap. */
264 #define LDLM_POOLS_FAST_SLV_CHANGE (50)
268 * Pool proc directory.
270 cfs_proc_dir_entry_t *pl_proc_dir;
272 * Pool name, should be long enough to contain compound proc entry name.
276 * Lock for protecting slv/clv updates.
280 * Number of allowed locks in in pool, both, client and server side.
284 * Number of granted locks in
290 atomic_t pl_grant_rate;
294 atomic_t pl_cancel_rate;
296 * Grant speed (GR-CR) per T.
298 atomic_t pl_grant_speed;
300 * Server lock volume. Protected by pl_lock.
302 __u64 pl_server_lock_volume;
304 * Current biggest client lock volume. Protected by pl_lock.
306 __u64 pl_client_lock_volume;
308 * Lock volume factor. SLV on client is calculated as following:
309 * server_slv * lock_volume_factor.
311 atomic_t pl_lock_volume_factor;
313 * Time when last slv from server was obtained.
315 time_t pl_recalc_time;
317 * Recalc and shrink ops.
319 struct ldlm_pool_ops *pl_ops;
321 * Planned number of granted locks for next T.
325 * Grant plan step for next T.
331 struct lprocfs_stats *pl_stats;
334 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
335 void *req_cookie, ldlm_mode_t mode, int flags,
338 struct ldlm_valblock_ops {
339 int (*lvbo_init)(struct ldlm_resource *res);
340 int (*lvbo_update)(struct ldlm_resource *res, struct lustre_msg *m,
341 int buf_idx, int increase);
345 LDLM_NAMESPACE_GREEDY = 1 << 0,
346 LDLM_NAMESPACE_MODEST = 1 << 1
350 * Default value for ->ns_shrink_thumb. If lock is not extent one its cost
351 * is one page. Here we have 256 pages which is 1M on i386. Thus by default
352 * all extent locks which have more than 1M long extent will be kept in lru,
353 * others (including ibits locks) will be canceled on memory pressure event.
355 #define LDLM_LOCK_SHRINK_THUMB 256
358 * Default values for the "max_nolock_size", "contention_time" and
359 * "contended_locks" namespace tunables.
361 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
362 #define NS_DEFAULT_CONTENTION_SECONDS 2
363 #define NS_DEFAULT_CONTENDED_LOCKS 32
365 struct ldlm_namespace {
367 * Namespace name. Used for logging, etc.
372 * Is this a client-side lock tree?
374 ldlm_side_t ns_client;
377 * Namespce connect flags supported by server (may be changed via proc,
378 * lru resize may be disabled/enabled).
380 __u64 ns_connect_flags;
383 * Client side orig connect flags supported by server.
385 __u64 ns_orig_connect_flags;
388 * Hash table for namespace.
390 struct list_head *ns_hash;
391 spinlock_t ns_hash_lock;
394 * Count of resources in the hash.
399 * All root resources in namespace.
401 struct list_head ns_root_list;
404 * Position in global namespace list.
406 struct list_head ns_list_chain;
409 * All root resources in namespace.
411 struct list_head ns_unused_list;
413 spinlock_t ns_unused_lock;
415 unsigned int ns_max_unused;
416 unsigned int ns_max_age;
421 unsigned int ns_ctime_age_limit;
424 * Lower limit to number of pages in lock to keep it in cache.
426 unsigned int ns_shrink_thumb;
429 * Next debug dump, jiffies.
431 cfs_time_t ns_next_dump;
435 ldlm_res_policy ns_policy;
436 struct ldlm_valblock_ops *ns_lvbo;
438 cfs_waitq_t ns_waitq;
439 struct ldlm_pool ns_pool;
440 ldlm_appetite_t ns_appetite;
443 * If more than @ns_contented_locks found, the resource considered
446 unsigned ns_contended_locks;
449 * The resource remembers contended state during @ns_contention_time,
452 unsigned ns_contention_time;
455 * Limit size of nolock requests, in bytes.
457 unsigned ns_max_nolock_size;
460 * Backward link to obd, required for ldlm pool to store new SLV.
462 struct obd_device *ns_obd;
464 struct adaptive_timeout ns_at_estimate;/* estimated lock callback time*/
467 static inline int ns_is_client(struct ldlm_namespace *ns)
470 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
471 LDLM_NAMESPACE_SERVER)));
472 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
473 ns->ns_client == LDLM_NAMESPACE_SERVER);
474 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
477 static inline int ns_is_server(struct ldlm_namespace *ns)
480 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
481 LDLM_NAMESPACE_SERVER)));
482 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
483 ns->ns_client == LDLM_NAMESPACE_SERVER);
484 return ns->ns_client == LDLM_NAMESPACE_SERVER;
487 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
490 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
495 * Resource hash table
499 #define RES_HASH_BITS 10
500 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
501 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
505 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
506 struct ldlm_lock_desc *new, void *data,
508 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
510 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
512 /* Interval node data for each LDLM_EXTENT lock */
513 struct ldlm_interval {
514 struct interval_node li_node; /* node for tree mgmt */
515 struct list_head li_group; /* the locks which have the same
516 * policy - group of the policy */
518 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
520 /* the interval tree must be accessed inside the resource lock. */
521 struct ldlm_interval_tree {
522 /* tree size, this variable is used to count
523 * granted PW locks in ldlm_extent_policy()*/
525 ldlm_mode_t lit_mode; /* lock mode */
526 struct interval_node *lit_root; /* actually ldlm_interval */
530 struct portals_handle l_handle; // must be first in the structure
533 /* internal spinlock protects l_resource. we should hold this lock
534 * first before grabbing res_lock.*/
537 /* ldlm_lock_change_resource() can change this */
538 struct ldlm_resource *l_resource;
540 /* protected by ns_hash_lock. FIXME */
541 struct list_head l_lru;
543 /* protected by lr_lock, linkage to resource's lock queues */
544 struct list_head l_res_link;
546 struct ldlm_interval *l_tree_node; /* tree node for ldlm_extent */
548 /* protected by led_lock */
549 struct list_head l_export_chain; // per-export chain of locks
551 /* protected by lr_lock */
552 ldlm_mode_t l_req_mode;
553 ldlm_mode_t l_granted_mode;
555 ldlm_completion_callback l_completion_ast;
556 ldlm_blocking_callback l_blocking_ast;
557 ldlm_glimpse_callback l_glimpse_ast;
559 struct obd_export *l_export;
560 struct obd_export *l_conn_export;
562 struct lustre_handle l_remote_handle;
563 ldlm_policy_data_t l_policy_data;
565 /* protected by lr_lock */
571 /* If the lock is granted, a process sleeps on this waitq to learn when
572 * it's no longer in use. If the lock is not granted, a process sleeps
573 * on this waitq to learn when it becomes granted. */
575 struct timeval l_enqueued_time;
577 cfs_time_t l_last_used; /* jiffies */
578 struct ldlm_extent l_req_extent;
580 /* Client-side-only members */
581 __u32 l_lvb_len; /* temporary storage for */
582 void *l_lvb_data; /* an LVB received during */
583 void *l_lvb_swabber; /* an enqueue */
585 spinlock_t l_extents_list_lock;
586 struct list_head l_extents_list;
588 struct list_head l_cache_locks_list;
590 /* Server-side-only members */
592 /* protected by elt_lock */
593 struct list_head l_pending_chain; /* callbacks pending */
594 cfs_time_t l_callback_timeout; /* jiffies */
596 __u32 l_pid; /* pid which created this lock */
598 /* for ldlm_add_ast_work_item() */
599 struct list_head l_bl_ast;
600 struct list_head l_cp_ast;
601 struct ldlm_lock *l_blocking_lock;
604 /* protected by lr_lock, linkages to "skip lists" */
605 struct list_head l_sl_mode;
606 struct list_head l_sl_policy;
609 struct ldlm_resource {
610 struct ldlm_namespace *lr_namespace;
612 /* protected by ns_hash_lock */
613 struct list_head lr_hash;
614 struct ldlm_resource *lr_parent; /* 0 for a root resource */
615 struct list_head lr_children; /* list head for child resources */
616 struct list_head lr_childof; /* part of ns_root_list if root res,
617 * part of lr_children if child */
620 /* protected by lr_lock */
621 struct list_head lr_granted;
622 struct list_head lr_converting;
623 struct list_head lr_waiting;
624 ldlm_mode_t lr_most_restr;
625 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
626 struct ldlm_res_id lr_name;
627 atomic_t lr_refcount;
629 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
631 /* Server-side-only lock value block elements */
632 struct semaphore lr_lvb_sem;
636 /* when the resource was considered as contended */
637 cfs_time_t lr_contention_time;
640 struct ldlm_ast_work {
641 struct ldlm_lock *w_lock;
643 struct ldlm_lock_desc w_desc;
644 struct list_head w_list;
650 /* ldlm_enqueue parameters common */
651 struct ldlm_enqueue_info {
652 __u32 ei_type; /* Type of the lock being enqueued. */
653 __u32 ei_mode; /* Mode of the lock being enqueued. */
654 void *ei_cb_bl; /* blocking lock callback */
655 void *ei_cb_cp; /* lock completion callback */
656 void *ei_cb_gl; /* lock glimpse callback */
657 void *ei_cbdata; /* Data to be passed into callbacks. */
660 extern struct obd_ops ldlm_obd_ops;
662 extern char *ldlm_lockname[];
663 extern char *ldlm_typename[];
664 extern char *ldlm_it2str(int it);
666 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
669 if (((level) & D_CANTMASK) != 0 || \
670 ((libcfs_debug & (level)) != 0 && \
671 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
672 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
673 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
675 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
680 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
681 struct libcfs_debug_msg_data *data, const char *fmt,
683 __attribute__ ((format (printf, 4, 5)));
685 #define LDLM_ERROR(lock, fmt, a...) do { \
686 static cfs_debug_limit_state_t _ldlm_cdls; \
687 ldlm_lock_debug(&_ldlm_cdls, D_ERROR, lock, \
688 __FILE__, __FUNCTION__, __LINE__, \
692 #define LDLM_DEBUG(lock, fmt, a...) do { \
693 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
694 __FILE__, __FUNCTION__, __LINE__, \
698 #define LDLM_DEBUG(lock, fmt, a...) ((void)0)
699 #define LDLM_ERROR(lock, fmt, a...) ((void)0)
702 #define LDLM_DEBUG_NOLOCK(format, a...) \
703 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
705 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
706 int first_enq, ldlm_error_t *err,
707 struct list_head *work_list);
713 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
714 #define LDLM_ITER_STOP 2 /* stop iterating */
716 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
717 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
719 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
721 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
723 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
724 ldlm_res_iterator_t iter, void *closure);
726 int ldlm_replay_locks(struct obd_import *imp);
727 void ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
728 ldlm_iterator_t iter, void *data);
731 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
734 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
738 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
739 void *data, int flag);
740 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
741 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
742 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
743 ldlm_blocking_callback, ldlm_glimpse_callback);
744 int ldlm_handle_convert(struct ptlrpc_request *req);
745 int ldlm_handle_cancel(struct ptlrpc_request *req);
746 int ldlm_request_cancel(struct ptlrpc_request *req,
747 const struct ldlm_request *dlm_req, int first);
748 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
749 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock);
750 void ldlm_revoke_export_locks(struct obd_export *exp);
751 int ldlm_get_ref(void);
752 void ldlm_put_ref(void);
755 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
756 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
757 void ldlm_lock2handle(const struct ldlm_lock *lock,
758 struct lustre_handle *lockh);
759 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
760 void ldlm_cancel_callback(struct ldlm_lock *);
761 int ldlm_lock_set_data(struct lustre_handle *, void *data);
762 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
763 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *,
764 const struct lustre_handle *);
766 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
768 return __ldlm_handle2lock(h, 0);
771 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
772 struct lustre_msg *m, int buf_idx,
775 if (res->lr_namespace->ns_lvbo &&
776 res->lr_namespace->ns_lvbo->lvbo_update) {
777 return res->lr_namespace->ns_lvbo->lvbo_update(res, m, buf_idx,
783 #define LDLM_LOCK_PUT(lock) \
785 /*LDLM_DEBUG((lock), "put");*/ \
786 ldlm_lock_put(lock); \
789 #define LDLM_LOCK_GET(lock) \
791 ldlm_lock_get(lock); \
792 /*LDLM_DEBUG((lock), "get");*/ \
796 #define ldlm_lock_list_put(head, member, count) \
798 struct ldlm_lock *_lock, *_next; \
800 list_for_each_entry_safe(_lock, _next, head, member) { \
803 list_del_init(&_lock->member); \
804 LDLM_LOCK_PUT(_lock); \
809 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
810 void ldlm_lock_put(struct ldlm_lock *lock);
811 void ldlm_lock_destroy(struct ldlm_lock *lock);
812 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
813 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
814 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
815 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
816 void ldlm_lock_allow_match(struct ldlm_lock *lock);
817 int ldlm_lock_fast_match(struct ldlm_lock *, int, obd_off, obd_off, void **);
818 void ldlm_lock_fast_release(void *, int);
819 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
820 const struct ldlm_res_id *, ldlm_type_t type,
821 ldlm_policy_data_t *, ldlm_mode_t mode,
822 struct lustre_handle *);
823 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
825 void ldlm_lock_cancel(struct ldlm_lock *lock);
826 void ldlm_cancel_locks_for_export(struct obd_export *export);
827 void ldlm_reprocess_all(struct ldlm_resource *res);
828 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
829 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
830 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
831 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
834 struct ldlm_namespace *
835 ldlm_namespace_new(struct obd_device *obd, char *name,
836 ldlm_side_t client, ldlm_appetite_t apt);
837 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
838 void ldlm_namespace_free(struct ldlm_namespace *ns,
839 struct obd_import *imp, int force);
840 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
841 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
842 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
843 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
844 void ldlm_namespace_get_locked(struct ldlm_namespace *ns);
845 void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup);
846 void ldlm_namespace_get(struct ldlm_namespace *ns);
847 void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup);
848 int ldlm_proc_setup(void);
850 void ldlm_proc_cleanup(void);
852 static inline void ldlm_proc_cleanup(void) {}
855 /* resource.c - internal */
856 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
857 struct ldlm_resource *parent,
858 const struct ldlm_res_id *,
859 ldlm_type_t type, int create);
860 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
861 int ldlm_resource_putref(struct ldlm_resource *res);
862 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
863 struct ldlm_lock *lock);
864 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
865 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
866 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
867 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
868 void ldlm_resource_dump(int level, struct ldlm_resource *);
869 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
870 const struct ldlm_res_id *);
872 struct ldlm_callback_suite {
873 ldlm_completion_callback lcs_completion;
874 ldlm_blocking_callback lcs_blocking;
875 ldlm_glimpse_callback lcs_glimpse;
879 int ldlm_expired_completion_wait(void *data);
880 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
881 void *data, int flag);
882 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
883 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
884 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
885 struct ldlm_enqueue_info *einfo,
886 const struct ldlm_res_id *res_id,
887 ldlm_policy_data_t *policy, int *flags,
888 void *lvb, __u32 lvb_len, void *lvb_swabber,
889 struct lustre_handle *lockh, int async);
890 int ldlm_prep_enqueue_req(struct obd_export *exp,
891 struct ptlrpc_request *req,
892 struct list_head *cancels,
894 int ldlm_prep_elc_req(struct obd_export *exp,
895 struct ptlrpc_request *req,
896 int version, int opc, int canceloff,
897 struct list_head *cancels, int count);
898 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
899 const struct ldlm_request *dlm_req,
900 const struct ldlm_callback_suite *cbs);
901 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
902 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
903 int *flags, void *lvb, __u32 lvb_len,
904 void *lvb_swabber, struct lustre_handle *lockh,
906 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
907 const struct ldlm_res_id *res_id,
908 ldlm_type_t type, ldlm_policy_data_t *policy,
909 ldlm_mode_t mode, int *flags,
910 ldlm_blocking_callback blocking,
911 ldlm_completion_callback completion,
912 ldlm_glimpse_callback glimpse,
913 void *data, __u32 lvb_len, void *lvb_swabber,
914 struct lustre_handle *lockh);
915 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
916 void *data, __u32 data_len);
917 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
918 int ldlm_cli_update_pool(struct ptlrpc_request *req);
919 int ldlm_handle_convert0(struct ptlrpc_request *req,
920 const struct ldlm_request *dlm_req);
921 int ldlm_cli_cancel(struct lustre_handle *lockh);
922 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
923 int flags, void *opaque);
924 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
925 const struct ldlm_res_id *res_id,
926 ldlm_policy_data_t *policy,
927 ldlm_mode_t mode, int flags, void *opaque);
928 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
929 int count, int flags);
930 int ldlm_cli_join_lru(struct ldlm_namespace *,
931 const struct ldlm_res_id *, int join);
932 int ldlm_cancel_resource_local(struct ldlm_resource *res,
933 struct list_head *cancels,
934 ldlm_policy_data_t *policy,
935 ldlm_mode_t mode, int lock_flags,
936 int cancel_flags, void *opaque);
937 int ldlm_cli_cancel_list(struct list_head *head, int count,
938 struct ptlrpc_request *req, int flags);
941 /* This has to be here because recursive inclusion sucks. */
942 int intent_disposition(struct ldlm_reply *rep, int flag);
943 void intent_set_disposition(struct ldlm_reply *rep, int flag);
946 /* ioctls for trying requests */
947 #define IOC_LDLM_TYPE 'f'
948 #define IOC_LDLM_MIN_NR 40
950 #define IOC_LDLM_TEST _IOWR('f', 40, long)
951 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
952 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
953 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
954 #define IOC_LDLM_MAX_NR 43
956 static inline void lock_res(struct ldlm_resource *res)
958 spin_lock(&res->lr_lock);
961 static inline void unlock_res(struct ldlm_resource *res)
963 spin_unlock(&res->lr_lock);
966 static inline void check_res_locked(struct ldlm_resource *res)
968 LASSERT_SPIN_LOCKED(&res->lr_lock);
971 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
972 void unlock_res_and_lock(struct ldlm_lock *lock);
975 void ldlm_pools_recalc(ldlm_side_t client);
976 int ldlm_pools_init(void);
977 void ldlm_pools_fini(void);
978 void ldlm_pools_wakeup(void);
980 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
981 int idx, ldlm_side_t client);
982 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
983 unsigned int gfp_mask);
984 void ldlm_pool_fini(struct ldlm_pool *pl);
985 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
986 int ldlm_pool_recalc(struct ldlm_pool *pl);
987 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
988 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
989 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
990 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
991 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
992 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
993 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
994 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
995 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);