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 /* Don't put lock into the LRU list, so that it is not canceled due to aging.
156 * Used by MGC locks, they are cancelled only at unmount or by callback. */
157 #define LDLM_FL_NO_LRU 0x400000
159 /* Immediatelly cancel such locks when they block some other locks. Send
160 * cancel notification to original lock holder, but expect no reply. This is
161 * for clients (like liblustre) that cannot be expected to reliably response
162 * to blocking ast. */
163 #define LDLM_FL_CANCEL_ON_BLOCK 0x800000
165 /* Flags flags inherited from parent lock when doing intents. */
166 #define LDLM_INHERIT_FLAGS (LDLM_FL_CANCEL_ON_BLOCK)
168 /* completion ast to be executed */
169 #define LDLM_FL_CP_REQD 0x1000000
171 /* cleanup_resource has already handled the lock */
172 #define LDLM_FL_CLEANED 0x2000000
174 /* optimization hint: LDLM can run blocking callback from current context
175 * w/o involving separate thread. in order to decrease cs rate */
176 #define LDLM_FL_ATOMIC_CB 0x4000000
178 /* was LDLM_FL_ASYNC until 2.0.0 0x8000000 */
180 /* It may happen that a client initiate 2 operations, e.g. unlink and mkdir,
181 * such that server send blocking ast for conflict locks to this client for
182 * the 1st operation, whereas the 2nd operation has canceled this lock and
183 * is waiting for rpc_lock which is taken by the 1st operation.
184 * LDLM_FL_BL_AST is to be set by ldlm_callback_handler() to the lock not allow
185 * ELC code to cancel it.
186 * LDLM_FL_BL_DONE is to be set by ldlm_cancel_callback() when lock cache is
187 * droped to let ldlm_callback_handler() return EINVAL to the server. It is
188 * used when ELC rpc is already prepared and is waiting for rpc_lock, too late
189 * to send a separate CANCEL rpc. */
190 #define LDLM_FL_BL_AST 0x10000000
191 #define LDLM_FL_BL_DONE 0x20000000
193 /* measure lock contention and return -EUSERS if locking contention is high */
194 #define LDLM_FL_DENY_ON_CONTENTION 0x40000000
196 /* These are flags that are mapped into the flags and ASTs of blocking locks */
197 #define LDLM_AST_DISCARD_DATA 0x80000000 /* Add FL_DISCARD to blocking ASTs */
199 /* Flags sent in AST lock_flags to be mapped into the receiving lock. */
200 #define LDLM_AST_FLAGS (LDLM_FL_DISCARD_DATA)
203 * --------------------------------------------------------------------------
204 * NOTE! Starting from this point, that is, LDLM_FL_* flags with values above
205 * 0x80000000 will not be sent over the wire.
206 * --------------------------------------------------------------------------
209 /* Used for marking lock as an target for -EINTR while cp_ast sleep
210 * emulation + race with upcoming bl_ast. */
211 #define LDLM_FL_FAIL_LOC 0x100000000ULL
213 /* Used while processing the unused list to know that we have already
214 * handled this lock and decided to skip it */
215 #define LDLM_FL_SKIPPED 0x200000000ULL
217 /* The blocking callback is overloaded to perform two functions. These flags
218 * indicate which operation should be performed. */
219 #define LDLM_CB_BLOCKING 1
220 #define LDLM_CB_CANCELING 2
222 /* compatibility matrix */
223 #define LCK_COMPAT_EX LCK_NL
224 #define LCK_COMPAT_PW (LCK_COMPAT_EX | LCK_CR)
225 #define LCK_COMPAT_PR (LCK_COMPAT_PW | LCK_PR)
226 #define LCK_COMPAT_CW (LCK_COMPAT_PW | LCK_CW)
227 #define LCK_COMPAT_CR (LCK_COMPAT_CW | LCK_PR | LCK_PW)
228 #define LCK_COMPAT_NL (LCK_COMPAT_CR | LCK_EX | LCK_GROUP)
229 #define LCK_COMPAT_GROUP (LCK_GROUP | LCK_NL)
230 #define LCK_COMPAT_COS (LCK_COS)
232 extern ldlm_mode_t lck_compat_array[];
234 static inline void lockmode_verify(ldlm_mode_t mode)
236 LASSERT(mode > LCK_MINMODE && mode < LCK_MAXMODE);
239 static inline int lockmode_compat(ldlm_mode_t exist_mode, ldlm_mode_t new_mode)
241 return (lck_compat_array[exist_mode] & new_mode);
246 * cluster name spaces
250 #define DLM_OST_NAMESPACE 1
251 #define DLM_MDS_NAMESPACE 2
254 - do we just separate this by security domains and use a prefix for
255 multiple namespaces in the same domain?
265 * waiting_locks_spinlock
280 struct ldlm_resource;
281 struct ldlm_namespace;
283 struct ldlm_pool_ops {
284 int (*po_recalc)(struct ldlm_pool *pl);
285 int (*po_shrink)(struct ldlm_pool *pl, int nr,
286 unsigned int gfp_mask);
287 int (*po_setup)(struct ldlm_pool *pl, int limit);
291 * One second for pools thread check interval. Each pool has own period.
293 #define LDLM_POOLS_THREAD_PERIOD (1)
296 * ~6% margin for modest pools. See ldlm_pool.c for details.
298 #define LDLM_POOLS_MODEST_MARGIN_SHIFT (4)
301 * Default recalc period for server side pools in sec.
303 #define LDLM_POOL_SRV_DEF_RECALC_PERIOD (1)
306 * Default recalc period for client side pools in sec.
308 #define LDLM_POOL_CLI_DEF_RECALC_PERIOD (10)
312 * Pool proc directory.
314 cfs_proc_dir_entry_t *pl_proc_dir;
316 * Pool name, should be long enough to contain compound proc entry name.
320 * Lock for protecting slv/clv updates.
322 cfs_spinlock_t pl_lock;
324 * Number of allowed locks in in pool, both, client and server side.
326 cfs_atomic_t pl_limit;
328 * Number of granted locks in
330 cfs_atomic_t pl_granted;
334 cfs_atomic_t pl_grant_rate;
338 cfs_atomic_t pl_cancel_rate;
340 * Server lock volume. Protected by pl_lock.
342 __u64 pl_server_lock_volume;
344 * Current biggest client lock volume. Protected by pl_lock.
346 __u64 pl_client_lock_volume;
348 * Lock volume factor. SLV on client is calculated as following:
349 * server_slv * lock_volume_factor.
351 cfs_atomic_t pl_lock_volume_factor;
353 * Time when last slv from server was obtained.
355 time_t pl_recalc_time;
357 * Recalc period for pool.
359 time_t pl_recalc_period;
361 * Recalc and shrink ops.
363 struct ldlm_pool_ops *pl_ops;
365 * Number of planned locks for next period.
371 struct lprocfs_stats *pl_stats;
374 typedef int (*ldlm_res_policy)(struct ldlm_namespace *, struct ldlm_lock **,
375 void *req_cookie, ldlm_mode_t mode, int flags,
378 typedef int (*ldlm_cancel_for_recovery)(struct ldlm_lock *lock);
380 struct ldlm_valblock_ops {
381 int (*lvbo_init)(struct ldlm_resource *res);
382 int (*lvbo_update)(struct ldlm_resource *res,
383 struct ptlrpc_request *r,
385 int (*lvbo_free)(struct ldlm_resource *res);
389 LDLM_NAMESPACE_GREEDY = 1 << 0,
390 LDLM_NAMESPACE_MODEST = 1 << 1
394 * Default values for the "max_nolock_size", "contention_time" and
395 * "contended_locks" namespace tunables.
397 #define NS_DEFAULT_MAX_NOLOCK_BYTES 0
398 #define NS_DEFAULT_CONTENTION_SECONDS 2
399 #define NS_DEFAULT_CONTENDED_LOCKS 32
401 struct ldlm_ns_bucket {
403 struct ldlm_namespace *nsb_namespace;
404 /** estimated lock callback time */
405 struct adaptive_timeout nsb_at_estimate;
409 /** ldlm namespace lock stats */
416 LDLM_NS_TYPE_UNKNOWN = 0,
431 struct ldlm_namespace {
433 * Backward link to obd, required for ldlm pool to store new SLV.
435 struct obd_device *ns_obd;
438 * Is this a client-side lock tree?
440 ldlm_side_t ns_client;
445 cfs_hash_t *ns_rs_hash;
450 cfs_spinlock_t ns_lock;
453 * big refcount (by bucket)
455 cfs_atomic_t ns_bref;
458 * Namespce connect flags supported by server (may be changed via proc,
459 * lru resize may be disabled/enabled).
461 __u64 ns_connect_flags;
464 * Client side orig connect flags supported by server.
466 __u64 ns_orig_connect_flags;
469 * Position in global namespace list.
471 cfs_list_t ns_list_chain;
474 * All root resources in namespace.
476 cfs_list_t ns_unused_list;
479 unsigned int ns_max_unused;
480 unsigned int ns_max_age;
481 unsigned int ns_timeouts;
485 unsigned int ns_ctime_age_limit;
488 * Next debug dump, jiffies.
490 cfs_time_t ns_next_dump;
492 ldlm_res_policy ns_policy;
493 struct ldlm_valblock_ops *ns_lvbo;
495 cfs_waitq_t ns_waitq;
496 struct ldlm_pool ns_pool;
497 ldlm_appetite_t ns_appetite;
500 * If more than \a ns_contended_locks found, the resource is considered
503 unsigned ns_contended_locks;
506 * The resource remembers contended state during \a ns_contention_time,
509 unsigned ns_contention_time;
512 * Limit size of nolock requests, in bytes.
514 unsigned ns_max_nolock_size;
517 * Limit of parallel AST RPC count.
519 unsigned ns_max_parallel_ast;
521 /* callback to cancel locks before replaying it during recovery */
522 ldlm_cancel_for_recovery ns_cancel_for_recovery;
526 struct lprocfs_stats *ns_stats;
528 unsigned ns_stopping:1; /* namespace cleanup */
531 static inline int ns_is_client(struct ldlm_namespace *ns)
534 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
535 LDLM_NAMESPACE_SERVER)));
536 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
537 ns->ns_client == LDLM_NAMESPACE_SERVER);
538 return ns->ns_client == LDLM_NAMESPACE_CLIENT;
541 static inline int ns_is_server(struct ldlm_namespace *ns)
544 LASSERT(!(ns->ns_client & ~(LDLM_NAMESPACE_CLIENT |
545 LDLM_NAMESPACE_SERVER)));
546 LASSERT(ns->ns_client == LDLM_NAMESPACE_CLIENT ||
547 ns->ns_client == LDLM_NAMESPACE_SERVER);
548 return ns->ns_client == LDLM_NAMESPACE_SERVER;
551 static inline int ns_connect_lru_resize(struct ldlm_namespace *ns)
554 return !!(ns->ns_connect_flags & OBD_CONNECT_LRU_RESIZE);
557 static inline void ns_register_cancel(struct ldlm_namespace *ns,
558 ldlm_cancel_for_recovery arg)
561 ns->ns_cancel_for_recovery = arg;
566 typedef int (*ldlm_blocking_callback)(struct ldlm_lock *lock,
567 struct ldlm_lock_desc *new, void *data,
569 typedef int (*ldlm_completion_callback)(struct ldlm_lock *lock, int flags,
571 typedef int (*ldlm_glimpse_callback)(struct ldlm_lock *lock, void *data);
572 typedef unsigned long (*ldlm_weigh_callback)(struct ldlm_lock *lock);
574 /* Interval node data for each LDLM_EXTENT lock */
575 struct ldlm_interval {
576 struct interval_node li_node; /* node for tree mgmt */
577 cfs_list_t li_group; /* the locks which have the same
578 * policy - group of the policy */
580 #define to_ldlm_interval(n) container_of(n, struct ldlm_interval, li_node)
582 /* the interval tree must be accessed inside the resource lock. */
583 struct ldlm_interval_tree {
584 /* tree size, this variable is used to count
585 * granted PW locks in ldlm_extent_policy()*/
587 ldlm_mode_t lit_mode; /* lock mode */
588 struct interval_node *lit_root; /* actually ldlm_interval */
591 #define LUSTRE_TRACKS_LOCK_EXP_REFS (1)
595 LCF_ASYNC = 0x1, /* Cancel locks asynchronously. */
596 LCF_LOCAL = 0x2, /* Cancel locks locally, not notifing server */
597 LCF_BL_AST = 0x4, /* Cancel locks marked as LDLM_FL_BL_AST
599 } ldlm_cancel_flags_t;
605 __u64 blocking_owner;
606 void *blocking_export;
611 struct ldlm_extent l_extent;
612 struct ldlm_flock l_flock;
613 struct ldlm_inodebits l_inodebits;
614 } ldlm_policy_data_t;
616 void ldlm_convert_policy_to_wire(ldlm_type_t type,
617 const ldlm_policy_data_t *lpolicy,
618 ldlm_wire_policy_data_t *wpolicy);
619 void ldlm_convert_policy_to_local(ldlm_type_t type,
620 const ldlm_wire_policy_data_t *wpolicy,
621 ldlm_policy_data_t *lpolicy);
625 * Must be first in the structure.
627 struct portals_handle l_handle;
629 * Lock reference count.
633 * Internal spinlock protects l_resource. we should hold this lock
634 * first before grabbing res_lock.
636 cfs_spinlock_t l_lock;
638 * ldlm_lock_change_resource() can change this.
640 struct ldlm_resource *l_resource;
642 * Protected by ns_hash_lock. List item for client side lru list.
646 * Protected by lr_lock, linkage to resource's lock queues.
648 cfs_list_t l_res_link;
650 * Tree node for ldlm_extent.
652 struct ldlm_interval *l_tree_node;
654 * Protected by per-bucket exp->exp_lock_hash locks. Per export hash
657 cfs_hlist_node_t l_exp_hash;
659 * Protected by lr_lock. Requested mode.
661 ldlm_mode_t l_req_mode;
663 * Granted mode, also protected by lr_lock.
665 ldlm_mode_t l_granted_mode;
667 * Lock enqueue completion handler.
669 ldlm_completion_callback l_completion_ast;
671 * Lock blocking ast handler.
673 ldlm_blocking_callback l_blocking_ast;
675 * Lock glimpse handler.
677 ldlm_glimpse_callback l_glimpse_ast;
678 ldlm_weigh_callback l_weigh_ast;
683 struct obd_export *l_export;
685 * Lock connection export.
687 struct obd_export *l_conn_export;
690 * Remote lock handle.
692 struct lustre_handle l_remote_handle;
694 ldlm_policy_data_t l_policy_data;
697 * Protected by lr_lock. Various counters: readers, writers, etc.
703 * Set for locks that were removed from class hash table and will be
704 * destroyed when last reference to them is released. Set by
705 * ldlm_lock_destroy_internal().
707 * Protected by lock and resource locks.
711 * flag whether this is a server namespace lock
715 * If the lock is granted, a process sleeps on this waitq to learn when
716 * it's no longer in use. If the lock is not granted, a process sleeps
717 * on this waitq to learn when it becomes granted.
722 * Seconds. it will be updated if there is any activity related to
723 * the lock, e.g. enqueue the lock or send block AST.
725 cfs_time_t l_last_activity;
728 * Jiffies. Should be converted to time if needed.
730 cfs_time_t l_last_used;
732 struct ldlm_extent l_req_extent;
735 * Client-side-only members.
739 * Temporary storage for an LVB received during an enqueue operation.
747 * Server-side-only members.
750 /** connection cookie for the client originated the operation. */
751 __u64 l_client_cookie;
754 * Protected by elt_lock. Callbacks pending.
756 cfs_list_t l_pending_chain;
758 cfs_time_t l_callback_timeout;
761 * Pid which created this lock.
766 * For ldlm_add_ast_work_item().
770 * For ldlm_add_ast_work_item().
774 * For ldlm_add_ast_work_item().
778 struct ldlm_lock *l_blocking_lock;
782 * Protected by lr_lock, linkages to "skip lists".
784 cfs_list_t l_sl_mode;
785 cfs_list_t l_sl_policy;
786 struct lu_ref l_reference;
787 #if LUSTRE_TRACKS_LOCK_EXP_REFS
788 /* Debugging stuff for bug 20498, for tracking export
790 /** number of export references taken */
792 /** link all locks referencing one export */
793 cfs_list_t l_exp_refs_link;
794 /** referenced export object */
795 struct obd_export *l_exp_refs_target;
799 struct ldlm_resource {
800 struct ldlm_ns_bucket *lr_ns_bucket;
802 /* protected by ns_hash_lock */
803 cfs_hlist_node_t lr_hash;
804 cfs_spinlock_t lr_lock;
806 /* protected by lr_lock */
807 cfs_list_t lr_granted;
808 cfs_list_t lr_converting;
809 cfs_list_t lr_waiting;
810 ldlm_mode_t lr_most_restr;
811 ldlm_type_t lr_type; /* LDLM_{PLAIN,EXTENT,FLOCK} */
812 struct ldlm_res_id lr_name;
813 cfs_atomic_t lr_refcount;
815 struct ldlm_interval_tree lr_itree[LCK_MODE_NUM]; /* interval trees*/
817 /* Server-side-only lock value block elements */
818 /** to serialize lvbo_init */
819 cfs_semaphore_t lr_lvb_sem;
821 /** protect by lr_lock */
824 /* when the resource was considered as contended */
825 cfs_time_t lr_contention_time;
827 * List of references to this resource. For debugging.
829 struct lu_ref lr_reference;
831 struct inode *lr_lvb_inode;
835 ldlm_ns_name(struct ldlm_namespace *ns)
837 return ns->ns_rs_hash->hs_name;
840 static inline struct ldlm_namespace *
841 ldlm_res_to_ns(struct ldlm_resource *res)
843 return res->lr_ns_bucket->nsb_namespace;
846 static inline struct ldlm_namespace *
847 ldlm_lock_to_ns(struct ldlm_lock *lock)
849 return ldlm_res_to_ns(lock->l_resource);
853 ldlm_lock_to_ns_name(struct ldlm_lock *lock)
855 return ldlm_ns_name(ldlm_lock_to_ns(lock));
858 static inline struct adaptive_timeout *
859 ldlm_lock_to_ns_at(struct ldlm_lock *lock)
861 return &lock->l_resource->lr_ns_bucket->nsb_at_estimate;
864 struct ldlm_ast_work {
865 struct ldlm_lock *w_lock;
867 struct ldlm_lock_desc w_desc;
874 /* ldlm_enqueue parameters common */
875 struct ldlm_enqueue_info {
876 __u32 ei_type; /* Type of the lock being enqueued. */
877 __u32 ei_mode; /* Mode of the lock being enqueued. */
878 void *ei_cb_bl; /* blocking lock callback */
879 void *ei_cb_cp; /* lock completion callback */
880 void *ei_cb_gl; /* lock glimpse callback */
881 void *ei_cb_wg; /* lock weigh callback */
882 void *ei_cbdata; /* Data to be passed into callbacks. */
885 extern struct obd_ops ldlm_obd_ops;
887 extern char *ldlm_lockname[];
888 extern char *ldlm_typename[];
889 extern char *ldlm_it2str(int it);
891 #define ldlm_lock_debug(cdls, level, lock, file, func, line, fmt, a...) do { \
894 if (((level) & D_CANTMASK) != 0 || \
895 ((libcfs_debug & (level)) != 0 && \
896 (libcfs_subsystem_debug & DEBUG_SUBSYSTEM) != 0)) { \
897 static struct libcfs_debug_msg_data _ldlm_dbg_data = \
898 DEBUG_MSG_DATA_INIT(cdls, DEBUG_SUBSYSTEM, \
900 _ldlm_lock_debug(lock, level, &_ldlm_dbg_data, fmt, \
905 void _ldlm_lock_debug(struct ldlm_lock *lock, __u32 mask,
906 struct libcfs_debug_msg_data *data, const char *fmt,
908 __attribute__ ((format (printf, 4, 5)));
910 #define LDLM_DEBUG_LIMIT(mask, lock, fmt, a...) do { \
911 static cfs_debug_limit_state_t _ldlm_cdls; \
912 ldlm_lock_debug(&_ldlm_cdls, mask, lock, \
913 __FILE__, __FUNCTION__, __LINE__, \
917 #define LDLM_ERROR(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_ERROR, lock, fmt, ## a)
918 #define LDLM_WARN(lock, fmt, a...) LDLM_DEBUG_LIMIT(D_WARNING, lock, fmt, ## a)
920 #define LDLM_DEBUG(lock, fmt, a...) do { \
921 ldlm_lock_debug(NULL, D_DLMTRACE, lock, \
922 __FILE__, __FUNCTION__, __LINE__, \
925 #else /* !LIBCFS_DEBUG */
926 # define LDLM_DEBUG(lock, fmt, a...) ((void)0)
927 # define LDLM_ERROR(lock, fmt, a...) ((void)0)
928 # define ldlm_lock_debuf(cdls, level, lock, file, func, line, fmt, a...) \
932 #define LDLM_DEBUG_NOLOCK(format, a...) \
933 CDEBUG(D_DLMTRACE, "### " format "\n" , ##a)
935 typedef int (*ldlm_processing_policy)(struct ldlm_lock *lock, int *flags,
936 int first_enq, ldlm_error_t *err,
937 cfs_list_t *work_list);
943 #define LDLM_ITER_CONTINUE 1 /* keep iterating */
944 #define LDLM_ITER_STOP 2 /* stop iterating */
946 typedef int (*ldlm_iterator_t)(struct ldlm_lock *, void *);
947 typedef int (*ldlm_res_iterator_t)(struct ldlm_resource *, void *);
949 int ldlm_resource_foreach(struct ldlm_resource *res, ldlm_iterator_t iter,
951 void ldlm_namespace_foreach(struct ldlm_namespace *ns, ldlm_iterator_t iter,
954 int ldlm_replay_locks(struct obd_import *imp);
955 int ldlm_resource_iterate(struct ldlm_namespace *, const struct ldlm_res_id *,
956 ldlm_iterator_t iter, void *data);
959 int ldlm_flock_completion_ast(struct ldlm_lock *lock, int flags, void *data);
962 __u64 ldlm_extent_shift_kms(struct ldlm_lock *lock, __u64 old_kms);
966 int ldlm_server_blocking_ast(struct ldlm_lock *, struct ldlm_lock_desc *,
967 void *data, int flag);
968 int ldlm_server_completion_ast(struct ldlm_lock *lock, int flags, void *data);
969 int ldlm_server_glimpse_ast(struct ldlm_lock *lock, void *data);
970 int ldlm_handle_enqueue(struct ptlrpc_request *req, ldlm_completion_callback,
971 ldlm_blocking_callback, ldlm_glimpse_callback);
972 int ldlm_handle_convert(struct ptlrpc_request *req);
973 int ldlm_handle_cancel(struct ptlrpc_request *req);
974 int ldlm_request_cancel(struct ptlrpc_request *req,
975 const struct ldlm_request *dlm_req, int first);
976 int ldlm_del_waiting_lock(struct ldlm_lock *lock);
977 int ldlm_refresh_waiting_lock(struct ldlm_lock *lock, int timeout);
978 void ldlm_revoke_export_locks(struct obd_export *exp);
979 int ldlm_get_ref(void);
980 void ldlm_put_ref(void);
981 int ldlm_init_export(struct obd_export *exp);
982 void ldlm_destroy_export(struct obd_export *exp);
985 ldlm_processing_policy ldlm_get_processing_policy(struct ldlm_resource *res);
986 void ldlm_register_intent(struct ldlm_namespace *ns, ldlm_res_policy arg);
987 void ldlm_lock2handle(const struct ldlm_lock *lock,
988 struct lustre_handle *lockh);
989 struct ldlm_lock *__ldlm_handle2lock(const struct lustre_handle *, int flags);
990 void ldlm_cancel_callback(struct ldlm_lock *);
991 int ldlm_lock_remove_from_lru(struct ldlm_lock *);
992 int ldlm_lock_set_data(struct lustre_handle *, void *);
994 static inline struct ldlm_lock *ldlm_handle2lock(const struct lustre_handle *h)
996 return __ldlm_handle2lock(h, 0);
999 #define LDLM_LOCK_REF_DEL(lock) \
1000 lu_ref_del(&lock->l_reference, "handle", cfs_current())
1002 static inline struct ldlm_lock *
1003 ldlm_handle2lock_long(const struct lustre_handle *h, int flags)
1005 struct ldlm_lock *lock;
1007 lock = __ldlm_handle2lock(h, flags);
1009 LDLM_LOCK_REF_DEL(lock);
1013 static inline int ldlm_res_lvbo_update(struct ldlm_resource *res,
1014 struct ptlrpc_request *r, int increase)
1016 if (ldlm_res_to_ns(res)->ns_lvbo &&
1017 ldlm_res_to_ns(res)->ns_lvbo->lvbo_update) {
1018 return ldlm_res_to_ns(res)->ns_lvbo->lvbo_update(res, r,
1024 int ldlm_error2errno(ldlm_error_t error);
1025 ldlm_error_t ldlm_errno2error(int err_no); /* don't call it `errno': this
1026 * confuses user-space. */
1027 #if LUSTRE_TRACKS_LOCK_EXP_REFS
1028 void ldlm_dump_export_locks(struct obd_export *exp);
1032 * Release a temporary lock reference obtained by ldlm_handle2lock() or
1033 * __ldlm_handle2lock().
1035 #define LDLM_LOCK_PUT(lock) \
1037 LDLM_LOCK_REF_DEL(lock); \
1038 /*LDLM_DEBUG((lock), "put");*/ \
1039 ldlm_lock_put(lock); \
1043 * Release a lock reference obtained by some other means (see
1046 #define LDLM_LOCK_RELEASE(lock) \
1048 /*LDLM_DEBUG((lock), "put");*/ \
1049 ldlm_lock_put(lock); \
1052 #define LDLM_LOCK_GET(lock) \
1054 ldlm_lock_get(lock); \
1055 /*LDLM_DEBUG((lock), "get");*/ \
1059 #define ldlm_lock_list_put(head, member, count) \
1061 struct ldlm_lock *_lock, *_next; \
1063 cfs_list_for_each_entry_safe(_lock, _next, head, member) { \
1066 cfs_list_del_init(&_lock->member); \
1067 LDLM_LOCK_RELEASE(_lock); \
1072 struct ldlm_lock *ldlm_lock_get(struct ldlm_lock *lock);
1073 void ldlm_lock_put(struct ldlm_lock *lock);
1074 void ldlm_lock_destroy(struct ldlm_lock *lock);
1075 void ldlm_lock2desc(struct ldlm_lock *lock, struct ldlm_lock_desc *desc);
1076 void ldlm_lock_addref(struct lustre_handle *lockh, __u32 mode);
1077 int ldlm_lock_addref_try(struct lustre_handle *lockh, __u32 mode);
1078 void ldlm_lock_decref(struct lustre_handle *lockh, __u32 mode);
1079 void ldlm_lock_decref_and_cancel(struct lustre_handle *lockh, __u32 mode);
1080 void ldlm_lock_allow_match(struct ldlm_lock *lock);
1081 void ldlm_lock_allow_match_locked(struct ldlm_lock *lock);
1082 ldlm_mode_t ldlm_lock_match(struct ldlm_namespace *ns, int flags,
1083 const struct ldlm_res_id *, ldlm_type_t type,
1084 ldlm_policy_data_t *, ldlm_mode_t mode,
1085 struct lustre_handle *, int unref);
1086 ldlm_mode_t ldlm_revalidate_lock_handle(struct lustre_handle *lockh,
1088 struct ldlm_resource *ldlm_lock_convert(struct ldlm_lock *lock, int new_mode,
1090 void ldlm_lock_downgrade(struct ldlm_lock *lock, int new_mode);
1091 void ldlm_lock_cancel(struct ldlm_lock *lock);
1092 void ldlm_reprocess_all(struct ldlm_resource *res);
1093 void ldlm_reprocess_all_ns(struct ldlm_namespace *ns);
1094 void ldlm_lock_dump(int level, struct ldlm_lock *lock, int pos);
1095 void ldlm_lock_dump_handle(int level, struct lustre_handle *);
1096 void ldlm_unlink_lock_skiplist(struct ldlm_lock *req);
1099 struct ldlm_namespace *
1100 ldlm_namespace_new(struct obd_device *obd, char *name,
1101 ldlm_side_t client, ldlm_appetite_t apt,
1102 ldlm_ns_type_t ns_type);
1103 int ldlm_namespace_cleanup(struct ldlm_namespace *ns, int flags);
1104 void ldlm_namespace_free(struct ldlm_namespace *ns,
1105 struct obd_import *imp, int force);
1106 void ldlm_namespace_register(struct ldlm_namespace *ns, ldlm_side_t client);
1107 void ldlm_namespace_unregister(struct ldlm_namespace *ns, ldlm_side_t client);
1108 void ldlm_namespace_move_locked(struct ldlm_namespace *ns, ldlm_side_t client);
1109 struct ldlm_namespace *ldlm_namespace_first_locked(ldlm_side_t client);
1110 void ldlm_namespace_get(struct ldlm_namespace *ns);
1111 void ldlm_namespace_put(struct ldlm_namespace *ns);
1112 int ldlm_proc_setup(void);
1114 void ldlm_proc_cleanup(void);
1116 static inline void ldlm_proc_cleanup(void) {}
1119 /* resource.c - internal */
1120 struct ldlm_resource *ldlm_resource_get(struct ldlm_namespace *ns,
1121 struct ldlm_resource *parent,
1122 const struct ldlm_res_id *,
1123 ldlm_type_t type, int create);
1124 struct ldlm_resource *ldlm_resource_getref(struct ldlm_resource *res);
1125 int ldlm_resource_putref(struct ldlm_resource *res);
1126 void ldlm_resource_add_lock(struct ldlm_resource *res,
1128 struct ldlm_lock *lock);
1129 void ldlm_resource_unlink_lock(struct ldlm_lock *lock);
1130 void ldlm_res2desc(struct ldlm_resource *res, struct ldlm_resource_desc *desc);
1131 void ldlm_dump_all_namespaces(ldlm_side_t client, int level);
1132 void ldlm_namespace_dump(int level, struct ldlm_namespace *);
1133 void ldlm_resource_dump(int level, struct ldlm_resource *);
1134 int ldlm_lock_change_resource(struct ldlm_namespace *, struct ldlm_lock *,
1135 const struct ldlm_res_id *);
1137 #define LDLM_RESOURCE_ADDREF(res) do { \
1138 lu_ref_add_atomic(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1141 #define LDLM_RESOURCE_DELREF(res) do { \
1142 lu_ref_del(&(res)->lr_reference, __FUNCTION__, cfs_current()); \
1145 struct ldlm_callback_suite {
1146 ldlm_completion_callback lcs_completion;
1147 ldlm_blocking_callback lcs_blocking;
1148 ldlm_glimpse_callback lcs_glimpse;
1149 ldlm_weigh_callback lcs_weigh;
1152 /* ldlm_request.c */
1153 int ldlm_expired_completion_wait(void *data);
1154 int ldlm_blocking_ast_nocheck(struct ldlm_lock *lock);
1155 int ldlm_blocking_ast(struct ldlm_lock *lock, struct ldlm_lock_desc *desc,
1156 void *data, int flag);
1157 int ldlm_glimpse_ast(struct ldlm_lock *lock, void *reqp);
1158 int ldlm_completion_ast_async(struct ldlm_lock *lock, int flags, void *data);
1159 int ldlm_completion_ast(struct ldlm_lock *lock, int flags, void *data);
1160 int ldlm_cli_enqueue(struct obd_export *exp, struct ptlrpc_request **reqp,
1161 struct ldlm_enqueue_info *einfo,
1162 const struct ldlm_res_id *res_id,
1163 ldlm_policy_data_t const *policy, int *flags,
1164 void *lvb, __u32 lvb_len, struct lustre_handle *lockh,
1166 int ldlm_prep_enqueue_req(struct obd_export *exp,
1167 struct ptlrpc_request *req,
1168 cfs_list_t *cancels,
1170 int ldlm_prep_elc_req(struct obd_export *exp,
1171 struct ptlrpc_request *req,
1172 int version, int opc, int canceloff,
1173 cfs_list_t *cancels, int count);
1174 int ldlm_handle_enqueue0(struct ldlm_namespace *ns, struct ptlrpc_request *req,
1175 const struct ldlm_request *dlm_req,
1176 const struct ldlm_callback_suite *cbs);
1177 int ldlm_cli_enqueue_fini(struct obd_export *exp, struct ptlrpc_request *req,
1178 ldlm_type_t type, __u8 with_policy, ldlm_mode_t mode,
1179 int *flags, void *lvb, __u32 lvb_len,
1180 struct lustre_handle *lockh, int rc);
1181 int ldlm_cli_enqueue_local(struct ldlm_namespace *ns,
1182 const struct ldlm_res_id *res_id,
1183 ldlm_type_t type, ldlm_policy_data_t *policy,
1184 ldlm_mode_t mode, int *flags,
1185 ldlm_blocking_callback blocking,
1186 ldlm_completion_callback completion,
1187 ldlm_glimpse_callback glimpse,
1188 void *data, __u32 lvb_len,
1189 const __u64 *client_cookie,
1190 struct lustre_handle *lockh);
1191 int ldlm_server_ast(struct lustre_handle *lockh, struct ldlm_lock_desc *new,
1192 void *data, __u32 data_len);
1193 int ldlm_cli_convert(struct lustre_handle *, int new_mode, __u32 *flags);
1194 int ldlm_cli_update_pool(struct ptlrpc_request *req);
1195 int ldlm_handle_convert0(struct ptlrpc_request *req,
1196 const struct ldlm_request *dlm_req);
1197 int ldlm_cli_cancel(struct lustre_handle *lockh);
1198 int ldlm_cli_cancel_unused(struct ldlm_namespace *, const struct ldlm_res_id *,
1199 ldlm_cancel_flags_t flags, void *opaque);
1200 int ldlm_cli_cancel_unused_resource(struct ldlm_namespace *ns,
1201 const struct ldlm_res_id *res_id,
1202 ldlm_policy_data_t *policy,
1204 ldlm_cancel_flags_t flags,
1206 int ldlm_cli_cancel_req(struct obd_export *exp, cfs_list_t *head,
1207 int count, ldlm_cancel_flags_t flags);
1208 int ldlm_cancel_resource_local(struct ldlm_resource *res,
1209 cfs_list_t *cancels,
1210 ldlm_policy_data_t *policy,
1211 ldlm_mode_t mode, int lock_flags,
1212 ldlm_cancel_flags_t cancel_flags, void *opaque);
1213 int ldlm_cli_cancel_list_local(cfs_list_t *cancels, int count,
1214 ldlm_cancel_flags_t flags);
1215 int ldlm_cli_cancel_list(cfs_list_t *head, int count,
1216 struct ptlrpc_request *req, ldlm_cancel_flags_t flags);
1219 /* This has to be here because recursive inclusion sucks. */
1220 int intent_disposition(struct ldlm_reply *rep, int flag);
1221 void intent_set_disposition(struct ldlm_reply *rep, int flag);
1224 /* ioctls for trying requests */
1225 #define IOC_LDLM_TYPE 'f'
1226 #define IOC_LDLM_MIN_NR 40
1228 #define IOC_LDLM_TEST _IOWR('f', 40, long)
1229 #define IOC_LDLM_DUMP _IOWR('f', 41, long)
1230 #define IOC_LDLM_REGRESS_START _IOWR('f', 42, long)
1231 #define IOC_LDLM_REGRESS_STOP _IOWR('f', 43, long)
1232 #define IOC_LDLM_MAX_NR 43
1235 * "Modes" of acquiring lock_res, necessary to tell lockdep that taking more
1236 * than one lock_res is dead-lock safe.
1238 enum lock_res_type {
1243 static inline void lock_res(struct ldlm_resource *res)
1245 cfs_spin_lock(&res->lr_lock);
1248 static inline void lock_res_nested(struct ldlm_resource *res,
1249 enum lock_res_type mode)
1251 cfs_spin_lock_nested(&res->lr_lock, mode);
1255 static inline void unlock_res(struct ldlm_resource *res)
1257 cfs_spin_unlock(&res->lr_lock);
1260 static inline void check_res_locked(struct ldlm_resource *res)
1262 LASSERT_SPIN_LOCKED(&res->lr_lock);
1265 struct ldlm_resource * lock_res_and_lock(struct ldlm_lock *lock);
1266 void unlock_res_and_lock(struct ldlm_lock *lock);
1269 void ldlm_pools_recalc(ldlm_side_t client);
1270 int ldlm_pools_init(void);
1271 void ldlm_pools_fini(void);
1273 int ldlm_pool_init(struct ldlm_pool *pl, struct ldlm_namespace *ns,
1274 int idx, ldlm_side_t client);
1275 int ldlm_pool_shrink(struct ldlm_pool *pl, int nr,
1276 unsigned int gfp_mask);
1277 void ldlm_pool_fini(struct ldlm_pool *pl);
1278 int ldlm_pool_setup(struct ldlm_pool *pl, int limit);
1279 int ldlm_pool_recalc(struct ldlm_pool *pl);
1280 __u32 ldlm_pool_get_lvf(struct ldlm_pool *pl);
1281 __u64 ldlm_pool_get_slv(struct ldlm_pool *pl);
1282 __u64 ldlm_pool_get_clv(struct ldlm_pool *pl);
1283 __u32 ldlm_pool_get_limit(struct ldlm_pool *pl);
1284 void ldlm_pool_set_slv(struct ldlm_pool *pl, __u64 slv);
1285 void ldlm_pool_set_clv(struct ldlm_pool *pl, __u64 clv);
1286 void ldlm_pool_set_limit(struct ldlm_pool *pl, __u32 limit);
1287 void ldlm_pool_add(struct ldlm_pool *pl, struct ldlm_lock *lock);
1288 void ldlm_pool_del(struct ldlm_pool *pl, struct ldlm_lock *lock);