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))
69 #define LDLM_CTIME_AGE_LIMIT (10)
74 ELDLM_LOCK_CHANGED = 300,
75 ELDLM_LOCK_ABORTED = 301,
76 ELDLM_LOCK_REPLACED = 302,
77 ELDLM_NO_LOCK_DATA = 303,
79 ELDLM_NAMESPACE_EXISTS = 400,
80 ELDLM_BAD_NAMESPACE = 401
84 LDLM_NAMESPACE_SERVER = 1 << 0,
85 LDLM_NAMESPACE_CLIENT = 1 << 1
88 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
90 /* If the server returns one of these flags, then the lock was put on that list.
91 * If the client sends one of these flags (during recovery ONLY!), it wants the
92 * lock added to the specified list, no questions asked. -p */
93 #define LDLM_FL_BLOCK_GRANTED 0x000002
94 #define LDLM_FL_BLOCK_CONV 0x000004
95 #define LDLM_FL_BLOCK_WAIT 0x000008
97 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
98 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
99 * queued for sending. */
100 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
101 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
103 /* Lock is being replayed. This could probably be implied by the fact that one
104 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
105 #define LDLM_FL_REPLAY 0x000100
107 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
108 #define LDLM_FL_LOCAL_ONLY 0x000400 /* see ldlm_cli_cancel_unused */
110 /* don't run the cancel callback under ldlm_cli_cancel_unused */
111 #define LDLM_FL_FAILED 0x000800
113 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
114 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
115 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
116 #define LDLM_FL_WARN 0x008000 /* see ldlm_cli_cancel_unused */
117 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
119 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
122 /* file & record locking */
123 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
124 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
126 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
127 * the LVB filling happens _after_ the lock has been granted, so another thread
128 * can match`t before the LVB has been updated. As a dirty hack, we set
129 * LDLM_FL_LVB_READY only after we've done the LVB poop.
130 * this is only needed on lov/osc now, where lvb is actually used and callers
131 * must set it in input flags.
133 * The proper fix is to do the granting inside of the completion AST, which can
134 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
135 * pretty high-risk, though, and would need a lot more testing. */
137 #define LDLM_FL_LVB_READY 0x100000
139 /* A lock contributes to the kms calculation until it has finished the part
140 * of it's cancelation that performs write back on its dirty pages. It
141 * can remain on the granted list during this whole time. Threads racing
142 * to update the kms after performing their writeback need to know to
143 * exclude each others locks from the calculation as they walk the granted
145 #define LDLM_FL_KMS_IGNORE 0x200000
147 /* Immediatelly cancel such locks when they block some other locks. Send
148 * cancel notification to original lock holder, but expect no reply. This is
149 * for clients (like liblustre) that cannot be expected to reliably response
150 * to blocking ast. */
151 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
153 /* Flags flags inherited from parent lock when doing intents. */
154 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
156 /* These are flags that are mapped into the flags and ASTs of blocking locks */
157 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
158 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
159 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
161 /* completion ast to be executed */
162 #define LDLM_FL_CP_REQD 0x1000000
164 /* cleanup_resource has already handled the lock */
165 #define LDLM_FL_CLEANED 0x2000000
167 /* optimization hint: LDLM can run blocking callback from current context
168 * w/o involving separate thread. in order to decrease cs rate */
169 #define LDLM_FL_ATOMIC_CB 0x4000000
171 /* Cancel lock asynchronously. See ldlm_cli_cancel_unused_resource. */
172 #define LDLM_FL_ASYNC 0x8000000
174 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
175 * such that server send blocking ast for conflict locks to this client for
176 * the 1st operation, whereas the 2nd operation has canceled this lock and
177 * is waiting for rpc_lock which is taken by the 1st operation.
178 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
179 * ELC code to cancel it.
180 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
181 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
182 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
183 * to send a separate CANCEL rpc. */
184 #define LDLM_FL_BL_AST 0x10000000
185 #define LDLM_FL_BL_DONE 0x20000000
187 /* measure lock contention and return -EUSERS if locking contention is high */
188 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
190 /* The blocking callback is overloaded to perform two functions. These flags
191 * indicate which operation should be performed. */
192 #define LDLM_CB_BLOCKING 1
193 #define LDLM_CB_CANCELING 2
195 /* compatibility matrix */
196 #define LCK_COMPAT_EX LCK_NL
197 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
198 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
199 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
200 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
201 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
202 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
204 extern ldlm_mode_t lck_compat_array[];
206 static inline void lockmode_verify(ldlm_mode_t mode)
208 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
211 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
213 return (lck_compat_array[exist_mode] & new_mode);
218 * cluster name spaces
222 #define DLM_OST_NAMESPACE 1
223 #define DLM_MDS_NAMESPACE 2
226 - do we just separate this by security domains and use a prefix for
227 multiple namespaces in the same domain?
237 * waiting_locks_spinlock
252 struct ldlm_resource;
253 struct ldlm_namespace;
255 struct ldlm_pool_ops {
256 int (*po_recalc)(struct ldlm_pool *pl);
257 int (*po_shrink)(struct ldlm_pool *pl, int nr,
258 unsigned int gfp_mask);
259 int (*po_setup)(struct ldlm_pool *pl, int limit);
263 * One second for pools thread check interval. Each pool has own period.
265 #define LDLM_POOLS_THREAD_PERIOD (1)
268 * 5% margin for modest pools. See ldlm_pool.c for details.
270 #define LDLM_POOLS_MODEST_MARGIN (5)
273 * Default recalc period for server side pools in sec.
275 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
278 * Default recalc period for client side pools in sec.
280 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
284 * Pool proc directory.
286 cfs_proc_dir_entry_t *pl_proc_dir;
288 * Pool name, should be long enough to contain compound proc entry name.
292 * Lock for protecting slv/clv updates.
296 * Number of allowed locks in in pool, both, client and server side.
300 * Number of granted locks in
306 atomic_t pl_grant_rate;
310 atomic_t pl_cancel_rate;
312 * Grant speed (GR-CR) per T.
314 atomic_t pl_grant_speed;
316 * Server lock volume. Protected by pl_lock.
318 __u64 pl_server_lock_volume;
320 * Current biggest client lock volume. Protected by pl_lock.
322 __u64 pl_client_lock_volume;
324 * Lock volume factor. SLV on client is calculated as following:
325 * server_slv * lock_volume_factor.
327 atomic_t pl_lock_volume_factor;
329 * Time when last slv from server was obtained.
331 time_t pl_recalc_time;
333 * Recalc period for pool.
335 time_t pl_recalc_period;
337 * Recalc and shrink ops.
339 struct ldlm_pool_ops *pl_ops;
341 * Number of planned locks for next period.
347 struct lprocfs_stats *pl_stats;
350 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
351 void *req_cookie, ldlm_mode_t mode, int flags,
354 struct ldlm_valblock_ops {
355 int (*lvbo_init)(struct ldlm_resource *res);
356 int (*lvbo_update)(struct ldlm_resource *res, struct lustre_msg *m,
357 int buf_idx, int increase);
361 LDLM_NAMESPACE_GREEDY = 1 << 0,
362 LDLM_NAMESPACE_MODEST = 1 << 1
366 * Default value for ->ns_shrink_thumb. If lock is not extent one its cost
367 * is one page. Here we have 256 pages which is 1M on i386. Thus by default
368 * all extent locks which have more than 1M long extent will be kept in lru,
369 * others (including ibits locks) will be canceled on memory pressure event.
371 #define LDLM_LOCK_SHRINK_THUMB 256
374 * Default values for the "max_nolock_size", "contention_time" and
375 * "contended_locks" namespace tunables.
377 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
378 #define NS_DEFAULT_CONTENTION_SECONDS 2
379 #define NS_DEFAULT_CONTENDED_LOCKS 32
381 struct ldlm_namespace {
383 * Namespace name. Used for logging, etc.
388 * Is this a client-side lock tree?
390 ldlm_side_t ns_client;
393 * Namespce connect flags supported by server (may be changed via proc,
394 * lru resize may be disabled/enabled).
396 __u64 ns_connect_flags;
399 * Client side orig connect flags supported by server.
401 __u64 ns_orig_connect_flags;
404 * Hash table for namespace.
406 struct list_head *ns_hash;
407 spinlock_t ns_hash_lock;
410 * Count of resources in the hash.
415 * All root resources in namespace.
417 struct list_head ns_root_list;
420 * Position in global namespace list.
422 struct list_head ns_list_chain;
425 * All root resources in namespace.
427 struct list_head ns_unused_list;
429 spinlock_t ns_unused_lock;
431 unsigned int ns_max_unused;
432 unsigned int ns_max_age;
437 unsigned int ns_ctime_age_limit;
440 * Lower limit to number of pages in lock to keep it in cache.
442 unsigned int ns_shrink_thumb;
445 * Next debug dump, jiffies.
447 cfs_time_t ns_next_dump;
451 ldlm_res_policy ns_policy;
452 struct ldlm_valblock_ops *ns_lvbo;
454 cfs_waitq_t ns_waitq;
455 struct ldlm_pool ns_pool;
456 ldlm_appetite_t ns_appetite;
459 * If more than @ns_contented_locks found, the resource considered
462 unsigned ns_contended_locks;
465 * The resource remembers contended state during @ns_contention_time,
468 unsigned ns_contention_time;
471 * Limit size of nolock requests, in bytes.
473 unsigned ns_max_nolock_size;
476 * Backward link to obd, required for ldlm pool to store new SLV.
478 struct obd_device *ns_obd;
480 struct adaptive_timeout ns_at_estimate;/* estimated lock callback time*/
483 static inline int ns_is_client(struct ldlm_namespace *ns)
486 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
487 LDLM_NAMESPACE_SERVER)));
488 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
489 ns->ns_client == LDLM_NAMESPACE_SERVER);
490 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
493 static inline int ns_is_server(struct ldlm_namespace *ns)
496 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
497 LDLM_NAMESPACE_SERVER)));
498 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
499 ns->ns_client == LDLM_NAMESPACE_SERVER);
500 return ns->ns_client == LDLM_NAMESPACE_SERVER;
503 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
506 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
511 * Resource hash table
515 #define RES_HASH_BITS 10
516 #define RES_HASH_SIZE (1UL << RES_HASH_BITS)
517 #define RES_HASH_MASK (RES_HASH_SIZE - 1)
521 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
522 struct ldlm_lock_desc *new, void *data,
524 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
526 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
528 /* Interval node data for each LDLM_EXTENT lock */
529 struct ldlm_interval {
530 struct interval_node li_node; /* node for tree mgmt */
531 struct list_head li_group; /* the locks which have the same
532 * policy - group of the policy */
534 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
536 /* the interval tree must be accessed inside the resource lock. */
537 struct ldlm_interval_tree {
538 /* tree size, this variable is used to count
539 * granted PW locks in ldlm_extent_policy()*/
541 ldlm_mode_t lit_mode; /* lock mode */
542 struct interval_node *lit_root; /* actually ldlm_interval */
547 * Must be first in the structure.
549 struct portals_handle l_handle;
551 * Lock reference count.
555 * Internal spinlock protects l_resource. we should hold this lock
556 * first before grabbing res_lock.
560 * ldlm_lock_change_resource() can change this.
562 struct ldlm_resource *l_resource;
564 * Protected by ns_hash_lock. List item for client side lru list.
566 struct list_head l_lru;
568 * Protected by lr_lock, linkage to resource's lock queues.
570 struct list_head l_res_link;
572 * Tree node for ldlm_extent.
574 struct ldlm_interval *l_tree_node;
576 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
579 struct hlist_node l_exp_hash;
581 * Protected by lr_lock. Requested mode.
583 ldlm_mode_t l_req_mode;
585 * Granted mode, also protected by lr_lock.
587 ldlm_mode_t l_granted_mode;
589 * Lock enqueue completion handler.
591 ldlm_completion_callback l_completion_ast;
593 * Lock blocking ast handler.
595 ldlm_blocking_callback l_blocking_ast;
597 * Lock glimpse handler.
599 ldlm_glimpse_callback l_glimpse_ast;
604 struct obd_export *l_export;
606 * Lock connection export.
608 struct obd_export *l_conn_export;
611 * Remote lock handle.
613 struct lustre_handle l_remote_handle;
615 ldlm_policy_data_t l_policy_data;
618 * Protected by lr_lock. Various counters: readers, writers, etc.
624 * Set for locks that were removed from class hash table and will be
625 * destroyed when last reference to them is released. Set by
626 * ldlm_lock_destroy_internal().
628 * Protected by lock and resource locks.
633 * If the lock is granted, a process sleeps on this waitq to learn when
634 * it's no longer in use. If the lock is not granted, a process sleeps
635 * on this waitq to learn when it becomes granted.
639 struct timeval l_enqueued_time;
642 * Jiffies. Should be converted to time if needed.
644 cfs_time_t l_last_used;
646 struct ldlm_extent l_req_extent;
649 * Client-side-only members.
653 * Temporary storage for an LVB received during an enqueue operation.
660 spinlock_t l_extents_list_lock;
661 struct list_head l_extents_list;
663 struct list_head l_cache_locks_list;
666 * Server-side-only members.
670 * Protected by elt_lock. Callbacks pending.
672 struct list_head l_pending_chain;
674 cfs_time_t l_callback_timeout;
677 * Pid which created this lock.
682 * For ldlm_add_ast_work_item().
684 struct list_head l_bl_ast;
686 * For ldlm_add_ast_work_item().
688 struct list_head l_cp_ast;
690 * For ldlm_add_ast_work_item().
692 struct list_head l_rk_ast;
694 struct ldlm_lock *l_blocking_lock;
698 * Protected by lr_lock, linkages to "skip lists".
700 struct list_head l_sl_mode;
701 struct list_head l_sl_policy;
704 struct ldlm_resource {
705 struct ldlm_namespace *lr_namespace;
707 /* protected by ns_hash_lock */
708 struct list_head lr_hash;
709 struct ldlm_resource *lr_parent; /* 0 for a root resource */
710 struct list_head lr_children; /* list head for child resources */
711 struct list_head lr_childof; /* part of ns_root_list if root res,
712 * part of lr_children if child */
715 /* protected by lr_lock */
716 struct list_head lr_granted;
717 struct list_head lr_converting;
718 struct list_head lr_waiting;
719 ldlm_mode_t lr_most_restr;
720 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
721 struct ldlm_res_id lr_name;
722 atomic_t lr_refcount;
724 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
726 /* Server-side-only lock value block elements */
727 struct semaphore lr_lvb_sem;
731 /* when the resource was considered as contended */
732 cfs_time_t lr_contention_time;
735 struct ldlm_ast_work {
736 struct ldlm_lock *w_lock;
738 struct ldlm_lock_desc w_desc;
739 struct list_head w_list;
745 /* ldlm_enqueue parameters common */
746 struct ldlm_enqueue_info {
747 __u32 ei_type; /* Type of the lock being enqueued. */
748 __u32 ei_mode; /* Mode of the lock being enqueued. */
749 void *ei_cb_bl; /* blocking lock callback */
750 void *ei_cb_cp; /* lock completion callback */
751 void *ei_cb_gl; /* lock glimpse callback */
752 void *ei_cbdata; /* Data to be passed into callbacks. */
753 short ei_async:1; /* async request */
756 extern struct obd_ops ldlm_obd_ops;
758 extern char *ldlm_lockname[];
759 extern char *ldlm_typename[];
760 extern char *ldlm_it2str(int it);
762 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
765 if (((level) & D_CANTMASK) != 0 || \
766 ((libcfs_debug & (level)) != 0 && \
767 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
768 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
769 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
771 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
776 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
777 struct libcfs_debug_msg_data *data, const char *fmt,
779 __attribute__ ((format (printf, 4, 5)));
781 #define LDLM_ERROR(lock, fmt, a...) do { \
782 static cfs_debug_limit_state_t _ldlm_cdls; \
783 ldlm_lock_debug(&_ldlm_cdls, D_ERROR, lock, \
784 __FILE__, __FUNCTION__, __LINE__, \
788 #define LDLM_DEBUG(lock, fmt, a...) do { \
789 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
790 __FILE__, __FUNCTION__, __LINE__, \
793 #else /* !LIBCFS_DEBUG */
794 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
795 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
796 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
800 #define LDLM_DEBUG_NOLOCK(format, a...) \
801 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
803 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
804 int first_enq, ldlm_error_t *err,
805 struct list_head *work_list);
811 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
812 #define LDLM_ITER_STOP 2 /* stop iterating */
814 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
815 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
817 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
819 int ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
821 int ldlm_namespace_foreach_res(struct ldlm_namespace *ns,
822 ldlm_res_iterator_t iter, void *closure);
824 int ldlm_replay_locks(struct obd_import *imp);
825 void ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
826 ldlm_iterator_t iter, void *data);
829 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
832 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
836 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
837 void *data, int flag);
838 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
839 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
840 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
841 ldlm_blocking_callback, ldlm_glimpse_callback);
842 int ldlm_handle_convert(struct ptlrpc_request *req);
843 int ldlm_handle_cancel(struct ptlrpc_request *req);
844 int ldlm_request_cancel(struct ptlrpc_request *req,
845 const struct ldlm_request *dlm_req, int first);
846 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
847 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock);
848 void ldlm_revoke_export_locks(struct obd_export *exp);
849 int ldlm_get_ref(void);
850 void ldlm_put_ref(void);
851 int ldlm_init_export(struct obd_export *exp);
852 void ldlm_destroy_export(struct obd_export *exp);
855 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
856 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
857 void ldlm_lock2handle(const struct ldlm_lock *lock,
858 struct lustre_handle *lockh);
859 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
860 void ldlm_cancel_callback(struct ldlm_lock *);
861 int ldlm_lock_set_data(struct lustre_handle *, void *data);
862 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
863 struct ldlm_lock *ldlm_handle2lock_ns(struct ldlm_namespace *,
864 const struct lustre_handle *);
866 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
868 return __ldlm_handle2lock(h, 0);
871 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
872 struct lustre_msg *m, int buf_idx,
875 if (res->lr_namespace->ns_lvbo &&
876 res->lr_namespace->ns_lvbo->lvbo_update) {
877 return res->lr_namespace->ns_lvbo->lvbo_update(res, m, buf_idx,
883 #define LDLM_LOCK_PUT(lock) \
885 /*LDLM_DEBUG((lock), "put");*/ \
886 ldlm_lock_put(lock); \
889 #define LDLM_LOCK_GET(lock) \
891 ldlm_lock_get(lock); \
892 /*LDLM_DEBUG((lock), "get");*/ \
896 #define ldlm_lock_list_put(head, member, count) \
898 struct ldlm_lock *_lock, *_next; \
900 list_for_each_entry_safe(_lock, _next, head, member) { \
903 list_del_init(&_lock->member); \
904 LDLM_LOCK_PUT(_lock); \
909 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
910 void ldlm_lock_put(struct ldlm_lock *lock);
911 void ldlm_lock_destroy(struct ldlm_lock *lock);
912 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
913 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
914 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
915 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
916 void ldlm_lock_allow_match(struct ldlm_lock *lock);
917 int ldlm_lock_fast_match(struct ldlm_lock *, int, obd_off, obd_off, void **);
918 void ldlm_lock_fast_release(void *, int);
919 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
920 const struct ldlm_res_id *, ldlm_type_t type,
921 ldlm_policy_data_t *, ldlm_mode_t mode,
922 struct lustre_handle *);
923 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
925 void ldlm_lock_cancel(struct ldlm_lock *lock);
926 void ldlm_cancel_locks_for_export(struct obd_export *export);
927 void ldlm_reprocess_all(struct ldlm_resource *res);
928 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
929 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
930 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
931 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
934 struct ldlm_namespace *
935 ldlm_namespace_new(struct obd_device *obd, char *name,
936 ldlm_side_t client, ldlm_appetite_t apt);
937 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
938 void ldlm_namespace_free(struct ldlm_namespace *ns,
939 struct obd_import *imp, int force);
940 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
941 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
942 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
943 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
944 void ldlm_namespace_get_locked(struct ldlm_namespace *ns);
945 void ldlm_namespace_put_locked(struct ldlm_namespace *ns, int wakeup);
946 void ldlm_namespace_get(struct ldlm_namespace *ns);
947 void ldlm_namespace_put(struct ldlm_namespace *ns, int wakeup);
948 int ldlm_proc_setup(void);
950 void ldlm_proc_cleanup(void);
952 static inline void ldlm_proc_cleanup(void) {}
955 /* resource.c - internal */
956 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
957 struct ldlm_resource *parent,
958 const struct ldlm_res_id *,
959 ldlm_type_t type, int create);
960 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
961 int ldlm_resource_putref(struct ldlm_resource *res);
962 void ldlm_resource_add_lock(struct ldlm_resource *res, struct list_head *head,
963 struct ldlm_lock *lock);
964 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
965 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
966 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
967 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
968 void ldlm_resource_dump(int level, struct ldlm_resource *);
969 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
970 const struct ldlm_res_id *);
972 struct ldlm_callback_suite {
973 ldlm_completion_callback lcs_completion;
974 ldlm_blocking_callback lcs_blocking;
975 ldlm_glimpse_callback lcs_glimpse;
979 int ldlm_expired_completion_wait(void *data);
980 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
981 void *data, int flag);
982 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
983 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
984 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
985 struct ldlm_enqueue_info *einfo,
986 const struct ldlm_res_id *res_id,
987 ldlm_policy_data_t *policy, int *flags,
988 void *lvb, __u32 lvb_len, void *lvb_swabber,
989 struct lustre_handle *lockh, int async);
990 int ldlm_prep_enqueue_req(struct obd_export *exp,
991 struct ptlrpc_request *req,
992 struct list_head *cancels,
994 int ldlm_prep_elc_req(struct obd_export *exp,
995 struct ptlrpc_request *req,
996 int version, int opc, int canceloff,
997 struct list_head *cancels, int count);
998 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
999 const struct ldlm_request *dlm_req,
1000 const struct ldlm_callback_suite *cbs);
1001 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1002 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1003 int *flags, void *lvb, __u32 lvb_len,
1004 void *lvb_swabber, struct lustre_handle *lockh,
1006 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1007 const struct ldlm_res_id *res_id,
1008 ldlm_type_t type, ldlm_policy_data_t *policy,
1009 ldlm_mode_t mode, int *flags,
1010 ldlm_blocking_callback blocking,
1011 ldlm_completion_callback completion,
1012 ldlm_glimpse_callback glimpse,
1013 void *data, __u32 lvb_len, void *lvb_swabber,
1014 struct lustre_handle *lockh);
1015 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1016 void *data, __u32 data_len);
1017 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1018 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1019 int ldlm_handle_convert0(struct ptlrpc_request *req,
1020 const struct ldlm_request *dlm_req);
1021 int ldlm_cli_cancel(struct lustre_handle *lockh);
1022 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1023 int flags, void *opaque);
1024 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1025 const struct ldlm_res_id *res_id,
1026 ldlm_policy_data_t *policy,
1027 ldlm_mode_t mode, int flags, void *opaque);
1028 int ldlm_cli_cancel_req(struct obd_export *exp, struct list_head *head,
1029 int count, int flags);
1030 int ldlm_cli_join_lru(struct ldlm_namespace *,
1031 const struct ldlm_res_id *, int join);
1032 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1033 struct list_head *cancels,
1034 ldlm_policy_data_t *policy,
1035 ldlm_mode_t mode, int lock_flags,
1036 int cancel_flags, void *opaque);
1037 int ldlm_cli_cancel_list(struct list_head *head, int count,
1038 struct ptlrpc_request *req, int flags);
1041 /* This has to be here because recursive inclusion sucks. */
1042 int intent_disposition(struct ldlm_reply *rep, int flag);
1043 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1046 /* ioctls for trying requests */
1047 #define IOC_LDLM_TYPE 'f'
1048 #define IOC_LDLM_MIN_NR 40
1050 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1051 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1052 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1053 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1054 #define IOC_LDLM_MAX_NR 43
1057 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1058 * than one lock_res is dead-lock safe.
1060 enum lock_res_type {
1065 static inline void lock_res(struct ldlm_resource *res)
1067 spin_lock(&res->lr_lock);
1070 static inline void lock_res_nested(struct ldlm_resource *res,
1071 enum lock_res_type mode)
1073 spin_lock_nested(&res->lr_lock, mode);
1077 static inline void unlock_res(struct ldlm_resource *res)
1079 spin_unlock(&res->lr_lock);
1082 static inline void check_res_locked(struct ldlm_resource *res)
1084 LASSERT_SPIN_LOCKED(&res->lr_lock);
1087 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1088 void unlock_res_and_lock(struct ldlm_lock *lock);
1091 void ldlm_pools_recalc(ldlm_side_t client);
1092 int ldlm_pools_init(void);
1093 void ldlm_pools_fini(void);
1095 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1096 int idx, ldlm_side_t client);
1097 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1098 unsigned int gfp_mask);
1099 void ldlm_pool_fini(struct ldlm_pool *pl);
1100 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1101 int ldlm_pool_recalc(struct ldlm_pool *pl);
1102 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1103 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1104 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1105 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1106 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1107 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1108 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1109 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1110 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);