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 /* These are flags that are mapped into the flags and ASTs of blocking locks */
158 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
159 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
160 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
162 /* completion ast to be executed */
163 #define LDLM_FL_CP_REQD 0x1000000
165 /* cleanup_resource has already handled the lock */
166 #define LDLM_FL_CLEANED 0x2000000
168 /* optimization hint: LDLM can run blocking callback from current context
169 * w/o involving separate thread. in order to decrease cs rate */
170 #define LDLM_FL_ATOMIC_CB 0x4000000
172 /* Cancel lock asynchronously. See ldlm_cli_cancel_unused_resource. */
173 #define LDLM_FL_ASYNC 0x8000000
175 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
176 * such that server send blocking ast for conflict locks to this client for
177 * the 1st operation, whereas the 2nd operation has canceled this lock and
178 * is waiting for rpc_lock which is taken by the 1st operation.
179 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
180 * ELC code to cancel it.
181 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
182 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
183 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
184 * to send a separate CANCEL rpc. */
185 #define LDLM_FL_BL_AST 0x10000000
186 #define LDLM_FL_BL_DONE 0x20000000
188 /* measure lock contention and return -EUSERS if locking contention is high */
189 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
191 /* The blocking callback is overloaded to perform two functions. These flags
192 * indicate which operation should be performed. */
193 #define LDLM_CB_BLOCKING 1
194 #define LDLM_CB_CANCELING 2
196 /* compatibility matrix */
197 #define LCK_COMPAT_EX LCK_NL
198 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
199 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
200 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
201 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
202 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
203 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
204 #define LCK_COMPAT_COS (LCK_COS)
206 extern ldlm_mode_t lck_compat_array[];
208 static inline void lockmode_verify(ldlm_mode_t mode)
210 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
213 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
215 return (lck_compat_array[exist_mode] & new_mode);
220 * cluster name spaces
224 #define DLM_OST_NAMESPACE 1
225 #define DLM_MDS_NAMESPACE 2
228 - do we just separate this by security domains and use a prefix for
229 multiple namespaces in the same domain?
239 * waiting_locks_spinlock
254 struct ldlm_resource;
255 struct ldlm_namespace;
257 struct ldlm_pool_ops {
258 int (*po_recalc)(struct ldlm_pool *pl);
259 int (*po_shrink)(struct ldlm_pool *pl, int nr,
260 unsigned int gfp_mask);
261 int (*po_setup)(struct ldlm_pool *pl, int limit);
265 * One second for pools thread check interval. Each pool has own period.
267 #define LDLM_POOLS_THREAD_PERIOD (1)
270 * 5% margin for modest pools. See ldlm_pool.c for details.
272 #define LDLM_POOLS_MODEST_MARGIN (5)
275 * Default recalc period for server side pools in sec.
277 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
280 * Default recalc period for client side pools in sec.
282 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
286 * Pool proc directory.
288 cfs_proc_dir_entry_t *pl_proc_dir;
290 * Pool name, should be long enough to contain compound proc entry name.
294 * Lock for protecting slv/clv updates.
298 * Number of allowed locks in in pool, both, client and server side.
302 * Number of granted locks in
308 atomic_t pl_grant_rate;
312 atomic_t pl_cancel_rate;
314 * Grant speed (GR-CR) per T.
316 atomic_t pl_grant_speed;
318 * Server lock volume. Protected by pl_lock.
320 __u64 pl_server_lock_volume;
322 * Current biggest client lock volume. Protected by pl_lock.
324 __u64 pl_client_lock_volume;
326 * Lock volume factor. SLV on client is calculated as following:
327 * server_slv * lock_volume_factor.
329 atomic_t pl_lock_volume_factor;
331 * Time when last slv from server was obtained.
333 time_t pl_recalc_time;
335 * Recalc period for pool.
337 time_t pl_recalc_period;
339 * Recalc and shrink ops.
341 struct ldlm_pool_ops *pl_ops;
343 * Number of planned locks for next period.
349 struct lprocfs_stats *pl_stats;
352 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
353 void *req_cookie, ldlm_mode_t mode, int flags,
356 struct ldlm_valblock_ops {
357 int (*lvbo_init)(struct ldlm_resource *res);
358 int (*lvbo_update)(struct ldlm_resource *res, struct lustre_msg *m,
359 int buf_idx, int increase);
363 LDLM_NAMESPACE_GREEDY = 1 << 0,
364 LDLM_NAMESPACE_MODEST = 1 << 1
368 * Default value for ->ns_shrink_thumb. If lock is not extent one its cost
369 * is one page. Here we have 256 pages which is 1M on i386. Thus by default
370 * all extent locks which have more than 1M long extent will be kept in lru,
371 * others (including ibits locks) will be canceled on memory pressure event.
373 #define LDLM_LOCK_SHRINK_THUMB 256
376 * Default values for the "max_nolock_size", "contention_time" and
377 * "contended_locks" namespace tunables.
379 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
380 #define NS_DEFAULT_CONTENTION_SECONDS 2
381 #define NS_DEFAULT_CONTENDED_LOCKS 32
383 struct ldlm_namespace {
385 * Namespace name. Used for logging, etc.
390 * Is this a client-side lock tree?
392 ldlm_side_t ns_client;
395 * Namespce connect flags supported by server (may be changed via proc,
396 * lru resize may be disabled/enabled).
398 __u64 ns_connect_flags;
401 * Client side orig connect flags supported by server.
403 __u64 ns_orig_connect_flags;
406 * Hash table for namespace.
408 struct list_head *ns_hash;
409 spinlock_t ns_hash_lock;
412 * Count of resources in the hash.
417 * All root resources in namespace.
419 struct list_head ns_root_list;
422 * Position in global namespace list.
424 struct list_head ns_list_chain;
427 * All root resources in namespace.
429 struct list_head ns_unused_list;
431 spinlock_t ns_unused_lock;
433 unsigned int ns_max_unused;
434 unsigned int ns_max_age;
438 unsigned int ns_ctime_age_limit;
441 * Lower limit to number of pages in lock to keep it in cache.
443 unsigned long ns_shrink_thumb;
446 * Next debug dump, jiffies.
448 cfs_time_t ns_next_dump;
452 ldlm_res_policy ns_policy;
453 struct ldlm_valblock_ops *ns_lvbo;
455 cfs_waitq_t ns_waitq;
456 struct ldlm_pool ns_pool;
457 ldlm_appetite_t ns_appetite;
460 * If more than @ns_contented_locks found, the resource considered
463 unsigned ns_contended_locks;
466 * The resource remembers contended state during @ns_contention_time,
469 unsigned ns_contention_time;
472 * Limit size of nolock requests, in bytes.
474 unsigned ns_max_nolock_size;
477 * Backward link to obd, required for ldlm pool to store new SLV.
479 struct obd_device *ns_obd;
481 struct adaptive_timeout ns_at_estimate;/* estimated lock callback time*/
484 static inline int ns_is_client(struct ldlm_namespace *ns)
487 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
488 LDLM_NAMESPACE_SERVER)));
489 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
490 ns->ns_client == LDLM_NAMESPACE_SERVER);
491 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
494 static inline int ns_is_server(struct ldlm_namespace *ns)
497 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
498 LDLM_NAMESPACE_SERVER)));
499 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
500 ns->ns_client == LDLM_NAMESPACE_SERVER);
501 return ns->ns_client == LDLM_NAMESPACE_SERVER;
504 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
507 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
512 * Resource hash table
516 #define RES_HASH_BITS 10
517 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
518 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
522 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
523 struct ldlm_lock_desc *new, void *data,
525 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
527 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
528 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
530 /* Interval node data for each LDLM_EXTENT lock */
531 struct ldlm_interval {
532 struct interval_node li_node; /* node for tree mgmt */
533 struct list_head li_group; /* the locks which have the same
534 * policy - group of the policy */
536 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
538 /* the interval tree must be accessed inside the resource lock. */
539 struct ldlm_interval_tree {
540 /* tree size, this variable is used to count
541 * granted PW locks in ldlm_extent_policy()*/
543 ldlm_mode_t lit_mode; /* lock mode */
544 struct interval_node *lit_root; /* actually ldlm_interval */
549 * Must be first in the structure.
551 struct portals_handle l_handle;
553 * Lock reference count.
557 * Internal spinlock protects l_resource. we should hold this lock
558 * first before grabbing res_lock.
562 * ldlm_lock_change_resource() can change this.
564 struct ldlm_resource *l_resource;
566 * Protected by ns_hash_lock. List item for client side lru list.
568 struct list_head l_lru;
570 * Protected by lr_lock, linkage to resource's lock queues.
572 struct list_head l_res_link;
574 * Tree node for ldlm_extent.
576 struct ldlm_interval *l_tree_node;
578 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
581 struct hlist_node l_exp_hash;
583 * Protected by lr_lock. Requested mode.
585 ldlm_mode_t l_req_mode;
587 * Granted mode, also protected by lr_lock.
589 ldlm_mode_t l_granted_mode;
591 * Lock enqueue completion handler.
593 ldlm_completion_callback l_completion_ast;
595 * Lock blocking ast handler.
597 ldlm_blocking_callback l_blocking_ast;
599 * Lock glimpse handler.
601 ldlm_glimpse_callback l_glimpse_ast;
602 ldlm_weigh_callback l_weigh_ast;
607 struct obd_export *l_export;
609 * Lock connection export.
611 struct obd_export *l_conn_export;
614 * Remote lock handle.
616 struct lustre_handle l_remote_handle;
618 ldlm_policy_data_t l_policy_data;
621 * Protected by lr_lock. Various counters: readers, writers, etc.
627 * Set for locks that were removed from class hash table and will be
628 * destroyed when last reference to them is released. Set by
629 * ldlm_lock_destroy_internal().
631 * Protected by lock and resource locks.
636 * If the lock is granted, a process sleeps on this waitq to learn when
637 * it's no longer in use. If the lock is not granted, a process sleeps
638 * on this waitq to learn when it becomes granted.
642 struct timeval l_enqueued_time;
645 * Jiffies. Should be converted to time if needed.
647 cfs_time_t l_last_used;
649 struct ldlm_extent l_req_extent;
652 * Client-side-only members.
656 * Temporary storage for an LVB received during an enqueue operation.
663 spinlock_t l_extents_list_lock;
664 struct list_head l_extents_list;
666 struct list_head l_cache_locks_list;
669 * Server-side-only members.
672 /** connection cookie for the client originated the operation. */
673 __u64 l_client_cookie;
676 * Protected by elt_lock. Callbacks pending.
678 struct list_head l_pending_chain;
680 cfs_time_t l_callback_timeout;
683 * Pid which created this lock.
688 * For ldlm_add_ast_work_item().
690 struct list_head l_bl_ast;
692 * For ldlm_add_ast_work_item().
694 struct list_head l_cp_ast;
696 * For ldlm_add_ast_work_item().
698 struct list_head l_rk_ast;
700 struct ldlm_lock *l_blocking_lock;
704 * Protected by lr_lock, linkages to "skip lists".
706 struct list_head l_sl_mode;
707 struct list_head l_sl_policy;
708 struct lu_ref l_reference;
711 struct ldlm_resource {
712 struct ldlm_namespace *lr_namespace;
714 /* protected by ns_hash_lock */
715 struct list_head lr_hash;
716 struct ldlm_resource *lr_parent; /* 0 for a root resource */
717 struct list_head lr_children; /* list head for child resources */
718 struct list_head lr_childof; /* part of ns_root_list if root res,
719 * part of lr_children if child */
722 /* protected by lr_lock */
723 struct list_head lr_granted;
724 struct list_head lr_converting;
725 struct list_head lr_waiting;
726 ldlm_mode_t lr_most_restr;
727 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
728 struct ldlm_res_id lr_name;
729 atomic_t lr_refcount;
731 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
733 /* Server-side-only lock value block elements */
734 struct semaphore lr_lvb_sem;
738 /* when the resource was considered as contended */
739 cfs_time_t lr_contention_time;
741 * List of references to this resource. For debugging.
743 struct lu_ref lr_reference;
746 struct ldlm_ast_work {
747 struct ldlm_lock *w_lock;
749 struct ldlm_lock_desc w_desc;
750 struct list_head w_list;
756 /* ldlm_enqueue parameters common */
757 struct ldlm_enqueue_info {
758 __u32 ei_type; /* Type of the lock being enqueued. */
759 __u32 ei_mode; /* Mode of the lock being enqueued. */
760 void *ei_cb_bl; /* blocking lock callback */
761 void *ei_cb_cp; /* lock completion callback */
762 void *ei_cb_gl; /* lock glimpse callback */
763 void *ei_cb_wg; /* lock weigh callback */
764 void *ei_cbdata; /* Data to be passed into callbacks. */
765 short ei_async:1; /* async request */
768 extern struct obd_ops ldlm_obd_ops;
770 extern char *ldlm_lockname[];
771 extern char *ldlm_typename[];
772 extern char *ldlm_it2str(int it);
774 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
777 if (((level) & D_CANTMASK) != 0 || \
778 ((libcfs_debug & (level)) != 0 && \
779 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
780 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
781 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
783 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
788 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
789 struct libcfs_debug_msg_data *data, const char *fmt,
791 __attribute__ ((format (printf, 4, 5)));
793 #define LDLM_ERROR(lock, fmt, a...) do { \
794 static cfs_debug_limit_state_t _ldlm_cdls; \
795 ldlm_lock_debug(&_ldlm_cdls, D_ERROR, lock, \
796 __FILE__, __FUNCTION__, __LINE__, \
800 #define LDLM_DEBUG(lock, fmt, a...) do { \
801 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
802 __FILE__, __FUNCTION__, __LINE__, \
805 #else /* !LIBCFS_DEBUG */
806 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
807 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
808 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
812 #define LDLM_DEBUG_NOLOCK(format, a...) \
813 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
815 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
816 int first_enq, ldlm_error_t *err,
817 struct list_head *work_list);
823 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
824 #define LDLM_ITER_STOP 2 /* stop iterating */
826 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
827 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
829 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
831 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
833 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
834 ldlm_res_iterator_t iter, void *closure);
836 int ldlm_replay_locks(struct obd_import *imp);
837 void ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
838 ldlm_iterator_t iter, void *data);
841 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
844 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
848 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
849 void *data, int flag);
850 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
851 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
852 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
853 ldlm_blocking_callback, ldlm_glimpse_callback);
854 int ldlm_handle_convert(struct ptlrpc_request *req);
855 int ldlm_handle_cancel(struct ptlrpc_request *req);
856 int ldlm_request_cancel(struct ptlrpc_request *req,
857 const struct ldlm_request *dlm_req, int first);
858 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
859 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock);
860 void ldlm_revoke_export_locks(struct obd_export *exp);
861 int ldlm_get_ref(void);
862 void ldlm_put_ref(void);
863 int ldlm_init_export(struct obd_export *exp);
864 void ldlm_destroy_export(struct obd_export *exp);
867 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
868 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
869 void ldlm_lock2handle(const struct ldlm_lock *lock,
870 struct lustre_handle *lockh);
871 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
872 void ldlm_cancel_callback(struct ldlm_lock *);
873 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
875 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
877 return __ldlm_handle2lock(h, 0);
880 #define LDLM_LOCK_REF_DEL(lock) \
881 lu_ref_del(&lock->l_reference, "handle", cfs_current())
883 static inline struct ldlm_lock *
884 ldlm_handle2lock_long(const struct lustre_handle *h, int flags)
886 struct ldlm_lock *lock;
888 lock = __ldlm_handle2lock(h, flags);
890 LDLM_LOCK_REF_DEL(lock);
894 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
895 struct lustre_msg *m, int buf_idx,
898 if (res->lr_namespace->ns_lvbo &&
899 res->lr_namespace->ns_lvbo->lvbo_update) {
900 return res->lr_namespace->ns_lvbo->lvbo_update(res, m, buf_idx,
906 int ldlm_error2errno(ldlm_error_t error);
907 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
908 * confuses user-space. */
911 * Release a temporary lock reference obtained by ldlm_handle2lock() or
912 * __ldlm_handle2lock().
914 #define LDLM_LOCK_PUT(lock) \
916 LDLM_LOCK_REF_DEL(lock); \
917 /*LDLM_DEBUG((lock), "put");*/ \
918 ldlm_lock_put(lock); \
922 * Release a lock reference obtained by some other means (see
925 #define LDLM_LOCK_RELEASE(lock) \
927 /*LDLM_DEBUG((lock), "put");*/ \
928 ldlm_lock_put(lock); \
931 #define LDLM_LOCK_GET(lock) \
933 ldlm_lock_get(lock); \
934 /*LDLM_DEBUG((lock), "get");*/ \
938 #define ldlm_lock_list_put(head, member, count) \
940 struct ldlm_lock *_lock, *_next; \
942 list_for_each_entry_safe(_lock, _next, head, member) { \
945 list_del_init(&_lock->member); \
946 LDLM_LOCK_RELEASE(_lock); \
951 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
952 void ldlm_lock_put(struct ldlm_lock *lock);
953 void ldlm_lock_destroy(struct ldlm_lock *lock);
954 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
955 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
956 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
957 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
958 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
959 void ldlm_lock_allow_match(struct ldlm_lock *lock);
960 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
961 const struct ldlm_res_id *, ldlm_type_t type,
962 ldlm_policy_data_t *, ldlm_mode_t mode,
963 struct lustre_handle *, int unref);
964 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
966 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
967 void ldlm_lock_cancel(struct ldlm_lock *lock);
968 void ldlm_cancel_locks_for_export(struct obd_export *export);
969 void ldlm_reprocess_all(struct ldlm_resource *res);
970 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
971 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
972 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
973 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
976 struct ldlm_namespace *
977 ldlm_namespace_new(struct obd_device *obd, char *name,
978 ldlm_side_t client, ldlm_appetite_t apt);
979 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
980 void ldlm_namespace_free(struct ldlm_namespace *ns,
981 struct obd_import *imp, int force);
982 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
983 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
984 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
985 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
986 void ldlm_namespace_get_locked(struct ldlm_namespace *ns);
987 void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup);
988 void ldlm_namespace_get(struct ldlm_namespace *ns);
989 void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup);
990 int ldlm_proc_setup(void);
992 void ldlm_proc_cleanup(void);
994 static inline void ldlm_proc_cleanup(void) {}
997 /* resource.c - internal */
998 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
999 struct ldlm_resource *parent,
1000 const struct ldlm_res_id *,
1001 ldlm_type_t type, int create);
1002 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1003 int ldlm_resource_putref(struct ldlm_resource *res);
1004 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
1005 struct ldlm_lock *lock);
1006 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1007 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1008 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1009 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1010 void ldlm_resource_dump(int level, struct ldlm_resource *);
1011 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1012 const struct ldlm_res_id *);
1014 #define LDLM_RESOURCE_ADDREF(res) do { \
1015 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1018 #define LDLM_RESOURCE_DELREF(res) do { \
1019 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1022 struct ldlm_callback_suite {
1023 ldlm_completion_callback lcs_completion;
1024 ldlm_blocking_callback lcs_blocking;
1025 ldlm_glimpse_callback lcs_glimpse;
1026 ldlm_weigh_callback lcs_weigh;
1029 /* ldlm_request.c */
1030 int ldlm_expired_completion_wait(void *data);
1031 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1032 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1033 void *data, int flag);
1034 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1035 int ldlm_completion_ast_async(struct ldlm_lock *lock, int flags, void *data);
1036 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
1037 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1038 struct ldlm_enqueue_info *einfo,
1039 const struct ldlm_res_id *res_id,
1040 ldlm_policy_data_t *policy, int *flags,
1041 void *lvb, __u32 lvb_len, void *lvb_swabber,
1042 struct lustre_handle *lockh, int async);
1043 int ldlm_prep_enqueue_req(struct obd_export *exp,
1044 struct ptlrpc_request *req,
1045 struct list_head *cancels,
1047 int ldlm_prep_elc_req(struct obd_export *exp,
1048 struct ptlrpc_request *req,
1049 int version, int opc, int canceloff,
1050 struct list_head *cancels, int count);
1051 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1052 const struct ldlm_request *dlm_req,
1053 const struct ldlm_callback_suite *cbs);
1054 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1055 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1056 int *flags, void *lvb, __u32 lvb_len,
1057 void *lvb_swabber, struct lustre_handle *lockh,
1059 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1060 const struct ldlm_res_id *res_id,
1061 ldlm_type_t type, ldlm_policy_data_t *policy,
1062 ldlm_mode_t mode, int *flags,
1063 ldlm_blocking_callback blocking,
1064 ldlm_completion_callback completion,
1065 ldlm_glimpse_callback glimpse,
1066 void *data, __u32 lvb_len, void *lvb_swabber,
1067 const __u64 *client_cookie,
1068 struct lustre_handle *lockh);
1069 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1070 void *data, __u32 data_len);
1071 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1072 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1073 int ldlm_handle_convert0(struct ptlrpc_request *req,
1074 const struct ldlm_request *dlm_req);
1075 int ldlm_cli_cancel(struct lustre_handle *lockh);
1076 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1077 int flags, void *opaque);
1078 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1079 const struct ldlm_res_id *res_id,
1080 ldlm_policy_data_t *policy,
1081 ldlm_mode_t mode, int flags, void *opaque);
1082 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1083 int count, int flags);
1084 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1085 struct list_head *cancels,
1086 ldlm_policy_data_t *policy,
1087 ldlm_mode_t mode, int lock_flags,
1088 int cancel_flags, void *opaque);
1089 int ldlm_cli_cancel_list(struct list_head *head, int count,
1090 struct ptlrpc_request *req, int flags);
1093 /* This has to be here because recursive inclusion sucks. */
1094 int intent_disposition(struct ldlm_reply *rep, int flag);
1095 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1098 /* ioctls for trying requests */
1099 #define IOC_LDLM_TYPE 'f'
1100 #define IOC_LDLM_MIN_NR 40
1102 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1103 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1104 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1105 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1106 #define IOC_LDLM_MAX_NR 43
1109 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1110 * than one lock_res is dead-lock safe.
1112 enum lock_res_type {
1117 static inline void lock_res(struct ldlm_resource *res)
1119 spin_lock(&res->lr_lock);
1122 static inline void lock_res_nested(struct ldlm_resource *res,
1123 enum lock_res_type mode)
1125 spin_lock_nested(&res->lr_lock, mode);
1129 static inline void unlock_res(struct ldlm_resource *res)
1131 spin_unlock(&res->lr_lock);
1134 static inline void check_res_locked(struct ldlm_resource *res)
1136 LASSERT_SPIN_LOCKED(&res->lr_lock);
1139 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1140 void unlock_res_and_lock(struct ldlm_lock *lock);
1143 void ldlm_pools_recalc(ldlm_side_t client);
1144 int ldlm_pools_init(void);
1145 void ldlm_pools_fini(void);
1147 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1148 int idx, ldlm_side_t client);
1149 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1150 unsigned int gfp_mask);
1151 void ldlm_pool_fini(struct ldlm_pool *pl);
1152 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1153 int ldlm_pool_recalc(struct ldlm_pool *pl);
1154 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1155 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1156 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1157 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1158 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1159 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1160 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1161 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1162 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);