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)
78 #define LDLM_DEFAULT_PARALLEL_AST_LIMIT 1024
83 ELDLM_LOCK_CHANGED = 300,
84 ELDLM_LOCK_ABORTED = 301,
85 ELDLM_LOCK_REPLACED = 302,
86 ELDLM_NO_LOCK_DATA = 303,
88 ELDLM_NAMESPACE_EXISTS = 400,
89 ELDLM_BAD_NAMESPACE = 401
93 LDLM_NAMESPACE_SERVER = 1 << 0,
94 LDLM_NAMESPACE_CLIENT = 1 << 1
97 #define LDLM_FL_LOCK_CHANGED 0x000001 /* extent, mode, or resource changed */
99 /* If the server returns one of these flags, then the lock was put on that list.
100 * If the client sends one of these flags (during recovery ONLY!), it wants the
101 * lock added to the specified list, no questions asked. -p */
102 #define LDLM_FL_BLOCK_GRANTED 0x000002
103 #define LDLM_FL_BLOCK_CONV 0x000004
104 #define LDLM_FL_BLOCK_WAIT 0x000008
106 #define LDLM_FL_CBPENDING 0x000010 /* this lock is being destroyed */
107 #define LDLM_FL_AST_SENT 0x000020 /* blocking or cancel packet was
108 * queued for sending. */
109 #define LDLM_FL_WAIT_NOREPROC 0x000040 /* not a real flag, not saved in lock */
110 #define LDLM_FL_CANCEL 0x000080 /* cancellation callback already run */
112 /* Lock is being replayed. This could probably be implied by the fact that one
113 * of BLOCK_{GRANTED,CONV,WAIT} is set, but that is pretty dangerous. */
114 #define LDLM_FL_REPLAY 0x000100
116 #define LDLM_FL_INTENT_ONLY 0x000200 /* don't grant lock, just do intent */
117 #define LDLM_FL_LOCAL_ONLY 0x000400
119 /* don't run the cancel callback under ldlm_cli_cancel_unused */
120 #define LDLM_FL_FAILED 0x000800
122 #define LDLM_FL_HAS_INTENT 0x001000 /* lock request has intent */
123 #define LDLM_FL_CANCELING 0x002000 /* lock cancel has already been sent */
124 #define LDLM_FL_LOCAL 0x004000 /* local lock (ie, no srv/cli split) */
125 /* was LDLM_FL_WARN until 2.0.0 0x008000 */
126 #define LDLM_FL_DISCARD_DATA 0x010000 /* discard (no writeback) on cancel */
128 #define LDLM_FL_NO_TIMEOUT 0x020000 /* Blocked by group lock - wait
131 /* file & record locking */
132 #define LDLM_FL_BLOCK_NOWAIT 0x040000 // server told not to wait if blocked
133 #define LDLM_FL_TEST_LOCK 0x080000 // return blocking lock
135 /* XXX FIXME: This is being added to b_size as a low-risk fix to the fact that
136 * the LVB filling happens _after_ the lock has been granted, so another thread
137 * can match`t before the LVB has been updated. As a dirty hack, we set
138 * LDLM_FL_LVB_READY only after we've done the LVB poop.
139 * this is only needed on lov/osc now, where lvb is actually used and callers
140 * must set it in input flags.
142 * The proper fix is to do the granting inside of the completion AST, which can
143 * be replaced with a LVB-aware wrapping function for OSC locks. That change is
144 * pretty high-risk, though, and would need a lot more testing. */
146 #define LDLM_FL_LVB_READY 0x100000
148 /* A lock contributes to the kms calculation until it has finished the part
149 * of it's cancelation that performs write back on its dirty pages. It
150 * can remain on the granted list during this whole time. Threads racing
151 * to update the kms after performing their writeback need to know to
152 * exclude each others locks from the calculation as they walk the granted
154 #define LDLM_FL_KMS_IGNORE 0x200000
156 /* Don't put lock into the LRU list, so that it is not canceled due to aging.
157 * Used by MGC locks, they are cancelled only at unmount or by callback. */
158 #define LDLM_FL_NO_LRU 0x400000
160 /* Immediatelly cancel such locks when they block some other locks. Send
161 * cancel notification to original lock holder, but expect no reply. This is
162 * for clients (like liblustre) that cannot be expected to reliably response
163 * to blocking ast. */
164 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
166 /* Flags flags inherited from parent lock when doing intents. */
167 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
169 /* completion ast to be executed */
170 #define LDLM_FL_CP_REQD 0x1000000
172 /* cleanup_resource has already handled the lock */
173 #define LDLM_FL_CLEANED 0x2000000
175 /* optimization hint: LDLM can run blocking callback from current context
176 * w/o involving separate thread. in order to decrease cs rate */
177 #define LDLM_FL_ATOMIC_CB 0x4000000
179 /* was LDLM_FL_ASYNC until 2.0.0 0x8000000 */
181 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
182 * such that server send blocking ast for conflict locks to this client for
183 * the 1st operation, whereas the 2nd operation has canceled this lock and
184 * is waiting for rpc_lock which is taken by the 1st operation.
185 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
186 * ELC code to cancel it.
187 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
188 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
189 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
190 * to send a separate CANCEL rpc. */
191 #define LDLM_FL_BL_AST 0x10000000
192 #define LDLM_FL_BL_DONE 0x20000000
194 /* measure lock contention and return -EUSERS if locking contention is high */
195 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
197 /* These are flags that are mapped into the flags and ASTs of blocking locks */
198 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
200 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
201 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
204 * --------------------------------------------------------------------------
205 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
206 * 0x80000000 will not be sent over the wire.
207 * --------------------------------------------------------------------------
210 /* Used for marking lock as an target for -EINTR while cp_ast sleep
211 * emulation + race with upcoming bl_ast. */
212 #define LDLM_FL_FAIL_LOC 0x100000000ULL
214 /* Used while processing the unused list to know that we have already
215 * handled this lock and decided to skip it */
216 #define LDLM_FL_SKIPPED 0x200000000ULL
218 /* The blocking callback is overloaded to perform two functions. These flags
219 * indicate which operation should be performed. */
220 #define LDLM_CB_BLOCKING 1
221 #define LDLM_CB_CANCELING 2
223 /* compatibility matrix */
224 #define LCK_COMPAT_EX LCK_NL
225 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
226 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
227 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
228 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
229 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
230 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
231 #define LCK_COMPAT_COS (LCK_COS)
233 extern ldlm_mode_t lck_compat_array[];
235 static inline void lockmode_verify(ldlm_mode_t mode)
237 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
240 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
242 return (lck_compat_array[exist_mode] & new_mode);
247 * cluster name spaces
251 #define DLM_OST_NAMESPACE 1
252 #define DLM_MDS_NAMESPACE 2
255 - do we just separate this by security domains and use a prefix for
256 multiple namespaces in the same domain?
266 * waiting_locks_spinlock
281 struct ldlm_resource;
282 struct ldlm_namespace;
284 struct ldlm_pool_ops {
285 int (*po_recalc)(struct ldlm_pool *pl);
286 int (*po_shrink)(struct ldlm_pool *pl, int nr,
287 unsigned int gfp_mask);
288 int (*po_setup)(struct ldlm_pool *pl, int limit);
292 * One second for pools thread check interval. Each pool has own period.
294 #define LDLM_POOLS_THREAD_PERIOD (1)
297 * ~6% margin for modest pools. See ldlm_pool.c for details.
299 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
302 * Default recalc period for server side pools in sec.
304 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
307 * Default recalc period for client side pools in sec.
309 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
313 * Pool proc directory.
315 cfs_proc_dir_entry_t *pl_proc_dir;
317 * Pool name, should be long enough to contain compound proc entry name.
321 * Lock for protecting slv/clv updates.
323 cfs_spinlock_t pl_lock;
325 * Number of allowed locks in in pool, both, client and server side.
327 cfs_atomic_t pl_limit;
329 * Number of granted locks in
331 cfs_atomic_t pl_granted;
335 cfs_atomic_t pl_grant_rate;
339 cfs_atomic_t pl_cancel_rate;
341 * Server lock volume. Protected by pl_lock.
343 __u64 pl_server_lock_volume;
345 * Current biggest client lock volume. Protected by pl_lock.
347 __u64 pl_client_lock_volume;
349 * Lock volume factor. SLV on client is calculated as following:
350 * server_slv * lock_volume_factor.
352 cfs_atomic_t pl_lock_volume_factor;
354 * Time when last slv from server was obtained.
356 time_t pl_recalc_time;
358 * Recalc period for pool.
360 time_t pl_recalc_period;
362 * Recalc and shrink ops.
364 struct ldlm_pool_ops *pl_ops;
366 * Number of planned locks for next period.
372 struct lprocfs_stats *pl_stats;
375 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
376 void *req_cookie, ldlm_mode_t mode, int flags,
379 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
381 struct ldlm_valblock_ops {
382 int (*lvbo_init)(struct ldlm_resource *res);
383 int (*lvbo_update)(struct ldlm_resource *res,
384 struct ptlrpc_request *r,
386 int (*lvbo_free)(struct ldlm_resource *res);
390 LDLM_NAMESPACE_GREEDY = 1 << 0,
391 LDLM_NAMESPACE_MODEST = 1 << 1
395 * Default values for the "max_nolock_size", "contention_time" and
396 * "contended_locks" namespace tunables.
398 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
399 #define NS_DEFAULT_CONTENTION_SECONDS 2
400 #define NS_DEFAULT_CONTENDED_LOCKS 32
402 struct ldlm_ns_bucket {
404 struct ldlm_namespace *nsb_namespace;
405 /** estimated lock callback time */
406 struct adaptive_timeout nsb_at_estimate;
410 /** ldlm namespace lock stats */
417 LDLM_NS_TYPE_UNKNOWN = 0,
432 struct ldlm_namespace {
434 * Backward link to obd, required for ldlm pool to store new SLV.
436 struct obd_device *ns_obd;
439 * Is this a client-side lock tree?
441 ldlm_side_t ns_client;
446 cfs_hash_t *ns_rs_hash;
451 cfs_spinlock_t ns_lock;
454 * big refcount (by bucket)
456 cfs_atomic_t ns_bref;
459 * Namespce connect flags supported by server (may be changed via proc,
460 * lru resize may be disabled/enabled).
462 __u64 ns_connect_flags;
465 * Client side orig connect flags supported by server.
467 __u64 ns_orig_connect_flags;
470 * Position in global namespace list.
472 cfs_list_t ns_list_chain;
475 * All root resources in namespace.
477 cfs_list_t ns_unused_list;
480 unsigned int ns_max_unused;
481 unsigned int ns_max_age;
482 unsigned int ns_timeouts;
486 unsigned int ns_ctime_age_limit;
489 * Next debug dump, jiffies.
491 cfs_time_t ns_next_dump;
493 ldlm_res_policy ns_policy;
494 struct ldlm_valblock_ops *ns_lvbo;
496 cfs_waitq_t ns_waitq;
497 struct ldlm_pool ns_pool;
498 ldlm_appetite_t ns_appetite;
501 * If more than \a ns_contended_locks found, the resource is considered
504 unsigned ns_contended_locks;
507 * The resource remembers contended state during \a ns_contention_time,
510 unsigned ns_contention_time;
513 * Limit size of nolock requests, in bytes.
515 unsigned ns_max_nolock_size;
518 * Limit of parallel AST RPC count.
520 unsigned ns_max_parallel_ast;
522 /* callback to cancel locks before replaying it during recovery */
523 ldlm_cancel_for_recovery ns_cancel_for_recovery;
527 struct lprocfs_stats *ns_stats;
529 unsigned ns_stopping:1; /* namespace cleanup */
532 static inline int ns_is_client(struct ldlm_namespace *ns)
535 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
536 LDLM_NAMESPACE_SERVER)));
537 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
538 ns->ns_client == LDLM_NAMESPACE_SERVER);
539 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
542 static inline int ns_is_server(struct ldlm_namespace *ns)
545 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
546 LDLM_NAMESPACE_SERVER)));
547 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
548 ns->ns_client == LDLM_NAMESPACE_SERVER);
549 return ns->ns_client == LDLM_NAMESPACE_SERVER;
552 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
555 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
558 static inline void ns_register_cancel(struct ldlm_namespace *ns,
559 ldlm_cancel_for_recovery arg)
562 ns->ns_cancel_for_recovery = arg;
567 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
568 struct ldlm_lock_desc *new, void *data,
570 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
572 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
573 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
575 /* Interval node data for each LDLM_EXTENT lock */
576 struct ldlm_interval {
577 struct interval_node li_node; /* node for tree mgmt */
578 cfs_list_t li_group; /* the locks which have the same
579 * policy - group of the policy */
581 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
583 /* the interval tree must be accessed inside the resource lock. */
584 struct ldlm_interval_tree {
585 /* tree size, this variable is used to count
586 * granted PW locks in ldlm_extent_policy()*/
588 ldlm_mode_t lit_mode; /* lock mode */
589 struct interval_node *lit_root; /* actually ldlm_interval */
592 #define LUSTRE_TRACKS_LOCK_EXP_REFS (1)
596 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
597 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
598 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
600 } ldlm_cancel_flags_t;
606 __u64 blocking_owner;
607 void *blocking_export;
612 struct ldlm_extent l_extent;
613 struct ldlm_flock l_flock;
614 struct ldlm_inodebits l_inodebits;
615 } ldlm_policy_data_t;
617 void ldlm_convert_policy_to_wire(ldlm_type_t type,
618 const ldlm_policy_data_t *lpolicy,
619 ldlm_wire_policy_data_t *wpolicy);
620 void ldlm_convert_policy_to_local(ldlm_type_t type,
621 const ldlm_wire_policy_data_t *wpolicy,
622 ldlm_policy_data_t *lpolicy);
626 * Must be first in the structure.
628 struct portals_handle l_handle;
630 * Lock reference count.
634 * Internal spinlock protects l_resource. we should hold this lock
635 * first before grabbing res_lock.
637 cfs_spinlock_t l_lock;
639 * ldlm_lock_change_resource() can change this.
641 struct ldlm_resource *l_resource;
643 * Protected by ns_hash_lock. List item for client side lru list.
647 * Protected by lr_lock, linkage to resource's lock queues.
649 cfs_list_t l_res_link;
651 * Tree node for ldlm_extent.
653 struct ldlm_interval *l_tree_node;
655 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
658 cfs_hlist_node_t l_exp_hash;
660 * Protected by lr_lock. Requested mode.
662 ldlm_mode_t l_req_mode;
664 * Granted mode, also protected by lr_lock.
666 ldlm_mode_t l_granted_mode;
668 * Lock enqueue completion handler.
670 ldlm_completion_callback l_completion_ast;
672 * Lock blocking ast handler.
674 ldlm_blocking_callback l_blocking_ast;
676 * Lock glimpse handler.
678 ldlm_glimpse_callback l_glimpse_ast;
679 ldlm_weigh_callback l_weigh_ast;
684 struct obd_export *l_export;
686 * Lock connection export.
688 struct obd_export *l_conn_export;
691 * Remote lock handle.
693 struct lustre_handle l_remote_handle;
695 ldlm_policy_data_t l_policy_data;
698 * Protected by lr_lock. Various counters: readers, writers, etc.
704 * Set for locks that were removed from class hash table and will be
705 * destroyed when last reference to them is released. Set by
706 * ldlm_lock_destroy_internal().
708 * Protected by lock and resource locks.
712 * flag whether this is a server namespace lock
716 * If the lock is granted, a process sleeps on this waitq to learn when
717 * it's no longer in use. If the lock is not granted, a process sleeps
718 * on this waitq to learn when it becomes granted.
723 * Seconds. it will be updated if there is any activity related to
724 * the lock, e.g. enqueue the lock or send block AST.
726 cfs_time_t l_last_activity;
729 * Jiffies. Should be converted to time if needed.
731 cfs_time_t l_last_used;
733 struct ldlm_extent l_req_extent;
736 * Client-side-only members.
740 * Temporary storage for an LVB received during an enqueue operation.
748 * Server-side-only members.
751 /** connection cookie for the client originated the operation. */
752 __u64 l_client_cookie;
755 * Protected by elt_lock. Callbacks pending.
757 cfs_list_t l_pending_chain;
759 cfs_time_t l_callback_timeout;
762 * Pid which created this lock.
767 * For ldlm_add_ast_work_item().
771 * For ldlm_add_ast_work_item().
775 * For ldlm_add_ast_work_item().
779 struct ldlm_lock *l_blocking_lock;
783 * Protected by lr_lock, linkages to "skip lists".
785 cfs_list_t l_sl_mode;
786 cfs_list_t l_sl_policy;
787 struct lu_ref l_reference;
788 #if LUSTRE_TRACKS_LOCK_EXP_REFS
789 /* Debugging stuff for bug 20498, for tracking export
791 /** number of export references taken */
793 /** link all locks referencing one export */
794 cfs_list_t l_exp_refs_link;
795 /** referenced export object */
796 struct obd_export *l_exp_refs_target;
800 struct ldlm_resource {
801 struct ldlm_ns_bucket *lr_ns_bucket;
803 /* protected by ns_hash_lock */
804 cfs_hlist_node_t lr_hash;
805 cfs_spinlock_t lr_lock;
807 /* protected by lr_lock */
808 cfs_list_t lr_granted;
809 cfs_list_t lr_converting;
810 cfs_list_t lr_waiting;
811 ldlm_mode_t lr_most_restr;
812 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
813 struct ldlm_res_id lr_name;
814 cfs_atomic_t lr_refcount;
816 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
818 /* Server-side-only lock value block elements */
819 /** to serialize lvbo_init */
820 cfs_semaphore_t lr_lvb_sem;
822 /** protect by lr_lock */
825 /* when the resource was considered as contended */
826 cfs_time_t lr_contention_time;
828 * List of references to this resource. For debugging.
830 struct lu_ref lr_reference;
832 struct inode *lr_lvb_inode;
836 ldlm_ns_name(struct ldlm_namespace *ns)
838 return ns->ns_rs_hash->hs_name;
841 static inline struct ldlm_namespace *
842 ldlm_res_to_ns(struct ldlm_resource *res)
844 return res->lr_ns_bucket->nsb_namespace;
847 static inline struct ldlm_namespace *
848 ldlm_lock_to_ns(struct ldlm_lock *lock)
850 return ldlm_res_to_ns(lock->l_resource);
854 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
856 return ldlm_ns_name(ldlm_lock_to_ns(lock));
859 static inline struct adaptive_timeout *
860 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
862 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
865 struct ldlm_ast_work {
866 struct ldlm_lock *w_lock;
868 struct ldlm_lock_desc w_desc;
875 /* ldlm_enqueue parameters common */
876 struct ldlm_enqueue_info {
877 __u32 ei_type; /* Type of the lock being enqueued. */
878 __u32 ei_mode; /* Mode of the lock being enqueued. */
879 void *ei_cb_bl; /* blocking lock callback */
880 void *ei_cb_cp; /* lock completion callback */
881 void *ei_cb_gl; /* lock glimpse callback */
882 void *ei_cb_wg; /* lock weigh callback */
883 void *ei_cbdata; /* Data to be passed into callbacks. */
886 extern struct obd_ops ldlm_obd_ops;
888 extern char *ldlm_lockname[];
889 extern char *ldlm_typename[];
890 extern char *ldlm_it2str(int it);
892 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
895 if (((level) & D_CANTMASK) != 0 || \
896 ((libcfs_debug & (level)) != 0 && \
897 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
898 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
899 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
901 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
906 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
907 struct libcfs_debug_msg_data *data, const char *fmt,
909 __attribute__ ((format (printf, 4, 5)));
911 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
912 static cfs_debug_limit_state_t _ldlm_cdls; \
913 ldlm_lock_debug(&_ldlm_cdls, mask, lock, \
914 __FILE__, __FUNCTION__, __LINE__, \
918 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
919 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
921 #define LDLM_DEBUG(lock, fmt, a...) do { \
922 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
923 __FILE__, __FUNCTION__, __LINE__, \
926 #else /* !LIBCFS_DEBUG */
927 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
928 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
929 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
933 #define LDLM_DEBUG_NOLOCK(format, a...) \
934 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
936 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
937 int first_enq, ldlm_error_t *err,
938 cfs_list_t *work_list);
944 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
945 #define LDLM_ITER_STOP 2 /* stop iterating */
947 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
948 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
950 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
952 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
955 int ldlm_replay_locks(struct obd_import *imp);
956 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
957 ldlm_iterator_t iter, void *data);
960 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
963 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
967 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
968 void *data, int flag);
969 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
970 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
971 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
972 ldlm_blocking_callback, ldlm_glimpse_callback);
973 int ldlm_handle_convert(struct ptlrpc_request *req);
974 int ldlm_handle_cancel(struct ptlrpc_request *req);
975 int ldlm_request_cancel(struct ptlrpc_request *req,
976 const struct ldlm_request *dlm_req, int first);
977 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
978 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
979 void ldlm_revoke_export_locks(struct obd_export *exp);
980 int ldlm_get_ref(void);
981 void ldlm_put_ref(void);
982 int ldlm_init_export(struct obd_export *exp);
983 void ldlm_destroy_export(struct obd_export *exp);
986 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
987 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
988 void ldlm_lock2handle(const struct ldlm_lock *lock,
989 struct lustre_handle *lockh);
990 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
991 void ldlm_cancel_callback(struct ldlm_lock *);
992 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
993 int ldlm_lock_set_data(struct lustre_handle *, void *);
995 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
997 return __ldlm_handle2lock(h, 0);
1000 #define LDLM_LOCK_REF_DEL(lock) \
1001 lu_ref_del(&lock->l_reference, "handle", cfs_current())
1003 static inline struct ldlm_lock *
1004 ldlm_handle2lock_long(const struct lustre_handle *h, int flags)
1006 struct ldlm_lock *lock;
1008 lock = __ldlm_handle2lock(h, flags);
1010 LDLM_LOCK_REF_DEL(lock);
1014 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1015 struct ptlrpc_request *r, int increase)
1017 if (ldlm_res_to_ns(res)->ns_lvbo &&
1018 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1019 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1025 int ldlm_error2errno(ldlm_error_t error);
1026 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
1027 * confuses user-space. */
1028 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1029 void ldlm_dump_export_locks(struct obd_export *exp);
1033 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1034 * __ldlm_handle2lock().
1036 #define LDLM_LOCK_PUT(lock) \
1038 LDLM_LOCK_REF_DEL(lock); \
1039 /*LDLM_DEBUG((lock), "put");*/ \
1040 ldlm_lock_put(lock); \
1044 * Release a lock reference obtained by some other means (see
1047 #define LDLM_LOCK_RELEASE(lock) \
1049 /*LDLM_DEBUG((lock), "put");*/ \
1050 ldlm_lock_put(lock); \
1053 #define LDLM_LOCK_GET(lock) \
1055 ldlm_lock_get(lock); \
1056 /*LDLM_DEBUG((lock), "get");*/ \
1060 #define ldlm_lock_list_put(head, member, count) \
1062 struct ldlm_lock *_lock, *_next; \
1064 cfs_list_for_each_entry_safe(_lock, _next, head, member) { \
1067 cfs_list_del_init(&_lock->member); \
1068 LDLM_LOCK_RELEASE(_lock); \
1073 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1074 void ldlm_lock_put(struct ldlm_lock *lock);
1075 void ldlm_lock_destroy(struct ldlm_lock *lock);
1076 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1077 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1078 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1079 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1080 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1081 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1082 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1083 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
1084 const struct ldlm_res_id *, ldlm_type_t type,
1085 ldlm_policy_data_t *, ldlm_mode_t mode,
1086 struct lustre_handle *, int unref);
1087 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1089 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1091 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
1092 void ldlm_lock_cancel(struct ldlm_lock *lock);
1093 void ldlm_reprocess_all(struct ldlm_resource *res);
1094 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1095 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
1096 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1097 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1100 struct ldlm_namespace *
1101 ldlm_namespace_new(struct obd_device *obd, char *name,
1102 ldlm_side_t client, ldlm_appetite_t apt,
1103 ldlm_ns_type_t ns_type);
1104 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
1105 void ldlm_namespace_free(struct ldlm_namespace *ns,
1106 struct obd_import *imp, int force);
1107 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1108 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1109 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1110 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1111 void ldlm_namespace_get(struct ldlm_namespace *ns);
1112 void ldlm_namespace_put(struct ldlm_namespace *ns);
1113 int ldlm_proc_setup(void);
1115 void ldlm_proc_cleanup(void);
1117 static inline void ldlm_proc_cleanup(void) {}
1120 /* resource.c - internal */
1121 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1122 struct ldlm_resource *parent,
1123 const struct ldlm_res_id *,
1124 ldlm_type_t type, int create);
1125 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1126 int ldlm_resource_putref(struct ldlm_resource *res);
1127 void ldlm_resource_add_lock(struct ldlm_resource *res,
1129 struct ldlm_lock *lock);
1130 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1131 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1132 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1133 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1134 void ldlm_resource_dump(int level, struct ldlm_resource *);
1135 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1136 const struct ldlm_res_id *);
1138 #define LDLM_RESOURCE_ADDREF(res) do { \
1139 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1142 #define LDLM_RESOURCE_DELREF(res) do { \
1143 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1146 struct ldlm_callback_suite {
1147 ldlm_completion_callback lcs_completion;
1148 ldlm_blocking_callback lcs_blocking;
1149 ldlm_glimpse_callback lcs_glimpse;
1150 ldlm_weigh_callback lcs_weigh;
1153 /* ldlm_request.c */
1154 int ldlm_expired_completion_wait(void *data);
1155 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1156 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1157 void *data, int flag);
1158 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1159 int ldlm_completion_ast_async(struct ldlm_lock *lock, int flags, void *data);
1160 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
1161 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1162 struct ldlm_enqueue_info *einfo,
1163 const struct ldlm_res_id *res_id,
1164 ldlm_policy_data_t const *policy, int *flags,
1165 void *lvb, __u32 lvb_len, struct lustre_handle *lockh,
1167 int ldlm_prep_enqueue_req(struct obd_export *exp,
1168 struct ptlrpc_request *req,
1169 cfs_list_t *cancels,
1171 int ldlm_prep_elc_req(struct obd_export *exp,
1172 struct ptlrpc_request *req,
1173 int version, int opc, int canceloff,
1174 cfs_list_t *cancels, int count);
1175 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1176 const struct ldlm_request *dlm_req,
1177 const struct ldlm_callback_suite *cbs);
1178 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1179 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1180 int *flags, void *lvb, __u32 lvb_len,
1181 struct lustre_handle *lockh, int rc);
1182 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1183 const struct ldlm_res_id *res_id,
1184 ldlm_type_t type, ldlm_policy_data_t *policy,
1185 ldlm_mode_t mode, int *flags,
1186 ldlm_blocking_callback blocking,
1187 ldlm_completion_callback completion,
1188 ldlm_glimpse_callback glimpse,
1189 void *data, __u32 lvb_len,
1190 const __u64 *client_cookie,
1191 struct lustre_handle *lockh);
1192 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1193 void *data, __u32 data_len);
1194 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1195 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1196 int ldlm_handle_convert0(struct ptlrpc_request *req,
1197 const struct ldlm_request *dlm_req);
1198 int ldlm_cli_cancel(struct lustre_handle *lockh);
1199 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1200 ldlm_cancel_flags_t flags, void *opaque);
1201 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1202 const struct ldlm_res_id *res_id,
1203 ldlm_policy_data_t *policy,
1205 ldlm_cancel_flags_t flags,
1207 int ldlm_cli_cancel_req(struct obd_export *exp, cfs_list_t *head,
1208 int count, ldlm_cancel_flags_t flags);
1209 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1210 cfs_list_t *cancels,
1211 ldlm_policy_data_t *policy,
1212 ldlm_mode_t mode, int lock_flags,
1213 ldlm_cancel_flags_t cancel_flags, void *opaque);
1214 int ldlm_cli_cancel_list_local(cfs_list_t *cancels, int count,
1215 ldlm_cancel_flags_t flags);
1216 int ldlm_cli_cancel_list(cfs_list_t *head, int count,
1217 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1220 /* This has to be here because recursive inclusion sucks. */
1221 int intent_disposition(struct ldlm_reply *rep, int flag);
1222 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1225 /* ioctls for trying requests */
1226 #define IOC_LDLM_TYPE 'f'
1227 #define IOC_LDLM_MIN_NR 40
1229 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1230 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1231 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1232 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1233 #define IOC_LDLM_MAX_NR 43
1236 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1237 * than one lock_res is dead-lock safe.
1239 enum lock_res_type {
1244 static inline void lock_res(struct ldlm_resource *res)
1246 cfs_spin_lock(&res->lr_lock);
1249 static inline void lock_res_nested(struct ldlm_resource *res,
1250 enum lock_res_type mode)
1252 cfs_spin_lock_nested(&res->lr_lock, mode);
1256 static inline void unlock_res(struct ldlm_resource *res)
1258 cfs_spin_unlock(&res->lr_lock);
1261 static inline void check_res_locked(struct ldlm_resource *res)
1263 LASSERT_SPIN_LOCKED(&res->lr_lock);
1266 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1267 void unlock_res_and_lock(struct ldlm_lock *lock);
1270 void ldlm_pools_recalc(ldlm_side_t client);
1271 int ldlm_pools_init(void);
1272 void ldlm_pools_fini(void);
1274 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1275 int idx, ldlm_side_t client);
1276 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1277 unsigned int gfp_mask);
1278 void ldlm_pool_fini(struct ldlm_pool *pl);
1279 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1280 int ldlm_pool_recalc(struct ldlm_pool *pl);
1281 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1282 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1283 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1284 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1285 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1286 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1287 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1288 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1289 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);