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 (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved.
30 * Use is subject to license terms.
32 * Copyright (c) 2010 Whamcloud, Inc
36 * This file is part of Lustre, http://www.lustre.org/
37 * Lustre is a trademark of Sun Microsystems, Inc.
40 #ifndef _LUSTRE_DLM_H__
41 #define _LUSTRE_DLM_H__
43 /** \defgroup ldlm ldlm
48 #if defined(__linux__)
49 #include <linux/lustre_dlm.h>
50 #elif defined(__APPLE__)
51 #include <darwin/lustre_dlm.h>
52 #elif defined(__WINNT__)
53 #include <winnt/lustre_dlm.h>
55 #error Unsupported operating system.
58 #include <lustre_lib.h>
59 #include <lustre_net.h>
60 #include <lustre_import.h>
61 #include <lustre_handles.h>
62 #include <interval_tree.h> /* for interval_node{}, ldlm_extent */
68 #define OBD_LDLM_DEVICENAME "ldlm"
70 #ifdef HAVE_BGL_SUPPORT
71 /* 1.5 times the maximum 128 tasks available in VN mode */
72 #define LDLM_DEFAULT_LRU_SIZE 196
74 #define LDLM_DEFAULT_LRU_SIZE (100 * cfs_num_online_cpus())
76 #define LDLM_DEFAULT_MAX_ALIVE (cfs_time_seconds(36000))
77 #define LDLM_CTIME_AGE_LIMIT (10)
82 ELDLM_LOCK_CHANGED = 300,
83 ELDLM_LOCK_ABORTED = 301,
84 ELDLM_LOCK_REPLACED = 302,
85 ELDLM_NO_LOCK_DATA = 303,
87 ELDLM_NAMESPACE_EXISTS = 400,
88 ELDLM_BAD_NAMESPACE = 401
92 LDLM_NAMESPACE_SERVER = 1 << 0,
93 LDLM_NAMESPACE_CLIENT = 1 << 1
96 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
98 /* If the server returns one of these flags, then the lock was put on that list.
99 * If the client sends one of these flags (during recovery ONLY!), it wants the
100 * lock added to the specified list, no questions asked. -p */
101 #define LDLM_FL_BLOCK_GRANTED 0x000002
102 #define LDLM_FL_BLOCK_CONV 0x000004
103 #define LDLM_FL_BLOCK_WAIT 0x000008
105 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
106 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
107 * queued for sending. */
108 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
109 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
111 /* Lock is being replayed. This could probably be implied by the fact that one
112 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
113 #define LDLM_FL_REPLAY 0x000100
115 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
116 #define LDLM_FL_LOCAL_ONLY 0x000400
118 /* don't run the cancel callback under ldlm_cli_cancel_unused */
119 #define LDLM_FL_FAILED 0x000800
121 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
122 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
123 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
124 /* was LDLM_FL_WARN until 2.0.0 0x008000 */
125 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
127 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
130 /* file & record locking */
131 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
132 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
134 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
135 * the LVB filling happens _after_ the lock has been granted, so another thread
136 * can match`t before the LVB has been updated. As a dirty hack, we set
137 * LDLM_FL_LVB_READY only after we've done the LVB poop.
138 * this is only needed on lov/osc now, where lvb is actually used and callers
139 * must set it in input flags.
141 * The proper fix is to do the granting inside of the completion AST, which can
142 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
143 * pretty high-risk, though, and would need a lot more testing. */
145 #define LDLM_FL_LVB_READY 0x100000
147 /* A lock contributes to the kms calculation until it has finished the part
148 * of it's cancelation that performs write back on its dirty pages. It
149 * can remain on the granted list during this whole time. Threads racing
150 * to update the kms after performing their writeback need to know to
151 * exclude each others locks from the calculation as they walk the granted
153 #define LDLM_FL_KMS_IGNORE 0x200000
155 /* Immediatelly cancel such locks when they block some other locks. Send
156 * cancel notification to original lock holder, but expect no reply. This is
157 * for clients (like liblustre) that cannot be expected to reliably response
158 * to blocking ast. */
159 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
161 /* Flags flags inherited from parent lock when doing intents. */
162 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
164 /* completion ast to be executed */
165 #define LDLM_FL_CP_REQD 0x1000000
167 /* cleanup_resource has already handled the lock */
168 #define LDLM_FL_CLEANED 0x2000000
170 /* optimization hint: LDLM can run blocking callback from current context
171 * w/o involving separate thread. in order to decrease cs rate */
172 #define LDLM_FL_ATOMIC_CB 0x4000000
174 /* was LDLM_FL_ASYNC until 2.0.0 0x8000000 */
176 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
177 * such that server send blocking ast for conflict locks to this client for
178 * the 1st operation, whereas the 2nd operation has canceled this lock and
179 * is waiting for rpc_lock which is taken by the 1st operation.
180 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
181 * ELC code to cancel it.
182 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
183 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
184 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
185 * to send a separate CANCEL rpc. */
186 #define LDLM_FL_BL_AST 0x10000000
187 #define LDLM_FL_BL_DONE 0x20000000
189 /* measure lock contention and return -EUSERS if locking contention is high */
190 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
192 /* These are flags that are mapped into the flags and ASTs of blocking locks */
193 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
195 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
196 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
199 * --------------------------------------------------------------------------
200 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
201 * 0x80000000 will not be sent over the wire.
202 * --------------------------------------------------------------------------
205 /* Used for marking lock as an target for -EINTR while cp_ast sleep
206 * emulation + race with upcoming bl_ast. */
207 #define LDLM_FL_FAIL_LOC 0x100000000ULL
209 /* Used while processing the unused list to know that we have already
210 * handled this lock and decided to skip it */
211 #define LDLM_FL_SKIPPED 0x200000000ULL
213 /* The blocking callback is overloaded to perform two functions. These flags
214 * indicate which operation should be performed. */
215 #define LDLM_CB_BLOCKING 1
216 #define LDLM_CB_CANCELING 2
218 /* compatibility matrix */
219 #define LCK_COMPAT_EX LCK_NL
220 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
221 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
222 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
223 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
224 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
225 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
226 #define LCK_COMPAT_COS (LCK_COS)
228 extern ldlm_mode_t lck_compat_array[];
230 static inline void lockmode_verify(ldlm_mode_t mode)
232 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
235 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
237 return (lck_compat_array[exist_mode] & new_mode);
242 * cluster name spaces
246 #define DLM_OST_NAMESPACE 1
247 #define DLM_MDS_NAMESPACE 2
250 - do we just separate this by security domains and use a prefix for
251 multiple namespaces in the same domain?
261 * waiting_locks_spinlock
276 struct ldlm_resource;
277 struct ldlm_namespace;
279 struct ldlm_pool_ops {
280 int (*po_recalc)(struct ldlm_pool *pl);
281 int (*po_shrink)(struct ldlm_pool *pl, int nr,
282 unsigned int gfp_mask);
283 int (*po_setup)(struct ldlm_pool *pl, int limit);
287 * One second for pools thread check interval. Each pool has own period.
289 #define LDLM_POOLS_THREAD_PERIOD (1)
292 * ~6% margin for modest pools. See ldlm_pool.c for details.
294 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
297 * Default recalc period for server side pools in sec.
299 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
302 * Default recalc period for client side pools in sec.
304 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
308 * Pool proc directory.
310 cfs_proc_dir_entry_t *pl_proc_dir;
312 * Pool name, should be long enough to contain compound proc entry name.
316 * Lock for protecting slv/clv updates.
318 cfs_spinlock_t pl_lock;
320 * Number of allowed locks in in pool, both, client and server side.
322 cfs_atomic_t pl_limit;
324 * Number of granted locks in
326 cfs_atomic_t pl_granted;
330 cfs_atomic_t pl_grant_rate;
334 cfs_atomic_t pl_cancel_rate;
336 * Server lock volume. Protected by pl_lock.
338 __u64 pl_server_lock_volume;
340 * Current biggest client lock volume. Protected by pl_lock.
342 __u64 pl_client_lock_volume;
344 * Lock volume factor. SLV on client is calculated as following:
345 * server_slv * lock_volume_factor.
347 cfs_atomic_t pl_lock_volume_factor;
349 * Time when last slv from server was obtained.
351 time_t pl_recalc_time;
353 * Recalc period for pool.
355 time_t pl_recalc_period;
357 * Recalc and shrink ops.
359 struct ldlm_pool_ops *pl_ops;
361 * Number of planned locks for next period.
367 struct lprocfs_stats *pl_stats;
370 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
371 void *req_cookie, ldlm_mode_t mode, int flags,
374 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
376 struct ldlm_valblock_ops {
377 int (*lvbo_init)(struct ldlm_resource *res);
378 int (*lvbo_update)(struct ldlm_resource *res,
379 struct ptlrpc_request *r,
381 int (*lvbo_free)(struct ldlm_resource *res);
385 LDLM_NAMESPACE_GREEDY = 1 << 0,
386 LDLM_NAMESPACE_MODEST = 1 << 1
390 * Default values for the "max_nolock_size", "contention_time" and
391 * "contended_locks" namespace tunables.
393 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
394 #define NS_DEFAULT_CONTENTION_SECONDS 2
395 #define NS_DEFAULT_CONTENDED_LOCKS 32
397 struct ldlm_ns_bucket {
399 struct ldlm_namespace *nsb_namespace;
400 /** estimated lock callback time */
401 struct adaptive_timeout nsb_at_estimate;
405 /** ldlm namespace lock stats */
412 LDLM_NS_TYPE_UNKNOWN = 0,
427 struct ldlm_namespace {
429 * Backward link to obd, required for ldlm pool to store new SLV.
431 struct obd_device *ns_obd;
434 * Is this a client-side lock tree?
436 ldlm_side_t ns_client;
441 cfs_hash_t *ns_rs_hash;
446 cfs_spinlock_t ns_lock;
449 * big refcount (by bucket)
451 cfs_atomic_t ns_bref;
454 * Namespce connect flags supported by server (may be changed via proc,
455 * lru resize may be disabled/enabled).
457 __u64 ns_connect_flags;
460 * Client side orig connect flags supported by server.
462 __u64 ns_orig_connect_flags;
465 * Position in global namespace list.
467 cfs_list_t ns_list_chain;
470 * All root resources in namespace.
472 cfs_list_t ns_unused_list;
475 unsigned int ns_max_unused;
476 unsigned int ns_max_age;
477 unsigned int ns_timeouts;
481 unsigned int ns_ctime_age_limit;
484 * Next debug dump, jiffies.
486 cfs_time_t ns_next_dump;
488 ldlm_res_policy ns_policy;
489 struct ldlm_valblock_ops *ns_lvbo;
491 cfs_waitq_t ns_waitq;
492 struct ldlm_pool ns_pool;
493 ldlm_appetite_t ns_appetite;
496 * If more than \a ns_contended_locks found, the resource is considered
499 unsigned ns_contended_locks;
502 * The resource remembers contended state during \a ns_contention_time,
505 unsigned ns_contention_time;
508 * Limit size of nolock requests, in bytes.
510 unsigned ns_max_nolock_size;
511 /* callback to cancel locks before replaying it during recovery */
512 ldlm_cancel_for_recovery ns_cancel_for_recovery;
516 struct lprocfs_stats *ns_stats;
519 static inline int ns_is_client(struct ldlm_namespace *ns)
522 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
523 LDLM_NAMESPACE_SERVER)));
524 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
525 ns->ns_client == LDLM_NAMESPACE_SERVER);
526 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
529 static inline int ns_is_server(struct ldlm_namespace *ns)
532 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
533 LDLM_NAMESPACE_SERVER)));
534 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
535 ns->ns_client == LDLM_NAMESPACE_SERVER);
536 return ns->ns_client == LDLM_NAMESPACE_SERVER;
539 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
542 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
545 static inline void ns_register_cancel(struct ldlm_namespace *ns,
546 ldlm_cancel_for_recovery arg)
549 ns->ns_cancel_for_recovery = arg;
554 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
555 struct ldlm_lock_desc *new, void *data,
557 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
559 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
560 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
562 /* Interval node data for each LDLM_EXTENT lock */
563 struct ldlm_interval {
564 struct interval_node li_node; /* node for tree mgmt */
565 cfs_list_t li_group; /* the locks which have the same
566 * policy - group of the policy */
568 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
570 /* the interval tree must be accessed inside the resource lock. */
571 struct ldlm_interval_tree {
572 /* tree size, this variable is used to count
573 * granted PW locks in ldlm_extent_policy()*/
575 ldlm_mode_t lit_mode; /* lock mode */
576 struct interval_node *lit_root; /* actually ldlm_interval */
579 #define LUSTRE_TRACKS_LOCK_EXP_REFS (1)
583 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
584 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
585 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
587 } ldlm_cancel_flags_t;
593 __u64 blocking_owner;
594 void *blocking_export;
599 struct ldlm_extent l_extent;
600 struct ldlm_flock l_flock;
601 struct ldlm_inodebits l_inodebits;
602 } ldlm_policy_data_t;
604 void ldlm_convert_policy_to_wire(ldlm_type_t type,
605 const ldlm_policy_data_t *lpolicy,
606 ldlm_wire_policy_data_t *wpolicy);
607 void ldlm_convert_policy_to_local(ldlm_type_t type,
608 const ldlm_wire_policy_data_t *wpolicy,
609 ldlm_policy_data_t *lpolicy);
613 * Must be first in the structure.
615 struct portals_handle l_handle;
617 * Lock reference count.
621 * Internal spinlock protects l_resource. we should hold this lock
622 * first before grabbing res_lock.
624 cfs_spinlock_t l_lock;
626 * ldlm_lock_change_resource() can change this.
628 struct ldlm_resource *l_resource;
630 * Protected by ns_hash_lock. List item for client side lru list.
634 * Protected by lr_lock, linkage to resource's lock queues.
636 cfs_list_t l_res_link;
638 * Tree node for ldlm_extent.
640 struct ldlm_interval *l_tree_node;
642 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
645 cfs_hlist_node_t l_exp_hash;
647 * Protected by lr_lock. Requested mode.
649 ldlm_mode_t l_req_mode;
651 * Granted mode, also protected by lr_lock.
653 ldlm_mode_t l_granted_mode;
655 * Lock enqueue completion handler.
657 ldlm_completion_callback l_completion_ast;
659 * Lock blocking ast handler.
661 ldlm_blocking_callback l_blocking_ast;
663 * Lock glimpse handler.
665 ldlm_glimpse_callback l_glimpse_ast;
666 ldlm_weigh_callback l_weigh_ast;
671 struct obd_export *l_export;
673 * Lock connection export.
675 struct obd_export *l_conn_export;
678 * Remote lock handle.
680 struct lustre_handle l_remote_handle;
682 ldlm_policy_data_t l_policy_data;
685 * Protected by lr_lock. Various counters: readers, writers, etc.
691 * Set for locks that were removed from class hash table and will be
692 * destroyed when last reference to them is released. Set by
693 * ldlm_lock_destroy_internal().
695 * Protected by lock and resource locks.
699 * flag whether this is a server namespace lock
703 * If the lock is granted, a process sleeps on this waitq to learn when
704 * it's no longer in use. If the lock is not granted, a process sleeps
705 * on this waitq to learn when it becomes granted.
710 * Seconds. it will be updated if there is any activity related to
711 * the lock, e.g. enqueue the lock or send block AST.
713 cfs_time_t l_last_activity;
716 * Jiffies. Should be converted to time if needed.
718 cfs_time_t l_last_used;
720 struct ldlm_extent l_req_extent;
723 * Client-side-only members.
727 * Temporary storage for an LVB received during an enqueue operation.
735 * Server-side-only members.
738 /** connection cookie for the client originated the operation. */
739 __u64 l_client_cookie;
742 * Protected by elt_lock. Callbacks pending.
744 cfs_list_t l_pending_chain;
746 cfs_time_t l_callback_timeout;
749 * Pid which created this lock.
754 * For ldlm_add_ast_work_item().
758 * For ldlm_add_ast_work_item().
762 * For ldlm_add_ast_work_item().
766 struct ldlm_lock *l_blocking_lock;
770 * Protected by lr_lock, linkages to "skip lists".
772 cfs_list_t l_sl_mode;
773 cfs_list_t l_sl_policy;
774 struct lu_ref l_reference;
775 #if LUSTRE_TRACKS_LOCK_EXP_REFS
776 /* Debugging stuff for bug 20498, for tracking export
778 /** number of export references taken */
780 /** link all locks referencing one export */
781 cfs_list_t l_exp_refs_link;
782 /** referenced export object */
783 struct obd_export *l_exp_refs_target;
785 /** export blocking dlm lock list, protected by
786 * l_export->exp_bl_list_lock.
787 * Lock order of waiting_lists_spinlock, exp_bl_list_lock and res lock
788 * is: res lock -> exp_bl_list_lock -> wanting_lists_spinlock. */
789 cfs_list_t l_exp_list;
792 struct ldlm_resource {
793 struct ldlm_ns_bucket *lr_ns_bucket;
795 /* protected by ns_hash_lock */
796 cfs_hlist_node_t lr_hash;
797 cfs_spinlock_t lr_lock;
799 /* protected by lr_lock */
800 cfs_list_t lr_granted;
801 cfs_list_t lr_converting;
802 cfs_list_t lr_waiting;
803 ldlm_mode_t lr_most_restr;
804 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
805 struct ldlm_res_id lr_name;
806 cfs_atomic_t lr_refcount;
808 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
810 /* Server-side-only lock value block elements */
811 /** to serialize lvbo_init */
812 cfs_semaphore_t lr_lvb_sem;
814 /** protect by lr_lock */
817 /* when the resource was considered as contended */
818 cfs_time_t lr_contention_time;
820 * List of references to this resource. For debugging.
822 struct lu_ref lr_reference;
824 struct inode *lr_lvb_inode;
828 ldlm_ns_name(struct ldlm_namespace *ns)
830 return ns->ns_rs_hash->hs_name;
833 static inline struct ldlm_namespace *
834 ldlm_res_to_ns(struct ldlm_resource *res)
836 return res->lr_ns_bucket->nsb_namespace;
839 static inline struct ldlm_namespace *
840 ldlm_lock_to_ns(struct ldlm_lock *lock)
842 return ldlm_res_to_ns(lock->l_resource);
846 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
848 return ldlm_ns_name(ldlm_lock_to_ns(lock));
851 static inline struct adaptive_timeout *
852 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
854 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
857 struct ldlm_ast_work {
858 struct ldlm_lock *w_lock;
860 struct ldlm_lock_desc w_desc;
867 /* ldlm_enqueue parameters common */
868 struct ldlm_enqueue_info {
869 __u32 ei_type; /* Type of the lock being enqueued. */
870 __u32 ei_mode; /* Mode of the lock being enqueued. */
871 void *ei_cb_bl; /* blocking lock callback */
872 void *ei_cb_cp; /* lock completion callback */
873 void *ei_cb_gl; /* lock glimpse callback */
874 void *ei_cb_wg; /* lock weigh callback */
875 void *ei_cbdata; /* Data to be passed into callbacks. */
876 short ei_async:1; /* async request */
879 extern struct obd_ops ldlm_obd_ops;
881 extern char *ldlm_lockname[];
882 extern char *ldlm_typename[];
883 extern char *ldlm_it2str(int it);
885 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
888 if (((level) & D_CANTMASK) != 0 || \
889 ((libcfs_debug & (level)) != 0 && \
890 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
891 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
892 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
894 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
899 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
900 struct libcfs_debug_msg_data *data, const char *fmt,
902 __attribute__ ((format (printf, 4, 5)));
904 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
905 static cfs_debug_limit_state_t _ldlm_cdls; \
906 ldlm_lock_debug(&_ldlm_cdls, mask, lock, \
907 __FILE__, __FUNCTION__, __LINE__, \
911 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
912 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
914 #define LDLM_DEBUG(lock, fmt, a...) do { \
915 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
916 __FILE__, __FUNCTION__, __LINE__, \
919 #else /* !LIBCFS_DEBUG */
920 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
921 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
922 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
926 #define LDLM_DEBUG_NOLOCK(format, a...) \
927 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
929 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
930 int first_enq, ldlm_error_t *err,
931 cfs_list_t *work_list);
937 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
938 #define LDLM_ITER_STOP 2 /* stop iterating */
940 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
941 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
943 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
945 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
948 int ldlm_replay_locks(struct obd_import *imp);
949 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
950 ldlm_iterator_t iter, void *data);
953 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
956 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
960 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
961 void *data, int flag);
962 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
963 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
964 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
965 ldlm_blocking_callback, ldlm_glimpse_callback);
966 int ldlm_handle_convert(struct ptlrpc_request *req);
967 int ldlm_handle_cancel(struct ptlrpc_request *req);
968 int ldlm_request_cancel(struct ptlrpc_request *req,
969 const struct ldlm_request *dlm_req, int first);
970 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
971 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
972 void ldlm_revoke_export_locks(struct obd_export *exp);
973 int ldlm_get_ref(void);
974 void ldlm_put_ref(void);
975 int ldlm_init_export(struct obd_export *exp);
976 void ldlm_destroy_export(struct obd_export *exp);
979 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
980 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
981 void ldlm_lock2handle(const struct ldlm_lock *lock,
982 struct lustre_handle *lockh);
983 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
984 void ldlm_cancel_callback(struct ldlm_lock *);
985 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
987 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
989 return __ldlm_handle2lock(h, 0);
992 #define LDLM_LOCK_REF_DEL(lock) \
993 lu_ref_del(&lock->l_reference, "handle", cfs_current())
995 static inline struct ldlm_lock *
996 ldlm_handle2lock_long(const struct lustre_handle *h, int flags)
998 struct ldlm_lock *lock;
1000 lock = __ldlm_handle2lock(h, flags);
1002 LDLM_LOCK_REF_DEL(lock);
1006 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1007 struct ptlrpc_request *r, int increase)
1009 if (ldlm_res_to_ns(res)->ns_lvbo &&
1010 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1011 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1017 int ldlm_error2errno(ldlm_error_t error);
1018 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
1019 * confuses user-space. */
1020 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1021 void ldlm_dump_export_locks(struct obd_export *exp);
1025 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1026 * __ldlm_handle2lock().
1028 #define LDLM_LOCK_PUT(lock) \
1030 LDLM_LOCK_REF_DEL(lock); \
1031 /*LDLM_DEBUG((lock), "put");*/ \
1032 ldlm_lock_put(lock); \
1036 * Release a lock reference obtained by some other means (see
1039 #define LDLM_LOCK_RELEASE(lock) \
1041 /*LDLM_DEBUG((lock), "put");*/ \
1042 ldlm_lock_put(lock); \
1045 #define LDLM_LOCK_GET(lock) \
1047 ldlm_lock_get(lock); \
1048 /*LDLM_DEBUG((lock), "get");*/ \
1052 #define ldlm_lock_list_put(head, member, count) \
1054 struct ldlm_lock *_lock, *_next; \
1056 cfs_list_for_each_entry_safe(_lock, _next, head, member) { \
1059 cfs_list_del_init(&_lock->member); \
1060 LDLM_LOCK_RELEASE(_lock); \
1065 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1066 void ldlm_lock_put(struct ldlm_lock *lock);
1067 void ldlm_lock_destroy(struct ldlm_lock *lock);
1068 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1069 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1070 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1071 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1072 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1073 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1074 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1075 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
1076 const struct ldlm_res_id *, ldlm_type_t type,
1077 ldlm_policy_data_t *, ldlm_mode_t mode,
1078 struct lustre_handle *, int unref);
1079 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1081 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
1082 void ldlm_lock_cancel(struct ldlm_lock *lock);
1083 void ldlm_reprocess_all(struct ldlm_resource *res);
1084 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1085 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
1086 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1087 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1090 struct ldlm_namespace *
1091 ldlm_namespace_new(struct obd_device *obd, char *name,
1092 ldlm_side_t client, ldlm_appetite_t apt,
1093 ldlm_ns_type_t ns_type);
1094 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
1095 void ldlm_namespace_free(struct ldlm_namespace *ns,
1096 struct obd_import *imp, int force);
1097 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1098 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1099 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1100 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1101 void ldlm_namespace_get(struct ldlm_namespace *ns);
1102 void ldlm_namespace_put(struct ldlm_namespace *ns);
1103 int ldlm_proc_setup(void);
1105 void ldlm_proc_cleanup(void);
1107 static inline void ldlm_proc_cleanup(void) {}
1110 /* resource.c - internal */
1111 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1112 struct ldlm_resource *parent,
1113 const struct ldlm_res_id *,
1114 ldlm_type_t type, int create);
1115 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1116 int ldlm_resource_putref(struct ldlm_resource *res);
1117 void ldlm_resource_add_lock(struct ldlm_resource *res,
1119 struct ldlm_lock *lock);
1120 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1121 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1122 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1123 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1124 void ldlm_resource_dump(int level, struct ldlm_resource *);
1125 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1126 const struct ldlm_res_id *);
1128 #define LDLM_RESOURCE_ADDREF(res) do { \
1129 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1132 #define LDLM_RESOURCE_DELREF(res) do { \
1133 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1136 struct ldlm_callback_suite {
1137 ldlm_completion_callback lcs_completion;
1138 ldlm_blocking_callback lcs_blocking;
1139 ldlm_glimpse_callback lcs_glimpse;
1140 ldlm_weigh_callback lcs_weigh;
1143 /* ldlm_request.c */
1144 int ldlm_expired_completion_wait(void *data);
1145 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1146 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1147 void *data, int flag);
1148 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1149 int ldlm_completion_ast_async(struct ldlm_lock *lock, int flags, void *data);
1150 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
1151 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1152 struct ldlm_enqueue_info *einfo,
1153 const struct ldlm_res_id *res_id,
1154 ldlm_policy_data_t const *policy, int *flags,
1155 void *lvb, __u32 lvb_len, struct lustre_handle *lockh,
1157 int ldlm_prep_enqueue_req(struct obd_export *exp,
1158 struct ptlrpc_request *req,
1159 cfs_list_t *cancels,
1161 int ldlm_prep_elc_req(struct obd_export *exp,
1162 struct ptlrpc_request *req,
1163 int version, int opc, int canceloff,
1164 cfs_list_t *cancels, int count);
1165 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1166 const struct ldlm_request *dlm_req,
1167 const struct ldlm_callback_suite *cbs);
1168 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1169 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1170 int *flags, void *lvb, __u32 lvb_len,
1171 struct lustre_handle *lockh, int rc);
1172 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1173 const struct ldlm_res_id *res_id,
1174 ldlm_type_t type, ldlm_policy_data_t *policy,
1175 ldlm_mode_t mode, int *flags,
1176 ldlm_blocking_callback blocking,
1177 ldlm_completion_callback completion,
1178 ldlm_glimpse_callback glimpse,
1179 void *data, __u32 lvb_len,
1180 const __u64 *client_cookie,
1181 struct lustre_handle *lockh);
1182 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1183 void *data, __u32 data_len);
1184 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1185 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1186 int ldlm_handle_convert0(struct ptlrpc_request *req,
1187 const struct ldlm_request *dlm_req);
1188 int ldlm_cli_cancel(struct lustre_handle *lockh);
1189 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1190 ldlm_cancel_flags_t flags, void *opaque);
1191 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1192 const struct ldlm_res_id *res_id,
1193 ldlm_policy_data_t *policy,
1195 ldlm_cancel_flags_t flags,
1197 int ldlm_cli_cancel_req(struct obd_export *exp, cfs_list_t *head,
1198 int count, ldlm_cancel_flags_t flags);
1199 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1200 cfs_list_t *cancels,
1201 ldlm_policy_data_t *policy,
1202 ldlm_mode_t mode, int lock_flags,
1203 ldlm_cancel_flags_t cancel_flags, void *opaque);
1204 int ldlm_cli_cancel_list_local(cfs_list_t *cancels, int count,
1205 ldlm_cancel_flags_t flags);
1206 int ldlm_cli_cancel_list(cfs_list_t *head, int count,
1207 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1210 /* This has to be here because recursive inclusion sucks. */
1211 int intent_disposition(struct ldlm_reply *rep, int flag);
1212 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1215 /* ioctls for trying requests */
1216 #define IOC_LDLM_TYPE 'f'
1217 #define IOC_LDLM_MIN_NR 40
1219 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1220 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1221 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1222 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1223 #define IOC_LDLM_MAX_NR 43
1226 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1227 * than one lock_res is dead-lock safe.
1229 enum lock_res_type {
1234 static inline void lock_res(struct ldlm_resource *res)
1236 cfs_spin_lock(&res->lr_lock);
1239 static inline void lock_res_nested(struct ldlm_resource *res,
1240 enum lock_res_type mode)
1242 cfs_spin_lock_nested(&res->lr_lock, mode);
1246 static inline void unlock_res(struct ldlm_resource *res)
1248 cfs_spin_unlock(&res->lr_lock);
1251 static inline void check_res_locked(struct ldlm_resource *res)
1253 LASSERT_SPIN_LOCKED(&res->lr_lock);
1256 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1257 void unlock_res_and_lock(struct ldlm_lock *lock);
1260 void ldlm_pools_recalc(ldlm_side_t client);
1261 int ldlm_pools_init(void);
1262 void ldlm_pools_fini(void);
1264 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1265 int idx, ldlm_side_t client);
1266 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1267 unsigned int gfp_mask);
1268 void ldlm_pool_fini(struct ldlm_pool *pl);
1269 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1270 int ldlm_pool_recalc(struct ldlm_pool *pl);
1271 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1272 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1273 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1274 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1275 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1276 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1277 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1278 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1279 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);