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 */
61 #define OBD_LDLM_DEVICENAME "ldlm"
63 #ifdef HAVE_BGL_SUPPORT
64 /* 1.5 times the maximum 128 tasks available in VN mode */
65 #define LDLM_DEFAULT_LRU_SIZE 196
67 #define LDLM_DEFAULT_LRU_SIZE (100 * num_online_cpus())
69 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
70 #define LDLM_CTIME_AGE_LIMIT (10)
75 ELDLM_LOCK_CHANGED = 300,
76 ELDLM_LOCK_ABORTED = 301,
77 ELDLM_LOCK_REPLACED = 302,
78 ELDLM_NO_LOCK_DATA = 303,
80 ELDLM_NAMESPACE_EXISTS = 400,
81 ELDLM_BAD_NAMESPACE = 401
85 LDLM_NAMESPACE_SERVER = 1 << 0,
86 LDLM_NAMESPACE_CLIENT = 1 << 1
89 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
91 /* If the server returns one of these flags, then the lock was put on that list.
92 * If the client sends one of these flags (during recovery ONLY!), it wants the
93 * lock added to the specified list, no questions asked. -p */
94 #define LDLM_FL_BLOCK_GRANTED 0x000002
95 #define LDLM_FL_BLOCK_CONV 0x000004
96 #define LDLM_FL_BLOCK_WAIT 0x000008
98 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
99 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
100 * queued for sending. */
101 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
102 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
104 /* Lock is being replayed. This could probably be implied by the fact that one
105 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
106 #define LDLM_FL_REPLAY 0x000100
108 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
109 #define LDLM_FL_LOCAL_ONLY 0x000400 /* see ldlm_cli_cancel_unused */
111 /* don't run the cancel callback under ldlm_cli_cancel_unused */
112 #define LDLM_FL_FAILED 0x000800
114 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
115 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
116 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
117 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
118 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
120 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
123 /* file & record locking */
124 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
125 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
127 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
128 * the LVB filling happens _after_ the lock has been granted, so another thread
129 * can match`t before the LVB has been updated. As a dirty hack, we set
130 * LDLM_FL_LVB_READY only after we've done the LVB poop.
131 * this is only needed on lov/osc now, where lvb is actually used and callers
132 * must set it in input flags.
134 * The proper fix is to do the granting inside of the completion AST, which can
135 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
136 * pretty high-risk, though, and would need a lot more testing. */
138 #define LDLM_FL_LVB_READY 0x100000
140 /* A lock contributes to the kms calculation until it has finished the part
141 * of it's cancelation that performs write back on its dirty pages. It
142 * can remain on the granted list during this whole time. Threads racing
143 * to update the kms after performing their writeback need to know to
144 * exclude each others locks from the calculation as they walk the granted
146 #define LDLM_FL_KMS_IGNORE 0x200000
148 /* Immediatelly cancel such locks when they block some other locks. Send
149 * cancel notification to original lock holder, but expect no reply. This is
150 * for clients (like liblustre) that cannot be expected to reliably response
151 * to blocking ast. */
152 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
154 /* Flags flags inherited from parent lock when doing intents. */
155 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
157 /* completion ast to be executed */
158 #define LDLM_FL_CP_REQD 0x1000000
160 /* cleanup_resource has already handled the lock */
161 #define LDLM_FL_CLEANED 0x2000000
163 /* optimization hint: LDLM can run blocking callback from current context
164 * w/o involving separate thread. in order to decrease cs rate */
165 #define LDLM_FL_ATOMIC_CB 0x4000000
167 /* Cancel lock asynchronously. See ldlm_cli_cancel_unused_resource. */
168 #define LDLM_FL_ASYNC 0x8000000
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 /* These are flags that are mapped into the flags and ASTs of blocking locks */
187 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
189 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
190 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
192 /* Used for marking lock as an target for -EINTR while cp_ast sleep situation
193 * emulation + race with upcoming bl_ast. */
194 #define LDLM_FL_FAIL_LOC 0x100000000ULL
196 /* The blocking callback is overloaded to perform two functions. These flags
197 * indicate which operation should be performed. */
198 #define LDLM_CB_BLOCKING 1
199 #define LDLM_CB_CANCELING 2
201 /* compatibility matrix */
202 #define LCK_COMPAT_EX LCK_NL
203 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
204 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
205 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
206 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
207 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
208 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
209 #define LCK_COMPAT_COS (LCK_COS)
211 extern ldlm_mode_t lck_compat_array[];
213 static inline void lockmode_verify(ldlm_mode_t mode)
215 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
218 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
220 return (lck_compat_array[exist_mode] & new_mode);
225 * cluster name spaces
229 #define DLM_OST_NAMESPACE 1
230 #define DLM_MDS_NAMESPACE 2
233 - do we just separate this by security domains and use a prefix for
234 multiple namespaces in the same domain?
244 * waiting_locks_spinlock
259 struct ldlm_resource;
260 struct ldlm_namespace;
262 struct ldlm_pool_ops {
263 int (*po_recalc)(struct ldlm_pool *pl);
264 int (*po_shrink)(struct ldlm_pool *pl, int nr,
265 unsigned int gfp_mask);
266 int (*po_setup)(struct ldlm_pool *pl, int limit);
270 * One second for pools thread check interval. Each pool has own period.
272 #define LDLM_POOLS_THREAD_PERIOD (1)
275 * 5% margin for modest pools. See ldlm_pool.c for details.
277 #define LDLM_POOLS_MODEST_MARGIN (5)
280 * Default recalc period for server side pools in sec.
282 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
285 * Default recalc period for client side pools in sec.
287 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
291 * Pool proc directory.
293 cfs_proc_dir_entry_t *pl_proc_dir;
295 * Pool name, should be long enough to contain compound proc entry name.
299 * Lock for protecting slv/clv updates.
303 * Number of allowed locks in in pool, both, client and server side.
307 * Number of granted locks in
313 atomic_t pl_grant_rate;
317 atomic_t pl_cancel_rate;
319 * Grant speed (GR-CR) per T.
321 atomic_t pl_grant_speed;
323 * Server lock volume. Protected by pl_lock.
325 __u64 pl_server_lock_volume;
327 * Current biggest client lock volume. Protected by pl_lock.
329 __u64 pl_client_lock_volume;
331 * Lock volume factor. SLV on client is calculated as following:
332 * server_slv * lock_volume_factor.
334 atomic_t pl_lock_volume_factor;
336 * Time when last slv from server was obtained.
338 time_t pl_recalc_time;
340 * Recalc period for pool.
342 time_t pl_recalc_period;
344 * Recalc and shrink ops.
346 struct ldlm_pool_ops *pl_ops;
348 * Number of planned locks for next period.
354 struct lprocfs_stats *pl_stats;
357 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
358 void *req_cookie, ldlm_mode_t mode, int flags,
361 struct ldlm_valblock_ops {
362 int (*lvbo_init)(struct ldlm_resource *res);
363 int (*lvbo_update)(struct ldlm_resource *res, struct lustre_msg *m,
364 int buf_idx, int increase);
368 LDLM_NAMESPACE_GREEDY = 1 << 0,
369 LDLM_NAMESPACE_MODEST = 1 << 1
373 * Default value for ->ns_shrink_thumb. If lock is not extent one its cost
374 * is one page. Here we have 256 pages which is 1M on i386. Thus by default
375 * all extent locks which have more than 1M long extent will be kept in lru,
376 * others (including ibits locks) will be canceled on memory pressure event.
378 #define LDLM_LOCK_SHRINK_THUMB 256
381 * Default values for the "max_nolock_size", "contention_time" and
382 * "contended_locks" namespace tunables.
384 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
385 #define NS_DEFAULT_CONTENTION_SECONDS 2
386 #define NS_DEFAULT_CONTENDED_LOCKS 32
388 struct ldlm_namespace {
390 * Namespace name. Used for logging, etc.
395 * Is this a client-side lock tree?
397 ldlm_side_t ns_client;
400 * Namespce connect flags supported by server (may be changed via proc,
401 * lru resize may be disabled/enabled).
403 __u64 ns_connect_flags;
406 * Client side orig connect flags supported by server.
408 __u64 ns_orig_connect_flags;
411 * Hash table for namespace.
413 struct list_head *ns_hash;
414 spinlock_t ns_hash_lock;
417 * Count of resources in the hash.
422 * All root resources in namespace.
424 struct list_head ns_root_list;
427 * Position in global namespace list.
429 struct list_head ns_list_chain;
432 * All root resources in namespace.
434 struct list_head ns_unused_list;
436 spinlock_t ns_unused_lock;
438 unsigned int ns_max_unused;
439 unsigned int ns_max_age;
440 unsigned int ns_timeouts;
444 unsigned int ns_ctime_age_limit;
447 * Lower limit to number of pages in lock to keep it in cache.
449 unsigned long ns_shrink_thumb;
452 * Next debug dump, jiffies.
454 cfs_time_t ns_next_dump;
458 ldlm_res_policy ns_policy;
459 struct ldlm_valblock_ops *ns_lvbo;
461 cfs_waitq_t ns_waitq;
462 struct ldlm_pool ns_pool;
463 ldlm_appetite_t ns_appetite;
466 * If more than @ns_contented_locks found, the resource considered
469 unsigned ns_contended_locks;
472 * The resource remembers contended state during @ns_contention_time,
475 unsigned ns_contention_time;
478 * Limit size of nolock requests, in bytes.
480 unsigned ns_max_nolock_size;
483 * Backward link to obd, required for ldlm pool to store new SLV.
485 struct obd_device *ns_obd;
487 struct adaptive_timeout ns_at_estimate;/* estimated lock callback time*/
490 static inline int ns_is_client(struct ldlm_namespace *ns)
493 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
494 LDLM_NAMESPACE_SERVER)));
495 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
496 ns->ns_client == LDLM_NAMESPACE_SERVER);
497 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
500 static inline int ns_is_server(struct ldlm_namespace *ns)
503 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
504 LDLM_NAMESPACE_SERVER)));
505 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
506 ns->ns_client == LDLM_NAMESPACE_SERVER);
507 return ns->ns_client == LDLM_NAMESPACE_SERVER;
510 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
513 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
518 * Resource hash table
522 #define RES_HASH_BITS 12
523 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
524 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
528 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
529 struct ldlm_lock_desc *new, void *data,
531 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
533 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
534 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
536 /* Interval node data for each LDLM_EXTENT lock */
537 struct ldlm_interval {
538 struct interval_node li_node; /* node for tree mgmt */
539 struct list_head li_group; /* the locks which have the same
540 * policy - group of the policy */
542 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
544 /* the interval tree must be accessed inside the resource lock. */
545 struct ldlm_interval_tree {
546 /* tree size, this variable is used to count
547 * granted PW locks in ldlm_extent_policy()*/
549 ldlm_mode_t lit_mode; /* lock mode */
550 struct interval_node *lit_root; /* actually ldlm_interval */
555 * Must be first in the structure.
557 struct portals_handle l_handle;
559 * Lock reference count.
563 * Internal spinlock protects l_resource. we should hold this lock
564 * first before grabbing res_lock.
568 * ldlm_lock_change_resource() can change this.
570 struct ldlm_resource *l_resource;
572 * Protected by ns_hash_lock. List item for client side lru list.
574 struct list_head l_lru;
576 * Protected by lr_lock, linkage to resource's lock queues.
578 struct list_head l_res_link;
580 * Tree node for ldlm_extent.
582 struct ldlm_interval *l_tree_node;
584 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
587 struct hlist_node l_exp_hash;
589 * Protected by lr_lock. Requested mode.
591 ldlm_mode_t l_req_mode;
593 * Granted mode, also protected by lr_lock.
595 ldlm_mode_t l_granted_mode;
597 * Lock enqueue completion handler.
599 ldlm_completion_callback l_completion_ast;
601 * Lock blocking ast handler.
603 ldlm_blocking_callback l_blocking_ast;
605 * Lock glimpse handler.
607 ldlm_glimpse_callback l_glimpse_ast;
608 ldlm_weigh_callback l_weigh_ast;
613 struct obd_export *l_export;
615 * Lock connection export.
617 struct obd_export *l_conn_export;
620 * Remote lock handle.
622 struct lustre_handle l_remote_handle;
624 ldlm_policy_data_t l_policy_data;
627 * Protected by lr_lock. Various counters: readers, writers, etc.
633 * Set for locks that were removed from class hash table and will be
634 * destroyed when last reference to them is released. Set by
635 * ldlm_lock_destroy_internal().
637 * Protected by lock and resource locks.
642 * If the lock is granted, a process sleeps on this waitq to learn when
643 * it's no longer in use. If the lock is not granted, a process sleeps
644 * on this waitq to learn when it becomes granted.
648 struct timeval l_enqueued_time;
651 * Jiffies. Should be converted to time if needed.
653 cfs_time_t l_last_used;
655 struct ldlm_extent l_req_extent;
658 * Client-side-only members.
662 * Temporary storage for an LVB received during an enqueue operation.
669 spinlock_t l_extents_list_lock;
670 struct list_head l_extents_list;
672 struct list_head l_cache_locks_list;
675 * Server-side-only members.
678 /** connection cookie for the client originated the operation. */
679 __u64 l_client_cookie;
682 * Protected by elt_lock. Callbacks pending.
684 struct list_head l_pending_chain;
686 cfs_time_t l_callback_timeout;
689 * Pid which created this lock.
694 * For ldlm_add_ast_work_item().
696 struct list_head l_bl_ast;
698 * For ldlm_add_ast_work_item().
700 struct list_head l_cp_ast;
702 * For ldlm_add_ast_work_item().
704 struct list_head l_rk_ast;
706 struct ldlm_lock *l_blocking_lock;
710 * Protected by lr_lock, linkages to "skip lists".
712 struct list_head l_sl_mode;
713 struct list_head l_sl_policy;
714 struct lu_ref l_reference;
717 struct ldlm_resource {
718 struct ldlm_namespace *lr_namespace;
720 /* protected by ns_hash_lock */
721 struct list_head lr_hash;
722 struct ldlm_resource *lr_parent; /* 0 for a root resource */
723 struct list_head lr_children; /* list head for child resources */
724 struct list_head lr_childof; /* part of ns_root_list if root res,
725 * part of lr_children if child */
728 /* protected by lr_lock */
729 struct list_head lr_granted;
730 struct list_head lr_converting;
731 struct list_head lr_waiting;
732 ldlm_mode_t lr_most_restr;
733 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
734 struct ldlm_res_id lr_name;
735 atomic_t lr_refcount;
737 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
739 /* Server-side-only lock value block elements */
740 struct semaphore lr_lvb_sem;
744 /* when the resource was considered as contended */
745 cfs_time_t lr_contention_time;
747 * List of references to this resource. For debugging.
749 struct lu_ref lr_reference;
752 struct ldlm_ast_work {
753 struct ldlm_lock *w_lock;
755 struct ldlm_lock_desc w_desc;
756 struct list_head w_list;
762 /* ldlm_enqueue parameters common */
763 struct ldlm_enqueue_info {
764 __u32 ei_type; /* Type of the lock being enqueued. */
765 __u32 ei_mode; /* Mode of the lock being enqueued. */
766 void *ei_cb_bl; /* blocking lock callback */
767 void *ei_cb_cp; /* lock completion callback */
768 void *ei_cb_gl; /* lock glimpse callback */
769 void *ei_cb_wg; /* lock weigh callback */
770 void *ei_cbdata; /* Data to be passed into callbacks. */
771 short ei_async:1; /* async request */
774 extern struct obd_ops ldlm_obd_ops;
776 extern char *ldlm_lockname[];
777 extern char *ldlm_typename[];
778 extern char *ldlm_it2str(int it);
780 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
783 if (((level) & D_CANTMASK) != 0 || \
784 ((libcfs_debug & (level)) != 0 && \
785 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
786 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
787 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
789 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
794 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
795 struct libcfs_debug_msg_data *data, const char *fmt,
797 __attribute__ ((format (printf, 4, 5)));
799 #define LDLM_ERROR(lock, fmt, a...) do { \
800 static cfs_debug_limit_state_t _ldlm_cdls; \
801 ldlm_lock_debug(&_ldlm_cdls, D_ERROR, lock, \
802 __FILE__, __FUNCTION__, __LINE__, \
806 #define LDLM_DEBUG(lock, fmt, a...) do { \
807 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
808 __FILE__, __FUNCTION__, __LINE__, \
811 #else /* !LIBCFS_DEBUG */
812 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
813 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
814 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
818 #define LDLM_DEBUG_NOLOCK(format, a...) \
819 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
821 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
822 int first_enq, ldlm_error_t *err,
823 struct list_head *work_list);
829 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
830 #define LDLM_ITER_STOP 2 /* stop iterating */
832 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
833 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
835 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
837 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
839 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
840 ldlm_res_iterator_t iter, void *closure);
842 int ldlm_replay_locks(struct obd_import *imp);
843 void ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
844 ldlm_iterator_t iter, void *data);
847 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
850 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
854 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
855 void *data, int flag);
856 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
857 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
858 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
859 ldlm_blocking_callback, ldlm_glimpse_callback);
860 int ldlm_handle_convert(struct ptlrpc_request *req);
861 int ldlm_handle_cancel(struct ptlrpc_request *req);
862 int ldlm_request_cancel(struct ptlrpc_request *req,
863 const struct ldlm_request *dlm_req, int first);
864 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
865 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
866 void ldlm_revoke_export_locks(struct obd_export *exp);
867 int ldlm_get_ref(void);
868 void ldlm_put_ref(void);
869 int ldlm_init_export(struct obd_export *exp);
870 void ldlm_destroy_export(struct obd_export *exp);
873 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
874 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
875 void ldlm_lock2handle(const struct ldlm_lock *lock,
876 struct lustre_handle *lockh);
877 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
878 void ldlm_cancel_callback(struct ldlm_lock *);
879 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
881 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
883 return __ldlm_handle2lock(h, 0);
886 #define LDLM_LOCK_REF_DEL(lock) \
887 lu_ref_del(&lock->l_reference, "handle", cfs_current())
889 static inline struct ldlm_lock *
890 ldlm_handle2lock_long(const struct lustre_handle *h, int flags)
892 struct ldlm_lock *lock;
894 lock = __ldlm_handle2lock(h, flags);
896 LDLM_LOCK_REF_DEL(lock);
900 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
901 struct lustre_msg *m, int buf_idx,
904 if (res->lr_namespace->ns_lvbo &&
905 res->lr_namespace->ns_lvbo->lvbo_update) {
906 return res->lr_namespace->ns_lvbo->lvbo_update(res, m, buf_idx,
912 int ldlm_error2errno(ldlm_error_t error);
913 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
914 * confuses user-space. */
917 * Release a temporary lock reference obtained by ldlm_handle2lock() or
918 * __ldlm_handle2lock().
920 #define LDLM_LOCK_PUT(lock) \
922 LDLM_LOCK_REF_DEL(lock); \
923 /*LDLM_DEBUG((lock), "put");*/ \
924 ldlm_lock_put(lock); \
928 * Release a lock reference obtained by some other means (see
931 #define LDLM_LOCK_RELEASE(lock) \
933 /*LDLM_DEBUG((lock), "put");*/ \
934 ldlm_lock_put(lock); \
937 #define LDLM_LOCK_GET(lock) \
939 ldlm_lock_get(lock); \
940 /*LDLM_DEBUG((lock), "get");*/ \
944 #define ldlm_lock_list_put(head, member, count) \
946 struct ldlm_lock *_lock, *_next; \
948 list_for_each_entry_safe(_lock, _next, head, member) { \
951 list_del_init(&_lock->member); \
952 LDLM_LOCK_RELEASE(_lock); \
957 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
958 void ldlm_lock_put(struct ldlm_lock *lock);
959 void ldlm_lock_destroy(struct ldlm_lock *lock);
960 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
961 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
962 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
963 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
964 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
965 void ldlm_lock_allow_match(struct ldlm_lock *lock);
966 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
967 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
968 const struct ldlm_res_id *, ldlm_type_t type,
969 ldlm_policy_data_t *, ldlm_mode_t mode,
970 struct lustre_handle *, int unref);
971 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
973 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
974 void ldlm_lock_cancel(struct ldlm_lock *lock);
975 void ldlm_cancel_locks_for_export(struct obd_export *export);
976 void ldlm_reprocess_all(struct ldlm_resource *res);
977 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
978 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
979 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
980 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
983 struct ldlm_namespace *
984 ldlm_namespace_new(struct obd_device *obd, char *name,
985 ldlm_side_t client, ldlm_appetite_t apt);
986 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
987 void ldlm_namespace_free(struct ldlm_namespace *ns,
988 struct obd_import *imp, int force);
989 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
990 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
991 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
992 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
993 void ldlm_namespace_get_locked(struct ldlm_namespace *ns);
994 void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup);
995 void ldlm_namespace_get(struct ldlm_namespace *ns);
996 void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup);
997 int ldlm_proc_setup(void);
999 void ldlm_proc_cleanup(void);
1001 static inline void ldlm_proc_cleanup(void) {}
1004 /* resource.c - internal */
1005 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1006 struct ldlm_resource *parent,
1007 const struct ldlm_res_id *,
1008 ldlm_type_t type, int create);
1009 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1010 int ldlm_resource_putref(struct ldlm_resource *res);
1011 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
1012 struct ldlm_lock *lock);
1013 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1014 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1015 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1016 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1017 void ldlm_resource_dump(int level, struct ldlm_resource *);
1018 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1019 const struct ldlm_res_id *);
1021 #define LDLM_RESOURCE_ADDREF(res) do { \
1022 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1025 #define LDLM_RESOURCE_DELREF(res) do { \
1026 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1029 struct ldlm_callback_suite {
1030 ldlm_completion_callback lcs_completion;
1031 ldlm_blocking_callback lcs_blocking;
1032 ldlm_glimpse_callback lcs_glimpse;
1033 ldlm_weigh_callback lcs_weigh;
1036 /* ldlm_request.c */
1037 int ldlm_expired_completion_wait(void *data);
1038 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1039 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1040 void *data, int flag);
1041 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1042 int ldlm_completion_ast_async(struct ldlm_lock *lock, int flags, void *data);
1043 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
1044 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1045 struct ldlm_enqueue_info *einfo,
1046 const struct ldlm_res_id *res_id,
1047 ldlm_policy_data_t *policy, int *flags,
1048 void *lvb, __u32 lvb_len, void *lvb_swabber,
1049 struct lustre_handle *lockh, int async);
1050 int ldlm_prep_enqueue_req(struct obd_export *exp,
1051 struct ptlrpc_request *req,
1052 struct list_head *cancels,
1054 int ldlm_prep_elc_req(struct obd_export *exp,
1055 struct ptlrpc_request *req,
1056 int version, int opc, int canceloff,
1057 struct list_head *cancels, int count);
1058 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1059 const struct ldlm_request *dlm_req,
1060 const struct ldlm_callback_suite *cbs);
1061 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1062 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1063 int *flags, void *lvb, __u32 lvb_len,
1064 void *lvb_swabber, struct lustre_handle *lockh,
1066 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1067 const struct ldlm_res_id *res_id,
1068 ldlm_type_t type, ldlm_policy_data_t *policy,
1069 ldlm_mode_t mode, int *flags,
1070 ldlm_blocking_callback blocking,
1071 ldlm_completion_callback completion,
1072 ldlm_glimpse_callback glimpse,
1073 void *data, __u32 lvb_len, void *lvb_swabber,
1074 const __u64 *client_cookie,
1075 struct lustre_handle *lockh);
1076 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1077 void *data, __u32 data_len);
1078 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1079 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1080 int ldlm_handle_convert0(struct ptlrpc_request *req,
1081 const struct ldlm_request *dlm_req);
1082 int ldlm_cli_cancel(struct lustre_handle *lockh);
1083 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1084 int flags, void *opaque);
1085 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1086 const struct ldlm_res_id *res_id,
1087 ldlm_policy_data_t *policy,
1088 ldlm_mode_t mode, int flags, void *opaque);
1089 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1090 int count, int flags);
1091 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1092 struct list_head *cancels,
1093 ldlm_policy_data_t *policy,
1094 ldlm_mode_t mode, int lock_flags,
1095 int cancel_flags, void *opaque);
1096 int ldlm_cli_cancel_list(struct list_head *head, int count,
1097 struct ptlrpc_request *req, int flags);
1100 /* This has to be here because recursive inclusion sucks. */
1101 int intent_disposition(struct ldlm_reply *rep, int flag);
1102 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1105 /* ioctls for trying requests */
1106 #define IOC_LDLM_TYPE 'f'
1107 #define IOC_LDLM_MIN_NR 40
1109 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1110 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1111 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1112 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1113 #define IOC_LDLM_MAX_NR 43
1116 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1117 * than one lock_res is dead-lock safe.
1119 enum lock_res_type {
1124 static inline void lock_res(struct ldlm_resource *res)
1126 spin_lock(&res->lr_lock);
1129 static inline void lock_res_nested(struct ldlm_resource *res,
1130 enum lock_res_type mode)
1132 spin_lock_nested(&res->lr_lock, mode);
1136 static inline void unlock_res(struct ldlm_resource *res)
1138 spin_unlock(&res->lr_lock);
1141 static inline void check_res_locked(struct ldlm_resource *res)
1143 LASSERT_SPIN_LOCKED(&res->lr_lock);
1146 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1147 void unlock_res_and_lock(struct ldlm_lock *lock);
1150 void ldlm_pools_recalc(ldlm_side_t client);
1151 int ldlm_pools_init(void);
1152 void ldlm_pools_fini(void);
1154 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1155 int idx, ldlm_side_t client);
1156 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1157 unsigned int gfp_mask);
1158 void ldlm_pool_fini(struct ldlm_pool *pl);
1159 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1160 int ldlm_pool_recalc(struct ldlm_pool *pl);
1161 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1162 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1163 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1164 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1165 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1166 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1167 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1168 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1169 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);