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
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
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 #ifdef HAVE_BGL_SUPPORT
63 /* 1.5 times the maximum 128 tasks available in VN mode */
64 #define LDLM_DEFAULT_LRU_SIZE 196
66 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
68 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
73 ELDLM_LOCK_CHANGED = 300,
74 ELDLM_LOCK_ABORTED = 301,
75 ELDLM_LOCK_REPLACED = 302,
76 ELDLM_NO_LOCK_DATA = 303,
78 ELDLM_NAMESPACE_EXISTS = 400,
79 ELDLM_BAD_NAMESPACE = 401
83 LDLM_NAMESPACE_SERVER = 1 << 0,
84 LDLM_NAMESPACE_CLIENT = 1 << 1
87 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
89 /* If the server returns one of these flags, then the lock was put on that list.
90 * If the client sends one of these flags (during recovery ONLY!), it wants the
91 * lock added to the specified list, no questions asked. -p */
92 #define LDLM_FL_BLOCK_GRANTED 0x000002
93 #define LDLM_FL_BLOCK_CONV 0x000004
94 #define LDLM_FL_BLOCK_WAIT 0x000008
96 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
97 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was sent */
98 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
99 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
101 /* Lock is being replayed. This could probably be implied by the fact that one
102 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
103 #define LDLM_FL_REPLAY 0x000100
105 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
106 #define LDLM_FL_LOCAL_ONLY 0x000400 /* see ldlm_cli_cancel_unused */
108 /* don't run the cancel callback under ldlm_cli_cancel_unused */
109 #define LDLM_FL_FAILED 0x000800
111 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
112 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
113 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
114 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
115 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
117 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
120 /* file & record locking */
121 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
122 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
124 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
125 * the LVB filling happens _after_ the lock has been granted, so another thread
126 * can match`t before the LVB has been updated. As a dirty hack, we set
127 * LDLM_FL_LVB_READY only after we've done the LVB poop.
128 * this is only needed on lov/osc now, where lvb is actually used and callers
129 * must set it in input flags.
131 * The proper fix is to do the granting inside of the completion AST, which can
132 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
133 * pretty high-risk, though, and would need a lot more testing. */
135 #define LDLM_FL_LVB_READY 0x100000
137 /* A lock contributes to the kms calculation until it has finished the part
138 * of it's cancelation that performs write back on its dirty pages. It
139 * can remain on the granted list during this whole time. Threads racing
140 * to update the kms after performing their writeback need to know to
141 * exclude each others locks from the calculation as they walk the granted
143 #define LDLM_FL_KMS_IGNORE 0x200000
145 /* Don't drop lock covering mmapped file in LRU */
146 #define LDLM_FL_NO_LRU 0x400000
148 /* Immediatelly cancel such locks when they block some other locks. Send
149 cancel notification to original lock holder, but expect no reply. */
150 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
152 /* Flags flags inherited from parent lock when doing intents. */
153 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
155 /* These are flags that are mapped into the flags and ASTs of blocking locks */
156 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
157 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
158 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
160 /* completion ast to be executed */
161 #define LDLM_FL_CP_REQD 0x1000000
163 /* cleanup_resource has already handled the lock */
164 #define LDLM_FL_CLEANED 0x2000000
166 /* optimization hint: LDLM can run blocking callback from current context
167 * w/o involving separate thread. in order to decrease cs rate */
168 #define LDLM_FL_ATOMIC_CB 0x4000000
170 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
171 * such that server send blocking ast for conflict locks to this client for
172 * the 1st operation, whereas the 2nd operation has canceled this lock and
173 * is waiting for rpc_lock which is taken by the 1st operation.
174 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
175 * ELC code to cancel it.
176 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
177 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
178 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
179 * to send a separate CANCEL rpc. */
180 #define LDLM_FL_BL_AST 0x10000000
181 #define LDLM_FL_BL_DONE 0x20000000
183 /* measure lock contention and return -EUSERS if locking contention is high */
184 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
186 /* 0x80000000 is occupied by LDLM_AST_DISCARD_DATA */
187 #define LDLM_FL_FAIL_LOC 0x100000000ULL
189 /* The blocking callback is overloaded to perform two functions. These flags
190 * indicate which operation should be performed. */
191 #define LDLM_CB_BLOCKING 1
192 #define LDLM_CB_CANCELING 2
194 /* compatibility matrix */
195 #define LCK_COMPAT_EX LCK_NL
196 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
197 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
198 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
199 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
200 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
201 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
203 extern ldlm_mode_t lck_compat_array[];
205 static inline void lockmode_verify(ldlm_mode_t mode)
207 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
210 static inline int lockmode_compat(ldlm_mode_t exist, ldlm_mode_t new)
212 return (lck_compat_array[exist] & new);
217 * cluster name spaces
221 #define DLM_OST_NAMESPACE 1
222 #define DLM_MDS_NAMESPACE 2
225 - do we just separate this by security domains and use a prefix for
226 multiple namespaces in the same domain?
236 * waiting_locks_spinlock
251 struct ldlm_resource;
252 struct ldlm_namespace;
254 struct ldlm_pool_ops {
255 int (*po_recalc)(struct ldlm_pool *pl);
256 int (*po_shrink)(struct ldlm_pool *pl, int nr,
257 unsigned int gfp_mask);
258 int (*po_setup)(struct ldlm_pool *pl, int limit);
262 * One second for pools thread check interval. Each pool has own period.
264 #define LDLM_POOLS_THREAD_PERIOD (1)
267 * 5% margin for modest pools. See ldlm_pool.c for details.
269 #define LDLM_POOLS_MODEST_MARGIN (5)
272 * Default recalc period for server side pools in sec.
274 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
277 * Default recalc period for client side pools in sec.
279 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
283 * Pool proc directory.
285 cfs_proc_dir_entry_t *pl_proc_dir;
287 * Pool name, should be long enough to contain compound proc entry name.
291 * Lock for protecting slv/clv updates.
295 * Number of allowed locks in in pool, both, client and server side.
299 * Number of granted locks in
305 atomic_t pl_grant_rate;
309 atomic_t pl_cancel_rate;
311 * Grant speed (GR-CR) per T.
313 atomic_t pl_grant_speed;
315 * Server lock volume. Protected by pl_lock.
317 __u64 pl_server_lock_volume;
319 * Current biggest client lock volume. Protected by pl_lock.
321 __u64 pl_client_lock_volume;
323 * Lock volume factor. SLV on client is calculated as following:
324 * server_slv * lock_volume_factor.
326 atomic_t pl_lock_volume_factor;
328 * Time when last slv from server was obtained.
330 time_t pl_recalc_time;
332 * Recalc period for pool.
334 time_t pl_recalc_period;
336 * Recalc and shrink ops.
338 struct ldlm_pool_ops *pl_ops;
340 * Number of planned locks for next period.
346 struct lprocfs_stats *pl_stats;
348 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
349 void *req_cookie, ldlm_mode_t mode, int flags,
352 struct ldlm_valblock_ops {
353 int (*lvbo_init)(struct ldlm_resource *res);
354 int (*lvbo_update)(struct ldlm_resource *res, struct ptlrpc_request *r,
355 int buf_idx, int increase);
359 LDLM_NAMESPACE_GREEDY = 1 << 0,
360 LDLM_NAMESPACE_MODEST = 1 << 1
363 /* default values for the "max_nolock_size", "contention_time"
364 * and "contended_locks" namespace tunables */
365 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
366 #define NS_DEFAULT_CONTENTION_SECONDS 2
367 #define NS_DEFAULT_CONTENDED_LOCKS 32
369 /* Default value for ->ns_shrink_thumb. If lock is not extent one its cost
370 * is one page. Here we have 256 pages which is 1M on i386. Thus by default
371 * all extent locks which have more than 1M long extent will be kept in lru,
372 * others (including ibits locks) will be canceled on memory pressure event. */
373 #define LDLM_LOCK_SHRINK_THUMB 256
375 struct ldlm_namespace {
377 ldlm_side_t ns_client; /* is this a client-side lock tree? */
378 __u64 ns_connect_flags; /* ns connect flags supported
379 * by server (may be changed via proc,
380 * lru resize may be disabled/enabled) */
381 __u64 ns_orig_connect_flags; /* client side orig connect
382 * flags supported by server */
383 struct list_head *ns_hash; /* hash table for ns */
384 spinlock_t ns_hash_lock;
385 __u32 ns_refcount; /* count of resources in the hash */
386 struct list_head ns_root_list; /* all root resources in ns */
387 struct list_head ns_list_chain; /* position in global NS list */
389 struct list_head ns_unused_list; /* all root resources in ns */
391 spinlock_t ns_unused_lock;
393 unsigned int ns_max_unused;
394 unsigned int ns_max_age;
395 unsigned int ns_timeouts;
397 /* Lower limit to number of pages in lock to keep it in cache */
398 unsigned int ns_shrink_thumb;
399 cfs_time_t ns_next_dump; /* next debug dump, jiffies */
403 ldlm_res_policy ns_policy;
404 struct ldlm_valblock_ops *ns_lvbo;
406 cfs_waitq_t ns_waitq;
407 struct ldlm_pool ns_pool;
408 ldlm_appetite_t ns_appetite;
410 /* if more than @ns_contented_locks found, the resource considered
412 unsigned ns_contended_locks;
413 /* the resource remembers contended state during @ns_contention_time,
415 unsigned ns_contention_time;
416 /* limit size of nolock requests, in bytes */
417 unsigned ns_max_nolock_size;
419 struct adaptive_timeout ns_at_estimate;/* estimated lock callback time*/
420 /* backward link to obd, required for ldlm pool to store new SLV. */
421 struct obd_device *ns_obd;
424 static inline int ns_is_client(struct ldlm_namespace *ns)
427 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
428 LDLM_NAMESPACE_SERVER)));
429 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
430 ns->ns_client == LDLM_NAMESPACE_SERVER);
431 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
434 static inline int ns_is_server(struct ldlm_namespace *ns)
437 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
438 LDLM_NAMESPACE_SERVER)));
439 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
440 ns->ns_client == LDLM_NAMESPACE_SERVER);
441 return ns->ns_client == LDLM_NAMESPACE_SERVER;
444 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
447 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
452 * Resource hash table
456 #define RES_HASH_BITS 10
457 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
458 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
462 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
463 struct ldlm_lock_desc *new, void *data,
465 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
467 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
469 /* Interval node data for each LDLM_EXTENT lock */
470 struct ldlm_interval {
471 struct interval_node li_node; /* node for tree mgmt */
472 struct list_head li_group; /* the locks which have the same
473 * policy - group of the policy */
475 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
477 /* the interval tree must be accessed inside the resource lock. */
478 struct ldlm_interval_tree {
479 /* tree size, this variable is used to count
480 * granted PW locks in ldlm_extent_policy()*/
482 ldlm_mode_t lit_mode; /* lock mode */
483 struct interval_node *lit_root; /* actually ldlm_interval */
487 struct portals_handle l_handle; // must be first in the structure
490 /* internal spinlock protects l_resource. we should hold this lock
491 * first before grabbing res_lock.*/
494 /* ldlm_lock_change_resource() can change this */
495 struct ldlm_resource *l_resource;
497 /* protected by ns_hash_lock. FIXME */
498 struct list_head l_lru;
500 /* protected by lr_lock, linkage to resource's lock queues */
501 struct list_head l_res_link;
503 struct ldlm_interval *l_tree_node; /* tree node for ldlm_extent */
505 /* protected by per-bucket exp->exp_lock_hash locks */
506 struct hlist_node l_exp_hash; /* per export hash of locks */
508 /* protected by lr_lock */
509 ldlm_mode_t l_req_mode;
510 ldlm_mode_t l_granted_mode;
512 ldlm_completion_callback l_completion_ast;
513 ldlm_blocking_callback l_blocking_ast;
514 ldlm_glimpse_callback l_glimpse_ast;
516 struct obd_export *l_export;
517 struct obd_export *l_conn_export;
519 struct lustre_handle l_remote_handle;
520 ldlm_policy_data_t l_policy_data;
522 /* protected by lr_lock */
528 /* If the lock is granted, a process sleeps on this waitq to learn when
529 * it's no longer in use. If the lock is not granted, a process sleeps
530 * on this waitq to learn when it becomes granted. */
532 struct timeval l_enqueued_time;
534 cfs_time_t l_last_used; /* jiffies */
535 struct ldlm_extent l_req_extent;
537 /* Client-side-only members */
538 __u32 l_lvb_len; /* temporary storage for */
539 void *l_lvb_data; /* an LVB received during */
540 void *l_lvb_swabber; /* an enqueue */
542 spinlock_t l_extents_list_lock;
543 struct list_head l_extents_list;
545 struct list_head l_cache_locks_list;
547 /* Server-side-only members */
549 /* protected by elt_lock */
550 struct list_head l_pending_chain; /* callbacks pending */
551 cfs_time_t l_callback_timeout; /* jiffies */
553 __u32 l_pid; /* pid which created this lock */
555 /* for ldlm_add_ast_work_item() */
556 struct list_head l_bl_ast;
557 struct list_head l_cp_ast;
558 struct ldlm_lock *l_blocking_lock;
561 /* protected by lr_lock, linkages to "skip lists" */
562 struct list_head l_sl_mode;
563 struct list_head l_sl_policy;
566 struct ldlm_resource {
567 struct ldlm_namespace *lr_namespace;
569 /* protected by ns_hash_lock */
570 struct list_head lr_hash;
571 struct ldlm_resource *lr_parent; /* 0 for a root resource */
572 struct list_head lr_children; /* list head for child resources */
573 struct list_head lr_childof; /* part of ns_root_list if root res,
574 * part of lr_children if child */
577 /* protected by lr_lock */
578 struct list_head lr_granted;
579 struct list_head lr_converting;
580 struct list_head lr_waiting;
581 ldlm_mode_t lr_most_restr;
582 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
583 struct ldlm_res_id lr_name;
584 atomic_t lr_refcount;
586 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
588 /* Server-side-only lock value block elements */
589 struct semaphore lr_lvb_sem;
593 /* when the resource was considered as contended */
594 cfs_time_t lr_contention_time;
597 struct ldlm_ast_work {
598 struct ldlm_lock *w_lock;
600 struct ldlm_lock_desc w_desc;
601 struct list_head w_list;
607 /* ldlm_enqueue parameters common */
608 struct ldlm_enqueue_info {
609 __u32 ei_type; /* Type of the lock being enqueued. */
610 __u32 ei_mode; /* Mode of the lock being enqueued. */
611 void *ei_cb_bl; /* Different callbacks for lock handling (blocking, */
612 void *ei_cb_cp; /* completion, glimpse) */
614 void *ei_cbdata; /* Data to be passed into callbacks. */
617 extern struct obd_ops ldlm_obd_ops;
619 extern char *ldlm_lockname[];
620 extern char *ldlm_typename[];
621 extern char *ldlm_it2str(int it);
623 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
626 if (((level) & D_CANTMASK) != 0 || \
627 ((libcfs_debug & (level)) != 0 && \
628 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
629 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
630 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
632 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
637 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
638 struct libcfs_debug_msg_data *data, const char *fmt,
640 __attribute__ ((format (printf, 4, 5)));
642 #define LDLM_ERROR(lock, fmt, a...) do { \
643 static cfs_debug_limit_state_t _ldlm_cdls; \
644 ldlm_lock_debug(&_ldlm_cdls, D_ERROR, lock, \
645 __FILE__, __FUNCTION__, __LINE__, \
649 #define LDLM_DEBUG(lock, fmt, a...) do { \
650 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
651 __FILE__, __FUNCTION__, __LINE__, \
655 #define LDLM_DEBUG_NOLOCK(format, a...) \
656 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
658 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
659 int first_enq, ldlm_error_t *err,
660 struct list_head *work_list);
666 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
667 #define LDLM_ITER_STOP 2 /* stop iterating */
669 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
670 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
672 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
674 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
676 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
677 ldlm_res_iterator_t iter, void *closure);
679 int ldlm_replay_locks(struct obd_import *imp);
680 void ldlm_resource_iterate(struct ldlm_namespace *, struct ldlm_res_id *,
681 ldlm_iterator_t iter, void *data);
684 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
687 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
691 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
692 void *data, int flag);
693 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
694 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
695 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
696 ldlm_blocking_callback, ldlm_glimpse_callback);
697 int ldlm_handle_convert(struct ptlrpc_request *req);
698 int ldlm_handle_cancel(struct ptlrpc_request *req);
699 int ldlm_request_cancel(struct ptlrpc_request *req,
700 struct ldlm_request *dlm_req, int first);
701 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
702 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
703 int ldlm_get_ref(void);
704 void ldlm_put_ref(void);
705 int ldlm_init_export(struct obd_export *exp);
706 void ldlm_destroy_export(struct obd_export *exp);
709 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
710 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
711 void ldlm_lock2handle(struct ldlm_lock *lock, struct lustre_handle *lockh);
712 struct ldlm_lock *__ldlm_handle2lock(struct lustre_handle *, int flags);
713 void ldlm_cancel_callback(struct ldlm_lock *);
714 int ldlm_lock_set_data(struct lustre_handle *, void *data);
715 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
716 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *,
717 struct lustre_handle *);
719 static inline struct ldlm_lock *ldlm_handle2lock(struct lustre_handle *h)
721 return __ldlm_handle2lock(h, 0);
724 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
725 struct ptlrpc_request *r, int buf_idx,
728 if (res->lr_namespace->ns_lvbo &&
729 res->lr_namespace->ns_lvbo->lvbo_update) {
730 return res->lr_namespace->ns_lvbo->lvbo_update(res, r, buf_idx,
736 #define LDLM_LOCK_PUT(lock) \
738 /*LDLM_DEBUG((lock), "put");*/ \
739 ldlm_lock_put(lock); \
742 #define LDLM_LOCK_GET(lock) \
744 ldlm_lock_get(lock); \
745 /*LDLM_DEBUG((lock), "get");*/ \
749 #define ldlm_lock_list_put(head, member, count) \
751 struct ldlm_lock *_lock, *_next; \
753 list_for_each_entry_safe(_lock, _next, head, member) { \
756 list_del_init(&_lock->member); \
757 LDLM_LOCK_PUT(_lock); \
762 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
763 void ldlm_lock_put(struct ldlm_lock *lock);
764 void ldlm_lock_destroy(struct ldlm_lock *lock);
765 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
766 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
767 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
768 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
769 void ldlm_lock_allow_match(struct ldlm_lock *lock);
770 int ldlm_lock_fast_match(struct ldlm_lock *, int, obd_off, obd_off, void **);
771 void ldlm_lock_fast_release(void *, int);
772 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
773 struct ldlm_res_id *, ldlm_type_t type,
774 ldlm_policy_data_t *, ldlm_mode_t mode,
775 struct lustre_handle *);
776 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
778 void ldlm_lock_cancel(struct ldlm_lock *lock);
779 void ldlm_cancel_locks_for_export(struct obd_export *export);
780 void ldlm_reprocess_all(struct ldlm_resource *res);
781 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
782 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
783 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
784 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
787 struct ldlm_namespace *
788 ldlm_namespace_new(struct obd_device *obd, char *name,
789 ldlm_side_t client, ldlm_appetite_t apt);
790 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
791 void ldlm_namespace_free(struct ldlm_namespace *ns,
792 struct obd_import *imp, int force);
793 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
794 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
795 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
796 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
797 void ldlm_namespace_get_locked(struct ldlm_namespace *ns);
798 void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup);
799 void ldlm_namespace_get(struct ldlm_namespace *ns);
800 void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup);
801 int ldlm_proc_setup(void);
803 void ldlm_proc_cleanup(void);
805 static inline void ldlm_proc_cleanup(void) {}
808 /* resource.c - internal */
809 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
810 struct ldlm_resource *parent,
811 struct ldlm_res_id, ldlm_type_t type,
813 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
814 int ldlm_resource_putref(struct ldlm_resource *res);
815 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
816 struct ldlm_lock *lock);
817 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
818 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
819 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
820 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
821 void ldlm_resource_dump(int level, struct ldlm_resource *);
822 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
826 int ldlm_expired_completion_wait(void *data);
827 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
828 void *data, int flag);
829 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
830 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
831 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **req,
832 struct ldlm_enqueue_info *einfo, struct ldlm_res_id res_id,
833 ldlm_policy_data_t *policy, int *flags,
834 void *lvb, __u32 lvb_len, void *lvb_swabber,
835 struct lustre_handle *lockh, int async);
836 struct ptlrpc_request *ldlm_prep_enqueue_req(struct obd_export *exp,
837 int bufcount, __u32 *size,
838 struct list_head *head, int count);
839 struct ptlrpc_request *ldlm_prep_elc_req(struct obd_export *exp, int version,
840 int opc, int bufcount, __u32 *size,
841 int bufoff, int canceloff,
842 struct list_head *cancels, int count);
843 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
844 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
845 int *flags, void *lvb, __u32 lvb_len,
846 void *lvb_swabber, struct lustre_handle *lockh,
848 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
849 struct ldlm_res_id *res_id,
850 ldlm_type_t type, ldlm_policy_data_t *policy,
851 ldlm_mode_t mode, int *flags,
852 ldlm_blocking_callback blocking,
853 ldlm_completion_callback completion,
854 ldlm_glimpse_callback glimpse,
855 void *data, __u32 lvb_len, void *lvb_swabber,
856 struct lustre_handle *lockh);
857 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
858 void *data, __u32 data_len);
859 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
860 int ldlm_cli_update_pool(struct ptlrpc_request *req);
861 int ldlm_cli_cancel(struct lustre_handle *lockh);
862 int ldlm_cli_cancel_unused(struct ldlm_namespace *, struct ldlm_res_id *,
863 int flags, void *opaque);
864 int ldlm_cli_cancel_req(struct obd_export *exp,
865 struct list_head *head, int count);
866 int ldlm_cli_join_lru(struct ldlm_namespace *, struct ldlm_res_id *, int join);
867 int ldlm_cancel_resource_local(struct ldlm_resource *res,
868 struct list_head *cancels,
869 ldlm_policy_data_t *policy, ldlm_mode_t mode,
870 int lock_flags, int cancel_flags, void *opaque);
871 int ldlm_cli_cancel_list(struct list_head *head, int count,
872 struct ptlrpc_request *req, int off);
874 /* This has to be here because recursive inclusion sucks. */
875 int intent_disposition(struct ldlm_reply *rep, int flag);
876 void intent_set_disposition(struct ldlm_reply *rep, int flag);
879 /* ioctls for trying requests */
880 #define IOC_LDLM_TYPE 'f'
881 #define IOC_LDLM_MIN_NR 40
883 #define IOC_LDLM_TEST _IOWR('f', 40, long)
884 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
885 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
886 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
887 #define IOC_LDLM_MAX_NR 43
889 static inline void lock_res(struct ldlm_resource *res)
891 spin_lock(&res->lr_lock);
894 static inline void unlock_res(struct ldlm_resource *res)
896 spin_unlock(&res->lr_lock);
899 static inline void check_res_locked(struct ldlm_resource *res)
901 LASSERT_SPIN_LOCKED(&res->lr_lock);
904 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
905 void unlock_res_and_lock(struct ldlm_lock *lock);
908 void ldlm_pools_recalc(ldlm_side_t client);
909 int ldlm_pools_init(void);
910 void ldlm_pools_fini(void);
912 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
913 int idx, ldlm_side_t client);
914 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
915 unsigned int gfp_mask);
916 void ldlm_pool_fini(struct ldlm_pool *pl);
917 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
918 int ldlm_pool_recalc(struct ldlm_pool *pl);
919 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
920 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
921 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
922 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
923 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
924 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
925 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
926 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
927 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);